Consumer response to retail stockouts

Journal of Business Logistics, 2001 by Zinn, Walter, Liu, Peter C

Possible explanations for the two statistically significant results are as follows. The perception of the quality of store merchandise is slightly lower among consumers who experienced a stockout. This may be explained by consumers who reacted to the stockout by substituting the product they sought to buy, but were not as happy with the substitute as they expected to be with the intended purchase. The second significant result is intuitive. Consumers who experienced a stockout are less likely to agree with the statement "the store usually has what I want." In general, however, the differences

between the two groups are relatively small, which seems to indicate that consumers are able to insulate the store's image from a single stockout incident.

Consumer SDL behavior in response to a stockout.

Recall that the principal objective of this research is to explore the relationship between SDL behavior and the independent variables listed in Table 2. One possible approach is to look at the effect of individual variables on SDL behavior using cross-tabulations or t-tests of difference. This approach, however, ignores the interaction effects among the independent variables. The observed effect of an individual independent variable on SDL behavior often changes when other independent variables change. We therefore selected a research tool that allows us to examine the simultaneous effect of multiple independent variables on SDL behavior.

The traditional tool for this type of analysis is multiple regression. This, however, is not an option in this research because multiple regression assumes a continuous dependent variable. SDL behavior is a discrete variable. The consumer either substitutes the item, delays the purchase, or leaves the store. Another option considered for the research is discriminant analysis, which assumes a discrete dependent variable. This option was eliminated because an alternative technique, multinomial logit modeling, is better. Multinomial logit modeling does not assume that the independent variables are normally distributed. The estimates produced by discriminant analysis when the independent variables are not normally distributed are not consistent." The specific model used in this research is detailed in the Appendix.

The multinomial logit model was first run with all 24 independent variables included in Table 2. This first run revealed a multicollinearity problem. Multicollinearity occurs when there is a strong correlation among two or more of the independent variables. To eliminate the multicollinearity problem, four variables were dropped from the analysis. These were: "Intention to Visit Specific Store" (colinear with Pre-Visit Agenda), "Perceived Difference Among Brands" (colinear with Brand Loyalty), "Expectation of Future Visits" (colinear with Upset with Stockout), and "Advertised Price" (colinear with four different variables).

The second run of the model, now with 20 independent variables, revealed that the demographic variables included in the research had no significant impact on SDL behavior. As a result, the nine demographic variables were also dropped from the model. With the purpose of confirming the above result, we ran a model with only the nine demographic variables on one side and SDL behavior as the dependent variable on the other side. The run confirmed no significant effect between demographic variables and SDL behavior.