Research Methods of Cluster and Regression Analysis in Literature - Essay Example

Quantitative Research Assignment Comments on the research method of Cluster Analysis used in: Allaway W. Allaway, Richard M. Gooner, David Berkowitz, & Lenita Davis (2006) "Deriving and Exploring Behavior Segments within a Retail Loyalty Card Program" European Journal of Marketing Vol. 40 No. 11/12 pp 1317 - 1339 'Cluster Analysis' - is a method of exploratory data analysis tool for sorting cases into groups or clusters so that the degree of association is strong between members of the same cluster and weak between members of different clusters. Each cluster thus describes on the basis of data collected, the class to which its members belong (Patent Storm). Cluster analysis technique in general is employed to study the relationships between variables which are normally distributed and which have an equal variance-covariance matrix in all groups (Kevin Voges et al) . In most of the marketing data sets like the one taken now for analysis none of these two conditions hold. Data Sources Data were collected from the database provided by the chain store, the loyalty program of which were researched upon. In any supermarket store loyalty, the location of the store plays a major role as a determinant for the loyalty of the customer (Messinger & Narasimhan, 1997; Engel et al., 1995). In this study the location of the three stores chosen would definitely have affected the purchasing behavior of the customers. Hence the study becomes weak in its identification of the data source. Another weakness of the data has been observed in the varying proportion of the different categories of buying which will have an influence on the application of the cluster analysis technique. Sample Size The study has not identified the total number of customers of the three stores and hence it would be difficult to comment on the randomness of the sample selected. A comparison of the total number of customers and the number constituting the sample size would have thrown some light on the comparability of the loyal customers between the customers who shopped generally during the period under study and the number of customers who opted to use the loyalty program. Basis for Collection of Information - Clustering Variables The collection of information and clustering considering the percentage of total share of wallet within product categories instead of taking into account the total purchases would have been a much better presentation of data under the research method of clustering analysis. The variables selected are far too general to form an opinion on the customer loyalty. The clustering lacks seriously because of the massiveness of data considered under the general clustering variables. For sure these clustering variables would have been subjected to behavioral benchmarking. It would be interesting to recall the behavioral factors like shopping frequency, tolerance of price increase etc. (Lacey, 2003) Validation of the Clusters The study has used the numerical taxonomy process to group the members into segments (Bunn, 1993; McKelvey, 1975; Punj and Stewart, 1983) However no clarity appears to be in sight in determining the range of potential market structures. Initially the number of groups ranged from two to eleven. Although the study has used the appropriate testing and analysis methods like scree testing, discriminant analysis and regression analysis to arrive at a particular number of groups as cluster groups, there is the lack of a scientific variation among the different groups evolved for study. There are possibilities that a slight change in the scaling would have vitiated the results especially in the middle range groups. This may be either due to problems of scaling as observed by Long (1997) or due to large volume of data analysed. According to Long (1997) "Scaling is a common cause of problems when numerical or interdependence methods are utilized, with the ratio between the largest standard deviation and the smallest standard deviation considered heuristically predictive of the likelihood and size of possible issues with cluster stability" This can be easily observed from the number of cardholders recorded in cluster group 2 and 3 in the case of 5 group clustering and 6 group clustering. Analysis Table VII of the study shows the predicted group memberships corrected to the percentages. There is absolutely no correlation in the analysis of the results of five cluster group variables though the different tables of results presented. The study has not explained the number disagreement in respect of the five group variables in Table II (Membership in five and six cluster solutions), Table III (Results of discriminant analysis on split-half validation), and Table VI (Mean characteristics of five and six cluster solutions) The following table illustrates this anomaly. Cluster # Table II Table III Table VI 1 23,973 23,973 23,973 2 10,600 1,305 10,600 3 19,927 19,947 19,927 4 1,305 1,846 1,308 5 1,846 10.600 1,846 The study has not sufficiently explained this variation in the split-half validation quantity between different cluster groups. Under normal circumstances the predicted membership in all the three tables should represent the same numbers. According to the International Development Research Centre which offers 'GENERAL HINTS WHEN CONSTRUCTING TABLES "Make sure that all the categories of the variables presented in the tables have been specified and that they are mutually exclusive (i.e. no overlaps and no gaps) and exhaustive. When making cross-tabulations, check that the column and row counts correspond to the frequency counts for each variable. Also check that the grand total in the table corresponds to the number of subjects in the sample. If not, an explanation is required. This could be presented as a footnote. (Missing data, for example.)" (IDRC) Table VII under the results presents a number of profile characteristics of the cluster variables. Some of these characteristics like the distance to the nearest billboard, distance to the nearest very early adopter, number of very early adopters do not largely influence the loyalty character of the card holders. Although the study claims that these form the geographical measures to arrive at the basis of clustering the memberships, these attributes are slightly away from the ability to influence the loyalty of the members. Since the parameters adopted for clustering the members like the distance to the nearest billboards, distance to the early adopters etc do not have much influence on the loyalty of cardholders, further statistical analysis of the results does not really presents any material evidence of loyalty. Assumptions The study has assumed to arrive at the customer loyalty behaviour on the basis of a single loyalty card program which is not a good base to proceed to make the analysis. Secondly, there are no comparable standards to verify the veracity of the results obtained from the study. The assumption that the customer loyalty behaviour could be assessed on the basis of numeric data itself is highly questionable. A perspective on quantitative research is based on using a proper research method or approach (Jenkins (1985) Conclusion For a study of this kind a quantitative research method to arrive at the loyalty behaviour cannot be advocated. Normally an online survey using a questionnaire method could be considered more appropriate where there is no need to synthesize a large volume of quantitative data. In this score this study loses its merit. The aim of the paper was to investigate the potential for deriving meaningful, managerially relevant customer segments within a retail loyalty-type program. But the study ended up with the segmentation of the customers in to different clusters as loyalty groups based on their purchasing behaviours which may not really contribute to the managerial decision making process from the marketing perspective. Thus the study has apparently not contributed significantly to the managerial decision making process. Comments on the research method of 'Regression Analysis' used in: David Holman, Claire Chissick, & Peter Totterdell (2002) 'The effects of Performance Monitoring on Emotional Labor and Well-being in Call Centers' Motivation and Emotion Vo. 26 No. 1 March 2002 pp 57 - 81 'Regression Analysis' - can be used to predict the outcome of a given key dependent variable on the interactions of other related explanatory variables. Thus regression analysis is recognised as a statistical tool for the examination of the relationships between variables (12Manage). Data Sources Lyberg and Kasprzyk (1991) observe that a researcher may decide on one or more data collection techniques within the general research approach. Data were collected in the form of questionnaire from a specified number of customer service agents in two UK call centers. The data collected represent the quantitative performance data on the time spent by the sample members on the actual job, average call time, and average call rate. Most of the data were collected using the method of 'remote monitoring' by the supervisor according to a specific guideline. Here the reliability of the data becomes questionable as the study was conducted in two call centers involving 347 call center agents. Again, the guideline in the case of 'mortgage call' one of the areas researched included eight variables like 'greeting', 'identifying and analysing the customer needs' and 'professionalism'. These attributes are highly subjective and hence are susceptible to the personal bias of the supervisor monitoring the performance. Hence the data is largely suffering from personal bias of the data collector. Sample Size The study has used 347 call center agents as the sample for the study. The samples have been selected from two divisions of a UK bank dealing with 'Mortgage Call' and 'Loan Call'. With the number of different purposes for which the call centers are established and with reference to the approximate total number of people working in the industry, the sample size cannot be used to make a fully justified generalization of the hypotheses selected by the study for research. Based on the sample size, the research question of "What are the relative effects of performance monitoring characteristics and work context variables on well-being in a call center environment" cannot be applied even to a particular segment of the industry. Basis of Data Collection According to Kerlinger (1986) it may be possible that a given research question may not be studied to the satisfaction of the researcher due to the absence of a proper data collection technique. The quantitative data were collected from both a questionnaire survey and onsite monitoring of the participants. The study is not very elaborate on the format or content of the questionnaire and the information or data gathered using the questionnaire. To this extent the study is vague on clarity in presentation. On the monitoring of the performance of the participants, a continuous electronic performance monitoring was used to collect the basic quantitative information like the average call time and average call rate. There were three different methods employed for collecting with most of the data collected by remote monitoring by supervisors. In the other methods the supervisor sits by the side of the employee and in the third method the conversation with the customer is taped for performance analysis. In all the methods of data collection because of the subjective attributes used for measuring the performance the data cannot be said to be one hundred percent unbiased. Variable - Performance Monitoring This variable includes four questions on the performance monitoring which are purely traditional. Both traditional and electronic forms of monitoring were used. However the use of a 5 - point scale will usually lead to the tendency of attempting to pick up the middle score and to this extent there is likelihood that the data collected is vitiated. "Sometimes Likert scales are used in a forced choice method where the middle option of "Neither agree nor disagree" is not available" (Lecture 6). Additionally, individuals tend to favor certain types of responses: Extreme responses, neutral responses, "agree" responses, or "disagree" responses (Kerlinger, 1973, p. 496). It is known that individuals with a "response bias" pick ratings towards the middle value on scales with an odd numb of alternatives, since the middle response is perceived as safe and comfortable. Individuals exhibit this response bias for a number of reasons: They may feel the question does not apply to them, they do not know the answer, or they do not want to reveal their true feelings, or they do not want to exert the energy to decide." Again usage of traditional scales like 'strongly agree' or 'strongly disagree' is bound to infuse some approximations in the data due to the tendency of the respondents. Reliability is a matter of assurance that the items posited to measure a construct are related to the requited extent (Cronbach 1951) Variable - Emotional Labor The subjectivity of the measures used has lead the data to lean more towards possessing an unreliable character. The collection of data here also suffers from the shortcoming of the scale adoption and description of the scale measures. Variable - Well-being The data collection with respect to this variable appears to be having some basis for more or less a correct quantification of the information. The measures that were being used to study the well-being have been well constructed to elicit the correct reporting by the respondents. Though subjectivity is still an element to be accounted for the nexus of the measures to the variable being studied makes the data more realistic and reliable. Results The statistical analysis of the findings especially in the case of the variable 'Well-being' has been done exactly to the theoretical requirements satisfying the conformity to the regression assumptions. The use of step-wise regressions of the measures with the control, content, purpose, and intent of monitoring was made perfectly to produce more reliable correlation between the dependent and the explanatory variables. Baron and Kenny (1986) proposed a four step approach in which several regression analyses are conducted and significance of the coefficients is examined at each step. Analysis Data analysis techniques use multivariate analysis simultaneously analysing multiple measurements of the different variables (Hair et al, 1995) The discussion part on the relationship of performance monitoring to well being is extensive. However the interrelationship of the performance monitoring and emotional exhaustion and anxiety and its negative association with depression ( = - .15, p < .01 from Table II) clearly indicates the negative association between the performance monitoring and well being. However the result that there is a positive association between job satisfaction and monitoring ( = .37, p < .01) is theoretically not possible. Any monitoring is likely to produce discontent among workers if unreliable measures are used for assessing their performance. "An unreliable measure of true performance levels may also be a source of much discontent among workers and supervisors (as also might, of course, a reliable but irrelevant or biased measure), which would further decrease the utility of the measure to the organization" (The National Academic Press). Hence the claim that the hypothesis 1 - performance related content of call monitoring will be positively associated with well being - is partially confirmed is not acceptable, even though there is a positive correlation shown by the statistical result. The analysis of hypothesis 2 and 3 has been well attempted. Cook and Campbell (1979) postulate that internal validity is a matter of causality. Straub (1989) points out that there are a number of ways in which the construct validity can be tested. However there does not appear to be a data validity test that has been conducted to test the reliability of the data. The analysis of the results relating to the establishment of relationships between performance monitoring and emotional labor with the results of the Hierarchical Regression as indicated by Table III can be taken to be an effort done in the proper direction and the results have more than represented the reality. However according to Gefen, Straub, and Boudreau (2000) the researcher should make sure that the assumptions related to the techniques are satisfied. Relationship of work context, performance monitoring, and well-being again are tested using hierarchical regressions and results presented in Table IV. A study and analysis of Table IV signifies the presence of significant interactions among the variables involved. However the applicability of the relationships is largely based on the reliability of the quantitative fed for statistical analysis as the measures presented were highly subjective. But still the data and results can be relied upon as the variables involved in this case are the actual feelings of the workers that were expressed in the actual work situation. Assumptions Since the study represents only a cross-section of the industry and workers it suffers from a basic limitation that the measures adopted by the study cannot generally be applied to other call center firms dealing with a product or other service industry as the degree of intensity of monitoring and the threshold for different attributes that affect the employee well being may differ. Again there is the question of generalization is not met by the study both by the sample size and also by the variables selected. Conclusion Although the study has made a reasonable attempt to justify meeting the objectives the reliability is greatly hindered by the method of data collection, the subjectivity of the variables and the traditional nature of the scales being used for answering the questionnaire. References 12 Manage 'Regression Analysis' Baron, R. M., & Kenny, D. A. (1986)The moderator-mediator variable distinction in social psychological research: Conceptual, strategic and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Bunn, M.D. (1993), "Taxonomy of buying decision approaches", Journal of Marketing, Vol. 57 No. 1, pp. 38-56. Cook, T. D. and D. T. Campbell (1979). Quasi Experimentation: Design and Analytical Issues for Field Settings Chicago, Rand McNally Cronbach, Lee J. (1951), "Coefficient Alpha and the Internal Structure of Tests," Psychometrika, 16, September, 297-334 Engel et al (1995) 'Consumer Behaviour' 4th Edition Dyeden Press Texas Gefen, D., Straub, D., and Boudreau, M.-C.(2000), "Structural Equation Modeling and Regression: Guidelines for Research Practice," Communications of AIS, Vol. 4, No. 7, pp. 1-80 Hair, Joseph H. Jr., Anderson, Rolph E., Tatham, Ronald L., Black, William C.(1995), Multivariate Data Analysis, Prentice Hall IDRC - International Development Research Centre 'Module 24: Cross-Tabulation of Quantitative Data' Jenkins, A. Milton, (1985)"Research Methodologies and MIS Research," In Research Methods in Information Systems, E. Mumford et al. (Ed.), Elsevier Science Publishers B.V., Amsterdam, Holland, 103-117. Kevin Voges, Nigel Pope & Mark Brown 'A Rough Cluster Analysis of Shopping Orientation Data' Kerlinger, F. N. (1973). Foundations of Behavioral Research (2nd ed) New York: Holt, Rinehart and Winston, Inc Kerlinger, Fred N. (1986) Foundations of Behavioral Research, Harcourt Brace Jovanovich Lacey, R.W. (2003), "Customer loyalty programs: strategic value to relationship marketing", unpublished PhD dissertation, The University of Alabama, Tuscaloosa, AL. Lecture 6 'Engaging the Stakeholder' Long, J.S. (1997), Regression Models for Categorical and Limited Dependent Variables, Sage Publications, Thousand Oaks, CA Lyberg. L. E. and Kasprzyk, D. (1991), "Data Collection Methods and Measurement Error: An Overview," Chapter 13 in Biemer, P.P., Groves, R.M., Lyberg. L.E., Mathiowetz, N.A. and Sudman, S. (eds.), in Measurement Errors in Surveys. New York: Wiley McKelvey, B. (1975), "Guidelines for the empirical classification of organizations", Administrative Science Quarterly, Vol. 20 No. 4, pp. 509-25. Messinger, Paul and C. Narasimhan, (1997)'A Model of REtail Formats Based on Consumers' Economizing on Shopping Time' Marketing Science 16 No. 1 pp 1-23. Patent Storm 'System and Method of Data Entry for a Cluster Analysis Program' Punj, G. and Stewart, D.W. (1983), "Cluster analysis in marketing research: review and suggestions for application", Journal of Marketing Research, Vol. 20 No. 2, pp. 134-48. Straub, Detmar W., "Validating Instruments in MIS Research," MIS Quarterly, 13, 2, June, (1989), 147-169 The National Academic Press 'Performance Assessment for the Whole Place Volume II: Technical Issues' (1991) Read More