Consumer response to retail stockouts

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

The 11 independent variables in the final version of the model are listed in Table 5. This reduced version was selected because the 11 selected variables are expected to have a significant effect on SDL behavior. An additional reason to use the reduced version of the model is that a model with fewer variables has more degrees of freedom. This is relevant because the sample size for the multinomial logit model, with 179 observations, is smaller than the number of collected cases. The multinomial logit model dropped every case that contained at least one missing value. While this problem is often resolved by replacing the missing values with the average for the variable, we elected not to do that because it artificially reduces the variance for that variable.

A run of the final version of the model indicated a strong overall significance at the .000058 level (X2 = 57.14). Table 6 presents the predicted and actual outcomes, which yield a hit ratio of 67%. In other words, although prediction is not the goal of this exploratory study, the behavior of two out of every three respondents was predicted correctly by the model. The actual response of consumers to stockouts was 62.0% for substitution, 15.1 % for delay, and 22.9% for leaving the store.

Tables 7, 8, and 9 present the results for each of the three SDL behaviors examined in this research. Substitution behavior is reported in Table 7, Delay in Table 8 and Leave in Table 9. The data are reported the same way in each of the three tables. For each independent variable, the reported coefficient represents the marginal effect of that variable on the probability of occurrence of the SDL behavior being examined, when all other variables are held constant. For example, in Table 7, the coefficient for the first variable (Store Prices) means that a one unit improvement in the average perception that the store offers good prices-say, from 3 to 4 on a 5-point scale-will increase by .05084 (or approximately 5%) the probability that the average consumer will react to the stockout by substituting the item. The above example also shows how Tables 7, 8 and 9 are linked. The one unit improvement in the average perception that the store offers good prices also impacts the probabilities that the consumer delays the purchase and leaves the store because the sum of the probabilities for the three types of SDL behavior must equal 1. Please note that the short version of variable names is reported in brackets in Table 5.

Table 7 indicates that four variables have a significant effect on the probability that a consumer will react to a stockout by substituting the intended purchase item. These are Store Prices (at the . 10 level), Urgency and Brand Loyalty (at the .05 level), and Upset (at the .01 level). The first three results are intuitive and will be described first. The stronger the consumer's perception that the store offers low prices, the more likely that the consumer will substitute the intended purchase. This seems to suggest that the consumer expects the substitute to be an equally good value.