The Alpha, Beta, Gamma Approach to Measuring Change and its use for Interpreting the Effectiveness of Service Quality Programs.

1. Introduction

Changing the attitudes and behaviour of customers is at the heart of much of what marketers pursue. Marketing managers try to influence various cognitive, affective and conative predispositions of consumers, and thus buyer behaviour. Self-report measures are the most widely used research technique to evaluate the effectiveness of various change programs.

Given the importance of these interventions, there is relatively little discussion in the marketing literature about how to measure the different types of change that may occur. For example, over the last decade Till and Priluck (2000) and others have studied how classical conditioning techniques may change people's attitudes toward products and brands. However, this research is not grounded in any formal change measurement framework. Without such a measurement model it is difficult to gauge the success or otherwise of marketing interventions. Also, in Swartz and Iacobucci's (2000) recent handbook on services marketing, not one of the 29 papers mentions how to measure the changes in service quality caused by a management intervention.

This paper argues that simple, common-sense ideas of how to measure change in marketing settings can often result in ambiguous (and possibly incorrect) conclusions being drawn. To illustrate the conceptual model of change advanced in this paper, a hypothetical example of measuring change in service quality (SQ) is provided. This is a topic of considerable interest to academic researchers, marketing managers, and market research firms. Over the last decade, researchers have refined the concept of SQ and the set of factors that a company can change to improve customers' perceptions of its service. Marketing managers have been quick to apply these ideas and many market research firms are now employed to track the effectiveness of these service improvement programs.

This paper has two objectives. First, to introduce a new conceptualisation of change to marketing researchers. This framework has been used in organisational research to interpret the effects of various types of interventions to change people's attitudes and behaviours. When applied in marketing contexts, it provides a logical, integrated framework to: (a) search for different types of change; and equally importantly (b) to report these changes. The second objective is to apply this more extensive conceptualisation of change to the very popular SERVQUAL measure of service quality developed by Parasuraman, Zeithaml and Berry (PZB) (1985; 1988; 1994a; b; c); Parasuraman, Berry and Zeithaml (1991; 1993); Zeithaml, Berry and Parasuraman (1990; 1996).

The contribution from the first objective is to provide a parsimonious review of an area of organisational research that has direct relevance for marketers. The change measurement model is then used as a broad framework to design, interpret, and communicate the results of change research, and thus enable marketing researchers to meta-analyse disparate research findings. The contribution from the second objective is to help academics and practitioners to better interpret measures of service quality.

2. Types of Change

While market research textbooks and review papers discuss measurement and scaling, they typically do not articulate how to measure change, especially on the self-reported scales used so often by academics and practitioners (e.g. Aaker, Kumar & Day 1995; Carroll & Green 1995; 1997). Despite its obvious importance, this topic only began to appear in the marketing journals in the 1990s (e.g. Peter, Churchill & Brown 1993). In psychology however, the measurement of change has been a topic of critical discussion for decades (e.g. Cronbach & Furby 1970).

Within the discipline of consumer behaviour, there is the temptation, and often the practice, of measuring changes in psychological variables with difference scores. Much of this research relies on the following methodology: respondents provide scale or item scores for pairs of observations that are considered to be conceptually linked. The difference between these linked observations represents a meaningful construct. For example, in the service quality paradigm, the difference between a person's service expectations and the subsequent service delivery, is said to measure SQ (Zeithaml, Parasuraman & Berry 1990, p. 18).

While many types of difference scores have a seductive face validity, it is well documented that their use is compromised by some basic technical problems, namely, potential unreliability, systematic correlation with their components, and spurious correlation with other variables (e.g. Cronbach & Furby 1970; Johns 1981; Peter, Churchill & Brown 1993). Their use for measuring SQ by instruments such as SERVQUAL has also been questioned (Cronin & Taylor 1992; Teas 1993; Brown, Churchill & Peter 1993), and defended (Parasuraman, Berry & Zeithaml 1993; PZB 1994a).

Arguing about the psychometric properties of scale measures and the `costs' of violating the assumptions underlying their use is not new in marketing (e.g. Dowling & Midgley 1991). And while these discussions are useful, new insights can sometimes be gained by looking at the issues from a different perspective. Here, the point of entry into the debate about measuring changes in consumer self-reports is by distinguishing among three different types of change that a marketing intervention such as a service improvement program may achieve. Based on these types of change, a set of measurement procedures is discussed and illustrated.

In an award-winning paper, Golembiewski, Billingsley and Yeager (1976) suggested that when researchers want to use self-report measures to calibrate the impact of an organisational intervention, then three types of change can occur:

1. Alpha Change--which occurs when the meaning of the construct to the respondent and the psychological interpretation of the units of measurement on the `measurement rule' stays the same, but the level of the measurement on this rule changes;

2. Beta Change--occurs where the meaning of the construct to the respondent stays the same, but the respondent subjectively recalibrates the `measurement rule' (Alpha change is assumed not to occur); and

3. Gamma Change--occurs where the conceptual domain of the construct (content and/or dimensionality) is radically altered in such a way as to make the previous meaning of the `measurement rule' irrelevant.

A marketing intervention such as a new advertising campaign, a price cut, or a service improvement can be designed to create alpha, beta or gamma change. However, many measures of change derived from consumer self-reports will be analysed in such a way that they can only detect alpha change. For example, many companies measure SQ change using difference scores across the items in a battery of scales (simple absolute differences, differences between profiles of attributes, or signed differences). The implicit assumption here is that beta and/or gamma change has not occurred. However, if the observed change is substantial, this assumption may be unfounded.

Because change in the measuring intervals of many marketing scales can be an intended effect of a company intervention, beta change goes beyond notions of the test-retest reliability of the measure. For example, Boulding, Kalra, Staelin & Zeithaml (1993) suggest that consumers continually update their expectations whenever they receive relevant information about a particular product or service. Thus, the expectations component of SERVQUAL, or any other expectations-based SQ measure of marketing effectiveness, is designed to change over time. This is an example of beta change. Hence, what some researchers may label as a defect of the measuring instrument might be better interpreted as a desired consequence of, in this case, improved service delivery. Because consumer self-reports are grounded in the socio-cultural context of the respondent, any marketing improvement program can easily lead to a redefinition of the components of its measure.

Unlike beta change that refers only to variations in the intervals measuring a relatively stable dimension of a marketing construct such as attitude to an advertisement or SQ, gamma change involves a basic redefinition of the construct. For example, many service quality programs introduced into companies try to create a new `social order' of service delivery. For example, Jan Carlzon's (1987) book Moments of Truth described such a program at Scandinavian Airlines. The development of such a new service culture would be intended gamma change. If this new service culture radically alters delivered customer service, then an old measure of SQ may now be inappropriate to measure both employees' and customers' new service experiences. The three types of change are summarised in table 1.

Table 1
Defining Alpha, Beta and Gamma Change

                Criteria                      Type of Change

                                           Alpha   Beta   Gamma

Change in the Level of the Measure          Yes    No      No
Change in the Calibration of the Measure    No     Yes     No
Change in the Meaning of the Construct      No     No      Yes

To the reader familiar with the Cook and Campbell (1979) tradition of quasi-experimental design and analysis of self-report data, alpha change would be interpreted as `true' change caused by the experimental manipulation, beta change as an instrumentation bias, and gamma change as a problem of construct validity.

This traditional interpretation assigns beta and gamma change to the error term in a measurement equation of the form:

(1) Observed Change = True Change + Error

When equation 1 is used as the model of change, researchers implicitly define alpha change as true change, and try to ensure that `measurement error' does not swamp their ability to detect this change. The approach advocated here however, separates out the three types of change so that they can be `designed into' research studies, and appropriate methods used to partition out their effects in the observed change scores. This suggests the expansion of equation 1:

(2) Observed Change = f(Alpha Change and/or Beta Change or Gamma Change) + Error

The `and/or' and `or' in equation 2 are important parts of this equation. As described in the next section, the presence of gamma change makes the search for alpha and beta change meaningless.

3. Measuring Alpha, Beta and Gamma Change

Theorists make the point that any measure is both conceptual and empirical (e.g. Bagozzi 1984). Hence, each marketing research study needs to stipulate what type of change the intervention is designed to achieve. This will help determine the appropriate research design and statistical techniques to partition the total observed change into alpha, beta and gamma change.

Figure 1 illustrates the general procedure for operationalising equation 2 for a SQ intervention. The first step is to stipulate which type(s) of change are desired from the organisational intervention. This, plus the context in which the research takes place, will suggest an appropriate (quasi)experimental design and measure of SQ. For example, to evaluate the effectiveness of the installation of a new PABX/PBX system to speed up answering customer calls would require only simple before--after measures of call waiting times. If however, the perceptions of customer call waiting times was the construct of interest, then an experiment would need to be designed which took account of contextual factors such as whether or not a customer had called before (the first call may set their expectations for subsequent calls), and how long ago had they last called (recent experiences may be more vivid than distant ones).

[ILLUSTRATION OMITTED]

The third step is to search for gamma change because if it is found, then by definition, alpha and beta change cannot be measured because of the noncomparability of SQ constructs. If gamma change is found the researcher can analyse the pre and post-measures separately and develop an explanation for why the SQ construct has so fundamentally changed.

If significant gamma change is not detected, the next step involves benchmarking the pre and post-test measures of SQ. For reflective measures various reliability procedures are available to estimate the amount of measurement error associated with operationalising the SQ construct before and after the intervention.(1) If both SQ constructs have similar measurement properties, then the magnitude of alpha change can be estimated. If they are dissimilar, then beta change can be assumed to have occurred. It may then be possible to adjust either the pre or post measure such that alpha change can also be estimated. (For a discussion of this issue see Bedeian, Armenakis & Gibson 1980.)

In the original Golembiewski, Billingsley and Yeager (1976) paper, the main focus was on detecting gamma change. This was measured as a change in the factor structure between multi-item pre and post measures. At this time, the authors stated that they were unsure about how to de-couple beta change from gamma change using their current approach. This shortcoming stimulated numerous researchers to propose a variety of alternatives procedures. While there are differences among the proposed methods, one such approach by Schmitt (1982) illustrated how LISREL could be used to measure and de-couple gamma and beta change. Since this time, confirmatory factor analysis has been a widely used general measurement methodology (e.g. Millsap & Hartog 1988; Schaubroek & Green 1989). (For reviews of the types of change and their measures, see Armenakis 1988; Dowling 1991; Thompson & Hunt 1996.)

Using Schmitt's (1982) confirmatory factor analysis approach, step 3 in figure 1 is operationalised as follows:

a. Test the equality of the variance-covariance matrices of the pre and post multi-item SQ measures. A significant difference implies the existence of gamma change, and possibly beta change and an unreliable SQ measure.

b. Test the equivalence of the before-after pattern of factor loadings. Nonequivalence signals beta change.

c. Evaluate differences in the factor means to detect alpha change.

For someone familiar with common and/or confirmatory factor analysis, the logic of this approach is straightforward. However, the use of longitudinal covariance structure models can be complicated (e.g. Joreskog 1979).

Because the contribution of this paper is conceptual rather than statistical, in the next section, that reviews some of the PZB empirical findings, the focus is on the interpretation of change rather than the intricacies of the estimation of such change.

4. Interpreting Changes in Service Quality

The PZB model defines service quality across five dimensions (factors), each of which is measured by a multi-item scale:

1. Tangibles--the appearance of facilities, equipment and personnel.

2. Reliability--the ability to perform the promised service and solve problems.

3. Responsiveness--the willingness to help customers and provide prompt service.

4. Assurance--the knowledge and courtesy of employees and their ability to convey trust and confidence.

5. Empathy--the provision of caring, individualised attention.

Since the introduction of this model of service quality, much of PZB's research and most challenges to their approach have used data about one or more companies at a single point in time (e.g. Mittal & Lassar 1996). Few studies report service quality measures at more than one point in time, and those that do (e.g. Newman & Cowling 1996; Triplett, Yau & Neal 1994) don't analyse their data in a way that could detect alpha, beta and gamma change.

To illustrate the interpretation of SQ change in figure 1, two PZB studies are used (viz., PZB (1988) and Parasuraman, Berry and Zeithaml (1991).(2) In each study, the authors analyse data from a number of companies, one of which is a retail bank. To develop a context for the following discussion, assume: (i) that it is the same bank; and (ii) that in the three years between the two reports, this company had embarked on a service improvement program. Table 2 reports a comparison of selected statistical results from each study.(3)

Table 2
SERVQUAL Changes Over Time

                   Factor Structure 1988, 1991

                          F1   F2     F3   F4   F5

Tangibles        Q1       34   28     --   --   --
                          99
                 Q2       64   --     --   --   --
                          52   29
                 Q3       39   --     --   28   --
                               81
                 Q4(*)    28   --     --   28   --
                               88

Reliability      Q5       --   72     --   --   --
                               76
                 Q6       30   63     --   --   --
                               (46)
                 Q7       --   71     --   --   --
                               92
                 Q8       --   80     --   --   --
                               8s
                 Q9       --   39     --   --   --
                               67

Responsiveness   Q10      --   --     37   --   --
                                      50
                 Q11      --   --     55   --   --
                                      87
                 Q12      --   --     62   --   --
                                      97
                 Q13      --   --     69   --   --
                                      82

Assurance        Q14      --   --     --   68   --
                                           56
                 Q15      --   --     --   84   --
                                      44        30
                 Q16      --   --     --   72   --

                                           60
                 Q17(*)   --   --     --   64   --
                                           61

Empathy          Q18      --   --     --   --   37
                                                79
                 Q19      --   --     --   --   48
                                                39
                 Q20      --   --     --   --   41
                                                89
                 Q21      --   --     --   --   33
                                                69
                 Q22      --   --     --   --   68
                                      28        51

Regression Equations--Overall Service Quality

1988             OSQ = 0.13T + 0.39Rel.(**) + 0.07Res. + 0.13A + 0.01E

1991             OSQ = 0.01T + 0.39Rel.(**) + 0.24Res.(**) +
                   0.14A + 0.13E(**)

                              Reliabilities

                          1988   # Items   1991   # Items

Tangibles        Q1       0.52   4         0.85   4(*)
                 Q2
                 Q3
                 Q4(*)

Reliability      Q5       0.80   5         0.92   5
                 Q6
                 Q7
                 Q8
                 Q9

Responsiveness   Q10      0.72   4         0.92   4
                 Q11
                 Q12
                 Q13

Assurance        Q14      0.84   4         0.90   4(*)
                 Q15
                 Q16
                 Q17(*)

Empathy          Q18      0.71   5         0.88   5
                 Q19
                 Q20
                 Q21
                 Q22

Regression Equations--Overall Service Quality

1988                                       [R.sup.2] = 0.28

1991                                       [R.sup.2] = 0.68

Note: (*) = New item in 1991; and
(**) = significant at p < 0.05

For the sake of discussion, let us also assume that the organisational intervention was designed to achieve a major change in customers' perceptions of service quality. Hence, in terms of step 1 in figure 1, we need to search for beta and gamma change. SERVQUAL is the chosen measure of service quality because it is widely used `generic' measure. To complete the description of step 2, let us assume that the quasi-experimental design to combine the two studies is a one-group, pretest--post-test design, namely, [O.sub.1] X [O.sub.2]. While this is a frequently used design in the social sciences, it has a number of weaknesses which, while they need not concern us here, would need to be ruled out before analysing data in a real study (Cook & Campbell 1979).

Following step 3 in figure 1, and the logic of Schmitt's (1982) approach, we first search for gamma change by comparing the factor structures in 1988 and 1991. The five-factor solution derived in 1988 is considered the `target' structure. It was derived from 22 items and explained 56.0% of the variance in SQ. Also, the average pairwise correlation among the five factors following an oblique rotation was 0.21.

In 1991, pre-testing of the self-completed questionnaire led to some changes in the way the SERVQUAL instrument was operationalised. First, the expectation statements were reworded to be less normative in orientation. Second, six negatively worded items in the 1988 instrument were changed to a positive format. Third, two new items were substituted for two of the original items, namely, Q4 and Q17 in table 2.

As seen in table 2, there is some difference between the factor structures derived in 1988 and 1991. Also, the revised items now explain more variance in SQ (71.6%), and they are more highly intercorrelated (approximately 0.37). In the 1991 paper, Parasuraman, Berry and Zeithaml note these differences and estimate a six factor solution to see if it provides a better interpretation.(4) Their conclusion is that `while the interdimensional overlap in the 1991 SERVQUAL scale is somewhat greater than in the 1988 scale, the refinement still reflects the basic five-dimensional structure of the original scale with one key exception--namely, the dichotomization of tangibles into two subdimensions' (1991, p. 431). The critical question for both the researcher and the bank, is whether they would agree with Parasuraman, Berry and Zeithaml and assume that the two SERVQUAL scales are essentially the same? That is, that gamma and beta change were negligible. If so, then a comparison of the scores of the five pre-post dimensions of SQ would signal whether or not service had improved, that is, had it induced alpha change?

Two types of criteria can be used to help assess the assumption that the SERVQUAL scale has remained essentially unchanged over the three-year period. First, the nature of the service improvement program and its communication to customers could help interpret the changes in the coefficients reported in table 2. For example, was the new service program designed to: (a) reposition the bank's services (gamma change); or (b) change consumers' definition of the meaning of one or more dimensions of SQ (beta change); or (c) was it merely designed to improve the current type(s) of service (alpha change)? As the PZB papers provide no information on these issues, no further comment is made about them, except that, it is this type of `contextual' information which guides much of the search for, and interpretation of different types of change.

Second, in the change measurement context outlined previously, the alterations made to the 1991 version of the SERVQUAL scale should alert the researcher to expect some beta change. Also, in this case, the theory of measuring service quality suggests that gamma change may occur. For example, while the rewording of the expectations questions was designed primarily to overcome ratings which clustered around six and seven on a seven point scale (as revealed in the questionnaire pretest), it is possible that stating expectations in a less normative tone may fundamentally change the way SQ is interpreted by customers (e.g. Boulding et. al. 1993; PZB 1994b).

Because most of the factor loadings are of the same sign and of a similar magnitude across the two scales, and the reliabilities of all five dimensions are significantly higher in 1991, Schmitt's (1982) confirmatory factory analysis approach would suggest that the differences in factor structure shown in table 2 are probably the result of a small amount of gamma change. This is associated primarily with the tangibles dimension of SQ, but is also reflected in a 15% increase in the average intercorrelation among the five dimensions. Based on this assessment, the two regression equations shown at the bottom of table 2 suggest that the (fictitious) service improvement program changed customers' perceptions of Overall SQ from being based primarily on reliability (the only significant variable in 1988) to now being based on reliability, responsiveness, and empathy. In a practical sense, this is a much different, and more important insight than a finding which merely indicates that the overall level of SQ has (say) improved, and/or that one or more of the five dimensions of SQ has improved (alpha change). For example, three dimensions of SQ could allow the bank to appeal to a wider range of customers than one.

In many applied settings, alpha change is the sole criteria used to assess the effectiveness of SQ. Had this been the change model used to evaluate the bank's service improvement program, it probably would have resulted in a poor evaluation. For example, it is likely that there would have been some debate about how the structure of the 1991 SERVQUAL scale differed `slightly' from that of the 1988 version. It is not unreasonable to assume that this discussion would have led to the conclusion that the 1991 scale was `better' than the earlier version (say based on the higher reliabilities). At this point, the pre-post average ratings of SQ across its five dimensions would (hopefully) show some significant improvement. The bank managers' confidence in the results of the SQ intervention would be confirmed.

However, with the alpha, beta, gamma change model as the methodological framework, a different, and more complex story would be emerge. First, the service improvement program, and/or the actions of competitors has changed the way customers now view SQ. Hence, rather than say that SQ has improved, a more accurate conclusion is that the bank's customers have (partially) redefined SQ. In 1988 reliability was the single key issue. In 1991, reliability, responsiveness and empathy are three related key issues. Also, each of these is defined (slightly) differently (as evidenced by the different pattern of factor loadings [is greater than] 0.5) than they were in 1988. Hence, the bank will need to modify its approach to delivering these aspects of customer service.

5. Conclusions

A review of the academic literature, and discussions with market research suppliers, suggests that the methodology used to track SQ changes is often based on a limited conceptualisation of change. Typically, market research firms measure change by calculating the difference between the same (multi-item) measure of SQ over time. As outlined earlier, this simple procedure is flawed and can camouflage whether or not change has Occurred. Also, one of the most popular approaches suggests a very simple approach to measuring the effects of an intervention to improve SQ (viz., Zeithaml, Parasuraman & Berry 1990, pp. 177-78).

Without a model of change (e.g. alpha, beta, gamma) and a systematic procedure for operationalising the model (e.g. fig. 1 and confirmatory factor analysis), it is easy to confuse theoretical and practical issues. For example, a primary theoretical issue of concern for PZB and other academic researchers is whether or not SERVQUAL has five dimensions, and whether these have remained stable over time. In an applied setting however, the more important question is whether or not the same number of (interpretable) dimensions reappear at the second point in time.

Because people use new information to change how they think about things and what they think about them, the alpha, beta, gamma change model is relevant for measuring change in a variety of consumer research settings other than SQ. For example, this approach can be used to measure changes in a wide range of self-reported constructs such as attitudes, beliefs, commitment, emotions, feelings, involvement, perceived risk, et cetera. These changes may result from the marketing interventions of companies or from government actions targeted at issues such as alcohol and cigarette consumption.

An area for future research is to investigate systematically the three types of change across differing circumstances and explore whether they typically occur through different cognitive processes. Thompson and Hunt (1996) outline a number of propositions that suggest this is the case for attitudes. These propositions could be tested in a marketing setting.

In a discipline often criticised for taking some methodological shortcuts, the review of change in this paper offers a logical, integrated framework to help consumer researchers search for different types of change, and report their findings in a way that will facilitate future meta-analyses. In psychology there are other models of change which also deserve scrutiny (e.g. Collins & Horn 1991; Hofmann, Jacobs & Baretta 1993). This paper is designed to broaden the focus of measuring change away from merely alpha change. This recommendation however, comes at a cost. It will place greater demands on researchers to articulate the explicit linkages between their theoretical and measurement models--a plea that has been made often over the years (e.g. Bagozzi 1994; Peter 1979).

(Date of receipt of final transcript: January, 2000. Accepted by Mark Uncles, Area Editor.)

(1) Reflective measures are derived from the researcher's theory of service quality, much like measures of intelligence are derived from a theory of intelligence. Formative measures are constructed by defining that a particular set of scales will be used to measure a construct. Rossiter (2000) argues that the various reliability measures commonly used by marketing researchers are irrelevant for this type of measure.

(2.) While it would be preferable to use actual data to illustrate how to find and interpret these three types of change, such data is expensive to collect. Also, the authors of completed studies would be reluctant to make their data available for a study which may lead to the reinterpretation of the original findings. Hence, the use of a hypothetical example.

(3.) The numbers used in this analysis refer to Bank 1 in Parasuraman, Berry and Zeithaml (1991).

(4.) This factor analysis was conducted across data from five companies.

References

Aaker, D.A., Kumar, V. & Day, G.S. 1995, Marketing Research, John Wiley & Sons, New York.

Armenakis, A.A. 1988, `A review of research on the change typology', in Research in Organisational Change and Development, eds. W.A. Pasmore & R.W. Woodman, JAI Press, Greenwich, CT, pp. 163-94.

Bagozzi, R.P. 1984, `A prospectus for theory construction in marketing', Journal of Marketing, vol. 48, no. 1, pp. 11-29.

Bagozzi, R.P. 1994, `Measurement in marketing research: Basic principles of questionnaire design', in Principles of Marketing Research, ed. R. Bagozzi, Blackwell, Oxford, pp. 1-49.

Bedeian, A.G., Armenakis, A.A. & Gibson, R.W. 1980, `The measurement and control of beta change', Academy of Management Review, vol. 5, no. 4, pp. 561-66.

Boulding, W., Kalra, A., Staelin, R. & Zeithaml, V.A. 1993, `A dynamic process model of service quality: From expectations to behavioural intentions', Journal of Marketing Research, vol. 30, no. 1, pp. 7-27.

Brown, T.J., Churchill, G.A. & Peter, J.P. 1993, `Improving the measurement of service quality', Journal of Retailing, vol. 69, no. 1, pp. 127-39.

Carlzon, J. 1987, Moments of Truth, Harper & Row, Cambridge, MA.

Carroll, D.J. & Green, P.E. 1995, `Psychometric methods in marketing research: Part 1, conjoint analysis', Journal of Marketing Research, vol. 32, no. 4, pp. 385-91.

Carroll, D.J. & Green, P.E. 1997, `Psychometric methods in marketing research: Part 2, multidimensional scaling', Journal of Marketing Research, vol. 32, no. 2, pp. 193-204.

Collins, L.M. & Horn, J.H. 1991, Best Methods for the Analysis of Change, American Psychological Association Washington.

Cook, T.D. & Campbell, D.T. 1979, Quasi-Experimentation: Design & Analysis for Field Settings, Rand McNally, Chicago.

Cronbach, L.J. & Furby, L. 1970, `How should we measure `change'--or should we?', Psychological Bulletin, vol. 74, no. 1, pp. 68-80.

Cronin, J.J. & Taylor, S.A. 1992, `Measuring service quality: A reexamination and extension', Journal of Marketing, vol. 56, no. 3, pp. 55-68.

Dowling, G.R. 1991, `Measuring change in self-report data: It's as easy as alpha, beta, gamma', Working Paper 91-003, Australian Graduate School of Management.

Dowling, G.R. & Midgley, D.F. 1991 `Using rank values as an interval scale', Psychology & Marketing, vol. 8, no. 1, pp. 37-41.

Golembiewski, R.T., Billingsley, K. & Yeager, S. 1976, `Measuring change and persistence in human affairs: Types of change generated by OD designs', Journal of Applied Behavioural Science, vol. 12, no. 2, pp. 133-57.

Hofmann, D.A., Jacobs, R. & Baratta, J.E. 1993, `Dynamic criteria and the measurement of change', Journal of Applied Psychology, vol. 78, no. 2, pp. 194-204.

Johns, G. 1981, `Difference score measures of organisational behaviour variables: A critique', Organizational Behaviour and Human Performance, vol. 27, no. 3, pp. 443-63.

Joreskog, K.G., 1979, `Longitudinal covariance models', in Advances in Factor Analysis and Structural Equation Models, eds. K.G. Joreskog & D. Sorbom, Abt Books, Cambridge, MA.

Millsap, R.E. & Hartog, S.B. 1988, `Alpha, beta and gamma change in evaluation research: A structural equation approach', Journal of Applied Psychology, vol. 73, no. 3, pp. 574-84.

Mittal, B. & Lassar, W.M. 1996, `The role of personalization in service encounters', Journal of Retailing, vol. 72, no. 1, pp. 95-109.

Newman, K. & Cowling, A. 1996, `Service quality in retail banking: The experience of two British clearing banks', International Journal of Bank Marketing, vol. 14, no. 6, pp. 3-11.

Parasuraman, A., Berry, L.L. & Zeithaml, V.A. 1991, `Refinement and reassessment of the SERVQUAL scale', Journal of Retailing, vol. 67, no. 4, pp. 420-50.

Parasuraman, A., Berry, L.L. & Zeithaml, V.A. 1993, `More on improving service quality measurement', Journal of Retailing, vol. 69, no. 1, pp. 140-47.

Parasuraman, A., Zeithaml, V.A. & Berry, L.L. 1985, `A conceptual model of service quality and its implications for future research', Journal of Marketing, vol. 49, no. 4, pp. 41-50.

Parasuraman, A., Zeithaml, V.A. & Berry, L.L. 1988, `SERVQUAL: A multiple-item scale for measuring consumer perceptions of service quality', Journal of Retailing, vol. 64, no. 1, pp. 12-40.

Parasuraman, A., Zeithaml, V.A. & Berry, L.L. 1994a, `Reassessment of expectations as a common standard in measuring service quality: Implications for further research', Journal of Marketing, vol. 58, no. 1, pp. 111-24.

Parasuraman, A., Zeithamal, V.A. & Berry, L.L. 1994b, `Moving forward in service quality research: Measuring different customer-expectation levels, comparing alternative scales, and examining the performance-behavioural intentions link', Working Paper, No. 94-114, Marketing Science Institute, Cambridge, Massachusetts.

Parasuraman, A., Zeithamal, V.A. & Berry, L.L. 1994c, `Alternative scales for measuring service quality: A comparative assessment based on psychometric and diagnostic criteria', Journal of Retailing, vol. 70, no. 3, pp. 201-30.

Peter, J.P. 1979, `Reliability: A review of psychometric basics and recent marketing practices', Journal of Marketing Research, vol. 16, no. 1, pp. 6-17.

Peter, J.P., Churchill, G.A. & Brown, T.J. 1993, `Caution in the use of difference scores in consumer research', Journal of Consumer Research, vol. 19, no. 4, pp. 655-62.

Rossiter, J.R. 2000, `Construct measurement in marketing', Working Paper 00-01, Department of Marketing, Faculty of Commerce, University of Wollongong.

Schaubroeck, J. & Green, S.G. 1989, `Confirmatory factor analytic procedures for assessing change during organizational entry', Journal of Applied Psychology, vol. 74, no. 6, pp. 892900.

Schmitt, N. 1982, `The use of analysis of covariance structures to assess beta and gamma change', Multivariate Behavioural Research, vol. 17, pp. 343-58.

Swartz, T. & Iacobucci, D. 2000, Handbook of Services Management, Sage Publications, Thousand Oaks, CA.

Teas, R.K. 1993, `Expectations, performance evaluation and consumers' perceptions of quality', Journal of Marketing, vol. 57, no. 4, pp. 18-34.

Thompson, R.C. & Hunt, J.G. 1996, `Inside the black box of alpha, beta and gamma change: Using a cognitive-processing model to assess attitude structure', Academy of Management Review, vol. 21, no. 3, pp. 655-90.

Till, B.D. & Priluck, R.L. 2000, `Stimulus generalisation in classical conditioning: An initial investigation and extension', Psychology & Marketing, vol. 17, no. 1, pp. 55-72.

Triplett, J.L., Yau, O.H.M. & Neal, C. 1994, `Assessing the reliability and validity of SERVQUAL in a longitudinal study: The experience of an Australian organisation', Asia Pacific Journal of Marketing and Logistics, vol. 6, no. 1, pp. 41-62.

Zeithaml, V.A., Parasuraman, A. & Berry, L.L. 1990, Delivering Quality Service: Balancing Customer Perceptions and Expectations, Free Press, New York.

Zeithaml, V.A., Parasuraman, A. & Berry, L.L. 1996, `The behavioural consequences of service quality', Journal of Marketing, vol. 60, no. 2, pp. 31-46.

Grahame R. Dowling, Australian Graduate School of Management, University of New South Wales, Sydney, NSW 2052. Email: [e-mail]