Data analysis Case Study 4: Theory-testing research: testing a probabilistic relation

Our hypotheses predicted the following pattern for each dimension of distribution strategy: Group 1 will have the lowest and the least steep line; Group 2 will have a higher and steeper line than Group 1; Group 3 will have a higher and steeper line than Group 2; and Group 4 will have the VEHICLE CAPACITY Costs – scenario C 1 small 2 3 4 large 1 small distance 2 3 4 large distance A B 1C 1 strict 2 normal 3 long STOPS PER ROUNDTRIP Costs – scenario C 1 f ew stops 2 3 4 many stops 10 20 30 40 50 60 20 40 60 80 100 10 20 30 40 50 60 70 80 Stores affected 20 40 60 80 100 120 10 20 30 40 50 60 70 80 Stores affected Stores affected 10 20 30 40 50 60 70 80 Stores affected Stores affected 1 br ief 2 3 4 long Percentage increase C 10 20 30 40 50 60 Percentage increase E 10 20 30 40 50 60 Percentage increase 10 20 30 40 50 60 70 80 Percentage increase D 5 10 15 20 25 30 35 40 Percentage increase STOPPING TIME PER VEHICLE Costs – scenario C DISTANCE BETWEEN DC AND STORES Costs – scenario C SELF-IMPOSED TIME WINDOWS Costs – scenario C Figure 7.1 Increase of distribution costs due to increase of time window pressure for different values of the dimension of strategy. Each graph A–E represents a dimension of strategy and each line within a graph represents a value of the dimension of strategy. All graphs are for stable time window length scenario C, see Table 6.2. highest and steepest line. Figure 7.1 window length 3 hours, scenario C shows that the observed pattern is as expected for the dimensions stops per roundtrip, distance between DC and shops, stopping time, and self-imposed time windows. For one dimension, vehicle capacity, the pattern in Group 1 in Figure 7.1B does not correspond with the expected pattern. The patterns that we observed for time window length scenarios A and B are similar to the pattern for scenario C, with an expected dif- ference; namely that the magnitude of the impacts differs. However, the corresponding data are not shown here. We concluded that hypotheses 1 and 3–5 were confirmed in this study and that hypothesis 2 was rejected.

7.2.11 Implications for the theory

Four hypotheses were confirmed. This is an indication that the propo- sition from which these hypotheses were derived is correct. However, the hypothesis on vehicle capacity was rejected. The proposition on vehicle capacity was based on the idea that, due to time windows, the possibilities of combining deliveries in one vehicle roundtrip would decrease, resulting in an increase in the number of vehicles and a corresponding decrease in the vehicle load factor. We hypothesized that, in particular, retailers using large trucks would have difficulties filling their vehicles and that they would be forced to make roundtrips with only partly filled trucks. As a result they would make trips with more empty than filled capacity, whereas a retailer using small vehicles would still make trips with well-filled smaller vehicles. From the rejection of the hypothesis we conclude that this reasoning was wrong. It appears that vehicle capacity in itself is not a determinant of the sensitivity to time windows. How can this be explained? We think now that there are other, much more important determinants of the underuse of the capacities of the available vehicles. If we take into account that the length of a vehicle roundtrip can either be deter- mined by the vehicle’s capacity actually used in that roundtrip, or by other factors e.g. the driver’s working times, opening hours of shops, etc., then we would be able to see how those other factors might explain the extent to which underuse of vehicle capacity contributes to total distribution costs. The effect of time window pressure might be that more vehicle roundtrips are getting constrained because of such non-capacity constraints. If such non-capacity constraints apply, this would imply that it is more cost efficient to use smaller vehicles, but it would not imply that retailers starting with smaller vehicles before 168 time window pressure increases are less sensitive to time windows. This reasoning can be formulated as a new proposition: Proposition 6: Retailers with a lower ability to use their full vehicle capacity are likely to have a higher increase in total distribution costs that occur with a given change in time access window pressure than retailers with a higher ability to use their full vehicle capacity.

7.2.12 Replication strategy

Since we formulated a new proposition, we need new theory-testing research. For such testing we need to define the concept “ability to use the full vehicle capacity” as well as to determine how its values could be measured.

7.3 Methodological reflection on Case Study 4

7.3.1 Theory

The object of study in Case Study 4 is the same as in Case Study 3: distribu- tion activities by retailers from a retailer’s distribution centre to the shops during one week. The concepts and their operationalization into variables were pre- cisely defined similar to the definitions in Case Study 3. However, the dependent variable is total distribution costs, which is the result of the separate distribution costs studied in Case Study 3. There are five propositions for each of the dimensions of distribution strategy. Each proposition states that there is a probabilistic relation between this dimension and total distribution costs. As in Case Study 3, the domain of the theory included all large retail- ers that distribute goods from a distribution centre to shops that are at least partly located in shopping areas in cities in which time access windows could be installed. The universe of instances of the object of study is bounded by the geographical boundary of Western Europe where retailers and local governments are familiar with the concept of time access windows. Implicitly it seems that the theory is applicable in different sectors as the study was done in supermarket, department store, fashion shop, and specialist shop sectors.