Model based interpretation of survey data: A case study of enterprise resource planning implementations

The selection of the appropriate analysis tools for survey data is an important decision for all researchers dealing with responses on questionnaires. Over the last two decades a number of approaches have been used for classifying variables, statistically measuring significance and developing predictions of outcomes. This paper compares and evaluates the use of regression analysis, logistic (logit) models, discriminate analysis and data envelopment analysis (DEA), for empirical data from a survey of enterprise resource planning (ERP) implementations in the US manufacturing sector. The data collected from this survey contains a mix of subjective and objective data, and provides an opportunity to assess the impact of these modeling techniques on measuring outcomes and a decision-maker’s acceptability of the results. The analysis illustrates that regression based tools are more valuable in developing predictive models, while logit and discriminate models are powerful in classifying the outcomes. The genetic search-based discriminate model is intuitively appealing, whereas DEA provides additional information with respect to understanding the process of arriving at the outcome over other tools. The analysis further shows that these techniques can be used in a complementary manner to insights that they cannot provide when used individually. In addition to the feasibility of these techniques, this analysis also provides important insights into ERP implementations.