AN EXAMINATION OF THE RELATIONSHIP BETWEEN
DEPTH OF INTERVENTION AND GAMMA CHANGE
by
Gary Robert Frank
B.A., Ohio University, 1973
M.P.A., University of Colorado, 1976
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado at Denver
in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Public Administration
1993
1993 by Gary Robert Frank
All rights reserved.
This thesis for the Doctor of Philosophy
degree by
Gary Robert Frank
has been approved for the
Graduate School of Public Affairs
by
Linda de Leon
Date
Frank, Gary Robert (Ph.D., Public Administration)
An Examination of the Relationship Between Depth of Intervention and
Gamma Change
Thesis directed by Professor Mark L. McConkie.
Abstract
This study explored the concept of plural change resulting from
Organization Development (OD) interventions. In particular, the study
examined the relationship between type of organizational intervention
and type of change generated. Interventions were differentiated on
the dimension of depth of emotional involvement. Type of change was
conceptualized by the alpha, beta, gamma change typology. These
designations correspond to change in condition (alpha), change in
metric (beta), and change in state (gamma).
The study examined time series survey data for two sub
populations of an organization that was engaged in a longterm OD
effort. Respondents were identified by their participation in two
different, systematic interventions. One group participated in survey
feedback activities only; the other participated in confrontation team
building meetings in addition to survey feedback. These interventions
were differentiated on the dimension of depth of emotional
involvement.
iv
A conceptual model relating depth of intervention to type of
change was developed. Four hypotheses concerning this relationship
were presented. Broadly, interventions shallow on the dimension of
emotional involvement like survey feedback will not generate gamma
change; interventions deep in emotional involvement like confrontation
team building will.
The survey data for these two groups were analyzed using a
comparison of factorial patterns. A 50% decrease in the common
variance between two factorial patterns defines gamma change. The
results support three of the four hypotheses. Specifically, the survey
feedback intervention showed a modest decrease in common variance
between factor patterns. Thus, a shallow intervention like survey
feedback does not evoke gamma change. Although no gamma
change was found for the team building intervention, a 32% decrease
in common variance supports the notion of a relationship between
depth of intervention and type of change.
Discussion of the results suggested further exploration of the
relationship between type intervention and type of change along other
distinguishing dimensions in addition to emotional depth. The
discussion also challenged the present operational definition of
gamma change and recommended a rigorous qualitative assessment
to complement quantitative assessment.
v
This abstract accurately
thesis. I recommend its
represents the content of the candidate's
publication,
Signed
Mark L. McConkie
To Barb
ACKNOWLEDGMENTS
There are several people whose assistance, support,
encouragement, and generosity I wish to acknowledge. This thesis
has been a long time in the making and its completion is a direct result
of their help.
I wish to acknowledge my family whose love and
encouragement kept me moving forward when the size of the task
overwhelmed me. There are innumerable evenings and weekends of
family time which were foregone understanding^ so that I might
become "Dr. Daddy." I especially wish to acknowledge my wife,
Barbara, to whom the thesis is dedicated. She has been long
suffering and unconditionally supportive beyond anything I believe
was implied in our marriage vows.
I wish also to acknowledge the members of my thesis
committee for their support and encouragement. I thank Mark
McConkie for his steadfast help throughout the effort. I appreciate his
willingness to stick with me through uneven progress. His cheerful
voice and ever positive regard were a motivating influence when the
struggle between writing the thesis and life in general resulted in
lapses in contact and doubt. I thank Linda de Leon for her willing
viii
support and especially for her sincere encouragement about the
quality of my work which caused me to switch tracks and pursue the
Ph.D. I thank Catherine Crawford for her willingness, as a new faculty
member, to take a chance on a student she did not know and for her
ready help in guiding the thesis to completion. I thank Tom lsgar for
his constant encouragement and support and critical feedback on the
manuscript that was instrumental in shaping and improving the quality
of the final product. I thank Wayne Boss who long ago recognized a
potential in me which he encouraged, nurtured, and supported to the
completion of this achievement.
Finally, I want to acknowledge two people without whose
assistance the idea for this research and its completion might never
have come about. I thank Kit Tennis for introducing me to the change
typology, helping me understand its potential as a research topic, and
for providing me with a basic literature to get started. And I thank
Keith Billingsley for the gracious support of his time, his interest in my
research, and his vitally important technical assistance and advice as
a fellow inquirer without which this thesis would be no more than a set
of untested hypotheses.
ix
CONTENTS
CHAPTER
1. INTRODUCTION, PURPOSE, AND ORGANIZATION OF
THE STUDY............................................ 1
Background........................................ 1
The Point of Departure............................ 3
Implications for OD and Response of the Academic
Community......................................... 5
Purpose and Significance of the Present Study..... 7
Outline of Thesis................................. 11
2. SURVEY OF LITERATURE................................. 13
Introduction...................................... 13
Alpha Change................................... 14
Beta Change.................................... 14
Gamma Change................................... 16
Discovering Gamma Change....................... 17
Conclusion..................................... 22
Summary and Structure for the Remainder of the
Chapter........................................ 24
Application of the Original Statistical Technique. 25
Randolph and Yanouzas (1977)................... 26
x
Randolph and Edwards (1978).................. 27
Randolph (1982)................................ 29
Summary........................................ 34
Attack on the Factorial Technique............... 35
Challenging the Change Typology................ 35
Golembiewski and Billingsley's Response...... 37
The Counter Response........................... 38
Summary........................................ 40
Alternative Approaches to Assessing Alpha, Beta, and
Gamma Change.................................... 41
Six Approaches That Employ Research Design.. 42
Summary of Research Design Approaches........ 65
Two Approaches That Employ Statistical Methods 69
Summary of Statistical Methods................. 76
Diverse Operational Approaches.................... 79
Com parison of T echn iq ues................... 79
Laboratory Studies and Simulations............. 81
Surrogate Variables............................ 83
Academic Response.............................. 87
General References............................. 88
xi
Summary........................................... 89
Unanswered Questions........................... 90
3. HYPOTHESES......................................... 93
Introduction...................................... 93
Interventions Used in the Change Effort........... 95
Overview of Intervention Activities............ 95
Description of Interventions................... 98
Designation of Primary Interventions.......... 102
Review of Primary Interventions................. 104
Survey Feedback............................... 104
Team Building Meetings........................ 110
Conceptual Framework for Relating Intervention to Type
of Change........................................ 116
Depth of Intervention......................... 116
A Conceptual Framework....................... 118
Hypotheses to Be Tested.......................... 119
Hypothesis 1.................................. 119
Hypothesis 2.................................. 120
Hypothesis 3.................................. 121
Hypothesis 4.................................. 122
xii
Summary.......................................... 123
4. METHODOLOGY........................................ 125
Introduction..................................... 125
Description of Subject Organization.............. 125
Description of Research Design................... 129
Research Design............................... 130
Interventions................................. 133
Method and Procedure for Creating Intervention and
Comparison Groups............................. 136
Testing Intervention and Comparison Group
Equivalence................................... 138
Research Instrument............................ 146
Description of the Likert Profile of Organizational
Characteristics............................... 146
How and Where the POC Has Been Used........... 150
Reliability and Validity of the POC........... 153
Operational Definition of Gamma Change and Data
Analysis Procedures............................. 156
Ahmavaara's T ransformation T echnique....... 159
Data Analysis Procedures...................... 166
Summary.......................................... 170
xiii
5. RESULTS.......................................... 171
Introduction.................................. 171
Results of Analysis........................... 171
Assessment of Gamma Change................. 172
Assessment of Alpha Change.................. 179
Results of Hypotheses Tested................... 187
Hypothesis 1................................ 187
Hypothesis 2................................ 190
Hypothesis 3............................... 191
Hypothesis 4.............................. 192
Summary........................................ 193
6. CONCLUSIONS....................................... 195
Introduction................................... 195
Conclusions and Implications Resulting from the
Research Question and Hypothesis Testing...... 196
A Brief Contextual Review................... 196
Interpreting Factorial Incongruence: How Much is
Enough to Say Yes?.......................... 198
Challenging the Operational Definition of Gamma
Change: Implications for Future Research... 205
Implications for Extending Quantitative
Assessment.................................. 210
xiv
Implications for Expanding the Definition of Gamma
Change.................................... 215
Unanswered Questions......................... 219
Question One.............................. 220
Question Two.............................. 222
Question Three............................ 224
Question Four............................. 225
An Unanticipated Consequence of Analysis...... 226
Summary...................................... 229
APPENDIXES
A. Likert Profile of Organizational Characteristics. 236
B. Comparison Matrix C............................. 237
C. Factor Comparison Solutions Using Principal Components
Unrotated Patterns and Varimax Rotated Patterns.. 242
REFERENCES........;...................................... 243
xv
CHAPTER 1
INTRODUCTION, PURPOSE, AND ORGANIZATION
OF THE STUDY
Background
Organization Development (OD) has been criticized over the
years for a variety of shortcomings. It has been criticized for
developing practice to the near exclusion of theory (Bennis 1969;
Kahn 1974; Alderfer 1977; French and Bell 1980). In fact, at least one
critic has labeled Organization Development atheoretical (Levinson
1972). Some have been critical of OD's overreliance on a single
method of intervention (Bennis 1969; Levinson 1972; Porras and Berg
1978). Others have been critical of OD's inattention to power and
politics in organizations (Bennis 1969; French and Bell 1980). OD has
been criticized for its artificial separation of structure from process
(Kahn 1974). It has also been criticized for its lack of impact on hard
measures of organizational success and performance (Friedlander
and Brown 1974; Kimberly and Nielsen1975; Nicholas 1982).
However, perhaps the greatest number of and sharpest criticisms have
focused on evaluation research.
OD has been criticized in general for its lack of rigor in research
methodology (Kahn 1974; White and Mitchell 1976; Porras and Berg
1978). More specifically, OD has been criticized for its inattention to
hypothesis testing and theory building (Alderfer 1977; Lundberg 1978;
French and Bell 1980); lack of definition, operationalization, and
differentiation of independent and dependent variables (White and
Mitchell 1976; Pate, Nielsen, and Bacon 1977; Nicholas and Katz
1985); the use of "soft" versus "hard" criteria, e.g. self reports
(Friedlander and Brown 1974; White and Mitchell 1976; Porras and
Berg 1978; Roberts and Porras 1982; and Nicholas and Katz 1985);
overreliance on case study and nonexperimental or weak research
designs (Kahn 1974; Friedlander and Brown 1974; Porras and Berg
1978, Nicholas and Katz1985); the brevity of intervention activities
(Porras and Berg 1978); overemphasis on the individual or the group
as the unit of analysis (Kahn 1974; White and Mitchell 1976; Porras
and Berg 1978); and insider evaluation (Alderfer 1977; Porras and
Patterson 1979; Huse 1980; Terpstra 1981). It is to this lamentable
state of affairs in OD evaluation research that the present dissertation
is addressed.
Armenakis, Bedeian, and Pond (1983) suggest that OD
evaluation research has evolved through three distinct but overlapping
phases. They are 1) identification of general evaluation problems and
development of evaluation guidelines, 2) demonstration of methods to
deal with commonly encountered evaluation problems, and 3)
resolution of specific methodological problems. The criticisms cited
2
above generally relate to Phase I. OD literature that might be
categorized as Phase II is really a response to these criticisms. Phase
III is a qualitative departure from the earlier phases. While Phases I
and II were concerned more with macrolevel research issues, e.g.
inadequate theoretical models, vaguely defined objectives, weak
research designs, inadequate or inconsistent definitions of
independent and dependent variables, difficulty in using control
groups, etc., Phase III has a microlevel focus. That is, Phase III is
focused on more specific research and methodological questions that
were ignored by the broader focus on macro level issues (Armenakis
et al. 1983, p. 325.) The work of Golembiewski, Billingsley, and
Yeager (1976a) on the change typology is cited as the point of
departure for Phase III. There is substantial agreement on this
milestone (Armenakis and Zmud 1979; Lindell and Drexler 1979;
Roberts and Porras 1982; Randolph and Elloy 1987).
The Point Of Departure
Golembiewski, Billingsley, and Yeager (1976a) contributed
significantly to the conceptual advancement of evaluating change
(Roberts and Porras 1982). They proposed a threetiered model of
change based on their experience of assisting with the installation of a
flexitime work system. In examining quantitative time series data and
comparing the results to respondents' experiences reported in
interviews, conclusions about the success of the intervention did not
3
match. This paradox caused the researchers to reexamine their data
and ultimately propose a threetiered model of change.
Golembiewski et al. (1976a) suggested that actually three kinds
of change can occur as a result of deliberate OD interventions. They,
are alpha, beta, and gamma. Alpha change was defined as first order
change (Watzlawick, Weakland, and Fisch 1974) dealing with
observed, "true" change along a stable conceptual domain. Beta
change was defined as scale re calibration also along a stable
conceptual domain. It does not refer to a defect of the measuring
instrument or measurement error (Upshaw 1962; Cronbach and Furby
1970). Gamma change was defined as second order change
(Watzlawick et al. 1974) dealing with a fundamental change in
conceptual domain or concept redefinition.
Golembiewski et al. (1976a) focused only on gamma change
based on the belief thatu... gamma changes are the prime intended
consequences of OD interventions" (p. 139). To do so, they required
an analytic technique which could assess the kind of instability of
parameters or dimensions associated with gamma change. For this
purpose, they used factor analysis and the comparison of factor
structures as the method for assessing gamma change. They found
evidence of gammalike change in their time series data set of self
report measures based on a rigorous comparison of factor structures.
They concluded their article by suggesting five issues and cautions for
4
OD evaluation research.
First, the assumption that alpha change is the only relevant
measure of change seems inappropriate where successful OD
interventions are concerned. Second, the existence of different types
of change makes the clear interpretation of OD intervention results
very difficult despite rigor and care in research design. Third, the
existence of types of change highlights the indispensability of time
series designs in OD research. Fourth, the analysis implies the value
of statistical methods for discovering more about the nature of change.
Fifth, the use of selfreport measures as a mainstay of OD research
may be problematic.
Implications for OD and Response of the Academic Community
Explication of the change typology had profound implications
for the OD world in particular and the evaluation of change in general.
If the change typology were conceptually correct, then evaluations of
previous change efforts were at best incomplete and at worst invalid.
Golembiewski et al. (1976a) were essentially arguing the fallacy of
alphaonly measurement. Further, they were suggesting that without
knowledge of which type of change is occurring, the clear
interpretation of OD evaluation research, no matter how rigorous and
careful the design, was suspect. In a later, major work on evaluation
research (Seashore, Lawler, Mirvis, and Cammann 1983), several
5
contributors (Seashore; Lawler, Nadler, and Mirvis; Macy and
Peterson) concur, noting that testing for type of change is necessary
to the proper evaluation of change efforts.
Responses to the typology were curious and mixed. There
were attempts at replication using the factor analytic approach
(Randolph and Yanouzas 1977; Randolph and Edwards 1978). There
were attacks on the factor comparison technique as the
operationalization for gamma change (Lindell and Drexler 1979,
1980). There was acknowledgment of the typology as a potential rival
hypothesis for unexplained results but no attempt to test for it (Gavin
and McPhail 1978; Porras and Wilkins 1980; Golembiewski 1982;
Gavin and Montgomery 1983). For the most part, however, the
literature has accepted of the change typology without challenge. In
fact, the bulk of the literature comes from a small band of contributors
who accept the concept of type of change and have worked
industriously to discover alternative methods for detecting and
evaluating beta and gamma changes (Armenakis and Smith 1978;
Zmud and Armenakis 1978; Armenakis and Zmud 1979; Terborg,
Howard, and Maxwell 1980; Bedeian, Armenakis, and Gibson 1980;
Schmitt 1982; Van de Vliert, Huismans, and Stok 1985). Much of the
remainder of the literature has been concerned with further
investigation or comparison of alternative operationalizations of beta
and gamma change (Armenakis, Randolph, and Bedeian 1982;
6
Schmitt, Pulakos, and Lieblein 1984; Terborg 1984; Buckley and
Armenakis 1986; Armenakis, Buckley, and Bedeian 1986; Randolph
and Elloy1987).
The change typology literature is indicative of OD evaluation
research in Phase III (Armenakis, Bedeian, and Pond 1983). It
represents the struggle to rethink the concept of change and its
operationalization. The change typology is still something of an
anomaly in the OD research literature. However, it has established
respectability as a legitimate research interest.
Purpose and Significance of the Present Study
As noted above, a major thrust of the change typology literature
has been the identification of alternative methods for assessing beta
and gamma change. For example, there are six contributions which
detail methods that use research design as a means to assess or
control for beta and gamma change (Armenakis and Smith 1978;
Zmud and Armenakis 1978; Armenakis and Zmud 1979; Bedeian,
Armenakis, and Gibson 1980; Terborg, Howard, and Maxwell 1980;
and Van de Vliert, Huismans, and Stok 1985). There are two
contributions which detail methods that use statistical methods to
assess beta and gamma change (Armenakis, Feild, and Wilmoth
1977; Schmitt 1982).
The literature, which will be reviewed in full in Chapter 2,
7
suggests an important observation which gives purpose to the present
research. The contributions noted above and others which applied
these alternative methods have a onedimensional perspective. They
are concerned only with the effect of the intervention. Is beta and/or.
gamma change present or not? Although this is an important question
to answer and an important first step in this line of inquiry, there is an
additional perspective of inquiry of equal or greater importance for OD
research and practice. Given that type of change such as beta or
gamma can be assessed, is there a relationship between type of
change and the intervention that "caused" it?
The present research proposes to examine this second
perspective. Is there a relationship between type of intervention and
type of change? Do different interventions result in different types of
change?
The research will examine time series survey data for two sub
populations of an organization that was engaged in a longterm OD
effort. The data for this research are from a large community hospital.
Complete quantitative observations of the hospital population exist for
1985 and 1987. Various interventions occurred between these
observations as a result of survey feedback.
Demographic data permit the identification of respondents in
two ways. First, respondents who participated in both observations
may be identified. Second, these respondents can be divided into two
8
groups based on their participation in two different interventions. One
group participated in survey feedback intervention activities only; the
other participated in confrontation team building meetings in addition
to survey feedback. These interventions can be distinguished from
one another on the dimension of depth of emotional (Harrison 1970).
Chapter 3 will develop this reasoning fully.
Using Harrison's model, survey feedback falls at the surface
end of the intervention spectrum. Confrontation team building with its
emphasis on the individual and their relationships to others falls at the
deeper end of the intervention spectrum. Thus, nominally, a
intervention group consisting of those who participated in
confrontation team building and a comparison group consisting of
those who participated only in survey feedback can be created from
the existing data. This creation of nominal groups permits an ex post
facto quasiexperimental design which will control for history and
maturation as sources of internal invalidity in the analysis. The
groups' data will be tested for the presence or absence of gamma
change in relation to the type of intervention experienced using the
original Golembiewski, Billingsley, and Yeager (1976a) factor
comparison technique.
The significance of the study rests on five points. First, the
research proposes to break new ground. To date, the development
and application of methods for assessing beta and gamma changes
9
have focused only on the presence or absence of nonalpha change.
The present research proposes to extend this line of inquiry by
examining the question, "Is there a relationship between type of
intervention and type of change?" An affirmative answer to this
question suggests two additional, important points of significance.
Second, an affirmative answer to the research question would
suggest that the potency of different interventions could be gauged. In
effect, a taxonomy of what kinds of interventions are likely to result in
what kind of change could be developed. From a practitioner's and/or
manager's perspective, this information would be powerful in helping
to choose interventions for change efforts where both time and
resources are scarce.
Third, knowledge of the relationship between intervention and
change would provide insight into how different interventions should
be appropriately assessed. From a researcher's perspective, the
knowledge would be useful in shaping the selection of appropriate
tools for assessing organizational change efforts. Such a conceptual
advancement would liberate assessment efforts from the constraint of
raw gain" scores and the comparison of means as the only
alternative.
Fourth, only four field studies of organizational change efforts
have investigated type of change (Randolph 1982; Macy and Peterson
1983; Randolph and Elloy 1987; Golembiewski and Munzenrider
10
1988,1989,1990). For all their prominence in the literature, the
contributions concerning alternative methods used simulated data to
test the efficacy of the method.
Fifth, no study has examined data from a largescale, longterm
change effort. Previous studies have examined only small
organizations or subunits over short periods of time. The examination
of interventions efforts within the context of largescale OD effort is an
additional uniqueness.
Outline of Thesis
This chapter has introduced the research and stated the
purpose and significance of the study. The significance of the study
was framed not only in terms of its contribution to Organization
Development research but its contribution to evaluation research as
well.
Chapter 2, will review the literature on the alpha, beta, gamma
change typology in its entirety. This literature will be examined in five
categories based on the suggestions of Golembiewski (1986) and
Tennis (1986). Since a major thrust of the literature is alternative
methods for assessing beta and gamma change, the review will
examine the methods in detail to convey the fullness of the approach.
Summary observations of the body of literature will conclude the
chapter.
11
Chapter 3 will describe the hypotheses to be tested and the
assumptions from the literature which underlie them. Specifically, this
chapter will the intervention approaches used in this particular change
effort and the resulting focal variables. A conceptual framework for
the research will be developed using Harrison's (1970) concept of
depth of emotional involvement which permits differentiation of the
interventions and Golembiewski et al's. (1976a) scheme for type and
order of change. Hypotheses about type of intervention and type of
change which follow from this framework will be described.
Chapter 4 chapter will describe all relevant methodological
matters. It will describe the subject organization and some of the
climate surrounding the OD effort. It will describe the research design,
the methods and procedures for creating the nominal intervention and
comparison groups, tests of the equivalence of the intervention and
comparison groups, and the research instrument, its reliability and
validity, and where it has been used. It will also describe in detail the
method and procedures used to compare factor structures.
Chapter 5 will present in full the results of the analysis in
relation to the postulated hypotheses. Chapter 6 will conclude the
presentation of the research with summary observations and
statements of implication for research.
12
CHAPTER 2
SURVEY OF LITERATURE
Introduction
The literature concerning the change typology can be traced to
a single journal article published by Golembiewski, Billingsley, and
Yeager (1976a). In that article, Golembiewski et al. proposed a three
tiered model of change based on their experience of assisting with the
installation of a flexitime work system. In examining quantitative time
series data and comparing the results to qualitative data collected in
interviews, conclusions about the success of the intervention did not
match. It was this paradox that caused the researchers to reexamine
their data set in a very different way and ultimately propose a three
tiered typology of organizational change.
Golembiewski et al. (1976a) suggested that actually three kinds
of change can occur as a result of deliberate Organization
Development (OD) interventions. Quoting their definition from the
original article,
ALPHA CHANGE involves a variation in the level of some
existential state given a constantly calibrated measuring
instrument related to a constant conceptual domain.
BETA CHANGE involves a variation in the level of some
existential state, complicated by the fact that some intervals of
the measuring continuum associated with a constant conceptual
domain have been recalibrated.
GAMMA CHANGE involves a redefinition or reconceptualization
of some domain, a major change in the perspective or frame of
reference within which phenomena are perceived and
classified, in what is taken to relevant in some slice of reality,
(authors' emphasis) (pp. 13435)
Each definition of type of change is elaborated.
Alpha Change
Drawing on concepts from the counseling literature,
Golembiewski et al. associate alpha change with "first order" change
(Watzlawick, Weakland, and Fisch 1974). First order change is
defined as change that "occurs within a given system which itself
remains unchanged" (p. 10). Golembiewski et al. observe that most
OD efforts and reports of change recognize only alpha change. Thus,
the usual pretest and posttest design (Oj X O2) suffices. Change
can be big or small, and random or nonrandom. The operational
assumption is that it is occurring along stable dimensions of reality
within a relatively fixed state (Golembiewski et al. 1976a, p. 135).
BetaChange
Beta change, like alpha change, deals with changes in
condition within a relatively stable state. However, unlike alpha
14
change, the intervals used to measure some stable dimension are no
longer constant and equal but have been recalibrated. Golembiewski
et al. (1976a) insist that such change in measuring intervals "is often
an intended effect of an OD intervention, as contrasted with some
defect of the measuring instrument" (p. 136). The authors offer an
item of the Likert Profile of Organizational Characteristics as an
example.
Based on the experience of an OD intervention, respondents
may stretch the psychological space of the Likert scale for an item like
"At what level in the organization are decisions formally made?" As
they learn that decisions are not made throughout the organization,"
"wellintegrated," and "overlapping" as they first thought (System 4),
but rather broad policy at the top, more specific decisions at lower
levels" (System 3), they may show a decrement from pretest to
posttest scores. In alpha change terms, this may seem like failure.
However, in beta terms what has changed is the respondents'
understanding of how organizational decisionmaking really functions
based on learning from the intervention. The concepts of Likert's
Systems 3 and 4 remain intact. Understanding of the scale has
changed and it has perhaps "lengthened" to accommodate
respondents' more realistic understanding of what decisionmaking is
now in relation to what it can be ideally.
15
Gamma Change
Gamma change is qualitatively different from either alpha or
beta. It represents a redefinition of the relevant psychological space
as a consequence of an OD intervention" (Golembiewski et al. 1976a,
p. 138). In Watzlawicks et al. (1974) terms, it is second order change;
change in the system itself. Golembiewski et al, offer the three states
of water as an analogy. Water can exist as a solid, a liquid, and a
gas. Each is a qualitatively different state based on temperature.
Within each state, changes in condition can occur. These may be
associated with the concepts of alpha or beta change. However, at
critical points, a discontinuous jump to a different state occurs and the
concepts which describe only condition do not readily apply to the new
state (e.g., hard vs. soft ice as compared to cold vs. hot water). The
measurement complication which the concept of gamma change poses
in evaluating OD intervention is formidable. The instrument used to
measure a preintervention state is no longer appropriate for the post
intervention state.
While Golembiewski et al. articulated three types of change,
they focused on gamma change. This focus was emphasized by a
statement of what the authors believe about the intention of OD
interventions.
Typical descriptions of OD that seek to induce a new 'social
order1 or 'culture' into an organization convince us that gamma
changes are the prime intended consequences of OD
16
interventions. They imply not only the recalibration of intervals,
but also new content for concepts describing the quality of
organization life. Yet gamma changes are also likely to be
masked by common measuring instruments whose
conceptualization and operationalization are typically rooted in
concepts of alpha change. (Golembiewski, Billingsley, and
Yeager 1976a, pp. 13940)
The authors attempted to distinguish gamma change from
either alpha or beta change by using an analytic technique which
assesses the kind instability of parameters or dimensions associated
with gamma change. Specifically, Golembiewski et al. applied factor
analysis, ex post facto, to a data set across three measures for which
ordinary analytic techniques could not explain the variation. Next,
using a specific technique for comparing factor structures (Ahmavaara
1954), they computed the amount of shared variance between the
three factor structures. They discovered that the amount of shared
variance decreased substantially from T to T2 and again from T2 to
T3. This profound change in factor structure signaled what they came
to call gamma change and was consistent with the interview reports
from intervention participants. This search for gamma change and the
steps involved bears closer examination.
Discovering Gamma Change
As noted above, the data were from an evaluation of a
structural intervention, the installation of a flexitime system of work
hours. The research design was as follows,
17
N Day 1 Day 15 Short post Long post
Experimentals 32 Oi X o2 Oa
Comparisons 18 Oi o2 Os
where O denotes data collections using a questionnaire and X
denotes the installation of flexible work hours. Data were collected
using a selfreport instrument of 18 items each on a Likert scale of
seven equal appearing intervals (Golembiewski et al. 1976a p. 143).
Two key assumption operated in the analysis. First,
incongruence between factor structures is interpreted as a change in
the dimensions of reality necessary to account for the variance in the
set of responses. Thus, gamma change was defined as "substantial
incongruence of pairs of betweenwave factorial structures"
(Golembiewski et al. 1976a p. 144). Second, nongamma change,
either alpha or beta, will not generate major incongruence between
pairs of factorial structures.
Six separate analyses of the data set were conducted to test
the assumption that gamma effects resulting from the OD effort explain
the variation in the data. The analyses consist of three waves of data
for Experimentals only and three for Experimentals and Comparisons
combined. The analyses were conducted painstakingly to thoroughly
assess rival or artifactual explanations for variance in the data set.
Further, it seems these analyses apparently were an accommodation
18
for the small N and the low respondent to questionnaire item ratio
(2.8:1).
1) Congruence of withinwave structures (N=50 and N=321.
Four different factor analytic techniques were used to assess factor
structures within each wave (Golembiewski, Billingsley, and Yeager
1976b). The four techniques yielded highly similar structures within
waves for both the experimental and comparison groups with the
average amount of common variance being 97%.
21 Congruence of betweenwave structures flVI=501.
Combinations of between wave structures were tested for
Experimentals plus Comparisons. The average amount of common
variance shared between any two wave structures was 67%, 30% less
than the variance accounted for within wave structures. Golembiewski
et al. contend that alpha and beta changes are inadequate to account
for a 30% "loss" of variance in comparisons of factor structures
congruence and point to gamma change as an explanation for the
fundamental variation in conceptual space (Golembiewski et al.
1976a, p. 148).
31 Congruence between waves. Experimentals only fN=321.
Testing their last premise more specifically, the authors examined the
betweenwave factor congruence for Experimentals only. With the
Comparisons removed from the analysis, the estimated average
common variance between waves fell to 51% from 67%. "If a 30%
19
loss in the interwave common variance is not sufficient to establish the
likelihood of gamma effects, a 50% loss provides very much more
formidable support for that conclusion11 (Golembiewski et al. 1976a, p.
149).
4) Congruence of withinwave structures for variable N. in
order to support their conclusion more fully, the authors returned to an
analysis of withinwave structures. Specifically, they tested the
question, "Is the lower percentage of common variance for
Experimentals only (N=32) versus Experimentals and Comparisons
(N=50) an artifact of the variation in N rather than an effect of
excluding the Comparisons?" To answer the question, respondents
were randomly eliminated so that five independent subpopulations of
35 were isolated for each of three waves. Factor analysis was applied
to these 15 subpopulations as well as the total population (N=50).
The analyses reveal common and stable factorial structures that are
substantially independent of major changes in N.
5) Congruence of betweenwave structures for variable number
of factors. To further test the sturdiness of their conclusion, the
researchers next structured their analysis of artifactual explanation for
the loss of betweenwave common variance to examine the effects of
the number of factors extracted. Prior analyses had used comparison
of the first seven factors extracted even where the eigen value was
less than 1.0. If the number of factors was allowed to vary to include
20
only those with eigen values greater than 1.0, then Wave 1 = 7, Wave
2 = 6, and Wave 3 = 5. Comparing betweenwave structures with
these different number of factors, the average common variance
increased to 70% from 67% for N=50, a small effect, and to 52.5%
from 51% for N=32, a negligible increase. The conclusion was very
little of the betweenwave incongruence can be reasonably assigned
to the convention of setting the number of factors to seven, the Wave
1 principal components factor structure.
61 Congruence of withinwave structures when iterations = 5 vs.
39. One last analysis was performed to test a potential artifact which
might explain the high degree of congruence of withinwave factor
structures. In SPSS, the default for the number of iterations performed
in factor analysis is five. In order to test the possibility that this
limitation was responsible for the high withinwave factor congruence,
ten tests were run. The tests were structured to include all three
waves, combined responses (N=50) and Experimental only (N=32),
and varying numbers of factors (7,6, and 5 as appropriate to each
wave). Further, the number of iterations was allowed to vary between
5 and 99. No meaningful difference was found in the common
variance estimated by 5 iterations vs. 99 iterations. The range of
estimated common variance for withinwave factor structures was 97%
to 100%. Thus, limiting the iterations to 5 in the initial analysis posed
no problem.
21
Golembiewski et al. conclude that their analysis "implies the
real possibility of gamma changes" (1976a, p. 152). The six tests
eliminate random causes, factorial procedure, the level of N, the
convention of using seven factors for each wave, and the number of
iterations as rival explanations for the dramatic decrease in shared
variance between waves. The analysis implies that the flexible work
hours intervention substantially changed "the psychological
dimensions that employees used to evaluate their work site" (p. 152).
Conclusion
In concluding their work, Golembiewski et al. (1976a) are clear
about their contribution and its limitations. While distinguishing
gamma changes from alpha and beta changes using the comparison
of factor structures and summary statistics, they do not describe the
specifics of gamma change. That description they place beyond the
scope of their research. They state that their analysis strongly
suggests further research to establish the existence of gamma
changes and to satisfactorily differentiate gamma from alpha and beta.
Specifically, they note seven issues for future research.
First. A focus of behavioral science research has been how to
measure change. Researchers have assumed that only alpha change
is relevant, an assumption which "seems clearly inappropriate for
successful OD interventions..." (p. 153). Thus, the first question
22
should be what kind of change is being measured before the question
of how to measure change is asked.
Second. The present analysis used the comparison of factor
structures to identify and distinguish change in the definition of
conceptual space. Further interpretation was not approached. While
comparison of factor structures may be a simple, useful way of
discovering the presence of gamma change, there is still the issue of
calculating the direction and magnitude of change scores.
Third. The existence of different types of change begs very
difficult questions for OD research in general. Interpreting clearly the
results of change efforts is indeed uchancy despite rigor and care in
research design without knowledge of type of change.
Fourth. The present analysis indicates the value of time series
designs in OD research. Simple prepost designs make the isolation
of beta and gamma changes more difficult.
Fifth. The analysis implies the value of statistical methods such
as factor analysis which seek to discover changes in definition of
psychological space. Such statistical methods help address more
fundamental questions about the nature of change as "continuous" or
"discontinuous."
Sixth. More attention to subjects responding to selfreport
measures is suggested. If their own perception is that the scale is a
"rubber yardstick," this perception could be used as an early warning
23
to beta or gamma changes.
Seventh. Since analysis of underlying structure appears to be
a method for detecting type of change, scaling techniques which are
less sensitive to metriclevel assumptions might be helpful.
Golem biewski et al. closed their report by suggesting "there is a
long trail ahead of this line of research..." While it is common place
and perhaps even trite to close inconclusive but compelling research
with a suggestion for further research, the challenge they laid down
was a formidable one. These researchers had potentially declared all
previous OD research (and more profoundly all behavioral science
research) findings questionable for not having answered the more
basic question, "What kind of change?" This declaration and
challenge would predictably have drawn response from researchers in
the field anyway. However, the challenge was made much more
formidable by the fact that the research was the 1975 Douglas
McGregor Memorial Award Winning Paper in the Journal of Applied
Behavioral Science. It could hardly be ignored.
Summary and Structure for the Remainder of the Chapter
The original article spawned a curious but active response in
the first several years following its publication. The literature
concerning the change typology has been characterized by Tennis
(1986) as a response by an "invisible college" borrowing Price and
24
deBeaver's (1966) phrase for its relatively closed fraternity in both
research and publication. Tennis (1986) further categorizes
responses five ways: application, avoidance, anomalies, alternatives
and integration. Golembiewski (1986) categorizes responses three
ways: (1) attack on the factorial technique while leaving the
conceptual distinctions intact, (2) alternative approaches to assessing
alpha, beta, and gamma change, and (3) diverse operational
approaches including comparison of techniques, laboratory studies
and simulations, the search for surrogate variables, and conceptual
analysis of the academic response. For the purpose of this review,
the Golembiewski categories shall structure the chapter with the
addition of Tennis's category concerning applications of the original
statistical techniques for assessing change at the beginning.
Applications of the Original Statistical Technique
As noted above, the change typology literature can be
categorized by type of response (Golembiewski 1986; Tennis 1986).
The earliest distinguishable response is applications of the original
statistical technique. This section reviews the three articles which
comprise this response (Randolph and Yanouzas 1977; Randolph and
Edwards 1978; Randolph 1982).
25
Randolph and Yanouzas (1977)
This first application of the original statistical technique reports
an analysis of performance variance in ad hoc organizations as
measured by the Likert Profile of Organizational Characteristics. The
purpose of the study was to identify the characteristic of high vs. low
performing ad hoc organizations and assess the Likert Profile's
suitability as an analytic tool for ad hoc organizations. Three
hypotheses concerning the Likert Profile's application to ad hoc
organizations were tested.
College students in an introductory management course were
formed into 21 "companies," each of which was required to complete
four projects. Organizational performance was assessed at the end of
each project by the organizational members using the Likert Profile.
Three administrations for all 21 companies were factor analyzed to
determine the underlying dimensions of the Profile in the ad hoc
situation. These factors were later used in a multiple regression
analysis to test a specific hypothesis.
The significance of this report to the change typology is that
Randolph and Yanouzas compared the factor structures across all
three administrations to determine stability. This was done with direct
reference to Golembiewski et al. (1976a) However, the specific
technique is not detailed. Randolph and Yanouzas do report the
productmoment correlations between factor structures which indicate
26
the similarity of factor patterns. Importantly, it is the first response to
the Golembiewski et al. report and the first application of the statistical
technique for type of change.
Randolph and Edwards M978)
The second application of the original statistical technique
reports the assessment of alpha, beta, and gamma change in an OD
intervention with a university student service center. The client
organization, or experimental group, consisted of 35 staff in three
centers. A student service center of similar size at a different
university served as a control group. A survey feedback approach to
intervention was used. The survey instrument used with both
organizations was developed through random, stratified interviews
with members of the experimental organization and addressed seven
areas of concern. The entire effort lasted ten months during which the
surveyfeedbackaction planning cycle occurred twice.
Randolph and Edwards report analyses and results for alpha,
beta, and gamma change. Alpha change was assessed by comparing
pretest and posttest means for both items and factor dimensions using
a simple ttest with the significance level at p<.05. Factors were
determined by performing factor analyses with Varimax rotation on the
combined pretest data and combined posttest data. Beta change was
assessed by Ftest comparisons at the p<.05 significance level of pre
27
and posttest variances. The assumption was made that decreased
variance, meaning greater respondent agreement in scale
interpretation, should occur in the experimental group but not the
control group. Gamma change was assessed using the factor
comparison technique described by Golembiewski et al. (1976a)
Results of the intervention are reported by type of change
analyzed, a first of its kind in the OD research literature. Significant
alpha change occurred in the experimental group in only one of the
areas hypothesized to show change. Five of 65 items and three of the
17 factors showed pre to post change at the p<.05 level. No alpha
change occurred in the control group.
Significant beta change, defined operationally as response
variance reduction, was demonstrated in seven of the 65 items and
one of the 17 factors for the experimental group and in one common
item for the control group (p<.05). All items were part of intervention
targets hypothesized to show change as a result of the effort.
Randolph and Edwards compared pretest and posttest response
standard deviations for items and factors to compute F ratios. (A
similar operational definition of beta change using the dispersion of
standard deviations will be reviewed later.) Their analysis represents
the first attempt and method to compute beta change for real research
results. Golembiewski et al. (1976a) concerned themselves only with
gamma change and offered no technique for assessing beta change.
28
Gamma change, using Ahmavaara's (1954) technique for
comparing factor structures, was indicated for three of the 17 factors in
the experimental group and for one of 17 factors in the control group.
For the experimental group, two of the three gamma changes occurred
in intervention target areas. The third gamma change, which was the
same for the control group, was demonstrated in a factor which was
not addressed by the intervention. Randolph and Edwards conclude
that an environmental force outside of the intervention created this
third gamma change.
It is important to note here that while the technique for
assessing gamma change, namely Ahmavaara's (1954) procedure for
comparing factor structures, is the same as Golembiewski et
al.(1976a), the criterion for concluding gamma change from the results
is not the same. Golembiewski et al. suggest a cut point" of 50%
reduction in shared variance between factor structures as indicative of
gamma change. Randolph and Edwards use a 20% reduction in
shared variances as their cut point. However, it is understandable to
observe that at this point in the development of the change typology,
with a single technique for assessing gamma change, there are no
established conventions for interpretation.
Randolph M9821
The final application represents a longitudinal extension of
29
Randolph and Edwards (1978). Using the same experimental and
control groups, Randolph reports on the outcome of additional survey
feedback cycles over a total of 33 months. As in the first report,
specific hypotheses related to the impact of the OD intervention were
formulated. Change in these hypothesized impact areas as opposed
to a generalized assessment of change is reported.
Results of this longer intervention are again reported by type of
change analyzed. However, the analyses are reported in reverse
order this time. In the five years between Randolph and Edwards
(1978) and Randolph (1982), several others argued for an analytic
protocol for assessing type of change (Armenakis and Zmud 1979;
Golembiewski and Billingsley 1980; Terborg, Howard, and Maxwell
1980). This protocol suggests assessing gamma change first since
beta and alpha changes cannot be assessed if concept redefinition
has occurred. If gamma change has not occurred, beta change is
assessed next. If it appears that no scale re calibration has occurred,
only then can alpha change be safely assessed.
Gamma change was again assessed using Ahmavaara's (1954)
technique as suggested by Golembiewski et al. (1976a) However,
since Randolph was interested in testing specific hypotheses and
since the item to respondent ration for factor analysis purposes would
have been too low, sectional analysis of the questionnaire was
performed. All six sections of the questionnaire were factor analyzed
30
separately for both the experimental and control groups and factor
structures were compared across administrations. Of particular
interest were the four sections of the questionnaire which were
hypothesized to address areas of intervention impact.
Randolph concluded there was no strong indication of gamma
change resulting from the intervention. Some gamma change was
indicated for the experimental group in three questionnaire sections
for T1 vs. T2. However, it was not sustained through subsequent
comparisons (2 vs. 3 and 1 vs. 3). Unintended gamma change was
indicated for both the experimental and control group in a section of
the questionnaire not addressed by the intervention, thus
demonstrating the value of control in the analysis. Moderate gamma
change was indicated in only one section of the questionnaire across
comparisons. However, the percentage of shared variance between
compared factor structures (58%) did not meet the stringent 50%
decrement suggested by Golembiewski et al. (1976a) and
Golembiewski and Billingsley (1980).
Beta change was assessed using a methodology suggested by
Zmud and Armenakis (1978). (This will be presented in full later.)
This method calls for the comparison of actual and ideal scores for
separate points in time. The assumption behind the method is that
respondents will respond to both actual and ideal measures at a
particular point in time on a consistent psychological scale even if
31
respondents have re calibrated that scale in the time between
measures. Difference scores are computed from each set of actual
and ideal scores. Combinations of equalities and inequalities for all
three scores are used to determine if beta change has occurred.
The analysis for beta change was limited by the fact that only a
portion of the questionnaire asked for ideal assessments. No beta
change was indicated for either the experimental group or the control
group for the questionnaire items able to be assessed. Randolph
concluded that since gamma and beta changes had not occurred, it
was safe to proceed to an analysis of alpha change. However, the
assessment of beta change was so limited and not related to items
associated with hypothesized intervention impact, the conclusion that
no beta change occurred is questionable.
Alpha change was assessed by using ttest comparisons of
mean differences of factors constructed from questionnaire items
loading .5 or greater in equal weighted fashion. In other words, Factor
1 consisting of three items which loaded at .5 or greater would have a
mean equal to the sum of the item scores divided by 3. Randolph
compared pretest scores and three posttest scores for the
experimental organization and pretest and two posttest scores for the
comparison organization. The level of significance for the ttests was
established at p<.05.
Modest alpha changes occurred. In the interval from Tj to T2,
32
three significant changes occurred in 12 factor means. In the interval
from T to T3, two factor means in 12 showed significant change. The
changes were all in factors which were hypothesized to be directly
affected by the OD interventions. No similar changes occurred for the
comparison organization.
Randolph discusses two major outcomes of the research. One
is support for the utility of the change typology. The other is the
importance of a sound research design which includes a comparison
organization and time series data.
While the results of the alpha change analysis were modest,
the most important finding was in a factor which showed no alpha
change. However, this factor showed sustained, moderate gamma
change (a shared variance of 58% from T to T3) for the experimental
organization in the gamma change analysis. That this factor showed a
substantial decrement in shared variance of factor structures across
time without demonstrating a statistically significant change in factor
means is a finding of importance. This finding counters empirically an
assertion by Lindell and Drexler (1979,1980) that alpha change
assessment alone is sufficient and that alpha change can create
gamma change. Randolph's discovery of gamma change without
alpha change disputes both points and demonstrates the value of
systematically assessing all three types of change before drawing
conclusions.
33
Further, the use of the comparison organization and time series
data assisted the correct interpretation of change that occurred in the
experimental organization. Randolph was able to correctly identify the
moderate gamma change which occurred in a target factor for the
experimental organization from unintended gamma change in another
factor by comparing results from both organizations across time. Also,
alpha change was reliably identified in the intended factors for the
experimental organization by comparing those factors with the
comparison organization.
Summary
These entries by Randolph et al. (1977,1978) and Randolph
(1982) are the only attempts in the literature to apply the original
statistical technique of Golembiewski et al. (1976a). Each successive
report demonstrates growing sophistication with both the technique
and its application. By the third report (Randolph 1982), a thorough
analysis using times series data and a comparison organization
demonstrates the value of the change typology and a technique for
assessing gamma change in evaluating change resulting from OD
intervention. During this period, however, developments which
pushed the matter of the change typology in very different directions
were happening concurrently. One such development was a strident
attack on the change typology and its value for evaluating change.
34
Attack on the Factorial Technique
Two entries stand as a direct challenge to the value of the
typology in general and the factor analytic technique for determining
gamma change in particular (Lindell and Drexler 1979,1980). This
section examines the issues surrounding the attack, the systematic
response from Golembiewski and Billingsley (1980), and the counter
response from Lindell and Drexler.
Challenging the Change Typology
An unquestionable attack on the change typology came from
Lindell and Drexler (1979). While they acknowledged the change
typology as a conceptual contribution to the basic theory of
organizational evaluation, they warned of its difficulties. They
criticized Golembiewski et al. for not defining a method for
operationalizing beta change. They further criticized the method for
assessing gamma change, namely comparing factor structures, as not
being an unequivocal test of gamma change.
Lindell and Drexler elaborate the distinction between alpha and
beta change and emphasize that the conceptual distinction is
theoretically important. Alpha change is defined as both change
on a measurement scale and corresponding change in the behavior
being measured. Drawing on earlier work, beta change is defined as
"change in the location of an object on a scale that is due to changes
in the properties of the scale itself" (Upshaw 1962, p. 14). Beta
35
change occurs when a respondent changes the psychological range
they ascribe to an object or property being measured, e.g. a
supervisor's supportive behavior. Due to some treatment or
intervention, the respondent's negative and positive endpoints may
change producing a corresponding change in category assignment.
Lindell and Drexler concede that this exact process could happen in
OD interventions.
The concept of gamma change, however, received a much less
charitable treatment. Comparing factors structures as an
operationalization of gamma change is rejected. Lindell and Drexler
demonstrate with hypothetical survey data that two simple changes in
the post rating, attributable to either alpha or beta change, can cause
a change in pre/post correlation. Since correlations are directly
related to factor structures, changes which affect correlation can also
affect factor structure. Thus, they argue, change in factor structure is
not an unequivocal operationalization of gamma change. They further
label the concept as suspect.
Lindell and Drexler conclude that the entire issue of interpreting
survey results is not as serious a problem as Golembiewski et al.
(1976a) claim. They argue that the use of behaviorally anchored
scales with multiple items overcomes the scale recalibration problem.
Further, if the possibility of beta change is eliminated through proper
psychometrics, the concern for gamma change is also eliminated since
36
it is probably a function of either alpha or beta change.
Golembiewski and Billingsley's Response
Golembiewski and Billingsley (1980) responded systematically
to the criticism. They listed five substantive disagreements.
First. They reject the suggestion that the exposition of the
change typology sought to denigrate the utility of survey methodology
for assessing organizational change. Instead, they sought to enrich
the analysis of change since it makes no sense to study means when
beta or especially gamma change has occurred.
Second. Golembiewski and Billingsley reject the assertion that
no method was provided for assessing "in an easy manner" the kind of
change that characterizes a data set. They counter that they
described and tested a workable method of assessment. Further, they
suggested conservative cut points for interpreting gamma change.
Third. Lindell and Drexler (1979) focus on beta change while
Golembiewski et al. (1976a) focused on gamma change.
Golembiewski and Billingsley agree with Lindell and Drexler's
restatement of beta change and the mechanics of re calibration, but
argue that a beta change conclusion can result"... only when gamma
change has been eliminated as a possibility" (p. 100).
Fourth. Golembiewski and Billingsley charge that Lindell and
Drexler's appreciation of gamma change is seriously inadequate as is
37
their understanding of factor analysis. Five concerns are stated. One,
while Lindell and Drexler fail to see the useful role of gamma change
in the measurement of organizational change, Golembiewski and
Billingsley argue that gamma change is the goal of many planned
interventions and thus is critical to measuring and explaining
organizational change. Two, psychometrically sound questionnaires
are not adequate to account for observed variations in data sets.
Evidence of gamma change implies redefinition of what constitutes a
psychometrically sound questionnaire." Three, their example of
changing two numbers in a post rating to change pre/post correlations
is irrelevant. One correlation does not make a structure and the
example simply confuses structure with correlation. Four, behaviorally
anchored scales do not obviate the need for considering beta and
gamma changes. OD tries to change the cultural meaning of objective
behaviors and thus one still needs to know how a respondent values a
given behavior in the context of other behaviors. Five, Golembiewski
et al. were unable to reproduce with computer simulations the kind of
factor structure change produced by respondents. Golembiewski and
Billingsley are convinced of the fundamental change which occurred
and reject the charge that alpha and beta change can produce change
in factor structure.
Fifth. Lindell and Drexler maintain that the issue of type of
change is not a serious concern and can be managed by more
38
conventional approaches to distinguishing change. Golembiewski and
Billingsley reply that they cannot think of a more serious problem than
type of change. While questionnaires consisting of multiple items or
measures are valuable, they may also hide our ignorance of what is
really occurring.
The Counter Response
Lindell and Drexler (1980) countered with a clarification of their
original objection. They rejected the connection that Golembiewski et
al. (1976a) make between the conceptual and methodological levels.
That is, they reject the assumption that factor incongruence
corresponds perfectly with concept redefinition. They argue that
gamma change could be a function of either alpha or beta change and
endorse some sort of confirmatory analysis such as cluster analysis
and maximum likelihood factor analysis as better choices of method.
They conclude by reiterating that the "nuisance" of beta and gamma
change could be addressed through proper instrumentation.
While there were no further exchanges between these parties
in the literature, there are subsequent references and responses to
the attack on the change typology. One direct reply to Lindell and
Drexler is Terborg, Howard, and Maxwell's (1980) summary of
research which demonstrates that psychometrically sound scales with
behavioral anchors do not eliminate the possibility of beta change.
39
Contrary to Lindell and Drexler's assertion that alpha change can
create gamma change, Randolph (1982) discovered evidence of
gamma change without alpha change. Randolph further concludes
that while questionnaires consisting of multiple items with behaviorally
anchored scales reduce the likelihood of beta change, one cannot
simply assume that beta or gamma change is impossible with
psychometrically sound instruments. Schmitt (1982) transformed the
responses in a data set to mimic alpha change in order to test the
sensitivity of his analytic procedure. He discovered that the
transformation had no effect on resulting factor structures and
concludes this contra indicates the LindellDrexler argument that
alpha change may at times manifest itself as gamma or beta change"
(Schmitt 1982, p. 356). With these responses, the matter ended.
Summary
The exchange between Lindell and Drexler (1979,1980) and
Golembiewski et al. (1980) constitutes the only serious rejection of the
change typology and specifically the factor incongruence method for
operationalizing gamma change. There were some earlier responses
to the typology that sought to employ research design to control for
beta and gamma change (Armenakis and Smith 1977,1978; Porras
and Berg 1978; Zmud and Armenakis 1978; Armenakis and Zmud
1979; Porras and Patterson 1979). None, however, challenged the
40
conceptual distinction of the typology or the factor incongruence
method for operationalizing gamma change.
This section has dealt with the only vocal, minority rejection of
the change typology. The next section deals with the bulk of the
change typology literature. In response to the Golembiewski et al.
(1976a) challenge, a spate of contributions sought to identify
alternative approaches to assessing alpha, beta, and gamma change.
Alternative Approaches to Assessing
Alpha. Beta, and Gamma Change
The third distinguishable response to the Golembiewski et al.
exposition of the change typology (1976a) is alternative approaches to
assessing alpha, beta, and gamma change. These alternative
approaches are a direct response to the challenge from Golembiewski
et al. to extend the research on the change typology and explore other
methods of operationalization. There are essentially seven
alternative approaches to assessing alpha, beta, and gamma change.
The approaches can be categorized in two ways (Armenakis 1988).
First, there are six techniques that employ research design as a
means to assess or control for beta and gamma change (Armenakis
and Smith 1978; Zmud and Armenakis 1978; Armenakis and Zmud
1979; Bedeian, Armenakis, and Gibson 1980; Terborg, Howard, and
Maxwell 1980; Van de Vliert, Huismans, and Stok 1985). Second,
there are two additional techniques that employ statistical methods to
41
detect beta and gamma change (Armenakis, Feild, and Wilmoth 1977;
Schmitt 1982). Since each of these contributions to the literature
describes a technique for assessing type of change, they will be
treated in detail to convey the fullness of their approach to
operationalizing change type.
Six Approaches That Employ Research Design
Armenakis and Smith M9781. The first alternative approach to
the Goiembiewski et al. (1976a) factor analytic technique is Armenakis
and Smith (1978). Noting that the most significant administrative
problem in evaluating change is the use of comparison groups
(Armenakis Feild, and Holley 1976), the authors make a case for
employing one group designs which control for or discount the sources
of internal invalidity. In addition to sources of internal invalidity, types
of change further complicate the measurement and interpretation of
change. In fact, alpha, beta, or gamma changes may be the result of
either sources of invalidity and/or the effects of an intervention. Thus,
it is the responsibility of the researcher to determine types of change
and causes of change in their analyses. Further, it is important to
select the optimal evaluation design to address these issues.
Armenakis and Smith recommend the abbreviated time series
design (ATSD) consisting of two pre measures and two post measures
at the group level of analysis in response to these issues. They
42
applied this design to an Army training brigade using as the
measurement instrument the Survey of Organizations. Observations
one and two occurred three months apart. The data were tested for
type of change. Consistent with Golembiewski et al. (1976a), they
tested the data set first for the presence of gamma change using
factor analysis but comparing factor structures with a different
algorithm, the coefficients of congruence (Harman 1967; Armenakis,
Feild, and Wilmoth 1977). They found no evidence of gamma change
from 0 to O2 in the four factors obtained. However, they discovered
a significant decrement in factor means between the first and second
administrations of the measures prior to any intervention. Thus, either
alpha or beta changes, which the authors argue are related to sources
of internal invalidity, had occurred.
Armenakis and Smith investigated the sources of invalidity
systematically using test statistics where possible. Maturation and
testing were noted as possible causes of the decrement. However,
the more important feature of this analysis is the use of the
abbreviated time series design. Without it, the observed changes
would not have been revealed and erroneous conclusions about a
subsequent intervention effect may have been reached.
Zmud and Armenakis. (19781. A second alternative approach
employing research design also at the group level of analysis is Zmud
and Armenakis (1978). The authors note the simultaneous concerns
43
of properly assessing types of change and determining or discounting
the causes of observed change. They acknowledge the importance of
the Golembiewski et al. (1976a) typology for its contribution to the
rigorous evaluation of change efforts. They cite four causes of
observed change: 1) the intervention itself, 2) internal invalidity, 3)
external invalidity, and 4) statistical artifacts.
Since intervention is a desirable cause, the researcher's task is
to discount the others. Zmud and Armenakis argue that research
design addresses sources of internal and external invalidity (Campbell
and Stanley 1963; Evans 1975; Armenakis and Smith 1978).
Statistical artifacts, which have to do with the improper application of
statistical procedures, are less easily addressed. Thus, the authors
offer specific procedures for assessing types of change to eliminate
this last concern.
Zmud and Armenakis recommend that gamma change be
assessed using the comparison of factor structure. This supports the
position of Golembiewski et al. (1976a) that concept redefinition can
be operationalized by factor structure incongruence. However, Zmud
and Armenakis recommend a different technique called coefficients of
congruence (Harman 1967; Armenakis, Feild, and Wilmoth 1977).
This technique yields coefficients which range from 1.0 to +1.0 and
are interpreted like correlation coefficients thus simplifying the
interpretation of factor structure similarity.
44
For assessing group level alpha and beta changes, Zmud and
Armenakis recommend a wholly different technique which relies on the
research design. The prime consideration of the technique is to
control for re calibration of the measurement scale (beta change)
without having to statistically examine scaling properties and test their
consistency.
This technique requires the collection of both pretest and
posttest "actual" and ideal" data. The operational assumption is that
respondents will respond to both sets of measures based on a
consistent psychological scale or metric at T and T2 even if that
scale has been re calibrated in the time between measures. With both
sets of measures, they compute difference scores, D=lAi and
D2=l2_A2 Then using a comparison of all three scores, they draw
conclusions about the presence of alpha and beta change based on
five cases.
1. If A1 = A2, H = I2. and Dj = D2, then no change has
occurred.
2. If Aj A2 and lj I2 even though D = D2, then beta
change has occurred.
3. If Aj = A2 but lj I2 and D D2, then beta change
has occurred but little can be said about alpha.
4. If Aj A2, lj I2. and D D2, then both alpha and
beta change have probably both occurred.
45
5. If Ai A2, H = I2. and Dj D2, then alpha change has
probably occurred.
In each case, equality or Inequality is established through the
use of ttests or analysis of variance. Zmud and Armenakis
recommend the use of factor analysis to detect gamma change first
and rule it out before proceeding with their own method to detect alpha
and beta change.
Three points should be made about the suggested technique.
First, it addresses the assessment of beta change directly. Up to this
point, the focus has been on gamma change with beta change as one
third of the typology but lacking an operational definition. Here beta
change is operationalized as a combination of equalities or
inequalities among "actual" and "ideal" scores. However, while the
technique permits a way to detect beta change, the second point is
that little or nothing can be said about the direction and magnitude of
the change. In this way, this technique for operationalizing beta
change is very much like the factor incongruence method for
operationalizing gamma change. Third, the analysis addresses group
level change only. The possibility of individual beta or gamma change
is left unexplored.
Armenakis and Zmud (1979T In the third alternative approach
employing research design, Armenakis and Zmud (1979) combined
the features of the abbreviated time series design (Armenakis and
46
Smith 1978) and their difference scores technique (Zmud and
Armenakis 1978). They examined types of change by administering
the Survey of Organizations to a sample of military trainers at two
different time without an intervention. They tested for gamma change
by factor analyzing the data and comparing the structures using the
Coefficients of Congruence method (Harman 1967; Armenakis, Feild,
and Wilmoth 1977). They tested for alpha and beta using the five
cases specified in Zmud and Armenakis (1978). They found no
gamma change, but beta change was indicated. Armenakis and Zmud
conclude that the research demonstrates the possibility of beta
change with no intervention and note that insensitivity to the change
typology may result in inappropriate inferences from an observed
decrement.
Terbora. Howard, and Maxwell 11980). A fourth alternative
research design approach to the Golembiewski et al. (1976a) factor
analytic technique is Terborg, Howard, and Maxwell (1980). Their
approach is a significant departure from the previous three
alternatives by Armenakis et al. in three ways. First, Terborg et al.
recommend a littleused design technique known as the retrospective
then. Second, their method is more rigorously quantitative than the
other proposals. Third, their method describes types of change at
both the individual and the group level using profile analysis.
Terborg et al. note several interesting points as they review a
47
portion of the literature to date which are salient to this review. In
reviewing the Zmud and Armenakis (1978) method using actual and
ideal scores, they note the problems of the ceiling effect on ideal
scores and conclude that the method has limited utility in determining
type of change. In reviewing Lindell and Drexler (1979), they question
their conclusion that the use of psychometrically sound scales will
reduce the likelihood of beta and gamma change. They criticize
Lindell and Drexler for defining beta change as a nonissue when
there is no technique for assessing beta change and hence no way to
test such an assertion.
Terborg et al. also cite the difficulties of the original
Golembiewski et al. factor analytic technique, noting three in
particular. First, the factor analytic technique focuses on gamma
change and assumes no beta change occurred. With no technique for
assessing beta change available, using factor analysis to test for
gamma change is a questionable procedure. Second, Terborg et al.
agree with Lindell and Drexler that either alpha or beta change could
produce change in factor structure and thus must be addressed as
part of any assessment of change type. Third, factor analysis
requires a 3:1 ratio of respondents of items. Since most OD
interventions involve small numbers of people, the factor analytic
technique is difficult or impossible to apply. It is upon these points
that they rest the case for their method.
48
Terborg, Howard, and Maxwell (1980) review the literature on
responseshift bias as a prelude to presenting their method for
assessing alpha, beta, and gamma change. A brief review of that
literature here is necessary to fully appreciate their method.
The responseshift bias is one of several issues surrounding
the problem of measuring change. The others problems, each
intertwined with the next, include measurement error of gain or
difference scores, the confounding effects of history and
instrumentation on measurement, and the further confounding effects
of history and instrumentation when selfreport measures are used.
These problems of measuring change are well documented in the
psychology and education literature (Campbell and Stanley 1963;
Harris 1963; Cronbach and Furby 1970; Linn and Slinde 1977).
The problem of measurement error surrounding gain or
difference scores is well treated in Cronbach and Furby (1970).
Researchers have traditionally relied on "raw gain" scores calculated
by subtracting the pretest score from the posttest score to draw
conclusions about the effectiveness of a treatment. Such conclusions
are often erroneous due to the random error of measurement in both
pretest and posttest scores which is exacerbated in the difference
score. This is further complicated by the troublesome question of
whether or not the pretest and posttest are measuring the same
variable.
49
The second problem of measuring change, related to the first, is
the confounding effects of history and instrumentation on
measurement. The classic experiment is thought to control most if not
all threats to internal validity (Campbell and Stanley 1963). However,
true experiments cannot be employed in all research situations,
especially field settings (Weiss 1972; Babbie 1979). Many
evaluations rely on one group pretestposttest designs or quasi
experimental designs using comparison groups in lieu of true control
groups. When these alternative, quasi experimental designs are
used, threats to internal validity surface. The most notable of these
are history and instrumentation. The latter, having to do with change
in the rater's calibration of the instrument scale, is particularly vexing.
In order for meaningful change to be described in terms of a pre/post
comparison, both measures must be on the same metric. To the
extent that they are not, the comparison is invalid.
The third problem of measuring change is related to the
second. It concerns the further confounding effect of history and
instrumentation when selfreport measures are used. It is one thing
for history and instrumentation to affect a rater's calibration of an
instrument under ordinary circumstances. It is quite another matter
when the treatment or intervention being measured is designed to
alter the rater's understanding of the target concept. The question of a
common metric is even greater. This interaction between
50
instrumentation and treatment is what Howard and others have
referred to as responseshift bias and is consistent with
Golembiewski's et al. concept of beta change.
In a series of studies, Howard et al. have examined response
shift bias and a method for combating it (Howard and Dailey 1979;
Howard, Ralph, Gulanick, Maxwell, Nance, and Gerber 1979; Howard,
Schmeck, and Bray 1979; Howard, Millham, Slaten, and O'Donnell
1979; Bray and Howard 1980). Howard, Ralph, Gulanick, Maxwell,
Nance, and Gerber (1979) conducted five separate studies assessing
the impact of assertiveness training, interview skills training, helping
skills training, and inter personal effectiveness skills training. Their
research confirmed the existence of the responseshift bias. Using
pretest, posttest, and retrospective then self measures, they
discovered that the then to posttest comparison was more similar to
objective behavioral ratings of skill acquisition than the pretest to
posttest comparison.
Howard and Dailey (1979) report similar findings in evaluating a
training program to improve interviewer skills. Using pretest, posttest,
retrospective then, and recall self measures, Howard and Dailey report
two important outcomes. First, then to posttest selfcomparisons
correlated higher and significantly with objective third party ratings of
pretest to posttest skill improvement than with the pretest posttest self
comparison. Second, recall scores were highly consistent with subject
51
pretest ratings indicating an ability to correctly discriminate a then
rating from a pretest rating.
Howard, Schmeck, and Bray (1979) report that retrospective
then self reports were significantly different from pretest self reports
and correlated much higher with actual pretest performance measures
of students in an educational psychology class. Bray and Howard
(1980) demonstrate that then to posttest comparisons of graduate
teaching assistants' selfperceived teaching ability correlated
significantly higher with actual behavioral changes as rated by
students.
These studies demonstrate that experimental interventions that
use pretest to posttest self report comparisons are subject to an
instrumentrelated source of contamination known as responseshift
bias. Further, they confirm the use of the retrospective then technique
as a means of guarding against this treat to internal validity.
Terborg, Howard, and Maxwell (1980) suggest that the
implications of the retrospective then method are straightforward for
assessing group level alpha and beta change. Pretest, posttest, and
then measures are taken from an intervention group. Comparable
measures are taken from a comparison group. If the group mean on
the pretest measure is different from the group mean on the then
measure for the intervention group but not the comparison group, then
beta change has occurred. Alpha change is then defined as the
52
difference between posttest and then means. When no beta change
is found, alpha can be assessed by comparing both pretest to posttest
and posttest to then ratings.
Terborg, Howard, and Maxwell continue by noting that while the
Howard retrospective then technique is superior to the idealactual
technique of Zmud and Armenakis (1978) and the psychometrically
sound scales suggestion of Lindell and Drexler (1979), when
combined with profile analysis it even more powerful for two reasons.
First, it adds the possibility of assessing gamma change. Second, it
allows the assessment at the individual level as well as group level.
Profile analysis is a method of examining differences between
two patterns of scores on the same set of items or scales. Pairs of
profiles can be compared according to level (similarity of means),
shape (similarity of correlations which are positive and significantly
different from zero), and dispersion (similarity of standard deviations).
The authors contend"... that alpha, beta, and gamma change for any
individual in an intervention or control group can be identified and
measured through selective comparison of profiles for pre, post, and
then ratings made by that individual to a set of items that make up a
single construct or dimension" (Terborg et al. 1980, p. 115).
Individual analysis takes place in the following way. For
individual beta change, a then rating taken at the same time as the
post rating should be on the same scale calibration. Thus, any re
53
calibration that has taken place from pretest to posttest should affect
the level (means) of the individual's profile not the shape or
dispersion. This assumes that scale recalibration is uniform for every
item in the scale. Beta change then is equal to the difference between
individual mean scores across all items on the pretest measure and
the then measure. In mathematical terms, the individual pretest mean
= (a+b+c+...n)/n (where a, b, c...n are items) and the then mean =
(a+b+c+...n)/n. The means across items are compared using a
dependent t test (independence is violated since all ratings come from
the same individual) and significant difference indicates beta change.
Individual alpha change assessment is straightforward. Alpha
change is computed in a similar manner by comparing profile means
for the then and posttest measures. A significant dependent t score
would indicate alpha change as well.
Individual gamma change assessment is more complicated.
Gamma change can be identified through two methods, profile shapes
(correlations) and profile dispersions (standard deviations). For
method one, calculate the correlations for pretest and posttest, pretest
and then, and posttest and then rating pairs. Gamma change would
be indicated by postest.fhen correlations being greater than either the
pretest,posttest or pretest,then correlations, i.e. gamma = rp0St^/7en >
rpre,postancJ Tpre,then
For method two, compare the standard deviations for the
54
pretest, posttest, and then ratings. Gamma change would be
indicated by the posttest standard deviations being approximately
equal to (not significantly different than) the then standard deviations
and both of these significantly different from the pretest standard
deviations, i.e. gamma = (SDp0St = SDthen) SDpre
A third method would be to simply examine both profile shapes
and dispersions.
This method of analysis can be extended to the group level by
aggregating individual level data for both an intervention group and a
comparison groups. To assess alpha change, calculate an index
number for each individual as the difference of the posttest mean and
then mean across items. Next, add the index numbers for all group
members and divide by the group N to arrive at the group index mean
(indexi+index2+...index\)/N = index meangr0Up). Then compare the
intervention group mean and the comparison group and do a ttest.
An alternative to this method is to directly use the t values from the
individual level analysis for comparison. Use the Mann Whitney U test
or the KruskalWallis H test on the ranked t scores for both groups. A
significant difference would suggest greater t values in the intervention
group and imply that the intervention had an effect.
To assess beta change, use the alternative alpha method with
the ranked t values from the comparison of the pretest and then profile
means. Again, apply the MannWhitney U test of the KruskalWallis H
55
test to the ranked t values to determine significance of difference
between the intervention and comparison groups.
The assessment of gamma change is conducted in the same
way. As in the individual level analysis, there are two methods
relating to profile shape and profile dispersion. In method one,
compute the correlations for each pair of pretest, posttest, and then
profile. If group level gamma change has occurred in the intervention
group, the correlation between pretest,posttest and posttest, then
should be similar, but the posttest,then correlation should be higher,
ier post, then > rpre,post rpre,then The comparison group
correlations should be equal.
In method two, compute the raw difference scores of standard
deviations for pairs of profiles, i.e. posttestsD thenÂ§Q, posttestsD
pretestsD. thenQ^ pretestgD Then do a MannWhitney U test on
each of the ranked difference scores for both the intervention group
and comparison group. Gamma change would be indicated by a
significant U value for both the posUhen and then.pretest ranked
differences for intervention group. The comparison group should
show little or no difference.
The Terborg, Howard, and Maxwell (1980) method has several
restrictions and advantages. There are three principal limitations.
First, it is impossible to perform meaningful significance tests at
the individual level. Individual level t values must be judged
56
descriptively not inferentially. This is not the case at the group level
where the assumption of independence is not violated.
Second, the assumption is made that items on the scale being
used to measure change are unidimensional. If they are not, then the
proposed assessments yield erroneous results.
Third, at the group level, a number of significance tests are
used. This could result in a higher probability of Type I error, rejecting
the null hypothesis when it is in fact true.
There are also three advantages of the Terborg, Howard, and
Maxwell method. First, it can assess different types of change
demonstrated by a single individual. Second, it can deal with small
sample sizes. Third, it can examine types of change in a more
independent manner and further examine what may have contributed
to or explains the change.
Bedeian. Armenakis. and Gibson (1980). At virtually the same
time of the publication of the Terborg et al. (1980) method, Bedeian,
Armenakis, and Gibson (1980) proposed a fifth alternative approach
using research design. Citing Howard and Dailey (1979), the authors
note the critical importance of a common metric between the pretest
and posttest when assessing change. Further, if the posttest rating
reflects a shift in the rater's standard of measurement in addition to
actual changes, then the posttest rating will be confounded by this
distortion of the internalized scale (Howard and Dailey 1979, p. 144).
57
The purpose of the Bedeian et al. method is to measure and control
such confounding.
Bedeian et al. build on the earlier contribution of Zmud and
Armenakis (1978) which used "Actual" and "Ideal" scores to detect
beta change. The use of "Actual" and Ideal" scales in combination
relies on two points. First, the use of "Actual" scores (conditions
existing reality) may reflect not only alpha change but also beta and
gamma. However, Ideal" scores (conditions existing in the form of an
idealized mental image) are susceptible to beta and gamma changes
but, by definition, not alpha. In fact, change in idealized perception is
more likely a gamma change than beta. Second, an operational
assumption is made that beta change occurs in an individual's
internalized standard of judging over time but is not expected to occur
between responses collected at the same time. Thus, measuring and
controlling the confounding influence of beta change requires two
things; one, the use of Actual" and "Ideal" scores in combination and
two, the analysis of individual responses.
The procedure for measuring and controlling beta change
consists of four steps.
Step 1. Array the Time 1 and Time 2 Ideal scores by
questionnaire item for each respondent.
Step 2. Compute a linear regression equation, Yj = a + bX\
where Yj is the T2 Ideal and Xj is the T1 Ideal. Examine the standard
58
error of estimate of the equation to see how good the fit is.
Step 3. To determine the presence of beta change, examine
the coefficient and the constant in the equation. If b is not significantly
different from 1 and a is not significantly different from 0 (using a t
test), then no beta change has occurred. If b = 1 but a 0, then scale
displacement has occurred. Scale displacements refers to situations
where the entire scale moves in equal magnitude. If b 1 regardless
of a, then scale stretching or interval sliding has occurred. Scale
stretching refers to lengthening of the psychological space between
intervals. Interval sliding refers to the shifting of some but not all
responses to higher or lower interval category.
As one last precaution to check for item to item shift (which is
one of the forms of beta change but may result in similar means),
compute the correlations between the T1 and Tg Ideals. If the
correlation is weak, then gamma change is suggested and the
procedure stops.
Step 4. Transform the "Actual" scores using the regression
equation. Insert the T1 "Actual" scores into the equation to compute
an adjusted or re calibrated actual. Then test the adjusted Tj against
the Tg to determine alpha change.
Despite the logical elegance of the Bedeian et al. technique, it
drew quick criticism from Terborg, Maxwell, and Howard (1982). Their
criticism rests on four points.
59
First, the Bedeian et al. (1980) technique is limited in its ability
to detect gamma change. The approach relies on the correlations
between pretest and posttest ideal measures. If the correlation is
"weak," then gamma change might have occurred. "Weak" is not
defined. Further, the restricted range of ideal measures may cause
low correlation when minor changes in score occur.
Second, the Bedeian et al. technique uses a restrictive
definition of alpha change. Bedeian et al. argue that alpha and beta
change cannot occur simultaneously. Terborg et al. argue that
Golembiewski et al. (1976a) do not impose this restriction. They
further assert that alpha and beta changes can occur simultaneously
and be correctly assessed.
Third, the Bedeian et al. technique uses "ideal" measures which
suffer from restricted range at the favorable or unfavorable end of a
scale. This range restriction make the detection of mean differences
difficult to detect. Also, slight changes in scores can dramatically
affect correlations and cause the Bedeian et al. gamma change
assessment to be incorrect.
Fourth, the use of regression analysis to correct for beta
change is problematic in three ways. First, since all scores used to
generate the regression equation are from the same individual,"...
the statistical assumption of independence is violated" (Terborg,
Maxwell, and Howard 1982, p. 294). Second, since the sample size in
60
the regression equation is equal to the number of measurement items
and the stability of regression weights depend on the sample size,
short measures will yield unstable regression weights. Third, the
assumption that no beta change has occurred when b= 1 and a=0
depends on a perfectly reliable measure with no error of measurement
in b and a.
Armenakis and Bedeian (1982) responded to the Terborg et al.
(1982) criticisms of their regression technique and offered criticism of
the retrospective then technique. Armenakis and Bedeian
systematically respond to the criticisms and refute them conceding
only that the interpretation of the statistics which indicate the presence
or absence of beta change (b= 1, a=0) should be inferential rather than
descriptive. The authors then mount evidence that the retrospective
then technique is suspect. They conclude that such designs are"...
subject to respondent rhetoric the omission, juxtaposition, and
embellishment of recollections" (Armenakis and Bedeian 1982, p.
297).
Two further responses to the Terborg et al. criticisms were
made indirectly and much later. While they will be reviewed in more
detail in a subsequent section, they bear on this exchange.
Buckley and Armenakis (1986) studied the ceiling effects of the
ideal scale technique in a laboratory environment. Using the
controlled stimulus of a videotaped performance appraisal, they
61
tested scale limitations of ideal perception. They concluded in
response to critics (Terborg, Howard, and Maxwell 1982) that the ideal
scale method does not restrict scale response and thus makes its use
as a means of detecting scale re calibration viable.
Armenakis, Buckley, and Bedeian (1986) conducted laboratory
research on the retrospective then technique. Respondents were
randomly assigned to treatment groups. Stimuli were replicated
exactly and time intervals were systematically varied. The findings
suggest that respondents could not accurately recall their pretest
ratings even after three weeks. The authors concluded the
retrospective then technique should not be used for evaluating
organizational change.
Van de Vliert. Huismans. and Stok (19851. The sixth and final
entry in the category of alternative approaches to assessing alpha,
beta, and gamma change which employ research design is from Van
de Vliert, Huismans, and Stok (1985). These authors propose a
design technique for detecting beta change both at the individual and
group level which they call the Criterion Approach. The technique
incorporates an analysis of a focal variable (F), the target variable of a
change effort, and a criterion variable (C), the variable which functions
as an indicator of change in the measurement scale. The "dynamic
correlations" between the changes in F and C are used to indicate
scale re calibration. Some elaboration will help.
62
Van de Vliert, Huismans, and Stok (1985) begin by touching on
a major unresolved issue in the assessment of types of change. They
assert that beta and alpha change.. preclude the occurrence of a
redefinition of the concepts measured, that is, gamma change" (p.
270). They go on to note that "the absence of gamma change is a
necessary but insufficient condition for measuring other kinds of
change." Terborg, Howard, and Maxwell (1980) and Lindell and
Drexler (1979,1980), for example, contend that alpha change can
indeed cause gamma change. However, Randolph (1982) appears to
demonstrate empirically that the phenomena are separate, an
orthodoxy Golembiewski et al. (1976a) initially stated and with which
Armenakis et al. (1978,1979) and Bedeian et al. (1980) concur. Van
de Vliert, Huismans, and Stok extend the orthodoxy by arguing that
since the absence of gamma change is indicated by the presence of
stable construct validity, repeated construct validation should always
precede the assessment of beta and alpha change.
Van de Vliert, Huismans, and Stok suggest that while the
assessment of gamma change requires independent validation
criteria, the assessment of beta change requires criteria which are
related or reveal variation in the intervals of the scale by which the
focal variable is measured (p. 270). There are three qualities of such
criteria.
First, a criterion variable must function as an indicator of a
63
changed standard of judgment by a respondent. For example, the
amount a time spent by a supervisor each month in conference with
subordinates discussing and solving work related problems may be a
good criterion for indicating change in judgment of a rating of
supervisor support. If preintervention rating of "moderately
supportive" equals 15 minutes of conference time per month and a
post intervention rating of "moderately supportive" equals one hour of
conference time per month, then in all probability the interval has been
recalibrated.
Second, the criterion variable should not be identical with the
focal variable. The issue here is to choose something that will not
become an alternative operationalization of the focal variable. Van de
Vliert, Huismans, and Stok recommend that the criterion variable (C)
and the focal variable (F) should not correlate higher that .60.
Third, the relationship between the criterion variable (C) and
focal variable (F) must bear up longitudinally. This must be tested
empirically. Where C may be a good indicator of F at T, C may not
possess the same goodness at T2.
The criterion analysis, at either the individual or group level,
proceeds in three steps. First, verify the validity of the focal variable
in both the pretest and posttest. Since validity testing requires
comparison, independent validation criteria must be established to
conduct this analysis. Stable construct validity is a prerequisite for
64
any further analysis. If the construct is not stable from the pretest to
the posttest, then gamma change has probably occurred and the
analysis is concluded.
Step two, identify the beta change criteria, which may be more
than one. In fact, the more suitable criteria that can be established,
the greater the likelihood of correctly identifying beta change. The
qualities of these criteria are noted above.
Step three, test the relationship between the criteria and the
focal variable empirically. The quality of the longitudinal relationship
noted above is critical to this analysis. It is the longitudinal shift in the
criterion variable as related to the longitudinal shift in the focal
variable that indicates beta change. Mathematically, Ccjiff=C2Cj and
Fdiff=F2F1 If the dynamic correlation Rcdiff.Fdiffis statistically
significant and alpha and beta change are uncorrelated, then beta
change has probably occurred. Further, Van de Vliert, Huismans, and
Stok suggest that the amount of beta change that has occurred is
indicated by the square of the dynamic correlation (Rcdiff.Fdiff)2 The
proportion of alpha change present can then estimated at 1
(RCdiff,Fdiff)2
Summarv of Research Design Approaches
This section has examined six alternative approaches to
assessing alpha, beta, and gamma change based on research design.
65
These alternative approaches are a direct response to the challenge
from Golembiewski et al. (1976a) to find other ways of operationalizing
the change typology. They extend understanding of the typology and
how it can be operationalized in several ways, but they also leave
some open questions. Six specific observations bear mention.
First. The focus of five of the six alternative design approaches
is beta change, not the gamma change focus that Golembiewski et al.
(1976a) emphasized. Only Terborg, Howard, and Maxwell (1980)
offer an alternative operationalization of gamma change in addition to
their recommendation for assessing the presence of beta change.
Perhaps beta change, which has to do with scale re calibration, is
much more a design than a statistical issue. Perhaps there is an
underlying belief that beta change is a more common and pernicious
threat to the accurate assessment of change than gamma change. Or
perhaps gamma change, as concept redefinition, is a more abstruse
matter than design can accommodate.
Second. Three of the alternative design approaches rely of the
orthodoxy of Golembiewski et al. (1976a) to assess the presence of
gamma change. This involves factor analysis and comparing factor
structure congruence at the group level to determine the presence of
gamma change. Armenakis and Smith (1978), Zmud and Armenakis
(1978), and Armenakis and Zmud (1979) all support the use of factor
comparisons, albeit by different method, to detect gamma change. As
66
mentioned above, only Terborg, Howard, and Maxwell (1980) offer a
truly different operationalization of gamma change. Van de Vliert,
Huismans, and Stok (1985) nearly treat gamma change as a nonissue
suggesting repeated construct validation to test for gamma change.
Bedeian, Armenakis, and Gibson (1980) favor the comparison of ideal
scale correlations, but offer no method.
Third. The six alternative approaches divide neatly in half
based on whether they focus on group or individuallevel analysis.
Those that support the use of factor analysis, even though they focus
on beta change, favor group level analysis. The remainder distinguish
themselves by emphasizing the importance of individual level change
as well as group level.
This dichotomy raises an observation and a question.
Alternative operationalizations of the change typology raise the issue
of level of change" as an analytic issue for the first time in the OD
literature. While individual level assessment of change has always
been possible, group level analysis of change has been the rule in OD
research. Suddenly, the concepts of beta and gamma change appear
to place at the forefront "level of change" as an issue. However, the
dichotomy in this literature leaves unanswered the question about
which is more important.
Fourth. The fourth observation is a paradoxical one. The six
contributions to the literature offer alternative ways to detect the
67
presence or absence of beta and gamma change. However, none
identifies the magnitude or direction of the change. The contribution
which comes closest to enabling a sense of magnitude or direction is
offered by Van de Vliert et al. (1985) who propose that the proportions
of beta and alpha change can be determined. In a less direct way, the
Bedeian et al. method can indicate the magnitude of beta change
through the dissimilarity of the intercept and slope coefficients across
time. However, no mention is made of this in that contribution. In
general, all that can be said of beta and gamma change is "Yes, we
have some."
Fifth. The fifth issue is an unanswered question. Can the
different types of change occur simultaneously? Golembiewski et al.
(1976a) contend that the types are conceptually distinct. Early reports
(Randolph and Edwards 1978; Randolph 1982) subscribe to the notion
of conceptual distinction and treat the assessment of type of change
sequentially. Others (Armenakis et al. 1978,1979; Bedeian et al.
1980) agree. However, Lindell and Drexler (1979,1980) contend that
alpha change can indeed cause gamma change and provide a
simulated example of how this could happen. Terborg, Howard, and
Maxwell (1980) agree with Lindell and Drexler on this point. But
Randolph (1982) demonstrates through careful analysis of real,
longitudinal data that gamma change can occur where no alpha
change is present. Van de Vliert et al. (1985) restate the orthodoxy
68
that the presence of alpha and beta change precludes the occurrence
of gamma change. But they go on to demonstrate through their
criterion method that alpha and beta change occur simultaneously and
the proportion of each can be indicated. Thus, through these
contributions we have no clear answer to the question, only point
counterpoint. There does appear, though, to be an allegiance to the
original Golembiewski et al. position.
Sixth. The literature reviewed here indicates three overall
observations about the change typology. First, it indicates a general
acceptance of the typology as an important contribution to the
conceptualization of change. There is no further argument about
whether or not the different types of change are real. Rather there is
an earnest search for alternative operationalizations. It is safe to say
that such an academic exchange legitimizes the subject. Second,
there is no certainty about what it all means. At this point, beta and
gamma change can be assessed in various ways, but knowledge of
their presence or absence is not meaningful or actionable. Third,
Golembiewski et al., who launched the entire exploration, are
conspicuously absent from this exchange.
Two Approaches That Employ Statistical Methods
Alternative approaches to assessing alpha, beta, and gamma
change, the second distinguishable response to Golembiewski et al.
69
(1976a), fall into two categories (Armenakis, 1988). The last section
reviewed in depth the approaches which use research design. This
section will review the statistical approaches proposed by Armenakis,
Feild, and Wilmoth (1977) and Schmitt (1982).
Golembiewski et al. (1976a) focused on gamma change,
defined as concept redefinition or change in the dimensions of
psychological space, because it is the prime intended consequence of
OD interventions (p. 139),. And since factor analysis is well suited to
"isolate the major dimensions of reality necessary to economically
account for the variance in scores on some large set of variables," (p.
143) factor analysis was also well suited to interpret change in the
dimensions of reality associated with gamma change. Thus, they
settled upon the comparison of factor structures as the statistical
technique for assessing the presence of gamma change.
The Golembiewski et al. (1976a) approach to comparing factor
structures used Ahmavaara's technique (1954). This particular
technique rotates one factor structure into the space of another to
compare them. Two statistics result: intraclass correlation and
productmoment correlation coefficients. The intraclass correlation
coefficient indicates pattern and magnitude similarity. The product
moment correlation coefficient indicates pattern similarity. The square
of the productmoment correlation coefficient indicates the percentage
of common variance. Ahmavaara's technique offers no test of
70
significance, so Golembiewski et al. suggested a cut point that would
indicate change in state, specifically a 50% decrement in common
variance.
Armenakis. Feild. and Wilmoth (1977V Armenakis, Feild, and
Wilmoth suggested the use of a different technique for comparing
factor structures. Named the Coefficients of Congruence, this method
for measuring the similarity of two factor analytic solutions is based on
the work of Harman (1967) and Korth and Tucker (1975). Its output is
a matrix of coefficients of congruence that shows the interrelationship
of the factors of one structure with the factors of another. These
coefficients can be positive, negative or zero like ordinary product
moment correlations. Further, the minimum values which a coefficient
of congruence must attain to reach statistical significance are
computed. Thus, factor congruence or incongruence may be stated
with greater specificity. The rationale and procedure for this method
are found in Armenakis, Randolph, and Bedeian (1982).
Step 1. Determine the determine the number of degrees of
freedom and obtain the minimum t value for the desired level of
significance. The degrees of freedom are equal to the difference
between the number of items factored in the analysis and the number
of factors extracted at Time 1. For example, 30 items yielding 6
factors would result in 24 degrees of freedom. The critical t value is
then determined by referring to a standard table.
71
Step Two. Compute the minimum value (MINVAL) that the first
coefficient of congruence must attain to be significant at the p<.05
level. This computation is based on a formula developed by Korth and
Tucker (1975) and is computed by a program developed by
Armenakis, Feild, and Wilmoth (1977). If the coefficient of congruence
is equal to or greater than the MINVAL, the compared factors are
considered to be similar at a 95% confidence level.
This method has two restrictions. First, the factor structures
being compared must have an equal number of factors. Second, the
items included in the factor analyses must be the same both times and
be arranged in the same order.
Several studies have employed the Coefficients of Congruence
technique rather than Ahmavaara's technique to test for gamma
change (Zmud and Armenakis 1978; Armenakis and Zmud 1979;
Bedeian, Armenakis, and Gibson 1980; Armenakis, Randolph, and
Bedeian 1982; Randolph and Elloy 1987). One study (Armenakis,
Randolph, and Bedeian 1982), which compared the results of both
techniques on the same data sets, will be reviewed later.
While the Armenakis, Feild, and Wilmoth (1977) technique
qualifies as an alternative statistical approach for assessing gamma
change, it is not conceptually different from Golembiewski et al.
(1976a). It still depends on the congruence of factor structures at the
group level to determine the presence or absence of gamma change.
72
What it does contribute is a more precise way of determining factor
congruence or incongruence than the gross "cut points" suggested in
the original research.
Schmitt (1982). Schmitt (1982) and Schmitt, Pulakos, and
Lieblein (1984) use a very different approach and a special program
called LISREL V to conduct their analysis of the change typology.
They use analysis of variance and covariance and examine the
similarity of the resulting matrices to assess the presence of alpha,
beta, and gamma change. The Schmitt approach involves the
sequential testing of five hypotheses on a model of measurement
represented by the equation X = Lf + u where X is the vector of
observed response, f is the common factor loadings, u is the unique
factors, and L is the set of factor loadings. Three of the hypotheses
test for gamma change while two test for beta. Alpha change can also
be computed from the means of the factor structures. The Schmitt
technique, while in some ways conceptually similar to the factor
structure comparison techniques of Golembiewski et al. (1976a) and
Armenakis et al. (1977), tests for both beta and gamma change at the
group level, an advantage neither of the other techniques shares. The
technique's five steps are outlined below.
Step 1. Test the hypothesis that the pre and post variance
covariance matrices are equal. If they are significantly different, there
are four implications.
73
1. The posttest factor pattern is different and some
unspecified gamma change has occurred.
2. The scale units are different and some unspecified beta
change has occurred.
3. The uniqueness of the matrices are different.
4. Some combination of the three has occurred.
Overall, analysis step indicates a gross presence or absence of
change types beyond simple alpha.
Step 2. Test the hypothesis that the number of factors and the
factor loadings are the same for both the pretest and posttest
measurements. This is the same premise as the Golembiewski et al.
use of Ahmavaara's (1954) technique for assessing gamma change.
This step tests the extent to which both measures are measuring the
same concepts. Differences in number of factors or factor loadings
would indicate gamma change.
Step 3. This step, also related to gamma change, tests the
hypothesis that the variancecovariance matrices of the common
factors are similar for the pre and post measures. This test is
accomplished by constraining the corresponding elements of the
variancecovariance matrix to be equal. Any difference in the
observed matrix and the model matrix would represent gamma
change. A Chisquare test is used to statistically assess the equality
constraint on the matrices.
74
Step 4. Test the hypothesis that the factor loadings are equal
across time. "Since the factor loadings are the maximum likelihood
estimates of the regression of observed scores on true scores, the
constraint of equality across time tests the equality of the scaling
units" (Schmitt 1982, p. 350). Schmitt notes that this test is not
concerned with the equality of scaling units for items, but rather for
factors. Therefore, multiple item measurement of constructs is
essential. A significant drop in the fit between the observed and the
reproduced matrices indicates beta change. Again, the significance of
the loss of fit can be assessed using a Chisquare test. The absence
of beta change implies that the construct variances do not change as
a result of the intervention.
Step 5. Test the hypothesis that the uniqueness associated
with the measured variables are invariant across time. This test
involves an equality constraint on similar pretestposttest elements of
the diagonal matrix of the variance of the unique factors. "A significant
Chisquare or a significant decrease in the variance accounted for by
the model would signal a difference with respect to the reliability of
measurement before and after intervention" (Schmitt 1982, p. 351).
Schmitt's technique is both attractive and troublesome at the
same time. It is attractive in three ways. First, it allows for concurrent
assessment of both gamma and beta change. Other methods require
sequential testing with different procedures. Second, the technique
75
appears to be sensitive and precise in its assessment of type of
change. The use of Chisquare and degrees of fit indices represent
more stringent tests by which type of change can be assessed. Third,
Schmitt's technique is very quantitatively rigorous. The requirement of
testing five sequential hypotheses using analysis of variance and
covariance in conjunction with factor analysis sets this technique apart
in its quantitative rigor. However, it is for this last reason that the
technique is also troublesome.
Schmitt's technique uses a special computer program named
LISREL. LISREL is not part of either SPSS of SAS. It is not readily
available on most university mainframe software menus. If it is
available, it is an ancillary program with separate documentation. All
of this aside, it is extremely difficult to use. Conversation with Dr.
Schmitt revealed that using LISREL requires that the user become
familiar with the nuances of the variancecovariance matrices report
and the attendant Chi square statistics and degrees of fit indices.
While LISREL is very powerful in its analytical utility, it is also very
formidable even forbidding in its use.
Summary of Statistical Methods
This section has examined the two alternative approaches to
assessing alpha, beta, and gamma change based on statistical
methods. Like the alternative approaches based on research design,
76
these approaches are also a direct response to the challenge by
Golembiewski et al. (1976a) to find other ways to operationalize the
change typology. The review of these approaches suggest five
observations regarding their responsiveness to the challenge
extended by Golembiewski et al.
First. There is really nothing new here. Both techniques rely
on factor analysis as the basis for assessment. Schmitt adds analysis
of variance and covariance to the analysis of factor structures thereby
enhancing rigor. Further, Schmitt's technique focuses on and permits
the concurrent assessment of both beta and gamma change.
However, in the end, the approaches suggested by Armenakis, Feild,
and Wilmoth (1977) and Schmitt (1982) are only refinements of the
Golembiewski et al. approach. They tacitly support the position that
factor analysis is the best choice of method to discern changes in the
dimensions of reality without engaging in the same kind of discourse
as Golembiewski et al. concerning choice of method.
Second. Both approaches are capable only of group level
analysis. Further, since they use factor analysis as the basis for
assessment they require a large N. Observations on the stability of
factor structures (Randolph 1982) indicate a minimum 2:1 ratio of
respondents to items to achieve stable analyses. Terborg, Howard,
and Maxwell (1980) suggested a ratio of 3:1. Such a criterion makes
the application of these approaches impractical for small groups and
77
ignores the possibility or reality of individual level change altogether.
Third A major advantage of both approaches is that they can
be applied to simple longitudinal data. They do not require additional
measurement such as repeated Ideals (Zmud and Armenakis 1978;
Bedeian, Armenakis, and Gibson 1980) or retrospective then (Terborg,
Howard, and Maxwell 1980). This significantly simplifies procedures
and even makes it possible to assess old, existing data sets.
Fourth. The application of the approaches requires a high level
of sophistication in statistics and their interpretation. The average OD
practitioner, while perhaps conversant in basic statistics, is probably
uninitiated and unskilled in the application of factor analysis. The
comparison of factor structures requires yet a another level of
sophistication. The use of LISREL to compare variancecovariance
matrices severely stretches the skill and understanding of anyone who
is not a statistician or psychometrician. What this means is that the
statistical approaches to assessing the change typology do not belong
to everyone. In fact, they begin to define the assessment of beta and
gamma change as the domain of academicians with specialized skill
and access to sophisticated computer applications. This does not
extend the cause of increasing the rigor of OD research (Kahn 1974;
Porras and Berg 1978; White and Mitchell 1976).
Fifth. Neither approach helps with understanding the
magnitude or direction of change. As in the case of alternative
78
approaches based on design, all that can be said of beta and gamma
change is "Yes, we have some."
This section has dealt with alternative approaches to
operationalizing the change typology employing statistical methods.
The final section of this review addresses diverse operational
approaches including comparison of techniques, laboratory studies
and simulations, the search for surrogate variables, and conceptual
analysis of the academic response.
Diverse Operational Approaches
Comparisons of Techniques
At present, there are three research studies which compare
techniques for assessing beta and gamma change. One study
(Armenakis, Randolph, and Bedeian 1982) compares two factor
analytic approaches to assessing gamma change. Armenakis (1988)
categorizes this as a comparison of statistical techniques. The other
two studies (Schmitt, Pulakos, and Lieblein 1984; Randolph and Elloy
1987) compare statistical techniques with design techniques and draw
conclusions about convergent validity.
Armenakis. Randolph, and Bedeian (1982V Armenakis,
Randolph and Bedeian (1982) analyzed two separate data sets of
before and after measures using both Ahmavaara's (1954)
Transformation method and the Coefficients of Congruence method.
79
For one data set yielding two factors, they found no gamma change
and perfect agreement between the methods for the similarity of the
before and after factor structures. For the second data set yielding
five factors, they found very little agreement between the methods.
The Coefficients of Congruence method found the before and after
factors structures similar. The Transformation method found a
substantial decrease in shared variance. The authors concluded that
the mathematical differences between the methods may be
exacerbated as the number of factors increases.
Schmitt. Pulakos. and Lieblein M984T Schmitt, Pulakos, and
Lieblein (1984) analyzed a single data set consisting of before and
after data from an experimental and a control group. They compared
the Zmud and Armenakis (1978) Ideal scale technique, the Terborg,
Howard, and Maxwell (1980) retrospective then technique, and
Schmitt's (1982) analysis of variance and covariance matrices
technique. They discovered no alpha, beta, or gamma changes using
the Ideal scale technique. They found some gamma change for both
the experimental and control groups, beta change in one factor for the
comparison group, and alpha in two factors for the comparison group
using the retrospective then technique. The analysis of variance and
covariance matrices detected gamma change in the control group but
not in the experimental group. The authors concluded there is little
agreement across methods for detecting gamma change.
80
Randolph and Elloy (1987) tested longitudinal data for beta and
gamma change using the Coefficients of Congruence method and the
retrospective then technique. Based on the goals of the intervention
effort, Randolph and Elloy hypothesized that gamma change should
be detected in five of the seven factors measured by the research
questionnaire. The factor analytic method detected gamma change in
four of the five factors where it was hypothesized and one where it
was not. The retrospective then aggregated correlations approach
(profile shape) detected gamma change for three of the five factors
where it was predicted and one where it was not. The aggregated
standard deviations approach (profile dispersion) detected gamma in
only one factor where is was predicted. Randolph and Elloy ask which
retrospective then technique is correct, but admit surprise in the
similarity of conclusions between the factor analytic method and the
then correlations technique given the fundamental differences in
statistical approach.
Laboratory Studies and Simulations
Terborg (1984). Terborg (1984) conducted a simulation of
change in pretest and posttest measures with two different scale
lengths (6 items and 16 items). Further, he manipulated the
responses to simulate beta and gamma changes and analyzed the
data at the individual level using the then technique and the Bedeian,
81
Armenakis, and Gibson (1980) regression technique. The results
were very mixed. One form of beta change, scale displacement, was
detected only 50% of the time. Gamma change was detected only
42% of the time where it was simulated. Terborg concluded that the
problem of correctly identifying types of change is more complicated
than first envisioned.
Buckley and Armenakis (1986T Buckley and Armenakis (1986)
studied the ceiling effects of the ideal scale technique in a laboratory
environment. Using the controlled stimulus of a videotaped
performance appraisal, they tested scale limitations of ideal
perception. They concluded in response to critics that the ideal scale
method does not restrict scale response and thus makes its use as a
means of detecting scale re calibration viable.
Armenakis. Bucklev. and Bedeian f1986T Armenakis, Buckley,
and Bedeian (1986) conducted laboratory research on the
retrospective then technique. Respondents were randomly assigned
to treatment groups. Stimuli were replicated exactly and time intervals
were systematically varied. The findings suggest that respondents
could not accurately recall their pretest ratings even after three weeks.
The authors concluded the retrospective then technique should not be
used for evaluating organizational change.
82
Surrogate Variables
Golembiewski (1986) and Golembiewski and Munzenrider
(1986) extended the research on the change typology in a very
different direction. They explored the phase approach to
psychological burnout as a surrogate variable for gamma change.
Using the Maslach Burnout Inventory (Maslach and Jackson 1982),
these researchers suggest that the phases of burnout, especially in
the extreme, imply differences in state (gamma change) rather that
differences in degree (alpha). If, they argue, such surrogacy can be
established, several of the difficulties associated with testing for
gamma change (e.g. a large N, group level vs. individual level
analysis, measurement technology, etc.) may be addressed.
Golembiewski and Munzenrider (1988.1989.19901. In a trio of
followup studies, Golembiewski and Munzenrider (1988,1989,1990)
examine the matter of surrogacy more closely. The studies represent
successively narrower examinations of the relationship among gamma
change, the phase model of burnout (Maslach and Jackson 1981), and
work site descriptors which covary with burnout.
The purpose of the first study (1988) was to examine evidence
that the MBI 8phase model measures at least one difference in state
rather than mere differences in degree. The MBI measures three sub
dimensions of burnout, depersonalization, personal accomplishment
(reversed scale), and emotional exhaustion. Combinations of high
83
and low scores on the three subdomains yield the eight phases of
burnout (Golembiewski and Munzenrider 1984). The researchers
tested a sample of 1590 respondents for gamma differences in phase
assignment on the MBI. A second measure, the General Satisfaction
scale from the Job Diagnostic Survey (Hackman and Oldham 1980),
was used as a co variant. Six pairs of subpopulations were
compared.
1. Top vs. bottom 10% on General Satisfaction scale,
Phases 1 vs. 8 on MBI
2. Top vs. bottom 20% on General Satisfaction scale,
Phases 1 vs. 8 on MBI
3. Top vs. bottom 30% on General Satisfaction scale,
Phases 1 vs. 8 on MBI
4. All responses in Phase 1 vs. Phase 8
5. Phase I responses vs. Phases 28 responses
6. Phase 8 responses vs. Phases 17 responses
Using Ahmavaara's (1954) technique for comparing factor
structures, Golembiewski and Munzenrider discovered gamma change
(defined as a 50% decrement in shared variance between factor
structures) in two comparisons. In a comparison of Phases 1 and 8
and decile rankings on the General Satisfaction scale, the top 10% in
Phase 1 vs. the bottom 10% in Phase 8 showed a shared variance of
47%. A comparison of all respondents in Phases 1 vs. all respondents
84
in Phases 28 showed a shared variance of 45%. The authors
conclude that the data suggest at least one discontinuous jump or
change in state along the eight phases.
In the second study, Golembiewski and Munzenrider (1989)
extended their investigation of state differences between phases with
four sets of comparisons. First, the seven adjacent phases were
compared. Second, the 21 other possible paired comparisons (Phase
1 vs. 3,1 vs. 4,1 vs. 5, etc.) were made. Third, some select
combinations of phases were compared. Fourth, aggregates of
phases were compared.
The results were dramatic. In the first comparison, the average
amount of shared variance between the seven adjacent phases was
33%, far below the required 50% defined to indicate gamma change.
In the second comparison, 18 of the 21 other possible comparisons
not covered in the first comparison achieved a decrement in shared
variance of 50% or better. The third comparison fared less well. Of
seven selected combinations of phases only one comparison, Phase 1
vs. Phases 28 (as in the first study) showed a decrement below 50%.
In the fourth analysis of aggregated phases, a gamma difference was
found for Phases 13 vs. Phases 45 (40% shared variance). No
gamma differences were found for comparisons of Phases 13 vs.
Phases 68 or Phases 45 vs. Phases 68 although the incongruence
in factor structure was substantial, 77% and 53% respectively. These
85
