Data Collection and Sample

The study was split up into two waves of data collection with an aim to collect 200 respondents per wave. First, data for the basic scenario were collected in May 2010. Next the interviews with the e-commerce scenario took place in October 2010. Different respondents were approached for each wave. The selection of respondents was restricted to three criteria: possession of a driver's license, aged between 18 and 60 years and having Dutch language skills. The former two characteristics were set up to ensure a real world reference for the experimental situations. The latter condition resulted from the interview language. A random sample of an existing national panel of 5000 Dutch households representative for the Netherlands was drawn (the panel originates from CentERdata, a research institute attached to Tilburg University and supported by the Dutch Organization for Scientific Research). In Table 6.1, sample characteristics are presented from all respondents who finished the online interview successfully, that is dropouts were excluded from further analysis. Statistical tests did not show significant differences between scenarios for any of the key socio-demographic variables.

Analysis

The presence of an online shopping alternative may have an impact on both the complexity of mental representations and the contents of the representations. In this section we describe the results of analyses regarding these two aspects successively.

The complexity of respondents' mental representations

In a first step, the mental representations were analysed structurally in terms of number of recalled considerations, number of attributes, number of benefits, number of benefits per attribute and number of cognitive subsets. Table 6.2 shows the results of descriptive analyses.

Table 6.1: Socio-demographics of the sample.

Table 6.2: Descriptive statistics of number of recalled considerations.

Number of recalled considerations

The variable number of recalled considerations (Table 6.2) counts in fact respondents' spontaneous recalls in the open question phase no matter of which category. The data were however aggregated so that variables that were recalled twice or thrice for different decision variables appear only once. It can be seen that e-commerce respondents could recall on average about half a variable more than basic-scenario respondents. The significance of the difference is confirmed by an Independent Samples t-Test (t = -2.500, df=473, p = .013). Minimum values of 0 are due to exclusions of non-interpretable inputs.

Number of attributes and number of benefits

When looking at descriptive statistics for number of attributes (Table 6.2) it can be seen that MRs for the e-commerce scenario on average consist of 0.5 attributes more than MRs for the basic scenario. An Independent Samples t-Test (t = -2.144, df = 473, p = .033) confirmed the significance of this difference. The finding that Min equals 0 is partly caused by the fact that some respondents considered benefits without linking them to attributes and partly by the fact that some considerations could not be interpreted and were thus excluded from further analysis. As for number of benefits, the figures show that on average e-commerce respondents considered more than one benefit more than basic respondents. An Independent Samples t-Test (t = -3.601, df=473, p< .001) confirmed the significance of this difference.

Benefits per attribute

As a further measure of compactness of MRs the ratio of benefits per attribute has been computed for each respondent. It describes how many needs (represented by benefits) a decision-maker wants to satisfy by the consideration of a characteristic of the choice alternatives (conceptualised as attribute). Although this ratio is slightly bigger in the basic scenario (Table 6.2) an Independent Samples t-Test indicates that there is no significant effect of scenario (t = 0.996, df= 465, p =.320).

Number of cognitive subsets

The fourth and final measure to describe and compare MRs is number of cognitive subsets. The term cognitive subset refers to a unique chain of linked variables. In other studies the terms (cognitive) subsets (Kusumastuti, 2011), hiesets (Farsari, 2006) or ladders (Reynolds & Gutman, 1988) are used to describe the same or similar constructs. In this analysis cognitive subsets can have two forms, namely 'decision variable – attribute – benefit' or 'decision variable – benefit'. As situational variables are treated like attributes, cognitive subsets with situational variables fall within the first case. However, the link between decision variables and situational variables is not of causal nature. Rather, it stands for a mental association respondents have between these two. That explains the choice for the name of this measure.

The basic scenario (10.05) resulted in less cognitive subsets per respondent than the e-commerce scenario (11.74) (Table 6.2). An Independent Samples t-Test (t = -1.759, df=473, p =.079) indicates however that the difference is not significant. Minimum values of 0 are again a result of excluding non-usable inputs. In fact, all respondents indicated at least three cognitive subsets. Even more surprising are the high maximum values which means that at least one respondent indicated (almost) all of the revealed benefits.

Conclusions on the complexity of respondents' MRs

Three findings were significant: Number of recalled considerations; number of attributes and number of benefits were significantly higher in the e-commerce than in the basic scenario. These findings confirm indeed the expectation that the introduction of additional choice alternatives which was online shopping in that scenario leads to an activation of additional needs which in turn results in an increase of attribute considerations. With a variation of the choice set as it has been tested in the experiment MRs become hence structurally more extended. Whether the considered attributes and benefits between the scenarios differ with regard to their nature will be investigated in the content analysis later.