To ascertain the individual effects of in-store and out-of-store factors on attention and evaluation levels, we conducted random-effects logistic regressions. We chose that method for two reasons (Hox 2010): First, calculating logistic regressions was called for because the dependent vari- ables of the regressions are binary categorical – attention and evaluation
levels are 0/1 coded. Second, the method needed to take into account that our data have a repeated measures structure. This was achieved through inclusion of random effects into the regressions. The data show repeated measured characteristics because every participant could look at multiple products. Fixation patterns in one participant have a higher likelihood of being correlated than fixation patterns between participants. To obtain reliable parameter estimates and standard errors, it was therefore neces- sary to separate the effects within participants from the between effects that are not mainly participant-induced. We controlled for individual heterogeneity by specifying the model as a panel model, with participants as the panel variable.
We specified the different stages of attention and evaluation as depend- ent variables and the in- and out-of-store factors as independent variables.
In sum, six binary random-effects logistic regressions were calculated:
one each for the attention levels “express fixation,” “pictorial fixation,”
“textual fixation,” “revisit,” and “test,” and for the evaluation level
“consideration.” For the other evaluation level “choice,” we estimated a conditional logistic regression as the study participants were bound to choose but one product, whereas they could guide their attention and consideration to several products (Verbeek 2008). No coefficients can be estimated for shopper-specific factors that are constant across products per participant (number of products, format type, willingness to pay). We believe that the separate analysis of the different fixation levels, revisit, test, consideration, and choice will yield important insights on the effects of in-store and out-of-store factors on a detailed set of behaviors. Table 3 summarizes the unstandardized coefficient estimates and standard errors of the regressions.
Table 3: Unstandardized Parameter Estimates and Standard Errors of Regressions
Results for in-store factors. We will first take a look at the effects of those in-store factors relating to in-store marketing.
Number of offered products does not have significant effects on attention.
It does have an impact, however, on the chance of a product ending up in the consideration set: In the smaller store with 16 SKUs, the likelihood of a product being considered is higher than in the larger store with 22 SKUs to choose from. The explanation for that is rather clear: Indifferent of the variety of SKUs offered, consideration sets in both stores contained on av- erage around 2.6 products. If the number of products considered remains fairly constant, the likelihood of a product making it into the consideration set shrinks with a larger number of competing SKUs around it on the shelf.
In terms of the impact of different store formats, the seemingly more customer-friendly information material and shelf design of the test format did not influence attention or consideration, leading us to reject H1. None of the coefficients was significant. Furthermore, signs of coefficients flip between different attention levels. It is fair to say that the standard format of the shelf setup performed just as well as the test format.
Vertical shelf position has a strong impact on all levels of attention, just as we expected. However, those effects do not carry through uniformly to evaluation. Therefore, H2 can only partly be upheld. Products on the lower of the two rows (the one between hand and eye level) are fixated on more often and are more likely to be revisited and tested. Yet, vertical row posi- tion does not have a significant effect on the likelihood of a product being considered, whereas it does have an impact on product choice, although this impact is less significant than the one it has on attention levels.
Partially contrary to our hypothesis H3a, which stated that all price in- formation was relevant for the shopper, lower shelf prices seem to enhance the attention stages pictorial and textual fixation. And entirely contrary to our hypothesis H3b, price does not drive evaluation levels. However, when looking at all the coefficients of price, which are uniformly negative and sometimes significantly so, it appears that lower shelf price drives up atten- tion and evaluation levels of a product at least slightly.
Let’s now turn to the product-specific in-store factors: Charging time appears to be an only modestly important determinant of attention and evaluation. Products with shorter charging times were textually fixated on
and revisited more often, which partly corroborates H4a. Charging time did not affect evaluation levels, however, so H4b is falsified.
The chief product-specific in-store factor in determining evaluation is product weight. In accordance with our expectations of H5b, this factor significantly affects both evaluation levels. Lighter products are significantly more likely to be considered and also to be chosen. The weight of the screwdriver, which obviously is a hand-operated tool, literally outweighs the other product characteristics in determining evaluation. Other than hypothesized in H5a, it does not drive attention, however.
We can conclude that the “classical” in-store marketing factor vertical shelf position is the only in-store factor that behaves entirely in line with what we expected and what the literature suggests (e.g., Van der Lans et al.
2008): It drives mainly attention, not evaluation. Another very significant in-store driver is product weight, but this factor drives mainly evaluation.
The other in-store factors appear to be only very modestly influencing at- tention and evaluation.
Results for out-of-store factors. The regression results show that, as expected, two out-of-store factors do indeed have strong effects on evalu- ation; the third one does not.
Higher market share does exercise positive effects on almost all attention levels and on evaluation, thus lending support to H6. Especially, higher market share products are much more likely to be chosen at the end of the decision process – perhaps because shoppers see a more established product as a safe choice.
The factor with the strongest impact on attention and evaluation levels is favorite brand. As we predicted in H7, if a product is marketed under the shopper’s favorite brand, every level of attention to, and evaluation of, the product is higher. It is worth noting that, although all coefficients are significant, the strength of the effect seems to increase for the later stages of the decision-making process.
The link between willingness to pay and attention and evaluation appears to be as weak as that of favorite brand was strong. In congruence with what we expected in H8, a higher reservation price leads to slightly higher attention levels. Coefficients for textual fixation and revisit are significant;
the others are not, but all coefficients are unanimously positive. This weak positive effect on attention does not carry through to higher consideration
at all, though. Hence overall, we have to conclude that H8 cannot really be upheld. As for choice, willingness to pay could not be included in the conditional logistic regression, because as a shopper-specific variable it is constant across all products.
Summing up the analysis of out-of-store factors’ effects, we have seen that favorite brand dwarfs the impact of all other factors on attention and evaluation; market share also enhances attention and evaluation; and will- ingness to pay helps attention a bit.
Summary. Through the regressions, we learnt about effects on individual levels of attention and evaluation. Some key findings stand out: We found three different patterns how drivers exert their influence on attention and/
or evaluation, and these do not follow the distinction “in-store versus out- of-store factors”: First, the “classic” in-store marketing lever vertical shelf position also shows the “classic” pattern one would expect – namely, ex- erting its influence mainly on attention, not evaluation. Second, product weight does not mainly influence attention, but evaluation, despite of be- ing an in-store factor. Third, two out-of-store factors in our study (market share and favorite brand) did not only significantly drive evaluation, but also attention.
We have now seen which in-store and out-of-store factors have effects on separate levels of attention and evaluation. The granularity in which attention was analyzed, together with evaluation, led to detailed insights.
Yet we still lack an understanding about whether higher levels of atten- tion elevate evaluation equally for all in-store and out-of-store factors, or whether some factors lead to higher attention that does not carry through to evaluation, whereas others influence evaluation directly without going the route through increased attention. These questions can best be answered through a mediation analysis, which we present in the next section.