Elevated smoking prevalence among Region VIII health center patients

Material Information

Elevated smoking prevalence among Region VIII health center patients
Levinson, Arnold Hal
Publication Date:
Physical Description:
xi, 203 leaves : ; 28 cm


Subjects / Keywords:
Smoking -- West (U.S.) ( lcsh )
Tobacco use -- West (U.S.) ( lcsh )
Smoking ( fast )
Tobacco use ( fast )
United States, West ( fast )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )


Includes bibliographical references (leaves 177-203).
General Note:
Department of Health and Behavioral Sciences
Statement of Responsibility:
by Arnold Hal Levinson.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
47058005 ( OCLC )
LD1190.L566 2000d .L48 ( lcc )

Full Text
Arnold Hal Levinson
B.A., University of California, 1977
M.J., University of California, 1980
A thesis submitted to the
University of Colorado at Denver
in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Health and Behavioral Sciences

2000 by Arnold Hal Levinson
All rights reserved.

This thesis for the Doctor of Philosophy
degree by
Arnold Hal Levinson
has been approved

Levinson, Arnold Hal (Ph.D., Health and Behavioral Sciences)
Elevated Smoking Prevalence Among Region VIII Health Center Patients
Thesis directed by Associate Professor Deborah Main
A large, cross-sectional sample survey to investigate tobacco-use behaviors and atti-
tudes found 40 percent smoking prevalence among health center patients ages 18-39 in
U.S. Public Health Service Region VIII (1645). Nearly half of non-Hispanic white
patients were current smokers, with essentially no gender difference in rates. Expected
smoking prevalence was more than twice the rate in the corresponding general
population (OR 2.2), adjusted for ethnicity, education, and marital status. Age, sex,
employment and number of co-resident children were not significant covariates. A
possible explanation, that smokers might use health facilities more often than non-
smokers due to smoking-attributable diseases, was not supported in secondary analysis
of National Health Interview Survey data. In the primary study, health center smokers
were widely concerned about smoking-related heath risks; were as ready to quit as
other populations (stage-of-change model), and widely expressed desire and ability to
quit. Compared to other populations, health center smokers reported multiple indi-
cators of greater addiction, including significantly more daily smoking, daily cigarette
consumption, heavy smoking (>20 cigarettes/ day), and smoking within 30 minutes of
waking. Prevalence of former smoking did not significantly increase with age and was
68 percent as common as it was among the general population. Further research should
determine whether health center patients have access to cessation-support medications
and, if not, whether enhanced access reduces smoking prevalence. Also, a sample sur-
vey is needed among the nations health center patients to determine whether the
smoking problem in Region VIII is isolated, multi-regional or national in scope.
This abstract accurately represents the content of the candidates thesis. I recom-
mend its publication.
Deborah S. Main

List of Figures ........................................... viii
List of Tables................................................ix
PREFACE ..............................................................x
1. THE STUDY QUESTION......................................... 1
2. BACKGROUND................................................. 3
The Tobacco Health Problem ............................. 3
Demographic Predictors of Tobacco Use .................. 6
States ........................................... 6
Age............................................... 7
Gender............................................ 7
Ethnicity......................................... 7
Education......................................... 8
Income............................................ 9
Marital Status................................... 10
Smokeless Tobacco................................ 10
Quitting .............................................. 10
Choice .......................................... 13
Addiction........................................ 15
Is Tobacco Addictive?.................................. 17
Nicotine Research ............................... 18
Alongside Nicotine: Behavioral Models.................. 20
Health Beliefs....................................22
Social Learning ..................................23
Reasoned Action ................................. 24
Transtheory and Stages of Change................. 24
Comparison of Theories........................... 25
Economic Approaches...............................26
Smoking Reasons.........................................27
Tobacco Research Contamination......................... 32
Summary ............................................... 34

3. METHODS ...................................................... 37
Development and Operation of a Clinical-research Network .... 37
Study Design .............................................. 41
The Instrument............................................. 42
Data Processing............................................ 45
Entry............................................... 45
Cleaning............................................ 46
Case Comparison, Excluded vs. Included Clinics.....46
Recoding and Transformation......................... 47
Analysis of Complex Samples................................ 48
Design Basis for Analysis of the Study Sample.............. 51
Case Weighting ..................................... 51
Imputation of Item-Missing Values................... 52
Population Representation........................... 53
Comparison Populations ............................. 54
Test of One-Way Trends.............................. 55
Analytic Methods, by Hypothesis ........................... 55
4. RESULTS ...................................................... 65
Description of the Sample.................................. 65
Hypothesis Test Results.................................... 71
HI. Sociodemographic Factors ....................... 71
H2. Health Concerns ................................ 80
H3. Readiness to Quit .............................. 82
H4. Addiction....................................... 83
H5. Desire to Quit ................................. 86
H6. Nonresponse Bias................................ 87
H7. Sample Frame Bias............................... 88
5. DISCUSSION ................................................... 91
Descriptive Findings ...................................... 91
Inferential Findings ...................................... 92
Sociodemographic Factors............................ 92
Addiction........................................... 95
Readiness, Desire, and Perceived Ability to Quit ....96
Health Concerns..................................... 97
Nonresponse Bias.................................... 97
Sample Frame Bias .................................. 99

Other Matters................................ 100
Limitations............................. 100
Practice-Based Research ................ 100
Sample Design........................... 102
Certain Analytic Tools under Complex Sampling. 103
Main Recommendations ................... 103
B. INITIAL LITERATURE REVIEW...................... 109
E. TEXT OF CLINIC STUDY AGREEMENT................. 119
G. SAMPLE REFUSAL LOG............................. 123
H. DIARY OF SITE COMMUNICATIONS .................. 124
I. ENGLISH QUESTIONNAIRE ......................... 137
J. SPANISH QUESTIONNAIRE ......................... 143
CITATIONS............................................... 149
BIBLIOGRAPHY ........................................... 177

1. U.S. Public Health Service Region VIII .................................. 1
2. Annual smoking-attributable diseases, U.S. 1990......................... 3
3. Annual adult per capita cigarette consumption-U.S., 1900-1998 .......... 6
4. Current smoking, general population vs. health centers ................. 76
5. Current smoking by ethnicity ........................................... 77
6. Current smoking by education............................................ 78
7. Current smoking by marital status....................................... 79

1. Demographic distribution and cigarette use by case inclusion status .47
2. Contrast of design-based vs. simple random sample results................... 51
3. Ever-user prompted reasons why I would like to stop or cut down. 58
4. Staging algorithm for readiness to quit smoking cigarettes.................. 59
5. Usual place of care......................................................... 64
6. Health center sample distribution, by state ................................ 65
7. Health center sample demographics .......................................... 66
8. Region VIII demographic comparison.......................................... 67
9. Region VIII tobacco use comparison.......................................... 69
10. Region VIII tobacco use by demographic characteristics .................... 70
11. White non-Hispanic current smoking prevalence ............................. 72
12. Hispanic current smoking prevalence ....................................... 73
13. Estimated adjusted odds ratios for current smoking ..................... 74,75
14. Current smokers, mean responses to reasons to stop or cut down............. 80
15. Current smokers, main reasons for wanting to quit smoking .... 81
16. Former smokers, mean responses to reasons to stop or cut down.............. 82
17. Smoker readiness to quit, health center vs. other populations.............. 83
18. Timing and frequency of smoking, by population............................. 84
19. Cessation attempts and beliefs, by population.............................. 85
20. Prevalence of former smoking by age group.................................. 86
21. Self-efficacy vs. desire to quit smoking................................... 87
22. No health facility last 12 months, by smoker status and usual place of care .. 88
23. Mean health facility visits in 12 months, by smoking status................ 89
24. Smoking prevalence, health facility use vs. no use......................... 90

In the Beginning, there arose a rescue mission. A few primary-care clinicians
serving an interstate regions community and migrant health centers had received a
small grant to initiate practice-based research. The group was interested in tobacco-
use cessation, but they were caught in a labyrinth of meaningless debate. While the
group groped and argued over research directions, their grant was half spent and
time was running out.
Enter the graduate student. An experienced facilitator and veteran tobacco-control
field director, he approached bearing Tablets of Science and challenging the
worship of theoretical idols (idle theory?). The clinicians embraced him as One True
Researcher, and together they pledged to design and implement a survey of tobacco-
use prevalence and reasons for use and non-use.
With considerable time and effort on all parts, they completed the survey reasonably
successfully and began preparing an invited proposal for an intervention study, only
to reach an impasse over divisions of labor and responsibility. The collaboration
ended, and the student returned to his doctoral studies.
Enter the wife, the daughter-child, and the few close friends, bearing respite, and
solace, and meaning ... and also interruption (Papa, my Barbies head came off),
and lamentation (I never see you anymore), and conflict (School! Home! Family!
School! Family!).
Enter the Wilderness. A National Cancer Institute review panel rejected the stu-
dents intended dissertation project on the effects of local ordinances prohibiting
tobacco possession and use by minors. Although encouraged to resubmit, he feared
to extend an already time-consuming idea beyond his endurance. So it came to pass
that he began to wonder, might the tobacco-use survey data provide the basis for a

dissertation? Wandering among the data, he found himself struggling to learn one
analytic method after another as he posed a question, found mistakes in the method,
tried again, pondered results, discarded and revised the question, and tested again,
all the while becoming humbler and perhaps a bit wiser.
The first year was unbearable, the ensuing years less so. Though seeming intermina-
ble, the time began to hasten as the reckoning drew near. At last, the research bore
fruit, and so the ending was not just a relief but also a rejoicing.
So Hallelujah! Enter the dissertation... please.

THE investigation of the truth is in one way hard, in another easy.
An indication of this is found in the fact that no one is able to attain
the truth adequately, while, on the other hand, we do not collectively
fail, but every one says something true about the nature of things,
and while individually we contribute little or nothing to the truth, by
the union of all a considerable amount is amassed.1
In 1996, a group of health center clinicians decided to
study tobacco use among community and migrant clinic
patients in U.S. Public Health Service (PHS) Region
VIII, which comprises Colorado, Utah, Montana, North
Dakota, South Dakota, and Wyoming (Figure 1). Specif-
ically, the clinicians wanted to measure prevalence rates
and patient attitudes toward tobacco use, in preparation
for a possible cessation intervention study. The baseline
study focused on young adults (ages 18-39), based on clinician beliefs that younger
adults might be more receptive to intervention than older adults. An at-risk popula-
tion, adolescents, did not visit the health centers often enough to populate the study.
On behalf of the clinicians, the current author directed a sample survey of patients
during fall 1997 and winter 1998, and preliminary results showed 40 percent overall
prevalence. Among non-Hispanic whites, prevalence was 47 percent nearly double
the rate among a comparable general population (same age group, ethnicity, state,
year). Similar patterns were found among other ethnic groups.
Figure 1

The current dissertation study undertook a multi-dimensional analysis of these
preliminary results in an effort to explain the elevated prevalence rate among these
health center patients. Answers to similar questions have previously been pursued
during the last four decades through a wide variety of perspectives, including
pharmacology, genetics, sociology, econometrics, and behavioral models that range
from deterministic theories of personality disorder to volitional, cognition-based
approaches. This dissertation combined several current perspectives and evaluated
the following hypothetical explanations for elevated smoking prevalence among
participant health-center patients:
HI. Sociodemographic Factors. The general population and participant health-
center patients have statistically indistinguishable current smoking preva-
lence after adjusting for known sociodemographic factors.
H2. Health Concerns. Smokers in the participant health-center patient population
report relatively little concern about the health risks of smoking.
H3. Readiness to Quit. Smokers in the participant health-center patient popula-
tion are less ready to quit than smokers in the general population.
H4. Addiction. Smokers in the participant health-center patient population are
more addicted than smokers in the general population.
H5. Desire to Quit. Smokers in the participant health-center patient population
dont wish to quit smoking.
H6. Nonresponse Bias. The estimate of participant health-center patient smoking
prevalence is biased upward because non-participating clinics had lower
H7. Sample Frame Bias. Health facility visitors exhibit higher smoking preva-
lence than the general population does, because smokers on average incur
more health-care needs than nonsmokers due to tobacco-related morbidity.

The Tobacco Health Problem
Put simply, tobacco makes people sick and kills them, and may (or may not) cost
everyone lots of money.
Chronic Lung
Regular tobacco use
among adults is the
leading cause of pre-
ventable serious disease
and premature death,
causing about 400,000
deaths a year in the
United States or about
one-sixth to one-fifth of
all mortality.2,3 People
whose deaths are
caused by cigarettes die
an average of 11-12
years prematurely, and
about one-fifth of these
years lost to smoking
occur before age 65.4,5,
Figure 2: Annual numbers of smoking-attributable Most smoki deaths
diseases, U.S. 1990.

involve cardiovascular, neoplastic, and/or respiratory diseases (Figure 2). In 1990,
43 percent of U.S. smoking deaths involved cardiovascular diseases, 36 percent
involved cancers, and 20 percent involved respiratory diseases such as emphysema.7
Nearly two-thirds of smoking deaths involved lung cancer, ischemic heart disease,
or emphysema.
Conversely, tobaccos responsibility for overall cancer deaths is estimated between
11 percent and 30 percent; its share of cardiovascular deaths, between 17 percent
and 30 percent, and its share of lung disease deaths, 30 percent.3 Lung cancer in
particular is almost exclusively a smoking disease,8 and up to 90 percent is caused
by direct smoking of cigarettes, cigars and pipes or by exposure to secondhand
smoke.9,10 Female lung cancer rates have risen in the wake of an epidemic uptake of
regular cigarette smoking by U.S. women, whose smoking prevalence rate in 1965
was two-thirds that for men (34 percent vs. 52 percent)11 but since 1983 has hovered
within six percentage points of the adult male rate2, n'12,13,14 and in 1997 was five-
sixths of the male rate (22.1 percent vs. 27.6 percent).15 As a result, U.S. female
lung cancer mortality rates increased 550 percent between 1950 and 1991, lagging
the rise in use by roughly 30 years in keeping with lung cancers latency period.16
The disease surpassed breast cancer in 1986 as the leading cause of cancer death
among women, and smoking now accounts for at least 80 percent of lung cancers in
women as well as men.17 One research team has cautiously suggested, based on
pooled Danish prospective data, that women may be more sensitive than men to the
health hazards of smoking.18
Simply but conservatively summarized, U.S. cigarette smoking causes nearly all
lung cancer deaths, about one-fourth of all cancer deaths, about one-fourth of all
cardiovascular deaths, and about one-third of all lung disease deaths. This grim
summary does not suggest, of course, that smokers would live forever if they didnt
smoke, but nonsmokers rarely die from lung cancer and typically die 11 to 12 years
later in life than smokers with similar histories. Although tobacco does not kill most
of its users, prematurely or otherwise (giving rise to well-worn anecdotes of ancient
Aunt Anne or Grandpa George who still smokes two packs a day), cigarettes do kill
between one in three19 and one in two20 regular users. At current uptake rates, one in
11 U.S. children (ages 0-17) now alive will die from smoking,21 a prospect that
persists in the face of prevention programs in every state22 supported by the NCI,
Centers for Disease Control and Prevention (CDC), Robert Wood Johnson Founda-
tion, National Center for Tobacco-Free Kids, earmarked state tobacco tax revenues,
and a limited pool of legislative allocations from litigation settlements.
These health consequences of tobacco are no longer seriously disputed. After

decades of denial, the largest U.S. cigarette manufacturing company, Philip Morris,
has posted the extraordinary acknowledgment that:
(T)here is an overwhelming medical and scientific consensus that cigarette
smoking causes lung cancer, heart disease, emphysema and other serious
diseases in smokers. Smokers are far more likely to develop serious diseases,
like lung cancer, than non-smokers. There is no safe cigarette,23
Controversy remains, though, over the economic costs of tobacco use. The CDC has
estimated annual U.S. medical expenses attributable to smoking at $22 billion for
1993 (slightly more than $2 per pack purchased), of which 45 percent was paid with
public funds.24 Other researchers have reached varying conclusions about the impact
of smoking on lifetime medical expenses or other costs to society. One analysis
concludes that, despite shortened lifespans, smokers incur higher lifetime medical
expenses an average of $6,239 per smoker in 1990 dollars.25 Another analysis
reaches an opposite conclusion, namely that while smokers are medically more
expensive than age-and-sex-matched nonsmokers, nonsmokers are medically more
expensive over their lifetime because of their greater average longevity.26 These
different conclusions may be due to the use of different analytic methods and
assumptions.27 Kenneth E. Warner, a respected analyst of the tobacco industrys
multi-faceted economic influences on society, has reviewed the same evidence and
reached the same conclusion:28
Whether smoking adds to or subtracts from aggregate medical expenditures
remains a matter of dispute.... Smoking may well impose a financial burden
on health care budgets, but its net impact is likely modest. Further research
on that burden is still clearly warranted.
Looking beyond medical expenses, one author has concluded that smokers pay
their own way through a combination of cigarette excise taxes, higher insurance
premium payments, and reduced use (due to earlier death) of social security bene-
fits.29* Others, including a formative author of methods for estimating lifetime
disease costs, conclude that smoking cost U.S. society $138 billion in 1995 for
direct medical care expenditures, indirect costs, and lost productivity for people who
were ill and disabled or died prematurely.30
A logical application of this argument is that nations may virtually eliminate medical and
pension costs by indoctrinating every citizen, upon reaching an age of majority, to commit

No literature was found that addresses health facility utilization by smokers and
nonsmokers in early or middle adulthood.
Demographic Predictors of Tobacco Use
Overall, per capita adult cigarette consumption in the United States generally
declined during the last third of the 20th Century (Figure 3). During the 1990s,
however, current cigarette smoking prevalence remained relatively constant among
adults,12, 14,31,32,33,34 measuring 25.5 percent in 1990 and 24.1 percent in 1998.35
Annual adult per capita cigarette consumption United States, 1900-1998
1st World Confortnct
on Smoking and Health
Sources: United States Department of Agriculture; 1986 Surgeon General's Report.
Figure 3
In 1998, adult prevalence rates outside Utah (an outlier) ranged from 18.0 percent in
Minnesota to 30.8 percent in Kentucky, with higher endpoints for men (19.7% to
36.5%) than for women (16.4% to 28.5%).36 Of the six states represented in the
current study, Colorado and South Dakota had rates above the median U.S. rate.36 In
1997, four study states had rates above the median,2 but mean prevalence across the
six study states was 24.6 percent, lower than the U.S. mean of 27.0 percent, due to
low rates in Utah and Montana.37

The highest U.S. smoking prevalence rates appear among young adults ages 18-44,35
an age group that includes the current study population (ages 18-39).
In 1965, a year after the first U.S. Surgeon Generals Report on Smoking and Health
was issued,15 52 percent of adult men were regular cigarette smokers, compared to
34 percent of adult women.38 Being male is no longer as strong a correlate of
smoking;2 in 1998, smoking prevalence among U.S. women was 83.3 percent of the
rate among men (22.0% vs. 26.4%).35
Although case-control studies have suggested smoking might affect the lungs of
men and women differently, a subsequent prospective study found no smoking-
related gender difference in lung cancer incidence,39 and a meta-analysis failed to
find widespread differences in smoking-related lung function by gender or
In 1998, U.S. current smoking prevalence was highest among American Indi-
ans/Alaska Natives, second highest among black and Southeast Asian men, and
lowest among Asian American and Hispanic women.36 These rankings have
remained steady for about two decades.41 Heavy smoking (>25 cigarettes per day)
was most prevalent in 1998 among non-Hispanic whites.
A 1998 U.S. government review of tobacco use and ethnicity41 found that during
Cigarette smoking prevalence declined among African American, Asian
American and Hispanic adults, but not among American Indian men during
1983-95 or women from 1978-95.
The prevalence declines were greater among African American, Hispanic,
and white men who were high school graduates than they were among those
with less formal education. Among women in these three groups, education-
related declines in cigarette smoking were less pronounced.
Compared with non-Hispanic white smokers, smokers in each of the four

racial/ethnic minority groups smoked fewer cigarettes each day and, with the
exception of American Indians, were less likely than whites to smoke daily.
Educational attainment accounted for only some of the differences in
smoking behaviors between whites and racial/ ethnic minority groups, while
other biological, social, and cultural factors were likely to further account for
these differences.
Among Mexican Americans, who comprise the largest Hispanic group in the
western United States,42 a 1991 study of Hispanic HANES data found that 1)
acculturation, especially for women, and 2) the presence of other smokers in the
immediate social environment, were more predictive of smoking than age, income,
marital status, education, or employment.43 The authors found more striking,
however, the similarity of results among Mexican Americans to predictors of
smoking in the general population.
The current study focused on non-Hispanic whites and Hispanics.
Smoking prevalence in the United States is strongly, inversely associated with
education,44,45 as it is in Europe, where a recent review across 12 nations found that
subjects ages 20-44 with low education were 1.7 times as likely to be smokers as
subjects with high education.46
In the last quarter century, differential declines across education groups have
widened the education-smoking gradient. In 1974, U.S. men with less than a high
school education were nearly twice as likely to smoke as those with a college
degree; by 1995, the least educated men were nearly 3 times as likely to smoke as
the most educated. Among U.S. women, the comparable ratios rose from 1.4 in
1974 to 2.4 in 1995.47
The inverse education-smoking relationship does not hold, however, among people
who never went to high school;48 prevalence rates in this group are comparable to
those for people who graduated high school. People with 9-11 years of education are
most likely to be ever, current and heavy smokers, and least likely to quit.49 (Half of
the current study population has attended or graduated from college.)
Also, secondary analysis across years using female-only data from the National
Health and Nutrition Survey (NHANES) found that ethnic and educational associa-

tions with smoking were mutually independent, but ethnic differences among
matched California white and Hispanic survey respondents become negligible with
increasing education levels.50
Some researchers have suggested that the education-smoking association is spurious
and due to non-behavioral factors associated with both education and mortality. In
one study, a combination of material factors financial problems, employment
status, and income proxy were more important than smoking plus three other
behavioral factors (alcohol, body mass index, and physical activity) in explaining
the inverse association of education and mortality.51 And in Scotland, a large,
prospective cohort study of working men found occupational social class more
predictive of smoking behavior than education.52 The authors suggest the finding
argues against interpretations that see cultural rather than material resources as
being the key determinants of socioeconomic differentials in health, and they
speculate that the association of education with cardiovascular mortality may be due
to educations indication of early socioeconomic circumstances.
The relationship of smoking to income, or to measures of socioeconomic status
(SES) that include income, has been studied extensively within and outside the
United States. An inverse relationship has consistently appeared in studies in the
United States,53 Canada,54 and New Zealand;55 among pregnant women in Maine56
and New Zealand57 and perigravid women in Australia,58 and among U.S.59 and
Dutch60 adolescents. Also consistent are findings that low-income individuals in the
United States61 and Britain62 are much more likely than high-income individuals to
quit smoking when cigarette prices increase.
In contrast, researchers have found no link between smoking and income in Bul-
garia,63 Austria,64 or China.65 The Chinese study a large-sample household survey
of a half-million-person district near Shanghai found 67 percent smoking preva-
lence among male adults, with smokers spending 60 percent of annual personal
income and 17 percent of annual household income on cigarettes. Such apparent
insensitivity to price warrants considerable public health concern, given that price
increases through taxation have been shown in other countries to be highly effective
policy strategies, and given that China is the worlds largest producer and consumer
of cigarettes, eclipsing the next largest producer, the United States, fourfold.66
In rural northern India, a study67 found smoking most prevalent among higher class
men and not associated with social class among women; the finding among women

may represent a prevalence curve found across Europe during the last half
century, where young, high-SES women led other women into smoking, then led the
way into cessation but left low-SES women behind.68
Marital Status
A strong relationship has been found between smoking and divorce in the United
States,69,70 Sweden,71 Germany among women,72 Great Britain among a large cohort
of men,73 the Netherlands,74 and Australia among hospitalized patients.75 The
relationship has not been found among U.S. Hispanics.43
One study reports that the association of smoking prevalence with divorce reflects
smoking-adoption decisions made during adolescence, and the authors propose that
both smoking adoption and divorce reflect prior personal characteristics or early
experiences.76 Similarly, social position in the first three decades of life has been
shown to predict a variety of sociodemographic risk factors, including smoking and
Smokeless Tobacco
In the United States, young white men comprise the bulk of smokeless tobacco
users,78 although this user group includes a small but growing number of young
adult females, perhaps due to (possibly mistaken) beliefs that it helps curb appetite
and control weight.79,80 Among high school males, prevalence is 20% among white
males, 6% among Hispanics males, and 4% among blacks males.34
Most tobacco users wish they could quit but find it hard to do. Almost always
initiated during adolescence,9,81,82 smoking during this physiologically83 and
socially84 tumultuous life stage does not necessarily compel regular use immedi-
ately.85, 86,87, 88 Perhaps this initial lability conditions a powerful first impression that
smoking is a choice, therefore so will quitting be; some authors have speculated that
the absence of immediate health consequences fosters a sense of invulnerability.89
An oft-encountered belief that adolescents feel immortal and/or underestimate
their own likelihood of harm faces evidence to the contrary.84 Indeed, one study of
adolescents found that smokers felt more vulnerable to smoking and other health
risks than nonsmokers, and that within the smoker subgroup of adolescents, per-
ceived invulnerability was associated with action to quit and lower consumption

Whatever the reason, nearly half of high-school seniors who are daily smokers
predict they wont be smoking five years hence.81 But among the estimated one-third
to one half of experimenters81 who become regular users81 and 70 percent of
high school students in 1997 had tried smoking91 most regret ever having started
and wish to quit;92 fully three-fourths of young daily smokers ages 10-22 say one
reason they smoke is because its hard to quit, and more than 90 percent who have
tried to quit experience at least one withdrawal symptom.93
Among U.S. adults, in surveys spanning two decades, a consistent two-thirds to
three-fourths of smokers say they would like to give up smoking and have made at
least one serious attempt.94 By the widely used stages of change measure, roughly
60 percent are considering or planning to quit.95 A 1993 survey in Victoria, Austra-
lia, that asked smokers whether they would quit or not quit if it could be done
painlessly, found that three-fourths said they would quit under such circumstances
and another 13 percent indicated potential interest.96
Some tobacco users do succeed in quitting. That nicotine fails to exercise an
absolute grip may be readily seen in the steady increase of former smokers in the
general population: from 13.6 percent of U.S. adults in 1965 to 24.1 percent in
1991 44 prom a more dramatic perspective, former smokers as a fraction of ever-
smokers increased from one-fourth in 1965 to one-half in 1991.
The process of quitting is complex. The MEDLINE research literature database
contains more than 700 articles in English since 1985 on smoking cessation meth-
ods; more than 600 articles on smoking cessation psychology, and roughly 200
review articles on smoking cessation.97 From this abundant literature, several
conclusions emerge:
Ninety percent of successful quitters have historically done without explicit
external assistance, most of them going cold turkey.98 Self-quitting appears
to remain the rule99 despite a steady march of technological advances, from
the experimental emergence of the nicotine patch in the late 1980s100,101 and
the determination of its generally safe (albeit dermally irritating) efficacy,102'
io3, km. io5, io6, io7, io8,109, no tQ the use 0f the anti-depressant bupropion
(Zyban)111l12-113 in doses below the therapeutic threshold for depression.112
The minority of smokers who use assistance in quit attempts has grown,
from roughly 8 percent in 1986 to 20 percent in 1996.114 For these smokers,

nicotine replacement therapy (NRT) by gum, patch, nasal spray or oral
inhaler, as well as bupropion alone or combined with NRT, enhance the
likelihood and longevity of successful cessation.115, U2,116 Less common and
unproven assistance methods include acupuncture,117 aversive smoking,118
and hypnotherapy.119 People who use one or more assisted methods are more
often women, middle-aged, formally educated beyond high school, and
heavier smokers.98 Although studies typically estimate the relative efficacy of
various methods, results might not be comparable across methods because
each one may have distinguishable types of adherents.
Physician advice to smokers is linked to slight increases in rates of success-
ful quitting,120 as is individual counseling by non-clinician specialists121 and,
less so, individual self-help materials,122 although tailoring the materials
appears to increase effectiveness.123 The increased quit rates produced by
these interventions, although slight, can yield large benefits measured at
population levels.124 At least two federal health care agencies the NCI and
the Agency for Healthcare Research and Quality (AHRQ, formerly the
Agency for Health Care Policy and Research) have developed and dissemi-
nated clinical guidelines, physician training programs and materials for
smoking cessation.125, 126 Physicians counsel identified smokers in only 23
percent of office visits, however; the level is 37 percent in general medical
examinations (vs. specific problem visits), but the rate did not improve
during 1990-1995.127
Smoking cessation is a common behavior, but so is relapse. As the U.S. Surgeon
General notes, Most former smokers have cycled several times through the process
of smoking cessation and relapse before attaining long-term abstinence.128 At least
half of former smokers quit three or more times before sustaining cessation.129 An
estimated two in five smokers quit for at least one day during a given year, but 86
percent of them resume smoking and only 2.5 percent of all smokers quit for good
each year.130
Despite the difficulty of quitting, two-thirds to three-fourths of adult U.S. smokers -
roughly the same number who say they want to quit but find it difficult also say
they could quit smoking if they made the decision.94 Smoking is a clearly paradoxi-
cal phenomenon, but perhaps no more so than a fundamental condition of human
existence choice.

Tobacco is a member of the 2,600-species nightshade plant family, solanaceae, that
includes the potato, the petunia, and other recognizable ornamental, edible, and
poisonous plants.131 Nicotine is a naturally occurring tobacco-leaf substance that
forms the base of many insecticides132 and is the addictive ingredient in tobacco.
Cigarettes are the most widely used tobacco product in the United States invented
in 1614 by beggars in Seville, Spain; remaining an obscure mode of consumption
until the Crimean War (1853-1856); growing in popularity in the United States after
the Civil War (1861-1865), and surpassing cigars in 1919 based on the 1880
invention of the cigarette rolling machine.133 Snuff, a fine tobacco powder inhaled
nasally or placed between the lip and gum, entered popular Western culture around
the turn of the 18th Century.134 The chewing of tobacco is believed to have emerged
alongside the smoking of it (as well as the anal insertion for hallucinogenic enemas)
among Native Americans around 1 B.C.E. The practice was common in the 18th
Century United States, gained great popularity in the South before the Civil War,
reached peak per capita U.S. consumption in 1890,135 and remains common today
primarily among young white men in the southeastern and Rocky Mountain western
United States.136
Tobacco, nicotine, their consumer forms these are straightforward facts. But with
the addition of the word use, a richly complex human behavior emerges, raising in
particular the issue of volitional vs. involuntary behavior, a philosophical problem
considered by probably every culture.* Starting with Plato and certain other ancient
Greek philosophers, Western views have generally radiated from belief that fate and
choice coexist, although the tension between these opposing forces has spawned
terrific debate over the ultimate primacy of determinism, nature, instinct, or neces-
sity over will, free will, nurture, spontaneity or freedom.
This problem of choice necessarily coexists with the problem of what is now often
called cognition, a perhaps reduced analog of what various ancestral cultures called
- in descending order of breadth spirit, soul, mind. Like choice, cognition has
generated prolonged debate over its origins: Was Descartes correct that a certain,
unshakable first principle is I think, therefore I am?137 Or should it be I am,
therefore I think? Tobacco use research grapples with a similar conundrum are
For practical reasons of manageability and application, this discussion concentrates
primarily on Western cultures, from Greek and Roman through European to U.S. and
Canadian. In no way is the focus meant to deny the relevance or contributions to the subject
of non-Westem cultures which, after all, represent a majority of the worlds people and

nicotines effects mainly physiologic or psychological, i.e., does body or mind
contribute most to dependent smoking?
Research in both areas is primarily positivist, although much might be learned about
the gestalt of individual tobacco use through thoughtfully designed and carefully
executed qualitative research. Indeed, the framing of constructivist and positivist
approaches as antithetical and mutually exclusive seems unnecessary. In a 1931
essay,138 German social psychologist Kurt Lewin modeled an epistemology of
experimental psychology after post-Galilean physics to argue against dichotomous
models of the psyche, the use of historic-geographic averages to establish psycho-
logical principles, and the dismissal of individual cases as aberrant or unexplainable.
No matter, he argued, how widely the net is cast within a community or across
nations, during one year or three decades the error of laws of averages lies in
asserting they are independent of the historic-geographic context in which they are
measured rather than partially or entirely a consequence of the context. Instead, he
posited that psychic life is continuous and serial rather than dichotomous; that it is
governed by unconditional general laws; that these psychic laws are dynamic,
person-environment forces measurable through differential calculus, and that
knowledge of these laws emerges from individual cases studied closely and com-
The tendency to comprehend the actual situation as fully and concretely as
possible, even in its individual peculiarities, makes the most precise possible
qualitative and quantitative determination necessary and profitable. ... it is
possible to go from the particular to the general without losing the particular
in the general and thereby making impossible the return from the general to
the particular.138 (p.35) (emphasis added)
Consistent with this last point, full, concrete study of individual phenomena need
not require renunciation of general applicability; instead, such study requires equally
thorough and careful explication of observational context so readers may properly
search for general principles operating within the case context. In this view, qualita-
tive and quantitative, positivist and constructivist methods are mutually indispens-
able and integral.*
One other philosophical problem deserves mention in the context of choice and tobacco use,
namely individual vs. social responsibility. Under typical conditions, the adverse health
sequelae of non-medical individual use of nicotine and other drugs increase pressure on
limited medical resources and pose unacceptable risks to children and adolescents. These
social implications justify intervention and its essential research foundation, but they do not

Similarly, a core of behavioral tobacco use research, the choice-compulsion conun-
drum, may be seen as a tense union of interactive opposites. This perspective is
consistent with social-cognitive psychology, as discussed by Albert Bandura:139
When viewed from a sociocognitive perspective, there is no incompatibility
between freedom and determinism. Freedom is not conceived negatively as
exemption from social influences or situational constraints. Rather, it is
defined positively as the exercise of self-influence to bring about desired
results. ...It is because self-influence operates deterministically on action [in
the sense of producing effects] that some measure of freedom is possible.
The choice of actions from among alternatives is not completely and invol-
untarily determined by environmental events. Rather, the making of choices
is aided by reflective thought, through which self-influence is largely exer-
cised. ...It is not that individuals generate thoughts that then become the
agents of action. The cognitive activities constitute the processes of self-
influence that are brought to bear on the courses of action to take (p.7).139
The view that individual behavior is simultaneously caused by and a cause of
thoughts, feelings, biology, and environment provides a useful theoretical frame-
work for much behavioral research. It provides an ample context for specific
investigations into tobacco use, where nicotine so profoundly affects thoughts,
feelings and biology that much research asks how, and how much, nicotine alters
biology and feelings and thereby interferes with cognitive processes essential to self-
The deceptively simple term for these effects is addiction.
Twentieth Century researchers have repeatedly wrestled over definitions of addic-
support a moral perspective that tobacco is intrinsically a societal matter. Some individuals,
fully informed and ready to take full fiscal and moral responsibility for potential health
consequences, might freely choose to use tobacco and use it in ways that dont expose others
to risk. Under these presumably rare conditions, use would reside within a private
behavioral estate. After a decade of professional tobacco-control work, I have found no
ground on which to stand and proclaim tobacco use is inherently wrong, evil, bad, or
deserving of crucifixion, any more than I have found a moral basis for those who dismiss it
as a purely private matter.

tion and dependence, the term the World Health Organization adopted for tobacco
use in 1965.140
Most contemporary researchers appear to use the two terms interchangeably, as does
the Institute of Medicines Committee on Opportunities in Drug Abuse Research
(CODAR), which considers dependence the preferred diagnostic nomenclature.141
The immediate past Chief of the U.S. Addiction Research Center has said addic-
tion is simply the more widely used term for what is more technically referred to as
The CODAR describes drug addiction (dependence) as drug-seeking behavior
involving compulsive use of high doses for no clear medical indication, with
substantial impairment of health and social function, tolerance (i.e., diminishing
response from a given drug level), and withdrawal (physical or motivational
disturbance when the drug is withdrawn). For self-report measurement, addiction to
tobacco use has been defined as craving, difficulty abstaining for a given period, and
projected or experienced difficulty in quitting.143
Earlier views included the proposition in 1975 by a prominent tobacco-use re-
searcher that addiction is a behavioral pattern of compulsive drug use, character-
ized by overwhelming involvement with the use of a drug, the securing of its
supply, and a high tendency to relapse after withdrawal.144 This definition suffers
from use of the imprecise term overwhelming and, by including the issue of supply,
seems confined to illicit drugs, possibly including underage tobacco use because a
minors act of obtaining tobacco typically involves a transaction that is illegal, at
least on the part of the provider.
Marlatt et al145 in 1988 defined addiction as a repetitive behavior pattern that
increases the risk of disease and/or associated personal and social problems. This
definition begs the question whether the actor must knowingly defy the risk of harm,
and seems to allow the awkward possibility that smoking was not addictive behavior
until harm was established. Moreover, a number of useful medications induce
withdrawal effects; such drugs include selective-serotonin-reuptake-inhibiting anti-
depressants146' 147,148 and corticosteroids.149 Nevertheless, the American Academy of
Addiction Psychiatry currently describes the classic criteria for drug dependence
as difficulty stopping, withdrawal, tolerance, and use despite knowledge of personal
From a behavioral psychological perspective, Serritella140 has described addiction as
a strong dependence engendering compulsive self-administration and withdrawal

effects. A co-editor, LAbate, emphasizes obsession, self-indulgence and self-
destruction. LAbate argues that one is an addict if one unilaterally focuses on any
behavior while diminishing more helpful behaviors and discounting loved ones; by
this definition, he says, half the U.S. population may well suffer from at least one
addiction.151 This inclusiveness may be useful for highlighting the universality and
the nonpharmacologic component of addiction, but it is too broad to guide research.
More recently, the Director of the National Institute on Drug Abuse defined the
essence of addiction as compulsive drug seeking and use, even in the face of
negative health and social consequences, and he termed addiction at its core, a
consequence of fundamental changes in brain function.... a brain disease for which
the social contexts in which it has both developed and is expressed are critically
Some or all of the following elements appear in many definitions:
Use of the substance is non-medical, i.e, dissociated from clinically detect-
able disease.
Use is repeated and self-administered.
Use entails both physiological and psychological pathways.
The user craves the substance.
Repeated use produces tolerance.
Absence of the substance engenders a state of physical and/or mental
distress or disorder (withdrawal).
Prolonged use harms the user.
Jaffe and Kanzler153 have raised three further questions:
Must all these elements be present for a behavior to qualify?
To what degree must each element be present?
What is gained and lost by applying the term to a substance?
The next questions for the current study, then, are whether and to what degree
tobacco use involves any or all of these elements.
Is Tobacco Addictive?
Those who term smoking an addiction do so for ideological not scientific -
reasons,154 Philip Morris, ca. 1996

Cigarette smoking is addictive, as that term is most commonly used today.23
Philip Morris, 1999
A recent turnaround in public stance by the worlds largest cigarette maker155
punctuated the 20th Century public debate over the nature of tobacco use. Consider-
ing whether it was an addiction, an early 20th Century British expert on opium
dismissed the notion as humorous because smoking failed to engender sufficient
craving and withdrawal.153 A half-century later, another British addiction researcher
agreed that strength of cravings and withdrawal symptoms were the key issues, but
he concluded that cigarette smoking thereby does qualify as addictive or dependent
As early as 1942, a British researcher successfully substituted nicotine injections for
smoking.157 Only relatively recently, however, did medicine and psychiatry define
smoking as a disorder. As late as 1962, WHOs Expert Committee on Drug Depend-
ence declared nicotine non-addictive, a view echoed by the first Surgeon Generals
Report on Smoking and Health.140 Neither the 8th International Classification of
Diseases (ICD-8) of the World Health Organization (WHO) nor the second Diag-
nostic and Statistical Manual of Mental Disorders (DSM-II) of the American
Psychiatric Association mentions tobacco smoking or tobacco dependence
disorder.158 Only in 1980 did tobacco use begin to be classified medically as
dependent behavior,159 and it has since gained general medical acceptance as a
treatable problem involving addiction or dependence, with current DSM and ICD
versions160'161 classifying it as such. The American Academy of Addiction Psychia-
try identifies nicotine dependence as the most common substance use disorder in
the United States and estimates that 60 percent to 80 percent of current smokers fit
classic criteria for drug dependence.150 The American Medical Association holds
that tobacco is a raw form of the drug nicotine and therefore is an addictive, crude
Nicotine Research
During the 1970s. Medical researchers explored nicotines withdrawal and
psychotropic effects163164'165,166167168'169170,171,172 and approached the conclusion
that smoking did generate addiction or dependence. Karl O. Fagerstrom produced a
scale for measuring nicotine dependence163 that was later found to be reliable.156
Investigators considered whether tobacco dependence is primarily pharmacologic or
psychologic. At least some smokers self-titrate blood nicotine levels to some
degree.173'174 Schachter found strong evidence for such a process, but only with

imprecise titration and considerable individual variation; he found as well that
withdrawal symptoms corresponded to nicotine availability, but again with individ-
ual variation.169 He suggested resolving the imprecision and variability by consider-
ing all long-time smoking as addicted but simultaneously influenced by personal
and social factors such as self-control.
During the 1980s. Difficulty remained over the diagnosis of tobacco
dependence. A population-based study of 1,005 smokers in 1987 found that two
leading diagnostic tools for tobacco use disorder DSM-III criteria and the Fager-
strom tobacco dependence scale163 yielded very different estimates of the preva-
lence of dependence (90 percent vs. 36 percent, respectively), and the authors
concluded DSM-III criteria were too inclusive.163 The Fagerstrom scale was
revised,175 and a reviewer found it improved but called for further development of
self-report measures of nicotine dependence.176
While misgivings regarding classification of tobacco use as an addiction continued
to be based largely on the relative nonseverity of withdrawal,177 a 1989 study among
a thousand smokers seeking treatment for other drug- or alcohol-dependence found
that more than half said cigarettes would be harder to stop using, and provided them
with less pleasure, than the substance they were seeking help to stop using.178 The
Director of the National Institute on Drug Abuse (NIDA) later observed that the
criterion of withdrawal severity was outdated.152
Mild debate emerged over these comparisons of nicotines pharmacologically
addictive potency to those of cocaine and heroin; such comparisons were based
mainly on the high rate of progression from curiosity to regular use and the high
prevalence of self-reported addicted users.179 A review article concluded that
nicotine cannot be considered as addictive as cocaine,180 and behavioral biologist
Jack Henningfield, former chief of NIDAs Addiction Research Center and presi-
dent of the Society for Research on Nicotine and Tobacco, concluded that nicotine
was not more addictive than cocaine.179 In earlier work, however, Henningfield
reported a possibly contrary finding, namely that current smokers who were also
former drug abusers often misidentified intravenous nicotine as cocaine, based on its
subjective effects.181
During the 1990s. A review challenged the Surgeon Generals conclusion182
that nicotine was addictive in the sense of cocaine and heroin, noting among other
objections that nicotine doesnt induce euphoria or intoxication and that not all
smokers are compulsive; the reviewers concluded that the proper comparison was to
caffeine in coffee.183 Others quickly challenged this view,184,185 dismissing the

objections as irrelevant,184 reiterating evidence of both animal and human tolerance
and withdrawal, and concluding that nicotines pharmacologic role is central (but
not exclusive) in smoking dependence.
As an alternative to the Fagerstrom scale,163 a Mayo Clinic group proposed a two
factor scale comprising self-efficacy and social skills deficit for assessing nicotine
dependence of cessation-initiating individuals.186
Cynthia S. and Ovide F. Pomerleau, nicotine-dependence researchers at the Univer-
sity of Michigan, proposed that constitutional differences in nicotine sensitivity
distinguish smokers and nonsmokers, i.e., that eventual smokers come from a
population who are simultaneously pleased with nicotines effect and who quickly
become tolerant, while eventual nonsmokers come from a population that physio-
logically responds to neither its potential pleasures nor its potential for tolerance-
In 1996, asserting jurisdiction over cigarettes as drug delivery devices, the FDA
reached an unequivocal conclusion that nicotine causes and sustains addiction by
exerting psychoactive (or mood-altering) effects on the brain that motivate re-
peated, compulsive use of the substance and create dependence in the user. The
pharmacological processes that cause this addiction to nicotine are similar to those
that cause addiction to heroin and cocaine.188
In recent years, much nicotine dependence research has focused on neuro-
pharmacologic pathways,189,190'191, 192,193, 194 nicotine replacement therapy,195,196,197,198,
199,2oo, 201 pr0perties 0f smokeless tobacco,202,203 and effects among adolescents204,205
and women.206,207 A potential genetic component of nicotine susceptibility has been
identified,208,209 which could provide biological support for a recent finding from a
small study that smokers like the psychophysiologic effects of intravenous nicotine
while nonsmokers dislike the effects.210
That nicotine is addictive seems well established. Is it sufficient to explain smoking
Alongside Nicotine: Behavioral Models
At different times and in different contexts, each of us uses multiple vocabu-
laries to describe and discuss drug abuse. Sometimes we use the vocabulary
of choice and responsibility. Sometimes we use the vocabulary of health and

disease. Sometimes we use the marketplace vocabulary of supply and
demand. Sometimes we use the vocabulary of crime and punishment. The list
goes on.... In the end, the debate about root cause turns as much on politi-
cal philosophy as it does on empirical evidence, and definitive answers are
unlikely to be produced by scientific investigation.141
Committee on Opportunities in Drug Abuse Research, Institute of Medicine
[Addiction research] is a field rooted historically in a variety of disciplines
that include medicine, psychology, physiology, sociology, social work,
biology, chemistry, politics, and witchcraft, to name only a few. The addic-
tive behaviors seem to defy classification and explanation.2U
Howard J. Shaffer, Harvard Medical School,
and Harvey B. Milkman, Metropolitan State College
One intriguing, less studied facet of tobacco use is the historic non-addicted reli-
gious use among some Native American tribes. Given an undeniable cultural
variability of nicotines addictiveness, as well as many other profound ways that its
addictiveness is not constant within cultures or even within individuals, explanatory
perspectives must consider other contributors to smoking behavior biology,
psychology, physical and social environment, culture and politics.212
Studies spanning several decades have shown that nicotines pharmacologic and
psychologic effects dont necessarily coincide. In the 1970s, a seminal and still
active tobacco-use investigator, Michael A.H. Russell of the British National
Addiction Centre, reviewed the state of the science and found little evidence to
support the nicotine dependence theory of smoking motivation;166 he concluded that
nicotine was a lesser cause of addicted smoking than nonpharmacologic factors.156
In 1980, an American research team found psychologic and pharmacologic factors
acting independently, and they showed older subjects were more psychologically
involved with smoking than were younger smokers.213 Craving, a self-reported
irresistible urge, has been shown to be dissociated with actual smoking in several
ways: Nicotine replacement can reduce tobacco intake but leave craving
unchanged;214 environmental cues to smoke and perceived opportunity to smoke can
provoke or influence craving.215 Under laboratory conditions, some individuals can
be induced to develop addictive behavior toward basic needs or desires such as
water, food, or cigarettes, if these things are made available only at limited times
and in sub-optimal amounts.141
The consistent demonstration of non-pharmacologic factors has drawn attention to

smoking from virtually every discipline that seeks to explain human behavior.
Consider a 1980 NIDA compendium216 of 43 theories of drug abuse: while the
authors identified psychiatry or psychology as a primary root of all but nine theories,
many theories had secondary roots in other disciplines, and six were primarily
rooted in sociology, four in neuroscience, three in biomedical sciences, and one each
in anthropology, biology, or genetics.
The abbreviated titles of these theories illustrate the breadth and scope of ideas
about the origins of drug abuse personality deficiency, social influence, addiction
to pleasure, disruptive environment, incomplete mourning, metabolic deficiency,
learned behavior, cognitive control, bad habit, existence, adaptation, social devi-
ance, biorhythm, problem behavior, drug subculture, bioanthropology, developmen-
tal stages, self-derogation, ego, general addiction, hyperactivity, coping, achieve-
ment anxiety, addictive experiences, social neurobiology, social interaction, natural
history, genetics, availability and proneness, perceived effects, life theme, family,
self-esteem, cyclical process, conditioning, role, defense structure, social control,
and combination of effects. No facet of the human condition seems to have
escaped scrutiny as a potential source of drug abuse.
A recent review217 that focused on health behavioral research identified a small
number of theoretical perspectives guiding a majority of investigations, including
many if not most tobacco use behavioral studies. At the level of the individual, the
most frequently used models included:
the Health Belief Model;
Social Cognitive/Leaming Theory;
Theory of Reasoned Action/Theory of Planned Behavior;
Stages of Change/The Transtheoretical Model.
Also widely used was the self-efficacy construct, which plays a central role in Social
Cognitive/ Learning Theory,139 has been added to the Health Belief Model,218 and is
often applied as a stand-alone perspective.217 Each of these dominant models is
briefly discussed here.
Health Beliefs
The Health Belief Model represents the earliest and thus most enduring theoretical
approach to health behavior.219 It posits six determinants:218
perceived susceptibility opinion of the likelihood of an adverse outcome;

perceived severity opinion of severity of the outcome and its sequelae;
perceived benefits opinion of effectiveness of the recommended action;
perceived barriers opinion of costs and difficulties of taking the action;
cues to action preparatory strategies ;
self-efficacy perception of ability to act under given circumstances.
According to the model, these constructs mediate demographic, sociopsychological
and structural variables, which exert only indirect influence on the models percep-
tual components (p.46).218 The model incorporated self-efficacy in 1988, reportedly
to address increasing professional emphasis on lifestyle interventions requiring
long-term change rather than single acts (p.47).218
Across studies and behaviors, perceived barriers are the Health Belief Models most
powerful single predictor of behavior (p.49).218 This finding jibes with the common
sense notion that humans resist change and that even in the presence of potential
benefits, the price (barriers) must be low before change seems worth it. Such
views are well-worn; witness an observation made four centuries ago by formative
Anglican cleric Richard Hooker:
If the benefit of that which is newly better devised be but small, [since] the
custom of easiness to alter and change is so evil, no doubt but to bear a
tolerable sore is better than to venture on a dangerous remedy.120
Authors of the Health Belief Model have noted that only a few smoking studies
have relied on any aspect of the model, and none have employed the model in
entirety (pp.50-52).218 The authors speculate that because most smokers continue
despite perceiving the health threat, a central element of the model seems irrelevant.
Social Learning
Social Learning Theory (SLT) presents perhaps the most comprehensive approach
among dominant research perspectives on smoking behavior. At its most abstract
level, the theory proposes that most human behavior is learned and can be explained
in terms of a continuous, three-way reciprocal interaction among behavior, environ-
mental determinants, and personal factors feelings, biology, and especially
thoughts.139,221 An especially well-developed construct within the theory is self-
efficacy, a belief about what one can do under different sets of conditions with
whatever skills one possesses (p.37).139 Efficacy beliefs are formed through four
processes direct mastery experiences, vicarious mastery experiences, verbal
persuasion, and physiological and affective states (chapt. 3) and are mediated by

cognitive, motivational, affective and selective processes (chapt. 4). The theorys
basic premise about smoking behavior is that it is learned and can be modified
through learning.222 That smoking is learned (as opposed to innate) seems self-
evident, and the construct of reinforcement that is central in both operant and
environmental learning theories223 provides a simple model for the behavioral role
of nicotine.
Reasoned Action
The Theory of Reasoned Action (TRA) recognizes that external factors are influen-
tial, but posits that they may be ignored in modeling because they must be mediated
by internal processes.224 The TRA posits intentions as immediate determinants of
behavior. Behind intentions are perceptions of the behavior as good or bad, termed
attitude, and perceptions of social pressures to perform or not perform the behavior,
termed subjective norm. The relative importance of attitude vs. subjective norm
varies among people. Attitude and subjective norm are in turn determined by
behavioral beliefs the behaviors perceived direct benefits and costs and
normative beliefs perceived social responses to the behavior. The closely related
Theory of Planned Behavior added the construct of perceived behavioral control,
which arises from two factors what behavioral resources/impediments the person
perceives, and how potent the person believes the resources/impediments are.225 (In
contrast, SLT proposes that self-efficacy arises from four factors.)
Self-efficacy alone, Social Learning Theory or social learning as a construct, and the
Theory of Reasoned Action/Planned Behavior are often used in whole or part to
discuss smoking behavior and to design interventions. They have been proposed,
alone or together with Health Beliefs and/or the PRECEDE-PROCEED model, to
design a five-year community heart health program seeking reduced risk factors
including smoking;226 to explain smoking initiation among adolescents;227 to predict
compliance with nonsmoking and other prescriptions among a hypertensive patient
population;228 to discuss nonpharmacological smoking factors;229 to predict adoles-
cent smoking status;230 to explore smoking motivations among low-SES African-
American women;231 to prevent tobacco use among American Indians and Alaska
Natives,232 and, especially for self-efficacy, to predict or prevent relapse among
cessation-oriented smokers.233,234,235,236
Transtheorv and Stages of Change
The Transtheoretical Model (TM) describes behavior change through four core
constructs stages of change, decisional balance, self-efficacy, and processes of

change.237 The self-efficacy construct here differs somewhat from the SLT version
by separating self-confidence from intensity of temptation. Decisional balance, a
persons weighing of pros and cons of changing, corresponds roughly to the TRAs
attitude toward behavior, and models one of the cognitive mediators of self-efficacy
in SLT. Not that the TM merely reflects or adapts other models: Its stage-of-change
construct is widely used to identify personal readiness to modify health risks or
adopt health promotion across the spectrum of lifestyle behaviors, especially
Clinical and other cessation-support program often tailor interventions to specific
stages of readiness to quit smoking.207,238'239,240 Three large population-based studies
in the United States have found that 40 percent of smokers are in precontemplation,
where they report having no thought of quitting smoking in the next six months; 40
percent are in contemplation, where they might quit smoking in the next six months
but not in the next 30 days, and 20 percent are in preparation, where they predict
they will quit smoking in the next 30 days.95 European studies have found much
higher proportions of smokers in precontemplation, roughly 70 percent.241,242 A
Dutch research team has suggested redefining precontemplation by asking about
five-year intentions to quit smoking, because smokers who do not plan to quit
within five years reportedly perceive significantly lower pros of quitting and are in
an immotive stage.243
The Transtheoretical Model has recently been challenged with respect to smoking.
California researcher John P. Pierce and colleagues have developed self-report
measures of addiction such as frequency and timing of smoking and found them
more predictive of cessation than are stages-of-change. 244,245 A Rhode Island group
that includes researchers David Abrams and Raymond Niaura has found that neither
decisional balance nor processes of change predicted stage movements in a worksite
cohort of smokers.246 These negative findings have not received published comment.
Comparison of Theories
Theories of Social Learning, Reasoned Action/Planned Behavior, and Transtheoreti-
cal/Stages of Change are psychologically embedded, and all propose that cognition
rather than the unconscious or impulsive emotion largely shapes human behavior. In
proponents words:
(Ajzen and Fishbein:) Human beings are usually quite rational... We do not
subscribe to the view that human social behavior is controlled by uncon-
scious motives or overpowering desires, nor do we believe that it can be

characterized as capricious or thoughtless,224 (p.5)
(Bandura:) Much human behavior, being purposive, is regulated by fore-
thought that embodies cognized goals.... Most courses of action are initially
shaped in thought. 139(p. 116)
These perspectives avoid trait theories of personality and behavior (as does the
proposed study), although smoking has been studied extensively in relation to its
uses among clinically mentally ill people, particularly in depression247,248,249 and
The three theories explicitly recognize that environment causally affects behavior,
and they rely on environmentally introduced learning opportunities to influence
unhealthy behavior. They are individual-focused, however. Efforts to explain or
affect health behavior at the community level might best exploit these intricate
approaches by embedding them within models of organizational and community
change. Support for this suggestion may be drawn from an empirical, cross-study
comparison in the early 1990s of Social Learning, Reasoned Action and Health
Beliefs.251 The meta-analysis found Reasoned Action was most predictive of
behavioral outcomes, Health Beliefs was least predictive, and Social Learning was
intermediate. The differences were not significant, however, and no model ex-
plained more than 35 percent of variance. Moreover, predictive power actually
decreased as methodological rigor of studies analyzed. One of many possible
interpretations is that individual differences, which these theories seek to predict,
might be less influential on behavior than trans-individual commonalities interacting
at community population levels with variable sociocultural and environmental
factors. The interface of individual- and community-level influences on behavior
remains opaque and disputatious.
Economic Approaches
Behavioral economics has abandoned its pre-1970 deference to psychology and
currently models addictive behaviors, including tobacco use, in several ways:252
as imperfectly rational behavior, in which conflicting preferences desire
to smoke, desire not to smoke co-exist, each representing a rational cost-
benefit preference but battling for control of immediate behavior;
as myopic behavior, in which an individuals response to price changes -
the sine qua non of microeconomic analysis, embodying all types of costs

and benefits are conditioned by previous experiences with the product, and
addictive products thus induce a myopic non-response to the price change;
as rational behavior, in which the individual treats the addictive commodity
as any other i.e., by seeking to maximize utility but tolerance from past
use diminishes current wealth and thus increases current utility of the
addictive commodity. This last view, termed rational addiction253 and
advanced by Nobel laureate Gary Becker254 and others, has been applied in
large data sets to argue that 1) consumption when cigarette prices change is
affected by past use rather than by the price change alone,252 and 2) a model
using rational addiction plus ability to quit relates unsuccessful cessation
attempts to socioeconomic status.255
Some behavioral economists extend the rational-consumer model of addiction and
argue that drug abuse can be reduced by increasing availability of acceptable
substitute commodities.256'257 One author proposes that social support is one such
commodity,258 taking the perspective that as an adjunctive behavior an alternative
to partly satisfy an otherwise thwarted need259,260 smoking is a substitute for social
support. Limited support for this view is found in a descriptive, social-interaction
analysis of smoking behavior: From lighting up to inhaling and exhaling smoke,
through holding the cigarette and tapping ashes, to the final butt-crush each mini-
act of smoking appeared to coincide with reduced social involvement, i.e., with
interruption of social discourse and distancing of smoker from others.261 *
Given the coexistence in tobacco use of choice and compulsion, and the interplay of
pharmacologic, psychologic and social forces, a challenge in studying smoking
behavior remains finding reliable ways to ask subjects why they smoke, and
interpreting the answers.
Smoking Reasons**
[T]he concept of motivation why people use drugs is far more compli-
cated than initially believed... 141
U.S. laws and policies now largely confine smoking to locations removed or isolated from
the general public. Psychological reasons notwithstanding, nonsmokers interact less and less
with smokers who are engaged in the act of smoking, and the act has thus become a
distancing mechanism by force majeure alone.
Instruments to measure smoking reasons are reviewed in Appendix A.

The riddle of smoking why smokers do it, and why they dont give it up has
puzzled observers since at least the 16th Century.153 Nicotine obviously plays a role,
as do environmental factors, but if both nicotine and promoting environments were
absent, would some individuals smoke anyway?
Concerted investigations emerged only during the last half of the 20th Century,
mainly in the United Kingdom and several of its former colonies Australia,
Canada, and the United States. By the early 1960s, researchers had generally
concluded that smoking experimentation, the onset of regular use, and maintenance
were three distinct behaviors,143 a conclusion that generally holds today; Pierce and
others have suggested further that experimentation itself involves distinct behavioral
stages.85,262 As the proposed study is limited to adults, the following discussion
focuses on reasons for regular or continuing use.
Theoretical and modeling approaches to regular smoking and smoking-cessation
intervention generally focus on outcomes and/or personality. Personality models
tend to rely on either trait theory, which assumes personality is a collection of traits
or characteristic ways of thinking, acting, feeling and reacting,263 or social learning
theory, which emphasizes situational or environmental influences on behavior.263
The addiction field in general has abandoned the search for an addictive personality,
however, in favor of interactive approaches that combine personality with various
other factors.263
During the 1960s, psychologist Silvan S. Tomkins, who wrote extensively on the
nature of human emotions,264,265,266 published a seminal model267,268 that described
smoking behavior in terms of its relation to positive and negative affect and addic-
tion. Tomkins argued against views that affective, motivational and behavioral
models were competing, on grounds that affective theory was still incomplete; he
described motives as combinations of affect, thoughts, perceived objects (of affect)
and mediators of affect.268
Also in the 1960s, Daniel Horn, a pioneer in establishing the dangers of smoking269,
270 and later a director of the CDCs National Clearinghouse on Smoking and
Health,271 proposed a learning model of smoking as a personal choice health
behavior271 that posited three ingredients for establishment of the behavior:
personal cost-benefit evaluation, smoking-role perception, and psychological
(personal and personality) factors. Horn believed smokings specific psychological
value was as an aid in managing difficult emotions.272
Among New York City college students, an early 1960s survey on smoking behav-

ior found that 72 percent of smokers gave relaxation and tension-release as reasons
for continuing to smoke; 32 percent said taste and smell were reasons; 9 percent said
smoker-friends were a reason to continue (compared to 40 percent who said they
had begun to smoke because most friends did).273 Among smokers who had quit, 54
percent gave as a reason that they no longer enjoyed it, 37 percent said they feared
adverse health effects, and 34 percent said it was too expensive. (The sum exceeds
100 percent because respondents could give multiple reasons.) The authors con-
cluded that interveners should offer substitute sources of relaxation and sensory
gratification and should study (in hopes of replicating) the process that reduces or
eliminates pleasure from smoking as had apparently occurred for many ex-smokers.
During the same period, British survey scientist Aubrey McKennell defined smokers
who expressed positive or negative attitudes toward smoking as consonant or
dissonant, respectively.143 In nationally representative survey samples, she found
fewer previous quit attempts and less self-reported addiction among consonant
smokers; she speculated that successful ex-smokers come mainly from previously
consonant smokers, and she cautioned against prevention messages that clash with
views of consonant smokers. A prominent British tobacco researcher, John Richard
Eiser, later faulted this classification,274,275 which begs the chicken-egg question
applied to addiction and attitude.
Since the 1960s, studies of smoking as a decisional process have used the term
decision in two ways: 1) the long-term choice to give up or to continue smoking,
and 2) the specific moment in which smokers decide to light up.276
Researchers have extensively studied smokers observed and self-reported, moment-
based craving for tobacco and have divided it into pharmacologic vs. nonpharmaco-
logic factors, or physiologic vs. psychologic, social, and/or psychosocial factors.
Some researchers, especially Canadian scientist Lynn T. Kozlowski, have advocated
abandoning the term as ambiguous to tobacco users,277'278'279 especially smokers,280
because users describe both weak and strong desires as cravings. Alternative
constructs have included urge 278,280,281,282 and expectancy'284 both of which have
traditions in behavioral psychology. Outcome expectancy is central to the perceived
behavioral control construct of the Theory of Planned Behavior, and in Social
Learning Theory it is one of several factors affecting motivation, which in turn is
one of several factors that mediate self-efficacy.
Another aspect of craving is its apparent mediation by both internal and external
factors. At least two studies have found that cues to smoke may be mediated by
expectancy of an opportunity to smoke. One study found that cues stimulated self-

reported craving only among subjects who knew they would be allowed to smoke
afterward.215 Another cue-exposure study found fewer self-reported urges among
smokers told they would not be allowed to smoke for three hours, compared to
smokers told they could smoke shortly. Such findings suggest that measures of
desire to smoke not be used in assessing levels of dependency for workplace
cessation intervention programs, because employees typically know their smoking-
opportunity schedules and may internally regulate desires to smoke.
Study of external cues to smoke has found mixed types of evoked responses in
smokers. Reactivity to smoking cues in laboratory environments was retrospectively
and prospectively predictive of relapse among smokers attempting to quit.285
Smoking messages, not unpleasantness of nonsmoking content, was responsible for
the effect of cues to smoke on desire to smoke and withdrawal.215 However, cues to
smoke in the context of experimental social interaction increased cardiovascular
activity but not urges or actual smoking.286 Another study of reactivity to smoking
cues found that desire, withdrawal symptoms and other measures of dependence or
craving increased with increasing deprivation of smoking opportunities (15, 90, and
180 minutes),287 although the three-hour maximum deprivation level and the 12-fold
gap between minimum and maximum levels makes the findings utility unclear.
Studies have found experimental value of cue-resistence conditioning to support
cessation efforts,288'289,290 but the findings have not been generalized.
A research perspective similar to cues considers situational factors that correspond
to times people smoke or use smokeless tobacco.291 This line of inquiry avoids
measuring craving but characterizes situations linked to smoking and attempts to
classify users accordingly. One study found that women reported smoking more in
emotional and social situations, while men reported smoking more in situations
requiring close task attention.292
A related concept is the influence of stress on smoking behavior. Exposure to
laboratory stressors has been associated with smoking,293 and smoking relieves self-
reported feelings of stress among some smokers.294295296297 Among Vietnam
veterans with post-traumatic stress disorder, symptom frequency and combat
memories are associated with smoking.298 The measured link of smoking to stressful
life events has been mixed, though. One study found stressful life events predicted
cessation-clinic dropout for men but not women.299 Another found no difference in
life event frequency between smokers and nonsmokers.300 Both studies were small,
and the findings are consistent if smokers generally experience little or no increased
life-event stress over nonsmokers but have difficulty managing the stressors of
cessation when also experiencing stressful life events.

Purdue professor Stephen T. Tiffany argues that the urge paradigm should replace
cravings as a preferred approach to behavioral research on smoking.301 Tiffany
developed a two-factor instrument (expected relief of withdrawal vs. desires/
intentions/expected pleasure) that measures hypothesized types of smoking urges in
order to explore each urge category.281 He has shown that urges respond both to
stimulated imagination302 and to external cues.282 Brian L. Carter and Tiffany have
concluded from a meta-analysis that cues to smoke (or to use other drugs) produce
stable responses and are useful as both a research method and a paradigmatic
alternative to psycho-physiological withdrawal, i.e., appetite stimulation.303 Others
have questioned the strength of the findings and the legitimacy of the conclusions.304
Several approaches to cessation emphasize recognition and accommodation of the
individual smokers gestalt rather than externally determined good-for-you goals.
Such approaches are loosely or perhaps unintentionally grounded in the phenomeno-
logical grouping of personality theories,263 in which the unique self or the individ-
uals private view of the world determines behavior. For example, a clinical
extension of educational self-determination theory305 describes smoking behav-
ioral factors as either controlled or autonomous, where controlled behavior responds
to internalized expectations and/or rewards and punishments while autonomous
behavior reflects what people find interesting and important. A somewhat similar
taxonomy emerges from a U.S. HMO-based research group that developed306 and
demonstrated307 a model for cessation motivation comprising the extrinsic factors of
immediate rewards and social influences, and the intrinsic factors of health concerns
and wishes for self-control. A New York medical research group has suggested that
clinicians engage patient autonomy when they diagnose and enhance tobacco-use
cessation motivation among patients.308
These otherwise credible approaches suffer from unnecessarily problematic termi-
nology. That behavior is either induced (seduced?) or autonomous begs the
question of origins of so-called autonomous behavior, unless one assumes genetic
and/or dispositional determinism; an intrinsic-extrinsic distinction may be adminis-
tratively convenient for sorting factors but is too general to be predictive and in
some sense is little more than a prescriptive, admonitory variation on the widely
used internal-external locus of control construct from social-learning psychology,
pioneered in 1966 by Julian B. Rotter,309 modified by Barry E. Collins,310 and
adapted to health contexts by Barbara S. and Kenneth A. Wallston.311,312 Kenneth
Wallston has led further refinement of the measure to make it adaptable to specific
disease conditions,313 and most recently extended it, with as yet unclear utility or
implications, to measure perceived spiritual control of health.314 A Smoking-specific
Locus of Control instrument, developed in 1991 using a small number of undergrad-

uates as subjects,315 has not reappeared in the literature.
Kenneth Wallston cautions that health locus of control beliefs are generally unreli-
able predictors if taken out of their proper context within social learning theory.316,
317 Rotter has likened external locus of control to the sociological concept of
alienation and relates it to concepts that involve passivity, impulsivity, dependency,
or learned helplessness,318 as smoking can be interpreted to do. A large British
survey sample study of the health locus of control scale in relation to smoking found
differences between smokers and never smokers and between ex-smokers and
smokers, but within the population of smokers, health locus of control dimensions
combined with subjects perceived value of health explained less than one percent
of the variance in smoking frequency.319 A small American study comparing self-
efficacy with locus-of-control measures found that self-efficacy was better than
health locus of control at predicting which smokers joined a cessation workshop
after an orientation session.320 A similar comparison in Australia found that only
self-efficacy predicted maintenance of smoking abstinence at 6 months post-
A related but less-problematic approach to accommodating smoker individuality,
controlled smoking, proposes that cessation assistance focus on strengthening two
skill-sets in the smoker, the ability to self-monitor smoking activity for discovery of
individual triggers and payoffs, and coping abilities relevant to the triggers and
payoffs stress control, relaxation training, behavioral management, etc.322
Tobacco Research Contamination
The literature-review phase of tobacco-related research must grapple with the
potential for non-scientific contamination of previous findings. The issue may be
illustrated by the topic of smoking and personality, which lies at the periphery of the
current study.
H. J. Eysenck, of the Institute of Psychiatry in the University of London, addressed
the question of why people smoke by presenting a voluminous case for genetic
origins of smoking in 1970323 and a condensed theory of smoking and personality in
1973.324 He concluded that genetic factors strongly distinguish nonsmokers from
persistent smokers; that parental smoking habits do not influence those of
offspring; that genetics account for about half the similarities in smoking habits
between twins, the other half attributable to non-genetic effects such as peer
influence, and that genetic differences are expressed in personality differences

between smokers and nonsmokers. Eysencks Personality Questionnaire, measuring
Neuroticism, Psychoticism and Extroversion, continues to appear in studies of
smoking and personality.325,326,327,328,329,330
Eysencks research was funded in part by the Council for Tobacco Research USA,
a tobacco manufacturing industry arm that for several decades funded and, to a
degree as yet unknown, manipulated scientific investigators and projects deemed
helpful to industry claims regarding tobacco and health. At least 305 reports on
nicotine pharmacology alone, published since 1945 but mostly after 1962, were
funded by the Council or other tobacco industry segments.331 As a result of litiga-
tion, the Council was dissolved in the late 1990s, and the industry may not reconsti-
tute it or its function.332
Eysenck made no secret of the Councils funding of his work. Whether the Council
exerted non-scientific influence is not known. The Council funded many other
investigators and projects, and the issue is thus present throughout the current
review for the proposed project. The question of whether stakeholder funds beget
suspect facts is not limited to tobacco but applies as well to scientific investigations
of pharmaceuticals funded largely by their manufacturers. Rather than attempt to
determine which, if any, of scores of studies received tobacco industry funding and,
further, whether the funding source influenced the work, the current study takes a
practical approach, consistent with the U.S. Food and Drug Administration (FDA)
review of drug studies, which assumes that some legitimate lines of scientific
inquiry coincide with industry interests but can still be evaluated on the merits. I
thus assume that undetectable data falsification does not correlate with the funding

1. Demographics: In the United States, education, income, and marital status
strongly predict tobacco use status, for reasons that are unclear; ethnicity,
gender, age and region weakly predict tobacco use status. Divorce and low
levels of income and education are strongly associated with current smoking.
2. Cessation:
a. During the last quarter-century, a consistent two-thirds to three-
fourths of U.S. smokers have said they would like to quit smoking.
b. At any given time, approximately three-fifths of U.S. smokers are
considering quitting or planning to quit smoking.
c. Nicotine-replacement therapy, bupropion and counseling have helped
some smokers quit and have demonstrated high efficacy.
d. Approximately three-fourths of U.S. smokers say that they could quit
if they decided to.
e. In any given year, about 2.5 percent of current U.S. smokers do quit.
f. Readiness to quit (stage-of-change model), previous quit attempts,
and timing and frequency of use are mild to moderate indicators of
cessation likelihood.
3. Dependence:
a. Dependent or addicted use may be defined as repeated, non-medi-
cal self-administration of a substance that harms the user over time,
that affects and is affected by both physiological and psychosocial
pathways, and that may invoke cravings, tolerance, and/or adverse
reactions to the substances withdrawal.
b. Nicotine alters human neurophysiology in many but not all individu-
als; primary adverse effects are induction (within variable durations
of use) of cravings, tolerance, and mild to severe discomfort upon

c. Much or most adult tobacco use is dependent, often involving one or
more psychosocial components that can be independent of nicotine
d. A simple, widely used measure of dependence is self-reported timing
of the first morning cigarette; use within 30 minutes of waking
suggests relatively strong dependence.
4. Choice: No operational approach currently exists to separate dependence
from rational choice in tobacco-use behavior.
5. Behavioral Models: Potential theoretical approaches to individual smoking
behavior include the Health Belief Model, Social Learning Theory (and Self-
Efficacy), Theory of Reasoned Action/Planned Behavior, and the
Transtheoretical (Stages-of-Change) Model.
a. The Health Belief Model is presumed to have limited relevance for
smoking cessation interventions because of widespread knowledge
among smokers of associated health risks.
b. The Health Belief Models most predictive component, barriers to
change, might be highly relevant to smoking cessation interventions
by suggesting an emphasis on removal of cessation barriers, e.g.,
through wider availability of NRT and/or bupropion.
c. The remaining models appear to offer only modest guidance for
smoking cessation interventions. To date, studies have shown that
cessation intervention outcomes are somewhat consistent with one or
more of the models, but no one model has emerged as either more
consistent with outcomes or highly predictive of them.
6. Reasons: A variety of paradigms guide studies of reasons for smoking:
a. Few researchers continue to focus solely or primarily on personality.
b. Some smokers appear to smoke mainly to manage negative emotions
(anger, sadness, helplessness, agitation, stress, etc.).

c. A largely untested paradigm of smoker consonance (favorable
toward smoking) vs. dissonance (negative about smoking) proposes
that consonant smokers can quit at will while dissonant smokers are
d. Several paradigms may be classified as mainly immediate (focused
on reasons for smoking the current or next cigarette) or mainly
enduring (focused on reasons for ongoing, repeated smoking events).
e. Studies of immediate reasons for smoking have shown that cravings
or urges that originate internally, and external cues or situa-
tional factors, are distinguishable reasons for smoking, can interact,
and have variable influence across individuals and across time.
The current study uses several of these findings to analyze the severely elevated
smoking prevalence found among young adult patients in participating Region VIII
community and migrant health centers. Specifically, the study tests the explanatory
power of sociodemographic factors; attitudes and past behaviors regarding cessa-
tion; dependence levels, and several study-design artifacts that might have biased
smoking prevalence estimates from the sample.
The current study measures proxies of several theoretical constructs, including
Social Learning Theorys self-efficacy (perceived ability to quit smoking) and
vicarious learning (household presence of other smoking behavior), the
Transtheoretical Models stages of change, and the Theory of Reasoned Actions
attitude (behavioral cons, desire to quit). The data are inadequate to support testing
of a full theoretical model.

Development and Operation of a Clinical-research Network
The Community Health Center (CHC) Program, funded under Section 330 of the
U.S. Public Health Service Act, provides for primary and preventive health care
services in medically underserved areas throughout the U.S. and its territories.333
The Migrant Health Program, under section 330g of the Act, provides medical and
support services to migrant and seasonal farm workers and their families.334 The
combined programs contract with not-for-profit organizations to operate the centers,
which are staffed primarily by physicians, dentists and mid-level clinicians who
repay professional-school loan obligations through their work in medically under-
served areas as part of the National Health Service Corps (NHSC).335
A total of 39 grantee organizations operate 167 health centers across Colorado,
Utah, Montana, South Dakota, North Dakota, and Wyoming, the states comprising
PHS Region VIII.336 A majority of health center patients in the region reported their
household income in 1997 at or below the federal poverty level.337
During Federal Fiscal Year (FFY) 1996, Region VIIIs health center association and
affiliated clinician network received a one-year grant for practice-based research
from the AHRQs Bureau of Primary Health Care (BPHC). The grant funded a
small study of smoking and pregnancy which, beyond its intended objectives,
revealed serious gaps in the networks research design and implementation skills.338
The oversight group organized a small research-training conference, which was
attended by fewer than 10 clinicians, and two nationally known experts on practice-
based research discussed design issues and strategies with those in attendance.338

Soon afterward, two clinicians proposed a study of patient beliefs about tobacco use,
more specifically, a contrast study between never-users and ever-users, guided by
the Health Belief Model and speculation that contrasts in ever/never-user beliefs
might inform design of a tobacco-use intervention.338
In May 1997,1 participated in a conference call to consider assisting the study. Also
present were several clinicians comprising the networks Practice-Based Research
(PBR) steering committee, the groups staff person, and a Medical Consultant-
Epidemiologist (MC-E) providing technical assistance to BPHC research grantees.
The discussion included:
discussion of a need for simple data-collection methods that clinicians
could administer quickly and easily and that health center staff could handle
within the constraints of low skills and competing demands;
the Health Belief Model as a potential theoretical foundation for the study.
The group agreed that I would review relevant theoretical literature and provide a
summary from which we would determine how to proceed. The review was limited
to recent articles regarding stages of change and the Health Belief Model; primary
findings were that level of addiction was emerging alongside stages of change as a
potential predictor of cessation,244 and the Health Belief Model produced mixed
results in smoking cessation. (The annotated and referenced summary is provided in
Appendix B).
Based on the summary review, the group agreed that the study should not be limited
to the Health Belief Model as its theoretical foundation.* We formed a study team -
myself as Principal Investigator (PI), the MC-E as Co-PI, the groups staff person as
Study Coordinator, and the clinicians as Study Advisory Committee for a two-
phase assessment of (Phase One) tobacco-use prevalence, associated attitudes, and
(Phase Two) clinician cessation-intervention based, if possible, on Phase One
attitudinal findings.
The study design, instrument and site protocol required negotiating acceptable
balances of scientific validity and practical feasibility. The study team weighed
clinic self-selection vs. random selection of clinics with aggressive recruitment of
those selected. To inform the decision, we conducted a pre-recruitment survey to
determine the clinics level of interest, willingness to commit staff resources to a
One clinician strongly dissented and later withdrew from participation, attributing his
decision to a view that the questionnaire was impractical.

study, and the potential role of financial incentives. The survey (Appendix C) was
distributed both to organizations (n=39) and to their individual sites (=167), and
38 responses were received, several of which represented multiple sites. Half the
respondents said they would participate, four declined, and the rest said they might
be willing or wanted more information.
The group was encouraged by these results, which were better than expected given
competing clinic demands and the studys inability to offer reimbursement for
participation. Also, the group lacked resources for intensive refusal-conversion
efforts. The team opted instead for clinic self-selection with persuasive recruitment,
believing participation would be high enough to produce useful results even if
findings might not be applicable to nonparticipant clinics.
The team then obtained time during otherwise established conference calls and
regional conferences to conduct highly interactive discussions with clinicians,
providing semi-structured facilitation that combined study promotion with research-
skills assessment and enhancement of clinic buy-in. Through these discussions the
team was seeking primarily to learn what subject-recruitment and data-collection
methods would succeed in health center environments, and to instill a sense of study
co-ownership by eliciting and incorporating clinician input and feedback. At one
regional conference,339 study team members facilitated a two-hour discussion with
about 30 clinicians, who unanimously agreed that:
tobacco use was a compelling problem in their clinics;
capture of demographic or other data from refusers was not feasible;
time and skill constraints on front-office staff necessitated a simple proto-
o the use of one questionnaire for ever-users and another for never-
users would be impractical;
o staff would generally be unavailable to administer questionnaires;
adolescents were a sparse population in health centers;
a Spanish translation of the questionnaire would be necessary;
most health center patients would be able to read and self-administer a
simple questionnaire written at a low literacy level (e.g., 6th to 8th
grade readability);
each clinics administrator and medical director should be contacted about
study participation. Either one could commit the clinic to the study,
but medical director commitment could ensure protocol compliance
while administrator commitment would not.

A sample questionnaire, clearly marked DRAFT, was circulated for discussion.*
After incorporating suggestions, the study team developed and distributed a clinic
recruitment letter (Appendix D) and memorandum of agreement (Appendix E). A
total of 31 clinic sites, serving roughly one-third of the Region VIII health center
patient population outside the Denver Health system, signed agreements. They were
sent study packets that included a cover memo and study instructions (Appendix F),
a survey refusal log (Appendix G), and the requested numbers of English and/or
Spanish questionnaires. No direct staff training was conducted, again due to lack of
funds for trainers and travel.
During data collection, the Study Coordinator conducted weekly or bi-weekly
telephone contact with designated liaisons at each participating clinic to assess
progress and offer technical assistance; notes of these communications (Appendix
H) were circulated monthly to the PI and Co-PI. Early contacts identified and
corrected mistakes, including three clinics that had been recruiting only smokers,
and two others that had photocopied and distributed the sample questionnaire
marked DRAFT. The clinics that had violated the protocol were instructed to
return all completed questionnaires, which were segregated by the study coordinator
and excluded from data entry and analysis. These clinics were also instructed to
return all materials not distributed in the official study packet, and to re-initiate data
collection using correct materials and eligibility criteria. The Study Coordinator also
broadcast a fax reminding all participating clinics of the correct eligibility protocol
and proper questionnaires, and she revisited the issue in the next round of telephone
Ongoing communication revealed that many clinics did not consistently determine
eligibility and request participation of every visitor on every day. Clinic liaisons
blamed the lapses primarily on unpredictable patient rush-hours and staff forget-
fulness. Especially damaging was the December holiday period, which created
staffing gaps due to vacations. None of the lapses appeared systematic; given that
each clinics sample represented an arbitrarily selected time period that was roughly
comparable across all clinics, no special adjustment was made during analysis.
Slightly more than half the clinics reported refusals orally and approximately; the
others did not provide information. In all but two reports, refusals were reportedly
infrequent, ranging from zero to five out of 50 to 100 completed surveys. Two
clinics reported high refusal rates; one estimated approximately 40 percent refusals,
the other, 66-75 percent refusals. No demographic or tobacco-use information was
Questionnaire development is described at p. 42 ff.

available about refusers or survey non-returners.
One clinic reported seeing non-English/Spanish-reading Vietnamese patients two
mornings each week; these patients were designated ineligible for the study due to
lack of resources to provide a Vietnamese-language questionnaire.
As the scheduled end of data collection approached, several clinics requested an
extension, which was granted, to compensate for slower-than-expected progress.
After the data-collection period ended, 30 of 31 participating clinics reported that
they had sent the completed questionnaires using the pre-paid return envelopes
provided by the study team; one clinic had collected fewer than 10 questionnaires
and not return them. Two packets failed to arrive; both clinics insisted the packets
had been mailed, and postal traces were initiated but failed to find them. The study
team had not asked clinics to make and retain photocopies of completed question-
naires, due to the time and expense that would have been required.
In retrospect, the study teams most consequential mistakes were 1) failing to
foresee and avoid disruption of data collection by the holidays, 2) not communicat-
ing directly or often enough to prevent clinic staff from misunderstanding eligibility
requirements, and 3) relying exclusively on postal mail to preserve completed
questionnaires during transport.
None of these mistakes appears to threaten study validity. Mistake No. 1 introduced
a non-calculable disruption of random (sequential) selection, but there is no evi-
dence or obvious reason to suspect systematic bias. Mistake No. 2 altered the data-
collection start date for three clinics, by several days to several weeks, but since the
original start date was arbitrary and the slippage is relatively slight, bias seems
unlikely or negligible. Mistake No. 3 eliminated data from two clinics, reducing the
sample size but leaving clinic participation by self-selection unaffected.
The most consequential site-based mistakes were 1) not identifying and recruiting
every eligible visitor, and 2) not tracking refusals. Both mistakes affect statistical
management of non-response and are discussed fully in that section.
Study Design
The study used a cross-sectional design with random (sequential) sampling of age-
eligible patients nested within self-selected clinics (primary sampling units, or

PSUs). The sample represents age-eligible patients of participating health centers
rather than all Region VIII health centers. This approach was chosen because it
allowed estimates and inferences to be used as baselines for comparison with
possible post-intervention data that might be collected in the same way from the
same clinics (repeated survey analysis340). Tracking of respondents was considered
but was ruled out as unnecessary and exceeding clinic capacity.
All health centers were eligible to join, and one of two sample sizes (50 or 100),
comprising equal numbers of men and women, was allocated to each participating
clinic according to staff-assessed recruitment ability. The decision to accrue equal
numbers of males and females was grounded in a belief that fundamental sex
differences exist in smoking prevalence and attitudes. Given that health center
visitors are disproportionately female, the design oversampled males by allotting
them a roughly equal proportion of the sample although they comprised consider-
ably less than half the patient population.
The study design underwent institutional review and was approved November 14,
The Instrument
Before starting to develop the survey questionnaire, the study team held several
discussions regarding content and format. The primary areas of interest were
tobacco-use status and reasons for smoking/not smoking. Consistent with guidance
from clinicians (see page 39), the team agreed the questionnaire would need to be
readable by low-literate individuals and would be produced in Spanish and English.
The Co-PI proposed, and the team agreed, that tobacco-use status would correspond
to the algorithm used in the CDCs Behavioral Risk Factor Surveillance System
(BRFSS), enabling prevalence rate comparisons with the general population.
Other development issues generated moderate disagreement, pitting concerns for
study validity against concerns for practical feasibility and raising such questions as
Could clinic staff administer the questionnaire, or would it have to be self-
Could clinic staff capture demographic and use-status data from refusers,
or would the study have to treat refusal cases as missing completely at

What was an acceptable compromise between asking all the study ques-
tions that seemed important, and asking so many questions that partial or
complete nonresponse would rise to unacceptable levels?
Which protocol would trigger fewer completion errors, having staff deter-
mine each persons smoking status and select the right questionnaire, or
respondents following smoking-status skip patterns in a one-size-fits-all
In general, practical considerations trumped scientific precision. Health center
clinicians had previously conducted almost no clinical research, and key-informant
clinicians and staff had said clearly that staff would not have time or knowledge to
do more than ask about age-eligibility and hand out questionnaires. The decision
was to produce a questionnaire that would 1) be primarily self-administered, with
clinic staff directed to assist if asked,* and 2) take five minutes or less to complete.
The number of questionnaire versions was left undecided pending pilot tests.
The agreed-on content included tobacco-use status and history, levels of addiction,
readiness to change, reasons for and against use and non-use, proximal (primary
social group) tobacco use, and demographic data.
Tobacco-use status measures were those used in the BRFSS. For smoking, they are:
current smoker smoked more than 100 cigarettes in lifetime and
smoked in last 30 days;
former smoker smoked more than 100 cigarettes in lifetime but did not
smoke in last 30 days;
never user did not smoke more than 100 cigarettes in lifetime.
Proxy addiction measures timing of first morning cigarette, daily vs. some-day
smoking, and number of cigarettes smoked per day also corresponded to BRFSS
questions. Current smokers readiness to change was measured by the method of
DiClemente and Prochaska.342
I relied on my general knowledge and experience, rather than targeted literature
reviews, to develop questions regarding pro-con reasons for smoking status, quit
In subsequent analysis, 5.1 percent of respondents reported assistance.

history, and desire and ability to quit smoking.
Pro-con lists for ever-smokers included 16 potential reasons for smoking and 10
potential reasons for quitting; a con-only list for never-smokers included 10 poten-
tial reasons for not starting. Each list of closed-ended items was preceded by an
open-ended request for reasons and was followed by an open-ended query about
reasons that were not listed.
Demographic measures were based on the teams general knowledge and included
age, ethnic group, number of children living at home, gender, marital status,
education, and employment status. Income was not asked, for fear it might cause
some potential respondents to decline to participate.
Proximal (primary social group) tobacco use was measured by questions about each
parents or guardians smoking status when respondent was a child, and number of
tobacco users in the respondents household.
Questionnaire wording was designed for low-literate populations, based on clinician
reports that low literacy was common among their patients. Multi-scale assessment
showed the questionnaire was readable at a 2nd to 7th grade level.*
Answer order was reversed on half the printed questionnaires (i.e., low-to-high or
negative-to-positive vs. high-to-low or positive-to-negative), and each participating
clinic received equal numbers of both versions, in order to assess the impact of
answer-order on responses.
The questionnaire was pilot-tested on three small samples of health center patients
and staff. First-round pilots showed that it was easily completed in less than five
minutes, but some wording, sequencing and formatting confused some participants.
Based on first-round results:
an initial distribution idea having staff give every clinic visitor an age-
eligibility and smoking-status card to be completed and exchanged for a
questionnaire matched to smoking status was abandoned because staff
found it hard to manage. Instead, a single questionnaire was produced for all
smoking statuses. A front-page question about ever use vs. never use di-
rected respondents to survey sections appropriate to their use status, and
Multiple methods were used to assess readability and produced varying results: Dale-Chall
(grade 5.2), Flesch (2.7), FOG (5.0) Powers (4.4), and SMOG (6.9).

simple skip patterns accommodated former vs. current users;
two-sided printing was replaced by single-sided printing because the Study
Coordinator observed pilot respondents overlooking some pages;
a question was added regarding whether survey participation made the
respondent feel more like quitting, prompted by a pilot participants report
of just such an experience and the teams curiosity about the frequency of
such feelings;
the strategy of asking respondents to number, in order of importance, the
top five listed reasons for smoking/not smoking/quitting was abandoned
because respondents found the instructions confusing. An ordinal scale was
used rather than rewritten instructions, because the Co-PI said analysis
would be easier if every respondent rated every item on a multi-level scale.
Three labeled levels were used, rather than 5- or 7-number endpoint-an-
chored Likert scales, because the other team members felt it would be easiest
for respondents to interpret;
the word Mexican was added to the ethnic category labeled Hispanic or
Latino or Chicano or Mexican-American;
The questionnaire was revised and retested successfully, then was translated into
Spanish (Appendix J) by a staff member of an Hispanic-serving clinic network
whom the network administrator represented as a native speaker. The translation
was done as a contribution; the study team did not apply formal translation proto-
cols or pilot-test the Spanish version, due to resource constraints (discussed further
at p. 100. Gender-specific color-coded formats were produced in both languages to
facilitate staff tracking toward recruitment goals.
Data Processing
Twenty-eight clinics returned a total of 1,916 surveys, 9 of which were blank. A
specialty firm (Unique Data Systems, Denver) entered data into ASCII comma-
delimited format using independent double-entry. The firm sequentially numbered
each questionnaire as it was entered to allow comparison of entered data with paper
originals. I translated ASCII files into Stata343 data files using StatTransfer.344

I excluded age-ineligible cases* and removed 103 (5.9%) of 1,748 age-eligible
questionnaires because they were not interpretable and could not be repaired
(reasons are listed in the next paragraph), or because they came from one of three
clinics that returned fewer than 25 questionnaires, which could confound analyses
requiring large-sample assumptions. Accepted, completed survey counts by clinic
ranged from 25 to 117, with a mean of 68.5 and standard deviation of 24.4.
The 55 non-interpretable surveys were removed for the following deficiencies:
blank attitudinal sections;
answers too inconsistent to be interpreted;
missing demographic data plus missing data in one other relevant section;
non-credible demographic data (e.g., age=6, birth date=3/16/98);
no ever-use status and nearly all blanks.
Of the three clinics with unacceptably small samples, two were in Colorado (n=14,
23) and one was in Utah (n=11).
Aggregate Case Comparison. Excluded vs. Included Clinics
Cases from the three excluded clinics were similar to the included sample in
sociodemographic distribution and cigarette use (Table 1), except that a greater
proportion of excluded-clinic cases were non-white and non-Hispanic (African
American, Asian American, or American Indian).
Age-ineligible cases (w=159) comprised 8.3% of completed, returned surveys; they included
14 cases younger than 18, and 145 cases older than 39.

Table 1. Sociodemographic distribution and cigarette use by clinic inclusion status (unweighted %)
Demographic Variable Included n=1645 Excluded n=A8
Language (Spanish) 11.8 4.3
Sex (female) 58.8 60.4
Ethnicity white 70.2 65.1
Hispanic 20.9 9.3
other 8.9 25.6***
Education >HS grad 15.0 11.6
high school grad 34.7 39.5
some college 27.6 37.2
college grad 22.7 11.6
Employment full-time 59.6 52.2
part-time 16.6 17.6
unemployed 23.8 27.3
Marital status married 48.2 45.8
divorced 21.3 27.1
single 30.5 27.1
Ever smoked (yes) 55.6 60.4
Currently smoke (yes) 42.6 44.2
*** /?<0.001
For fewer than 10 questionnaires where tobacco use status was missing, I completed
them if possible based on clear and consistent answers throughout the survey,
including correct completion of smoker, chewer, or never-user portions, as well as
written comments (e.g., never smoked).
Recoding and Transformation
The following sections were coded missing for the subgroups indicated (the
questionnaire is provided in Appendix I):

tobacco-use questions (3-24) for never-users;
never-user questions (26-30) for tobacco-users;
chew instead of smoking questions (5n-p) for never-chewers;
reason-to-quit questions (9-10) for those not wanting to quit (no on Q7);
chew-related addiction/readiness questions (A 11-21) for never-chewers;
smoking-related addiction/readiness questions (B11-21) for never-smokers;
current-chewer questions (A 15-22) for former chewers;
current-smoker questions (B15-22) for former smokers;
former-chewer/smoker questions (23-24) for current users.
If self-reported never-smokers/users answered smoking-attitudinal sections and
consistently marked not a reason or not true for me, I recoded answers missing.
Where two subjects answered that they had quit tobacco at age 2,1 interpreted the
answers to mean two years ago and adjusted the quit ages accordingly.
Data-entry staff reported that, for never smokers-chewers, attitudinal answers on
yes-first surveys numbered 0001 to 0925 were inadvertently entered with reverse-
coded values, i.e., a value of 3 should have been 1 and vice versa. This error
occurred because the presentation sequence of answers to questions 29a-j had not
been reversed from the no-first order. I recoded a total of 135 such surveys by
reversing 1 and 3 values.
Where respondents (42 ever-users and 25 never-users) answered big reason to
every question of motivation not to smoke, I created a dummy variable to flag them
as possibly representing acquiescence response bias,345 in order to allow analysis
with or without them.
Extreme values to the questions number of cigarettes/chews per day (n=l for
cigarettes, n=2 for smokeless tobacco) were recoded slightly above the next highest
value to make normal transformations possible. Both variables were still highly
skewed but were successfully transformed to near-normal distributions (square root
for cigarettes, log adjusted to zero-skewness for chew).
Analysis of Complex Samples
The process of selecting individuals to represent a population often yields a sample
that misrepresents portions of the population in relation to others, and contains
correlation structures unlike those in the population. These issues are typically

present even when the process involves randomness and probability-based sam-
pling. Then the use of statistical methods intended for elements selected independ-
ently and with equal probability will produce biased estimates of central tendencies
(such as means), model parameters (such as regression coefficients) and the preci-
sion (variance or standard error) of these estimated quantities.
Statistical approaches have been developed to handle these not simple, hence
complex samples, and such methods are routinely used to produce approximately
unbiased estimates from survey data.346,347'348,349 In general, the approaches involve
(1) the assignment of a calculated weight for each observation to compensate for
unequal selection probabilities arising from a variety of sources, and (2) computa-
tional methods that account for design-induced correlation from sampling by strata
and/or clusters.
Simple weights to account for selection probability are the inverse of the probabil-
ity, which may sometimes but not always be calculated as the inverse of the sam-
pling fraction.350 If n elements are selected randomly with equal probability from a
population size N, the sampling fraction is
which is easily shown to be each elements probability p of being drawn into the
sample. Under this condition, each selected element would receive a weight
W = f-'=. (3.2)
Such weights are only necessary if elements in a sample have unequal selection
probabilities. If 10 men and 10 women will be selected, for example, to represent a
population containing 100 men and 200 women, each female is half as likely as each
male to be selected and thus receives twice the weight relative to male sample
members. Such weights not only adjust for intentional or designed differences in
selection probabilities, they are often also used to adjust for differing response rates
of individuals across sampling strata and clusters.351 Weighting is also used to make
sample subgroups reflect their proportional presence in the population, in which
case the process is known as post-stratification. Final weights are the product of the
various types of weights and are usually rescaled to sum either to sample size or to
population size, depending on the statistical approach used to calculate estimates
and their precision.
In addition to weighting, complex-sample analysis addresses correlation structures
introduced during the selection process by stratification (sampling from conceptu-

ally designed population divisions or strata) and clustering (selecting a sample in
groups or clusters rather than individual by individual). These structures or design
elements can bias estimates of central tendencies but exert their strongest effects on
estimation of variance, due mainly to similarities or homogeneity among individuals
within subgroups (also called intraclass correlation or ICC, p, or Kishs synthetic
rate of homogeneity, ro/i346).
Two other issues arise in complex-sampling. First, a samples final size is often
unpredictable, especially in survey samples, where precise population numbers may
be unavailable, lists used for sampling are usually inaccurate, response rates are
uncontrollable, and PSUs are often unequal sizes. PSU sample sizes themselves
then becomes random variables, and parameter estimates require computation of
ratio-means, or division of one random variable by another. The process is not
addressed by methods based on simple random sampling (srs). Rather, specialized
computational programs estimate variance by first-order linearization (i.e., Taylor
Series approximation) or resampling (jackknife, repeat replication, and variants of
these approaches). To avoid significant bias, linearization is used only when the
coefficient of variation of the ratio-mean denominator is less than 0.1 or 0.15.*
Second, survey and other population samples are based on finite populations rather
than theoretically infinite populations of real numbers. A simple finite population
#c = l-/ = l--^, (3.3)
is used when sample sizes are large relative to the population (e.g.,/>0.1) to
prevent overestimation of variance.
In short, complex-sampling or design-based analysis addresses the impact of
stratification, clustering, all sources of case-weights and the effect of the weights,
division of random variables, and finite populations. Programs that take these
factors into account estimate parameters using a variety of formulas derived for each
set of conditions.
Coefficients of variation for ratio-mean estimates from complex-sampling data are computed
generally as the quotient of the standard error of the weighted mean PSU subgroup size
divided by the weighted mean PSU subgroup size; when PSUs are clusters, the result is
divided by the square root of the number of PSUs in which the subgroup appears. For
widely distributed subgroup counts, the Poisson distribution may offer a better
approximation. See Kish,346 p. 220.

To illustrate the impact of complex sampling, I arbitrarily selected covariates (age,
sex, college education) and contrasted design-based results with assumed srs results
under univariate regression of current smoking using the Stata svylogit command.343
The contrasted approaches produced slightly different results in two estimated odds
ratios (OR) and the respective standard errors, with no overall directional trend by
analytic method (Table 2). The most consequential difference was the loss of
statistical significance for male sex as a covariate under design-based estimation.
Design-based estimation clearly alters analytic results, often more markedly than in
the present illustration,352 and even small effects can lead to different interpretations
and conclusions.
Table 2. Contrast of univariate logistic regression results
under design-based analysis vs. simple random sample (srs) analysis
srs design-based
covariate OR of current smoking s.e.(OR) P OR of current smoking s.e.(OR) P
age (one-year increments) 1.03 0.0085 0.001 1.03 0.0088 0.006
any college 0.41 0.0421 <0.001 0.45 0.0308 <0.001
male 1.30 0.1349 0.013 1.34 0.2084 0.082
Design Basis for Analysis of the Study Sample
Case Weighting
True refusal nonresponse is unknown (see description of tracking, p. 40) but appears
to be less than 20 percent, missing completely at random,353,354 and ignorable. On
this basis, post-stratification of the sample to participant-clinic population data
adjusts for nonrespondents as well as over- or under-sampled subgroups.
Ten analytic pseudo-strata were formed based on the state, the community setting,
and the clinic size; these pseudo-strata were temporarily collapsed and combined as
needed to create cells with at least 15 observations, producing five weighting
Colorado mountains, Wyoming, and Montana;

the Dakotas;
Colorado plains;
Denver Health.
Within each weighting group, respondents were sorted by clinic, sex and three
ethnic categories (non-Hispanic white, Hispanic, other). Weights were the inverse of
apparent selection probability, f'1 = NJnac, where a represents clinics, c represents
cross-class categories (sex by ethnicity), and N and n represent population and
sample sizes, respectively. Data from 1997 for N were used for weighting.
Some clinics provided Nac in age-groups that matched the sample, but others could
provide data only in the format required of health center grantees by the federal
Uniform Data System (UDS),355 which uses age-groups 15-19, 20-24, and 25-44.
The study teams lack of foresight in this regard necessitated estimation of Nac for
these clinics, which was produced by applying age-by-state-specific factors derived
from Census Bureau age-by-state population estimates.
Denver Health clinics could only provide systemwide UDS data from which clinic-
specific information could not be extracted; cases from these clinics (Eastside,
Lowry, Park Hill) were assigned approximately mid-level weights.
Final case-weights were rescaled to sum to sample size. No finite population
correction was used because, at roughly 0.05, its effect would have been negligible.
Imputation of Item-Missing Values
Although many social or behavioral studies analyze only complete cases those
where subjects answered every item under study the approach can bias estimates
and reduces precision in proportion to the number of subjects excluded.354 To avoid
these problems, a number of techniques are widely used to impute item-missing
values based on all available information from complete and partially complete
For the current study, cases with item-missing data on both sex and ethnicity (=76)
or either variable alone (n=20 each) were imputed using random numbers and clinic
population proportions. (Where respondents had completed surveys in Spanish,
missing ethnicity was automatically imputed as Hispanic.) This process made
possible the development of post-stratification weights for all cases based on sex
and ethnicity.

Additional demographic item-missing values (7.1 percent of demographic variables)
were imputed using IVEWare,357 which performs best-subset, pseudo-maximum
likelihood regression with a stochastic component. The demographically imputed
data set was then split into four subsets (ever-smokers, ever-chewers, ever-users of
both forms, and never-users of either form), and IVEWare was used again to impute
item-missing attitude and addiction-level values appropriate to use-status for each
subset. The subsets were merged, and item-missing stage-of-change values were
imputed for current users only using Stata impute, a best-subsets regression proce-
Robust variance estimation in data sets imputed this way requires generation of
multiple data sets, each imputed using a different random seed, and calculation of
the variance of the variance for each estimated parameter across the sets. Generation
of five such sets is sufficient to prevent analyses from overestimating power based
on imputed data. For the present study, a single data set was imputed to simplify
analysis, and results with only marginal statistical significance were subjectively
Design-based variance was estimated using first-order linearization; the coefficient
of variation was <0.1 for parameters estimated.
Population Representation
The sample includes 1,645 cases weighted to represent the combined patient
populations, ages 18-39, seen in 25 participating clinics during 1997. UDS data for
1998, which recently became available, show that the total number of patients aged
18-39 grew by just 1 percent in organizations that administer the participating
clinics. The use of weights based on 1997 clinic populations are unlikely to bias
results significantly, especially since the recruitment period spanned both years.
The sample was collected during a three-month fall-winter period, and although it
has been adjusted by clinic, sex and ethnicity to represent the annual visiting
population, seasonal patients are presumably over- or under-represented. One may
speculate that agricultural migrant workers might populate the region only during
April-October; construction labor, a common migrant occupation, also tends to be
most available during fair-weather months. Nonetheless, the sample includes a
sizeable proportion of Hispanic respondents (21%) and a relatively large percentage
of Spanish-language respondents (12% of the total). No definitive answer is possible
to the question whether the variables of interest differ non-randomly by seasonality
of patient visits.

Although clinics chose to participate or abstain, the age-eligible patient population
of participating clinics comprised nearly one-third of the entire population using
Region VIII health centers other than Denver Health.
Comparison Populations
In preparation for health center vs. general population comparisons and related
analyses, survey data external to the current study were obtained, including:
1997 results of the Behavioral Risk Factor Surveillance System (BRFSS)37
surveys for Region VIII (=5139, ages 18-39) and the nation (=52837, ages
18-39). BRFSS is a random-digit-dialing (RDD) telephone survey that uses
state-based sampling, with states selecting multi-stage cluster designs,
disproportionate stratified designs, or simple random sampling of all house-
holds with telephones in the states. Primary sampling units (PSUs) are
hundred-blocks of phone numbers, e.g. (800) 555-10xx. BRFSS documen-
tation reports that 93.8 percent of U.S. households, and 93.7 percent to 96.8
percent of households in Region VIII states, had telephones in 1994-96. The
BRFSS provides smoking prevalence estimates comparable to the Census
Bureaus Current Population Survey Tobacco Supplement, a national survey
that includes households without telephones;358
1997 results of the National Health Interview Survey (NHIS; =15526,
ages 18-39).359 NHIS is a household survey that uses stratified multistage
area sampling, with states and certain large metropolitan areas serving as
strata and PSUs defined using Census boundaries;
1997 results of the National Household Survey on Drug Abuse (NHSDA;
n=\ 1858, ages 18-39).360 NHSDA is a household survey that uses multistage
area sampling, stratified by ethnic concentration to facilitate oversampling of
Hispanic and black households;
1996 results of the California Tobacco Survey (CTS; =9089, ages 18-
39).361 CTS is a list-assisted RDD telephone survey; PSUs are hundred-
blocks of phone numbers containing at least one working residential number,
a technique known as 1-plus working banks;
Tobacco-use and demographic items are highly similar across all surveys; differ-
ences and comparison methods are indicated where necessary. For all data, design-
based descriptive statistics were produced using Stata wy commands.343

Test of One-Way Trends
Where Mantel-Haenszel trend tests were needed, variance was approximated using
design-based cross-tabular counts under simulated reductions of sample size. This
approach was necessary because no published method exists for computing design-
based estimates under tests, like the Mantel-Haenszel trend, that involve a single
degree of freedom.362 The simulations applied the concept of design effect, deff, as a
factor by which a complex sample is converted to an effective sample size, i.e., an
srs sample size of comparable efficiency and power for the statistical computation
involved;150 a complex-design sample of 100 with deff= 2 under a given estimation
procedure has equivalent statistical efficiency of an srs sample of 50 under that
For the current study, design-based cross-tabulations with initially significant
Mantel-Haenszel trend tests were uniformly reduced by the maximum pseudo-
design effect,
pseudo deff = -esign , (3.4)
that retained significance. Where a tenfold reduction in sample size retained
statistical significance of the Mantel-Haenszel trend, for example, the probability of
chance,/?, is reported along withpseudodeff= 10. This approach, although not yet
methodically studied, is likely to be conservative, even overly so, because complex
statistics tend to have smaller deffs than simple summaries such as means;362 the
largest true deff observed in the current study was 9.
Analytic Methods, by Hypothesis
HI. Sociodemographic Factors. The general and participant health-center patient
populations have statistically indistinguishable current smoking prevalence
after adjusting for known sociodemographic factors.
Test: A designed variable representing health center cases has a non-
significant coefficient in a sequential logistic regression model of
current smoking, fitted with potential covariates of sex; white,
Spanish- or English-language Hispanic ethnicity; age; education;
marital status, and state.
For this comparison, data were limited to white and Hispanic observations due to

small numbers of other ethnic groups in the health center data set and highly
dissimilar non-white, non-Hispanic ethnic mixes across the health center and
reference population data sets.
General population data were extracted from the 1997 BRFSS data set for the six
states of PHS Region VIII (Colorado, Montana, North Dakota, South Dakota, Utah,
and Wyoming.) Each of five Region VIII states constituted a single stratum in the
BRFSS data set; Utah sampled disproportionately from strata corresponding to
sub-state regions in order to provide adequate samples for smaller geographically
defined populations of interest.
Because states use a variety of BRFSS protocols in racial coding of Hispanics,363
recoded variables were constructed to identify BRFSS respondents as non-Hispanic
white or Hispanic (any race), using the surveys two questions regarding ethnicity
(race and Spanish or Hispanic origin).
Final BRFSS weights, the product of selection probability and post-stratification,
sum to population estimates for each state364 and were rescaled to sum to BRFSS
subsample size. The health center and BRFSS data subsets were then merged.
Health center cases comprised 1.1 percent of the combined total.
Design-based prevalence rates of current smoking were compared across the two
populations matched on ethnicity, education level and marital status. Using design-
based sequential logistic regression, in which the researcher specifies the order of
entry of covariates,365 models of current smoking were tested using (in order of
entry) sex, ethnicity, language,* age, education, marital status**, state, and health
center case-origin. The sequence was based on speculation that it represents the
order of average importance of each covariate to smoking behavior. Covariates that
tested non-significant in sequence were retested alone and in other positions to
avoid basing the final model on speculated order of importance. The reference
population was white, married females with some college education, each character-
istic representing a majority or plurality of the respective covariate.
Sex and age were non-significant alone and in all multivariate entry positions.
Ethnicity was non-significant alone but significant when entered after education;
interaction effects disappeared after other covariates were entered. When Utah was
available for health center data only
excluding BRFSS cases where respondent indicated member of unmarried couple (=213,
4.1% of sample)

entered, only it and Montana remained significantly different from other states. The
final model included ethnicity, education, marital status, two states (Utah or
Montana vs. all others), and health center case origin.
To consider covariate influences in each population without the others influence,
similar models were fitted separately; in the health center data, ethnicity was
significant in univariate analysis, becoming marginally non-significant (p=0.084)
when language was added, and Wyoming alone was significant among states. Final
models included the three sociodemographic covariates, language in the health
center model, and respectively significant states.
Each data subset (health center vs. BRFSS) was additionally modeled using ethnic-
ity, marital status and education as covariates, and adjusted rates were estimated
(Stata adjprop366) for each covariate in each population, setting other covariates to
the specific populations mean values. An alternative adjustment method, using
reference-category values, produced similar relative rate differences across the two
populations but lower smoking prevalence rates overall. The choice of reference
categories was based on majority or plurality representation and for most compari-
sons was, again, non-Hispanic white, married females with some college education.
As a final comparison, current smoking prevalence was predicted for each popula-
tion based on the others demographic distribution, computed as the sum of the
products of prevalence rate in each education/marital status/ethnicity category and
the other populations proportion in the same category. The results are prevalence
rates for each population adjusted to the other populations demographic distribu-
H2. Health Concerns. Smokers in the participant health-center patient population
report relatively little concern about the health risks of smoking.
Test: Among current smokers, health risk avoidance (1) ranks in the
bottom half of estimated mean reasons to quit; (2) is classified on
average as not a reason to quit, or (3) is not the leading reason for
wanting to quit.
Both current smokers 100 cigarettes in lifetime, smoked within last 30 days) and
former smokers 100 cigarettes, did not smoke within last 30 days) were offered a
list of 10 potential reasons to quit (Table 3) and asked whether each item was
never a reason or not true for me, a reason but not a big one for me, or a big
reason for me. Responses were scored 0 (not a reason/not true) to 2 (big reason),

and each groups mean scores were estimated after excluding respondents who had
coded every reason a big reason (4.9% of current smokers, 3.7% of former
smokers). To determine whether mean responses were significantly different,
variables were designed for paired differences between health risk avoidance and
each other reason, and each difference was tested for significant difference from
zero under design-based estimation.
Table 3. Ever-user prompted reasons why I would like to stop or cut down
a. to avoid getting cancer, a heart attack, or another deadly disease.
b. because it has caused me health problems already or has made them worse.
c. because it makes your clothes, hair or breath smell bad, or it stains your teeth.
d. because it makes you short of breath when you work, exercise or play sports.
e. because it costs too much money.
f. to please my partner or spouse.
g. to please my family.
h. to set a good example for children.
i. to stop being addicted.
j. because I stopped enjoying it.
An open-ended question preceding these questions asked for the biggest reason
why I would like to quit, and 78 percent of current smokers wrote in one or more
reasons (responses by former smokers were not analyzed). Written responses were
coded into 10 specific categories derived from the responses themselves. Multiple
responses (up to three were present) were entered into separate variables in the order
given. Responses coded as health reasons included current ill-health complaints
(other than fatigue), wishes to avoid premature death, and smoking-attributed
morbidity or mortality of family members. Responses such as to feel better and
want more energy were coded in other categories.
H3: Readiness to Quit. Smokers in the participant health-center patient popula-
tion are less ready to quit than smokers in the general population.
Test: The distribution of health center smokers along the stages-of-change
model includes significantly greater proportions in precontemplation
or contemplation, and a significantly smaller proportion in prepara-
tion, than stage distributions reported for general populations.
Current smokers were staged with a three-question rubric (Table 4), and a designed
variable was constructed with values for precontemplation (precon), contemplation

(con) and preparation (prep). Design-based proportional distribution was estimated,
and results were compared with published distributions.95,367,368
Table 4. Staging algorithm for readiness to quit smoking cigarettes precon con prep
1. In the last six months, I stopped smoking for a day or longer. -s' m Vjc. t
2.1 might stop smoking in the next six months. no yes
3.1 am planning to stop smoking in the next 30 days. EZiTJ no yes
H4. Addiction. Smokers in the participant health-center patient population are
more addicted than smokers in the general population.
Tests: Timing and Frequency. Compared to other smoker populations,
health center current smokers are more likely to: 1) smoke within 30
minutes of waking, 2) smoke every day, and 3) smoke more ciga-
rettes per day (daily smokers only).
Cessation Attempts and Beliefs. Compared to other smoker popula-
tions, health center current smokers are less likely to 1) have quit for
at least 24 hours in the previous six months, 2) affirm that they are
able to quit smoking, and 3) report that they are addicted.
Cessation Rate. Compared to other smoker populations, health center
ever-smokers are less likely to be former smokers.
Design-based proportions were estimated (Stata svyprop) for current smokers
responses to each dichotomous test question. For Region VIII populations (health-
center vs. general), cigarettes-per-day responses were transformed (square-root) to
normal distribution; design-based proportions were estimated and grouped in half-
pack steps. Results were compared across Region VIII populations and against
published reports from other populations. Where possible, significance of differ-
ences between population rates was assessed using unpaired t-tests of means with
unequal variances.
Most questions were identical or reasonably comparable across surveys. Exceptions
were NHSDA questions regarding daily smoking and cigarettes per day, and the
time period for reporting recent quit attempts (12 months in comparison surveys, 6
months in the health center survey). Responses to non-comparable NHSDA ques-

tions were not used; the possible impact of differing quit-attempt questions is
addressed in Chapter 5. Significance of differences was not assessed where re-
sponses being compared were to non-identical questions.
H5. Desire to Quit. Smokers in the participant health-center patient population
dont wish to quit smoking.
Test: Fewer than 75 percent of health center current smokers say they want
to quit or cut down (in contrast to consistent findings since 1990 that
75 percent of smokers in the general population want to give up
Design-based binomial proportions were estimated for responses to the
statements I would like to stop or cut down ..., and I believe I can stop
smoking cigarettes. Outcomes were assessed using the 99% confidence
lower bound for estimated proportions answering yes. The potential impact
of self-efficacy (can quit) on desire (want to quit) was assessed using
design-based Pearson %1 2 3-
H6. Nonresponse Bias. The estimate of health center smoking prevalence is
biased upward because non-participating centers had lower prevalence.
Test: The predicted rate of smoking prevalence across Region VIII health
centers is not significantly different than the rate in the general
population, assuming the lowest reasonable rate of smoking preva-
lence in nonparticipant centers.
No information is available regarding clinic reasons to participate or abstain, nor
about smoking prevalence in abstaining clinics. A conservative set of assumptions
is that across all clinics:
1) clinicians accurately perceived their patient populations smoking preva-
2) clinicians viewed smoking prevalence above a critical but unknown level
(to be estimated) as a severe health issue meriting study and inter-
vention, and viewed prevalence below the same level as not warrant-
ing the effort that a study and intervention would entail;
3) smoking prevalence among patients older than 39 was not systematically
lower in abstaining clinics (which otherwise might confound clini-
cian perceptions of the issues severity in the age-eligible popula-

tion), and
4) other health issues were neither more prevalent nor more severe in ab-
staining clinics (which otherwise might make equal smoking preva-
lence appear relatively less urgent in abstaining clinics).
Under these assumptions, study abstention would have meant smoking prevalence
fell below the critical level, while participation meant prevalence exceeded the
This hypothesis is explored using sensitivity analysis, which seeks to bound an
unknown parameter under one or more assumptions of interest. The current analysis
seeks to estimate the level below which abstention would have been reasonable.
Observations from three Denver Health clinics (n=172) were excluded from the
analysis, as were corresponding Region VIII health center population data, because
the System was inadequately represented in the sample and differs from more rural
clinics in the Region. The remaining sample (n=1473) represents approximately
31,500 of 104,000 age-eligible patients seen in Region VIII health centers during
1997 (excluding Denver Health clinics). General population smoking prevalence
was estimated from the BRFSS 1997 survey results among age-comparable adults
(ages 18-39).
In order to propose possible critical levels, overall smoking prevalence in Region
VIII health centers (excluding the Denver system) is defined by
p = prevalence among participant-clinic patients,
n s proportion of all patients from participant clinics,
7rn = proportion of all patients from nonparticipant clinics, and
pn = prevalence to be estimated among nonparticipant-clinic patients.
The overall rate follows, then, from supposing an average smoking prevalence rate
in abstaining health centers, based on extrinsic information and judgment. The
lowest reasonable prevalence rate was judged to be the general population rate.
The predicted health center rate and the general population rate were tested for
nonequivalence using unpaired t-tests with unequal variances and one-sided 95%
confidence limits; sample sizes were reduced in proportion to design effects
(deff= 6.0 for the health center sample, effective n = 274; deff= 2.8 for the general

population sample, effective =T835).
H7. Sample Frame Bias. Health facility visitors exhibit higher smoking preva-
lence than the general population does, because smokers on average incur
more health-care needs than nonsmokers due to tobacco-related morbidity.
Test: On average, smokers visit health care facilities more often than non-
Smoking causes a variety of diseases, but this excessively well-established fact
leaves unanswered the question whether smokers visit health care facilities for
treatment more often than non-smokers (regardless of different disease incidence or
prevalence rates between the two groups). If smokers present as patients more
frequently on average than do non-smokers, their elevated facility attendance alone
might account for elevated smoking prevalence found in health center clinics
compared to corresponding general population rates.
The central question is whether smokers are more likely than nonsmokers to visit a
health facility, specifically a health center, and thus be more likely to be sampled.*
For this to occur, either or both of two possibilities must be present:
1. Smokers are more likely than nonsmokers to visit a health facility at
least once in a specified period. Then:
a. The proportion of smokers who have visited a health facility at least
once during the specified period is greater than the proportion of
nonsmokers who have done so (ti5 > uj.
b. Smoking prevalence rates differ between the group reporting no
health facility visits and the group reporting any health facility visits.
Letting n represent the proportion of nonsmokers reporting any
health facility visit, the table illustrates the situation:
The analysis inherently accounts for factors that might affect smoker visitation, such as
increased health problems due to smoking or less adequate insurance coverage, by
comparing the outcome such factors produce, namely the visitation rates themselves.

any visit(s) no visit
smoker 7t+ S 1 -(7T-S)
nonsmoker 7T 1 n
prevalence n +s 1-71 +S
rate 2n +s 2 2n + s
If smokers were more likely than nonsmokers
to visit health facilities at least once in time t.
The null hypothesis is
7U +S 1-71 +S
2n + s 2 2n + s
which is (approximately) true only if s is zero (or negligible).
Smokers visited a health facility on average more frequently than non-
smokers did. To compute the effect on sampling probability, a relatively
nonrestrictive assumption is helpful: Multiple visits are assumed to be
randomly distributed over the recruitment period, or alternatively, repeat
visits may arise in clusters that are assumed to be randomly distributed.
The selection probability pr for person i who had initial selection proba-
bility Pi and made r, visits during time t is then
o, o
Pi*~, r. = 1,2,3
Pi Al SC*
,t~ 1.
The mean selection probability for a given subgroup in a sample is then
E(Pr)=Y,*,P, O-T)
where is the proportion of the subgroup making r, visits during time t.
No publications were located discussing the issue of differential health-facility
visitation by smoking status. A secondary analysis was conducted on data from the
1997 NHIS,359 which includes a series of questions about whether respondents had a
usual place for sick care and for preventive care, the type of usual place(s), and
frequency of health care visits in the previous 12 months. A new variable was

designed to group respondents ages 18-39 according to the place in which they
obtained either sick care or preventive care, or both types of care. Respondents
reporting no type of place for either type of care were grouped as no usual place,
and those with missing values for both types of care, or who reported different
places for each type of care, were coded missing (Table 5).
Table 5. Usual Place of Care for sickness and advice, or prevention, or both types of care. Ages 18-39, U.S. 1997
n proportion s.e.
no usual place 2104 12.7% 0.3%
clinic or health center 3191 19.4% 0.5%
doctor's office or HMO 8750 59.2% 0.6%
hospital ER 241 1.6% 0.1%
hospital outpatient dept. 351 2.0% 0.2%
some other place 803 5.0% 0.2%
Visit frequency during the previous 12 months was recoded so that four or more
visits was the maximum level, since the current study enrollment period was
roughly three months or one-fourth the NHIS measured period.

Description of the Sample
The participant health-center sample contains 1645 complete observations, weight-
ed to represent the visiting patient populations of 25 Region VIII clinics. Fourteen
clinics and half of respondents were in Colorado and Utah (Table 6). Because
similar-sized subsamples were drawn from each clinic, several large Utah clinics
were undersampled, producing a large difference between weighted and unweighted
proportions from that state.
Table 6. Health center sample distribution, by state
State clinics n unweighted % weighted %
CO 8 485 29.5 29.7
UT 6 343 20.9 30.2
MT 4 324 19.7 15.6
SD 4 295 17.9 14.4
WY 2 118 7.2 5.5
ND 1 80 4.9 4.6
Total 25 1645 100 100
Although the sample is more female than male and more than one-fifth Hispanic,

females and Hispanics were undersampled and thus were up-weighted. Weighting
did not substantially affect other demographic distributions (Table 7).
Three-fourths of respondents reported being employed part-time or full-time, and
more than half reported having attended or graduated from college. Among His-
panic respondents (n=344), 80 percent were surveyed in Colorado and Utah and
more than half responded in Spanish; many or most Spanish-language respondents
were presumably migrants or recent immigrants, since in 1990, 96 percent of
Region VIII residents spoke only English,369 and in 1999, only 9 percent of the
western U.S. population was bom in Latin America.370
Table 7. Health center sample demographics
Variable unweighted (%) weighted (%)
female 58.6 64.0
ethnicity white, non-Hispanic 70.2 61.6
Hispanic 21.0 29.7
other/missing 8.9 9.0
Spanish language 11.9 17.4
education High school grad 34.5 33.0
>HS 50.4 51.0
marital sta- tus never married 30.6 28.3
divorced* 21.2 21.2
married 48.2 50.4
employment full-time 59.7 58.9
part-time 16.5 17.0
unemployed 23.8 24.2
father smoked 58.6 56.4
mother smoked 44.9 43.1
* includes small numbers of separated or widowed individuals.

Table 8. Region VIII demographic comparison, general population* vs. participant health centers (ages 18-39; design-based [weighted] percent)
Characteristic genl. pop. C/MHC
State: CO 47.8 29.7
UT 22.4 30.2
MT 9.1 15.6
SD 8.2 14.4
WY 5.1 5.5
ND 7.4 4.6
Age groups: 18-23 24.2 25.7
24-29 27.5 29.0
30-34 22.2 21.9
35-39 26.2 23.5
Sex:** female 49.6 64.0
male 50.4 36.0
Ethnicity:** non-Hispanic white 83.1 61.7
Hispanic 11.0 29.4
other 5.9 8.9
Education:** < H.S. graduate 8.5 15.9
H.S. graduate 29.3 33.0
some college 35.0 27.1
college graduate 27.3 24.0
Employment status: employed 77.26 75.84
unemployed 22.74 24.16
Marital Status:**11 never married 36.9 28.3
divorcedt 9.5 21.3
married 53.6 50.4
Children at home:** none 45.5 31.9
1 21.7 20.1
2 21.3 20.1
3 or more 11.6 27.9
* Source: Behavioral Risk Factor Surveillance System, 1997. Survey data, National Center
for Chronic Disease Prevention and Health Promotion, CDC, DHHS. April 1999.
** /><0.001 for distributional difference between populations
n BRFSS value member of unmarried couple omitted (4.2% of BRFSS total),
t includes small numbers of separated or widowed individuals

Compared to the same-age general population in Region VIII (Table 8), the popula-
tion of participant health-center clinics was:
less concentrated in Colorado but not significantly differently distributed
across the states;
slightly, non-significantly younger;
significantly more female (approximately two-thirds vs. half);
significantly more Hispanic and non-white;
significantly less educated (more than 60 percent of the general population
in this age range has attended or graduated from college, compared
to about 50 percent of the health center population);
employed at a similar rate;
more than twice as likely to be divorced;
living with significantly more children at home.
The divergent distributions of sex and ethnicity between the health-center and
general populations in Region VIII mirror those found nationally: Compared to the
U.S. general population, the U.S. health-center patient population is more Hispanic
(31.1% vs. 10.5%), more non-white (65.0% vs. 26.7%), and more female
(59.2%).337 National health-center data are unavailable regarding patients educa-
tion, marital status, employment status, or children at home. During FFY 1998,
when data for the current study were collected, 86 percent of patients seen in U.S.
health centers were at or below 200 percent of the federal poverty level.371
In tobacco use, the participant health-center population was significantly more
likely than the age-comparable Region VIII general population to ever have smoked
(52.9% vs. 38.6%, p<0.001), to currently smoke (40.0% vs. 24.6%, p<0.001), to
smoke every day (84.2% vs. 75.0%, p<0.001), and to smoke more per day (Table 9).
Former smokers comprised more than one-third of ever-smokers in the general
population, compared to slightly less than one-fourth in the health-center population
(36.1% vs. 24.4%, pO.0001).

Table 9. Region VIII tobacco use comparison, general population* vs. participant health centers (ages 18-39; design-based [weighted] percent)
Use Characteristic general C/MHC
ever smoked >100 cigs.
all 38.6 52.9**
men 42.0 57.5**
women 35.2 50.3**
currently smoke cigs.
all 24.6 40.0**
men 26.2 44.8**
women 22.9 37.3**
current smokers, smoke daily 75.0 84.2**
daily smokers, amount per day***
<.Vi pack 37.5 26.1
>1/4to1 pack 52.7 51.5
>1 to VA packs 5.6 13.7
>1Vzto2 packs 3.3 6.7
>2 (avg. 2 V*) packs 0.9 2.1
* Source: Data for respondents ages 18-39 from Colorado, Montana, North Dakota, South
Dakota, Utah, Wyoming, extracted from: Behavioral Risk Factor Surveillance
System (BRFSS), 1997. Survey data, National Center for Chronic Disease Preven-
tion and Health Promotion, CDC, DHHS. April 1999.
** p < 0.001, design-corrected Pearson x2
*** p < 0.0001, design-corrected Pearson x2; trend p < 0.05, pseudodeff=\0.
The participant health-center population had higher lifetime and current smoking
rates across every sociodemographic category (Table 10) except Hispanic ethnicity,
where a large proportion of Spanish-language health center respondents brought
rates below those for the general population. English-language Hispanic health
center respondents reported higher lifetime and current smoking prevalence than the
Hispanic general population, which may have included an unidentified but small
number of Spanish-language respondents (on the order of 0-10 percent based on
population proportions and survey protocols).

Table 10. Region VIII tobacco use by demographic characteristics,
general population" vs. participant health centers
(ages 18-39; design-based percentages)
Ever smoked cigarettes (more than 100 in lifetime) Currently smoke cigarettes
characteristic: genl. C/MHC characteristic: genl. C/MHC
category pop. category pop.
State: CO 42.4 54.5 State: CO 27.4 39.8*
UT 26.3 40.0 UT 15.4 26.3
MT 38.0 62.1* MT 21.7 47.2*
SD 45.9 51.0 SD 30.4 43.0
WY 44.6 87.7t WY 29.6 77.7*
ND 40.1 60.4* ND 28.1 52.7*
age group: 18-23 32.4 46.5** age group: 18-23 26.3 36.3*
24-29 38.0 49.3* 24-29 23.9 37.5**
30-34 36.7 54.2** 30-34 21.3 41.8*
35-39 46.7 63.0* 35-39 26.4 45.5*
sex: male 42.0 57.5* sex: male 26.2 44.8*
female 35.2 50.3** female 22.9 37.3***
ethnicity: 60.7t ethnicity: 47.0*
non-Hispanic white 38.9 non-Hispanic white 24.2
Hispanic, any lang." 36.4 33.8 Hispanic, any lang."" 23.6 22.8
Hispanic, Engl, survey"" n/a 50.7 Hispanic, Engl, survey"" n/a 36.5
other 47.5 62.0 other 32.6 48.7
education: education:
< H.S. graduate 53.6 61.2 < H.S. graduate 40.4 48.3
H.S. graduate 49.5 61.9*** H.S. graduate 35.4 49.7*
some college 38.0 50.2* some college 23.0 36.4**
college graduate 23.1 38.0*** college graduate 10.0 25.2*
employment status: employment status: 38.9*
employed 40.1 52.1** employed 25.2
unemployed 34.1 55.4t unemployed 22.7 43.6*
marital status: marital status:
never married 36.0 56.4t never married 26.3 44.7*
divorced""" 62.7 67.7 divorced""" 44.9 56.8*
married 36.6 44.7* married 20.7 30.3***
* p<0.05
** p<0.01
t /KO.OOOl
Source: BRFSS 1997 survey data.
BRFSS data do not identify Spanish inter-
includes small numbers of separated or
widowed individuals

The direction and magnitude of associations between smoking rates and covariate
categories was generally the same in both populations, with a few exceptions:
Wyoming health center respondents had extremely high rates of ever
smoking and current smoking;
health center current smoking rates increased with age, while the reverse
was true among general respondents except for the oldest group;
high school graduation was not associated with a decline in smoking
prevalence among health center respondents;
the slight association of current smoking and employment was direct for
health center respondents and inverse for the general population;
never-married health center respondents had higher smoking rates than
currently married respondents.
Among non-Hispanic white respondents, current smoking prevalence approached
half the population for both genders (47.6% for males [95% c.i. 38.9, 56.3], 46.5 for
females [39.0, 54.0]).
Hypothesis Test Results
HI. Sociodemographic Factors
The general and participant health-center patient populations have statisti-
cally indistinguishable current smoking prevalence after adjusting for
known sociodemographic factors.
Test: A designed variable representing health center cases has a non-
significant coefficient in a sequential logistic regression model of
current smoking fitted with potential covariates of sex; white,
Spanish- or English-language Hispanic ethnicity; age; education;
marital status, and state.
Among non-Hispanic whites, smoking prevalence was higher among health center
patients than in the general population at every combination of education level and
marital status (Table 11). The largest relative difference was among married college
graduates (n=877), where health center respondents were nearly three times as likely
as their general population counterparts to be current smokers (19.3% vs. 6.7%,
p<0.05). Roughly twofold differences in prevalence rates were found among
divorced college graduates (70.2% vs. 33.4%, p<0.0001; =155), divorcees with

some college (66.0% vs. 37.2%, p<0.001; =271), never-married people with some
college (42.7% vs. 21.5%, p<0.001, =609), and never-married high school gradu-
ates (60.2% vs. 32.6%, pO.OOOl, =504). The most similar rates were among
married people with some college (25.8% vs. 20.1%, n.s., n=1083) and divorcees
with less than high school education (83.8% vs. 79.5%, n.s., n=78).
Table 11. White non-Hispanic current smoking prevalence,
general population" vs. health center patients
with similar education and marital status
(design-adjusted estimates)
education marital population prevalence (%) s.e. n P<
less than high school graduation married general 37.0 8.4 104 0.01
health ctr. 64.3 5.9 58
never married general 49.6 7.8 98 n.s.
health ctr. 68.6 8.2 38
divorced general 79.5 9.9 42 n.s.
health ctr. 83.8 6.3 36
high school graduate married general 31.0 2.9 793 0.05
health ctr. 43.6 4.1 187
never married general 32.6 3.9 354 0.0001
health ctr. 60.2 2.3 150
divorced general 58.3 5.8 177 0.05
health ctr. 75.2 4.1 85
some col- lege married general 20.1 2.0 929 n.s.
health ctr. 25.8 4.9 154
never married general 21.5 3.1 480 0.001
health ctr. 42.7 5.2 129
divorced general 37.2 5.8 195 0.001
health ctr. 66.0 5.0 76
college graduate married general 6.7 1.4 730 0.05
health ctr. 19.3 5.5 147
never married general 15.0 3.1 354 n.s.
health ctr. 23.6 5.9 55
divorced general 33.4 7.1 112 0.0001
health ctr. 70.2 4.3 43
Source: BRFSS 1997 survey data.

Among Hispanics, smoking was not consistently associated with population
membership across combinations of education level and marital status (Table 12).
In the largest single category married high school graduates (n= 112) current
smoking prevalence was higher among health center patients than in the general
population (25.5% vs. 19.0%, n.s.); the only significant difference was higher health
center prevalence among married college graduates (24.1% vs. 0.0%, p<0.001).
Table 12. Hispanic current smoking prevalence,
general population0 vs. health center patients
with similar education and marital status
___________(design-adjusted estimates)__________
education marital population prevalence (%) s.e. n
less than high school graduation married general 24.2 9.6 29
health ctr. 22.5 4.3 49
never married general 32.6 18.2 11
health ctr. 15.2 9.5 20
divorced general 24.8 14.6 11
health ctr. 42.3 13.6 14
high school graduate married general 19.0 7.3 62
health ctr. 25.5 6.3 50
never married general 40.7 11.3 35
health ctr. 34.9 11.9 22
divorced general 57.1 16.4 26
health ctr. 30.8 5.7 26
some col- lege married general 15.1 7.8 37
health ctr. 12.9 7.0 34
never married general 16.1 7.2 28
health ctr. 27.0 15.5 11
divorced general 35.2 16.8 15
health ctr. 29.1 18.1 9
college graduate married general 0.0 0.0 30
health ctr. 24 -j *** 6.7 67
never married general 22.2 13.4 15
health ctr. 9.3 6.3 20
divorced general 7.6 8.5 5
health ctr. 12.5 7.7 22
Source: BRFSS 1997 survey data.
/?<0.001; other comparisons not significant (p>0.1)

In separate regression models for each population, sociodemographic factors
predicted similar odds ratios of current smoking (Tables 13a-b). When the two
populations were modeled together, the health-center population had more than a
twofold likelihood of smoking (adjusted odds ratio = 2.2, p<0.001) after adjustment
for education, ethnicity, marital status, and the states of Utah and Montana (Table
13c). The adjusted odds ratio (OR) for current smoking was 2.6 for people with less
than high school education and 3.3 for divorcees; the reference category was
married, non-Hispanic whites with some college. As discussed in Chapter 3, age,
sex, employment, number of co-resident children, and the four other states were not
significant covariates and were omitted from the model.
Table 13a: Estimated odds ratios (OR)* for current smoking among whites and Hispanics ages 18-39. Region VIII Participant Health Center Patients
education ethnicity marital status state
OR 2.5 1.8 0.8 0.3 1.4 3.0 3.4
95% c.i. 1.7-3.7 1.2-2.7 0.5-1.4 0.2-0.5 1.1-1.8 2.3-3.9 2.1- 5.5
P< .001 .01 .4 .001 .01 .001 .001
Table 13b: Estimated odds ratios (OR)* for current smoking among whites and Hispanics ages 18-39. Region VIII General Population
education ethnicity marital status state
OR 2.5 1.9 0.4 0.6 1.4 3.4 0.4 0.6
95% c.i. 1.6-4.0 1.5-2.5 0.3-0.6 0.4-0.9 1.1-1.8 2.4-4J 0.3-0.5 0.5-0.8
P< .001 .001 .001 .03 .02 .001 .001 .002
* adjusted for other factors in the table
** Reference population: married non-Hispanic whites with some college education,
surveyed outside state(s) indicated.

Table 13c: Estimated odds ratios (OR)* for current smoking among whites and Hispanics ages 18-39. Region VIII Combined (general & health center) Populations
education ethnicity marital status state Health
OR 2.6 1.9 0.5 0.5 1.4 3.3 0.5 0.7 2.2
95% c.i. 1.9-3.5 1.5-2.4 0.4-0.7 0.4-0.7 1.1-1.7 2.6-4.1 0.4-0.6 0.6-0.9 1.8-2.6
P< .001 .001 .001 .001 .002 .001 .001 .01 .001
* adjusted for other factors presented in the table
** Reference population: married non-Hispanic whites in the general population, with some
college education, surveyed outside Utah and Montana.
The participant health-center population had significantly higher predicted, adjusted
rates of current smoking than the general population in every modeled category
(Figure 4). The largest differences between population rates were among college
graduates (25.7% vs. 10.6%) and non-Hispanic whites (47.3% vs. 22.6%).
Each estimated prevalence rate changed little when based on each others demo-
graphic distribution. For the health center population, the adjusted rate was 36.9
percent compared to the estimated rate of 39.2 percent, a relative decline of less
than 6 percent. For the general population, the adjusted rate was 26.8 percent
compared to 23.7 percent estimated, a 13 percent relative increase.

Figure 4: Current smoking, general population vs. C/MHC
(predicted rates, adjusted for other modeled covariates)
Education T Marital Status T Ethnicity
"L i
3* i. __ i )-|i c3=l *
=3=* at i£j

general pop C/MHC general pop C/MHC general pop C/MHC
rate (95% c.i.), %
general pop. health center
1 didnt complete HS 44.6 (39.0, 50.4) 58.1 (51.7,64.3)
Education HS graduate 32.9 (30.5, 35.4) 51.1 (46.9, 55.3)
some college 22.1 (20.2, 24.2) 35.9 (31.5, 40.5)
college graduate 10.6 (9.0, 12.4) 25.7 (21.4, 30.5)
Marital Status | divorced 39.4 (35.4, 48.0) 62.0 (56.6, 67.2)
single 23.3(21.1,26.4) 43.1 (38.7, 47.6)
married 18.5(17.0, 19.8) 32.1 (28.8, 35.5)
Ethnicity white, non-Hispanic 22.6 (21.3, 23.9) 47.3 (44.4, 50.3)
|| Hispanic 19.1 ( 29.4 (25.4. 33.71
* reference category
The similarity of several categorical odds ratios across the health center and general
populations suggested an additional question: Do the odds ratios represent patterns
only within Region VIII, or do similar patterns appear nationally? To address this
question I analyzed BRFSS national data and data from the 1997 NHIS and
NHSDA surveys. Across the three national data sets, directions and relative magni-
tudes of predicted adjusted smoking rates were identical with Region VIII results for
categories of ethnicity (Figure 5), education (Figure 6), and marital status (Figure
7). Two data sets (NHIS and BRFSS) had statistically equivalent or nearly equiva-
lent smoking rates in every sociodemographic category except less than high school
education. In contrast, NHSDA rates diverged significantly from the other national

adjusted rates, with no consistent directional trend.
Smoking prevalence rates among the Region VIII general population were non-
significantly lower than or overlapping national (BRFSS and NHIS) rates except
among Hispanics. Region VIII health center smoking prevalence was higher,
however, than all other rates in every category.
Figure 5: Current smoking by ethnicity
national vs. Region VIII, adjusted for education and marital status
white non-Hispanic * Hispanic
:i :

I Region VIII I Region VIII
national national
rate (95% c.i.), %
non-Hisp. white* Hispanic
NHSDA 38 (37, 40) 26 (24, 27)
National NHIS 30 (29, 31) 15(14, 16)
BRFSS 28 (27, 28) 17(17. 18)
Region VIII Health Centers 47 (44, 50) 29 (25, 34)
BRFSS 23 (21.241 19 (15. 241
* reference population

Figure 6: Current smoking by education
national vs. Region VIII, adjusted for marital status and ethnicity
national national national national
ip l
S -r

Region VIII I Region VIII I Region VIII I Region VIII
rate (95% c.i.), %
< HS grad HS grad some college* college graduate
BRFSS 49 (48, 51) 35 (34, 36) 25 (24, 25) 12(12, 13)
National | NHSDA 48 (46, 50) 38 (36, 39) 30 (28, 31) 18(17, 20)
L NHIS 40 (38, 42) 35 (34, 37) 25 (23, 26) 11 (10,12)
Region VIII ^ Health Centers 58 (52, 64) 51 (47, 55) 36 (32, 41) 26(21,31)
BRFSS 45 (35. 501 33 (31.351 22 (20. 231 11 ( 9.121
* reference population

Figure 7: Current smoking by marital status
national vs. Region VIII, adjusted for education and ethnicity
national national national
married j never married divorced ^
i i i i i .!
1 -i j
i J 1 ^ b =
^ "F 1 i Tj i| 1 |
1 i' 1 i |
Region VIII I Region VIII I Region VIII
rate (95% c.i.), %
married* never married divorced
NHSDA 29 (28, 30) 37 (36, 38) 46 (43, 50)
National | NHIS 22 (21,23) 27 (26, 28) 38 (36, 40)
r BRFSS 21 (20, 21) 27 (26, 27) 38 (37, 40)
Region VIII ^ Health Centers 32 (29, 36) 43 (39, 48) 62 (57, 67)
BRFSS 19117. 201 23 121. 261 39 (35. 48)
* reference population
HI Results Summary. Non-Hispanic white patients in participating health
centers exhibited significantly higher current smoking prevalence than the general
population after applying multiple adjustment methods for differences in education
and marital status; other available sociodemographic covariates showed no effect on
rates. When applied to Hispanic respondents, the same analytic methods yielded no
clear trend between the populations. Results for non-Hispanic whites do not support
the hypothesis; the test for Hispanics is inconclusive.

H2. Health Concerns
Smokers in the participant health-center patient population report relatively
little concern about the health risks of smoking.
Test: Among current smokers, health risk avoidance (1) ranks in the
bottom half of estimated mean reasons to quit; (2) is classified on
average as not a reason to quit, or (3) is not the leading reason for
wanting to quit.
Current Smokers. Contrary to the first two parts of the hypothesis, current
smokers ranked health risk avoidance as the leading reason to stop or cut down
(Table 14). The mean of 1.62 on the 0-2 scale fell nearest the numerical rank that
signified a big reason, and was significantly higher than means for every other
reason except cost. The lowest-ranked reasons were primary social influences from
family and spouses.
Table 14. Current smokers, mean responses to reasons to stop or cut down
(0=not a reason, 1=reason, 2=big reason) mean
a. to avoid getting cancer, a heart attack, or another deadly disease. 1.62
e. because it costs too much money. 1.54
d. because it makes you short of breath at work, exercise or sports. 1.50
i. to stop being addicted. 1.50
c. because it makes clothes, hair or breath smell bad, or stains teeth. 1.41
h. to set a good example for children. 1.30
b. because it has caused or worsened health problems for me. 0.94
j. because I stopped enjoying it. 0.83
g. to please my family. 0.70
f. to please my partner or spouse. 0.43

Table 15. Main reasons for wanting to quit smoking, health center current smokers
reason 1'or only reason (n=548) s.e. 2nd reason (n=132) s.e. 3rd reason (n=20) s.e.
health 68.1 2.2 22.6 3.5 19.4 7.5
cost 7.3 17.3 4.3 8.5 9.5
children 6.7 lil 29.6 5.5 33.2 3.6
distaste 6.1 ..1.T 15.4 3.3 27.3 10.7
pregnancy 2.7 o.9;- 0.0 0.0
feel better 1.9 0.4 2.8 1.8
fatigue 1.2 0.4 4.4 1.8
social pressure 1.1 0.6 4.8 1.9 11.7 8.0
addiction 0.1 0.1 2.6 0.6
religious/spiritual 0.1 0.1 0.6 0.5
all other 4.7 1.1
More than two-thirds of current smokers volunteered health concerns as their main
reason for wanting to quit smoking (Table 15). Cost and concerns about children
(health or setting a proper example) were distant secondary reasons.
Former smokers. Listed reasons received rankings similar to those given by
current smokers (Table 16). Health risk avoidance had the highest mean score
(1.49), which was slightly but not significantly lower than the mean score from
current smokers. As with current smokers, primary social influences received the
lowest mean score, and the health risk avoidance ranking was significant against all
but the second-highest reason, unpleasant odor and stains, which former smokers
ranked higher than breathing problems and addiction.

Table 16. Former smokers, mean responses to reasons to stop or cut down
(0=nota reason, 1=reason, 2=big reason) mean
a. to avoid getting cancer, a heart attack, or another deadly disease. 1.49
c. because it makes clothes, hair or breath smell bad, or stains teeth. 1.41
d. because it makes you short of breath at work, exercise or sports. 1.22
i. to stop being addicted. 1.21
e. because it costs too much money. 1.19
h. to set a good example for children. 1.16
j. because I stopped enjoying it. 0.86
b. because it has caused or worsened health problems for me. 0.78
g. to please my family. 0.61
f. to please my partner or spouse. 0.54
H2 Results Summary. Current smokers gave health risk avoidance the highest
ranking among reasons to quit smoking, and more than two-thirds volunteered
health as their main reason for wanting to quit. Results do not support the hypothe-
H3. Readiness to Quit
Smokers in the participant health-center patient population are less ready to
quit than smokers in the general population.
Test: The distribution of health center smokers along the stages-of-change
model includes significantly greater proportions in precontemplation
or contemplation, and a significantly smaller proportion in prepara-
tion, than stage distributions reported for general populations.
Slightly more than one-fourth of participant health-center smokers reported prepar-
ing to quit smoking (25.7%), and nearly half reported seriously thinking about
quitting (45.6%); the others (28.6%) reported not thinking about quitting (Table 17).
Greater proportions were in contemplation or preparation to quit, and a smaller
proportion in precontemplation, than had been found roughly six to eight years

earlier among generally similar-age respondents to large U.S. survey studies of work
sites, individual states, and a health maintenance organization (HMO). The propor-
tion of smokers from the current study who were in precontemplation was similar to
the proportion reported from another health-center patient population, adults seen in
a southeastern U.S. community health center serving a largely African American
Table 17. Smoker readiness to quit, health-center and other populations (design-based percent) precon con prep
Health Region VIII sample, ages 18-39, 1997-98 (n=700) 28.6 (24.0, 33.7) 45.7 (41.3,50.1) 25.7 (20.5,31.7)
Centers Lincoln (NC) Community Health Center, ages 18+, ca. 1995 (n=379)36? 30.3 (25.7,34.9) 69.7 (65.1,74.3)
114 work sites, ages 25-44, ca. 1990 (n=2,887)95 40.1 (38.3,41.9) 38.7 (36.9,40.5) 21.3 (19.8,22.8)
Other California general population, ages 25-44, 1990-91 (n=4,877)95 32.8 (31.4,34.2) 50.6 (49.1,52.1) 16.6 (15.5,17.7)
Sites Rhode Island general population ages 25-44, ca. 1990 (n=2,120)95 43.0 (40.9,45.1) 40.4 (38.3,42.5) 16.6 (15.0,18.2)
U.S. HMO members ages 18+, 1991 (n=4,748)368 42.4 (not avail.) 38.0 (not avail.) 19.6 (not avail.)
H3 Results Summary. A smaller proportion of health center smokers were
precontemplators, and a larger proportion preparers, than corresponding proportions
of general, work-site and HMO smokers. Results do not support the hypothesis.
H4. Addiction
Smokers in the participant health-center patient population are more
addicted than smokers in the general population.
Tests: Timins and Frequency. Compared to other smoker populations,
health center current smokers are more likely to: 1) smoke within 30
minutes of waking, 2) smoke every day, and 3) smoke more cigarettes
per day (daily smokers only).
Cessation Attempts and Beliefs. Compared to other smoker popula-
tions, health center current smokers are less likely to 1) have quit for

at least 24 hours in the previous six months, 2) affirm that they are
able to quit smoking, and 3) report that they are addicted.
Cessation Prevalence. Compared to other smoker populations,
health center ever-smokers are less likely to be former smokers.
Timing and Frequency. Within Region VIII, health-center smokers were
more likely to smoke daily than were smokers in the general population (88.5% vs.
75.0%, p<0.0001). Health center smokers also smoked more cigarettes per day
(mean 15.8 vs. 11.7, p<0.0001), and a greater proportion smoked more than 20
cigarettes per day (21.6% vs. 9.7%, p<0.0001) (Table 18).
Nearly three-fourths of health center smokers reported smoking within 30 minutes
of waking, compared to roughly half of national and California smokers (72.7% vs.
47.3% and 48.2%, respectively; p<0.0001). Health center smokers were also
significantly more likely than national and California smokers to be daily smokers
Table 18. Tir current sm ning and Fr okers ages % (95% smoke within 30 minutes of waking equency of 18-39, by pc ci) smoke every day Smok jpulat s/2 ing, ion pact >'A to 1 ;s per >1 to r/2 lay* >114 to 2 >2
Region VIII Health centers (n=700) 72.8 (68.1,77.6) 88.5 (84.3,92.6) 24.9 53.5 14.2 6.0 1.4
General pop. (n=1289) n/a 75.0 (70.7,79.3) 37.5 52.7 5.6 3.3 0.9
National general population BRFSS (n=14442) n/a 77.6 (76.5,78.8) 35.8 49.1 9.7 4.4 1.0
NHIS (n=4405) n/a 70 1 (77.8,80.7) 33.8 49.5 11.3 4.5 1.0
NHSDA (n=4060) 47.3 (44.4,50.2)
California general pop. (n=3323) 48.2 (46.0,50.3) 61.7 (59.4,64.0) 42.2 47.1 7.3 2.8 0.7
* daily smokers only
** question worded differently, data not used

Cessation Attempts and Beliefs. In most comparisons to other populations,
health center smokers were less likely to report a serious quit attempt in the previ-
ous six months than other smokers reported for the previous 12 months (Table 19).
This was true in Region VIII, one of two national surveys, and the California survey
(45.9% vs. 54.2%, 56.5%, and 53.4%, respectively). There was no difference
between the health center rate and the other national survey rate (45.9% vs. 46.5%).
The health center proportion reporting that addiction was a big reason for their
smoking was similar to the California proportion saying they believed they were
addicted (67.3% vs. 70.1%), and a greater proportion of health center smokers
(85.1%) said addiction was either a big reason or a reason for their smoking. No
comparable question was asked in national surveys.
A slightly greater proportion of health center smokers said they believed they could
quit smoking than the proportion of California smokers who said they were very
sure or somewhat sure they could quit for a month (76.9% vs. 68.0%).
Table 19. current sr Cessation At nokers ages % (95% smoke due big reason tempts and B 8-39, by popi s-i.) to addiction reason eliefs, jlation quit for ;> 24 hr in last year* can quit smoking
Health center respondents 67.3 17.8 AtL Q** (40.7, 51.2) 76.9
Region VIII general pop. n/a n/a 54.2 (49.1, 59.4) n/a
National gen- eral popula- tion BRFSS n/a n/a 56.5 (55.1, 57.9) n/a
NHIS n/a n/a 46.5 (44.7, 48.4) n/a
California general pop. 70.1* 53.4 (51.3, 55.6) 68.0*
daily smokers only
** in last 6 months
t question: I believe I am addicted.
} percent very sure or somewhat sure can quit for at least one month
Cessation Prevalence. More than one of three ever-smokers in the general
population was a former smoker, compared to roughly one of four in the health

center patient population (36.0% vs. 24.3%, p<0.0001); the relative difference
represents a 48 percent greater prevalence of former smoking in the general popula-
tion (Table 20). Former smoking prevalence increased with age in the general
population (trend p<0.02,pseudodeff=\Q), but not in the health center population.
Table 20. Prevalence of former smoking
by age group,
general vs. health center populations
health ctr.
24.4 %
H4 Results Summary. Compared to other populations, health center smokers
reported higher prevalence of addiction-related behaviors (daily smoking, daily
cigarette consumption, heavy smoking, and smoking upon waking), similar preva-
lence of perceived addiction, and lower prevalence of successful cessation. These
results individually and collectively support the hypothesis; quit-attempt compari-
sons produced mixed results.
H5. Desire to Quit
Smokers in the participant health-center patient population dont wish to
quit smoking.
Test: Fewer than 75 percent of health center current smokers say they
want to quit or cut down (in contrast to consistent findings since
199094 that 75 percent of smokers in the general population want to
give up smoking.).
The proportion of health center smokers affirming a desire to quit or cut down was
80.9 percent, with a 99%-confidence lower bound of 74.4 percent. Respondents
belief or disbelief in ability to quit made no difference in expressed desire to quit or
cut down (Table 21); the proportion affirming that they could quit smoking was
77.2 percent, with a 99%-confidence lower bound of 72.8 percent.

Table 21. Self-efficacy
and desire to quit smoking
(design-based estimates)
I want to quit
or cut down
yes 410 96 506
81.0% 19.0% 100%
23.5% 22.9% 23.1%
I believe
I can quit
no 122 30 152
80.4% 19.6% 100%
76.5% 77.1% 76.9%
532 126 658
80.9% 19.1%
100% 100%
H5 Results Summary. Health center smokers express desires to quit,
independent of their belief in their ability to quit, in the same proportion as U.S.
smokers have expressed during the last decade. Results do not support the hypothe-
H6. Nonresponse Bias
The estimate of health center smoking prevalence is biased upward because
non-participating centers had lower prevalence.
Test: The predicted rate of smoking prevalence across Region VIII health
centers is not significantly different than the rate in the general
population, assuming the lowest reasonable rate of smoking preva-
lence in nonparticipant centers.
The known parameters of Equation 3.5 are smoking prevalence among participant-
clinic patients (pp = 40.2%); the proportion of patients from participant clinics (7tp =
30.3%), and the proportion of patients from nonparticipant clinics(7rn = 69.7%). The
relationship of nonparticipant prevalence (/>) to overall health center prevalence
(pa) is then approximately
Settingp at the general population rate of 24.6 percent, overall smoking prevalence
in Region VIII health centers would be 29.2 percent, not significantly different from
the true general population rate (p=0.06). The smallest p generating significantly
different population and health-center prevalence rates is 25.0 percent, when overall
health center smoking prevalence would be 29.5 percent, a relative gap of 20

percent above the general population smoking rate (p<0.05).
H6 Results Summary. Supposing smoking prevalence in nonparticipant
health centers was no higher than the general population rate, the smoking rate
across all Region VIII health centers was not significantly different from the rate in
the general population. Results support the hypothesis.
H7. Sample Frame Bias
Health facility visitors exhibit higher smoking prevalence than the general
population does, because smokers on average incur more health-care needs
than nonsmokers due to tobacco-related morbidity.
Test: On average, smokers visit health care facilities more often than non-
In secondary analysis of NHIS 1997 data, smokers were significantly less likely than
nonsmokers to have attended a health facility at least once in the previous 12
months (27.0% vs. 22.9%, p<0.0001). This result held, but was not significant, for
each listed type of care facility (Table 22), including clinics and health centers
(22.5% vs. 20.4%), doctors offices and HMOs (18.1% vs. 16.5%), hospital
emergency rooms (54.7% vs. 44.8%), and hospital outpatient departments (26.6%
vs. 23.2%). Among respondents reporting no usual care place or naming an unlisted
type of care place, smokers were slightly more likely than nonsmokers to have
visited at least once in the 12 months.
Table 22. Percent without a health facility visit
in the previous 12 months,
by current smoking status and usual care place
smokers nonsmokers
(n=3768) (n=11551)
all respondents 27.0* 22.9*
no usual care place 57.6 60.7
clinic/health center 22.5 20.4
doctors office, HMO 18.1 16.5
hospital ER 54.7 44.8
hospital output, dept. 26.6 23.2
some other place 18.2 21.9
* /X0.0001

Among smokers who did use a health facility in the previous 12 months, smokers
were slightly but significantly less frequent visitors than nonsmokers (mean 2.05
visits vs. 1.96 visits, p<0.01). Smokers were slightly, nonsignificantly more frequent
visitors than nonsmokers in the group reporting regular care at clinics or health
centers (mean 2.14 visits vs. 2.11 visits). The pattern of relative frequency in other
facility types was mixed (Table 23).
Table 23. Mean health facility visits in 12 months, by smoking status
nonsmokers smokers
all respondents 2.05* 1.96*
no usual care place 0.84 0.88
clinic/health center 2.11 2.14
doctors office, HMO 2.26 2.26
hospital ER 1.27 1.11
hospital output, dept. 2.22 1.97
some other place 2.05 2.28
When smoking prevalence was estimated by visitation status, the proportion of
smokers was significantly higher among respondents who had not visited a health
facility in the previous 12 months than among those who had visited once or more
times (32.2% vs. 27.6%, pO.OOOl). The same rate-relationship was found for each
type of facility listed, while the reverse relationship was found among respondents
with no usual care place or a care place other than those listed (Table 24). Among
those reporting clinics and health centers as usual care places, smoking prevalence
was 32.3 percent among nonvisitors and 29.6 percent among visitors.