Theory Methodological reflection on Case Study 4

7.3.2 Research objective

The study’s objective was to test five new propositions about the effect of five dimensions of distribution strategy on the retailer’s sensitivity to time access windows.

7.3.3 Research strategy

Each proposition specifies a probabilistic relation between time window pressure and distribution costs. A probabilistic proposition can best be tested in an experiment. However, an experiment was not feasible for this study, because it would require that retail chains would vary their distribution strategy for the purpose of this research which is not pos- sible in practice. Because an experimental research strategy was not feasible in this study, a survey was the second-best strategy. However, for a survey, a large number of instances must be available in order to be able to make the required statistical analyses. In this study the number of instances is too small for a statistical analysis. Therefore the third-best strategy was cho- sen in the comparative case study.

7.3.4 Candidate cases

The universe of instances of the object of study to which the theory was applicable consists of all distribution activities of all large retailers in Western Europe that are at least partly located in shopping areas in cities in which time access windows could be installed. Dutch retailers are instances from this domain.

7.3.5 Case selection

We recommend selecting a very small population in order to avoid problems regarding the representativeness of the sample used for the test see 7.1.3. In 6.2.6, it was explained how a sample of 14 Dutch retailers with different competitive strategies was selected for this study, which could be considered representative for Dutch distributors in terms of type of retail, including discounters lower end of the market, retailers that focus on cost middle segment of the market, and retail- ers that focus on response or differentiation higher end of the mar- ket. It was mentioned in the methodological reflection with Case Study 3 in 6.3.5 that such representativeness was not needed for that study. For the current quasi-survey, a probability sample was preferred. However, it is clear particularly also from the discussion in 6.2 that such sampling was not realistic for the current study.

7.3.6 Hypotheses

Because the propositions in this study specified probabilistic relations, the hypothses stated that the rise in total distribution costs is steeper “on average” for subgroups of retailers with higher values on the dimensions of distribution strategy than for subgroups of retailers with lower values on these dimensions.

7.3.7 Measurement

Cost data were generated by the same model that was used in Case Study 3, which in its turn used empirical data as input.

7.3.8 Data presentation

Figure 7.1 presents the data generated for one scenario, as an example. The mode of presentation is such that it directly provides for the com- parison of the steepness of the increase in average total costs between the subgroups that were constructed for each independent variable.

7.3.9 Data analysis

Averages per subgroup three or four for each independent variable should be perfectly rank ordered. Such a perfect rank order is repre- sented in each graph in Figure 7.1 by four lines above each other in the right order subgroup 1 the lowest, and subgroup 4 the highest line and without any crossings between these lines. Figure 7.1 allows for a visual inspection in which it can be assessed whether this is actually true. This is the case for the independent variables stops per roundtrip, Chapter 7