The content of respondents' mental representations

Besides on the structural variables mental representations may differ on their content components between the scenarios. Thus, the nature of attributes, benefits and cognitive subsets is examined on possible substantial shifts between scenarios. The first subsection will however report how respondents ranked the decision variables.

Ranking of decision variables

At the beginning of the interview respondents were asked to rank the in random order shown decision variables time of shopping (TS), transport mode (TM) and shopping location (SL) according to the preferred sequence of decisions. In both scenarios respondents prefer to consider time of shopping before shopping location and transport mode. The mean ranks are 2.52 (TM choice), 1.89 (SL choice) and 1.59 (TS choice) in the base scenario and 2.23 (TM choice), 2.00 (SL choice) and 1.77 (TS choice) in the e-commerce scenario. Thus, respondents prefer to consider time of shopping before shopping location and transport mode. While Kusumastuti et al. (2009) report also a first order rank for time of (leisure) shopping, their findings for location and mode choice are reversed. Yet, according to Davidson et al. (2007) many activity-based models assume that the location choice is made before the mode choice which is herewith supported. A MANOVA confirms the significance of the rank order of decision variables across scenarios (F = 6262.1, df=2, p < .001). As a second effect, the MANOVA also reveals a significant rank order difference between scenarios (F=8.0, df=2, p< .001). Although the rank order of the average scores does not change, respondents of the e-commerce scenario rank TS and SL choice more often lower and TM choice more often higher compared to basic respondents. Especially the latter finding is unexpected as the introduction of online shopping would not suggest an impact on transport mode choice on first sight. Possibly, respondents perceive more freedom by the increased number of shopping alternatives along with less travel- related restrictive consequences which in turn allows them to decide on their transport mode before considering the other choices.

The frequency of elicited attributes

Figure 6.2 shows all attributes that were elicited from at least 5% of respondents in any scenario. There is quite strong agreement between both scenarios in the sense that the available product assortment, distance from current location and number of bags to carry are most frequently considered attributes. While spatial accessibility attributes such as accessibility of the store and parking opportunities appear next frequent in the basic scenario, e-commerce respondents considered rather product-related attributes (the price level of the assortment) or personal/situational availability attributes such as available time to shop and leisure time more often. In general there is however only little variation between both scenarios with regard to the nature of attributes. For each of the top 10 attributes of either scenario cross tables have been set up with number of respondents who considered them and number of respondents who did not as rows and scenario as columns. In total, 11 Chi-Square tests have been performed. The results (not shown) indicate that there are no significant differences for the frequency of the 11 investigated attributes. The biggest difference was observed for

Figure 6.2: Frequency of elicited attributes.

accessibility of the store which failed however the level of significance slightly (X2 = 3.469, df = 1, p = 063).

The frequency of elicited benefits

Figure 6.3 lists the frequencies of all elicited benefits for the scenarios. Ease of travelling, Time savings and Ease of shopping are the most frequently considered benefits in both scenarios while safety aspects, social acceptance and personal care receive little attention. Furthermore, it is remarkable that except for Health, Course of Fitness/Well-being and Safety in the shopping location all benefits yield higher frequencies in the e-commerce scenario. As for attributes, cross tables with scenario as column and number of considers versus number of non-considers as rows have been performed. It turned out that Time savings (X2 = 11.259, df = 1, p = .001), Ease of shopping (X2 = 10.196, df = 1, p = .001), Diversity in product choice (X2 = 7.682, df = 1, p = .006) and Travel comfort (X2 = 4.628, df = 1, p = .031) are considered significantly more often in the e-commerce scenario than in the basic scenario.

The frequency of elicited cognitive subsets

A frequent item set analysis has also been performed to investigate the cognitive subsets. None in the top 10% of most

Figure 6.3: Frequency of elicited benefits.

frequent cognitive subsets included the decision for Time of shopping (TS) although most respondents started with this consideration, as discussed. As for attributes and benefits also the most frequent cognitive subsets were compared separately between scenarios by means of cross tables. Of the top eight cognitive subsets only TM – ease of travelling was considered significantly more often (X2 = 4.118, df=l, p = .042) in the basic scenario. The subset SL – distance – time savings missed the level of significance slightly (X2 = 3.415, df= 1, p = .065).

Centrality of variables

The previous three sections analysed MRs in the extent to which there is agreement among respondents in the considered attributes, benefits and cognitive subsets. While these three descriptors give insight in the salience of some components of MRs and how they differ between scenarios they say little about the role of these components within the causal network. For instance, it remains yet unclear whether the most frequent considered attribute is linked to one decision variable (DV) and benefit only or interlinked to several DVs and benefits. In the latter case, the attribute would have a central role for the decision-maker. Possibly, the role of some attributes undergoes shifts for different situational decision contexts.

In order to determine the centrality of variables an implication matrix has been set up for each respondent or MR, respectively, where all variables which can be a parent node (DYs and attributes) are represented as rows and all variables which can serve as a child node (attributes and benefits) are represented as columns. All indicated causal links between them were coded as 1. All other cells were filled up with a 0. Adding then the row and column sum of a variable and dividing it by the matrix sum results in the centrality value of this variable which can take on values from the range between 0 and 1 (Knoke & Burt, 1982). In other words, the centrality c of a variable V represents the sum of its incoming (Z, V) and outgoing (V, Y) links over the sum of occurring (Z, Y) links in the MR of respondent j. In formula, the measure is defined as:

(6.1)

Table 6.3 lists means for the top 10 central variables per scenario. DVs are highlighted in dark grey and benefits in light grey. Attributes are shown in normal style. Despite of the fact that DVs have no incoming links they score most central in both scenarios. This circumstance is not surprising when considering the fact that each cognitive subset has a DV as origin. Due to the limited number of decision variables their centrality values are this high. In both scenarios the benefits ease of shopping, time savings and ease of travelling belong to the top 10 central variables which speaks to the stability and general validity of these variables as underlying benefits. This holds also for attributes as available assortment, distance from current location and number of bags to carry do not differ considerably in their centrality value. The fact that the accessibility of the store fell out of the top 10 central variables in

Table 6.3: Top 10 central variables in both scenarios.

the e-commerce scenario confirms the shift for this attribute as shown in Figure 6.2. What however is obvious is the fact that all shopping-related variables no matter of which category increased in their centrality values in the e-commerce scenario. All transport-related variables lost centrality in comparison to the basic scenario. A multivariate analysis of variance (MANOVA) has been performed with the top 11 central variables as dependent variables and scenario as factor which did not show a significant effect (F = 0.783, df = 11, p = .658).