A.1 Linking sales and path data
One important features of our data set is the linkage of sales to trip records. As part of the RFID tracking process, the data report when the consumer arrives at the checkout. Independently, the sales data also have a time stamp for each shopper’s transaction at the checkout. Comparing the time stamp of a particular path with the sales data allows us to define a set of “candidate” checkout product baskets that occurred at a similar point in time.21 Matching which trip goes with which specific transaction involves considering the physical location of all the UPCs in each candidate basket. Based on how many of those locations lie on the path we are trying to match, a score is created for the baskets and the highest-scoring one is matched to the path.22 The matches do not necessarily yield a perfect score, because consumers might occasionally leave the cart and pick up an item. Therefore, we might not see the path of the consumer going past a specific item, even if the item was in her matched purchase basket.
A.2 (Lack of a) spatial correlation in feature advertising
In this section, we explore spatial correlation patterns in feature advertising activity in different categories. Correlation in feature advertising could have an impact on our results with regards to the lack of an effect of advertising on category traffic. Specifically, if feature advertising in categories that are stocked near each other is negatively correlated over time, such a correlation could mask an effect of advertising on traffic for any individual category.
To study spatial correlation, we first compute correlations between pairs of categories that are stocked in the same aisle. Among the 21 categories in our sample, 11 such pairs exist, and no systematic patterns emerges regarding the pairwise correlations. Out of 11 correlations, 5 are positive and 6 are negative.
Next, to assess the relationship between categories more systematically, we calculate the distance between each pair of categories in our sample. We then estimate a regression at the category-pair level where we regress the correlation coefficient (of features) for the category pair on the distance between the categories. Doing so, we find a small and insignificant coefficient for the distance variable. A one-standard-deviation change in distance (51 feet) leads to an (insignificant) increase in the correlation coefficient of 0.027. This number corresponds to a 0.05-standard-deviation increase in the correlation coefficient. We also implement regression specifications that include a “same-aisle” dummy and higher-order terms for the distance variable. Across all such specifications, we find consistently small and insignificant effects of distance (and other measures of vicinity) on the correlation in features between category pairs.
A.3 Intertemporal effects of advertising
Our main analysis of advertising impact on product sales in Section 3.2 investigates the effect of advertising on category-level sales in the same time period. It is conceivable that any increase in contemporaneous purchases is offset by lower levels of purchases in subsequent periods. Such intertemporal demand effects are well documented for price promotions (see Erdem et al. 2003; Hendel and Nevo 2006 and Osborne 2011) and could also occur in response to advertising.
To look at intertemporal advertising effects, we amend our regression framework in a simple way. Namely, we add lagged feature advertising, as well as similar terms for the other marketing variables, to our main regression, which regresses category-level sales on marketing variables (feature advertising, display, promotion dummy, and average price), and category and day fixed effects. Such a regression will show a “post-advertising dip” in sales if intertemporal effects are important, and hence a negative effect of lagged advertising would provide evidence for intertemporal substitution.
In Table 8
, we present results for the two sales measures on which advertising has a significant impact: the number of consumer/UPC pairs and total quantity (the dependent variables used in columns (3) and (4) of Table 3
). The baseline regressions without lagged variables are replicated in the first two columns, followed by the corresponding regressions with lagged terms.23
For both outcome variables, we find the effect of lagged advertising to be insignificant and small in magnitude. The magnitude of the contemporaneous advertising effects do not change significantly relative to the specifications without lagged terms. However, adding the lagged variables makes the effect of contemporaneous advertising insignificant in the specification based on total quantity (column (4)). Results stay significant when using consumer/UPC pairs as the dependent variable. We also note that when we run the traffic regressions with lagged terms (not reported), both contemporaneous and lagged advertising effects are insignificant.
The impact of lagged advertising on purchases
We take the results from these regressions as evidence that intertemporal advertising effects do not occur in our setting.
A.4 The impact of feature advertising on visit timing
In this section, we describe in more detail the analysis of category-visit timing summarized briefly in Section 3.4. To analyze the timing of visits, we compute for each shopping trip the point in time at which the consumer is for the first time walking past a specific product category. We then compute the average time since the start of the trip during which a specific category was visited at the category/day level.24 We first regress the time of the visit (measured in minutes since the start of the trip) and fraction of total shopping time elapsed on the number of featured products in a particular category. Both regressions include category and day fixed effects and marketing controls, and hence mirror the traffic regression (Eq. 1).
We start by implementing the analysis based on all product locations for each category. In other words, we define visit timing as the point in time at which a consumer first passes any location in the store associated with the particular category. The results using both minutes elapsed and the fraction of shopping time elapsed are reported in columns (1) and (2) of Table 9
. Columns (3) and (4) replicate the same regressions, but base the visit timing only on the primary locations of each category. Across all four specifications, we find effects of feature advertising that are consistently small in magnitude and mostly insignificant. Take, for example, the results in column (1). According to the (insignificant) point estimates, a one-standard-deviation increase in the number of features (eight additional features) in a particular category delays the visit to the category by 0.016 minutes (i.e., about 1 second) or shifts the visit timing back by 0.05 percentage points relative to the total time spent in the store.25
The impact of advertising on visit timing and dwell-time
The marginally significant effect in column (4) is similarly small in magnitude and does not constitute an economically meaningful shift in the timing of the category visit.
Finally, advertising might only affect a small set of consumers who are planning to purchase within the category due to the feature ad. When analyzing the visit timing of all consumers in the store, the unaltered behavior of the majority of visitors to the store might mask a significant effect for this group of consumers. We hence isolate the group of consumers who are most likely to be affected, by computing the daily average time of a category visit based only on consumers who purchase in the specific category. The results from regressions based on this measure of visit timing are reported in columns (5) and (6) of Table 9. We again find a null effect of feature advertising on visit timing, and the confidence intervals do not contain economically large effect sizes.26
We hence conclude feature advertising does not influence when consumers visit a specific category.
A.5 The impact of feature advertising on dwell-time
In this section, we provide further details on the impact of advertising on dwell-time in front of the category. Based on the path data, we calculate the total time a consumer spends on traffic points belonging to the specific category for each category in which she purchased during a given shopping trip. Similar to other parts of our analysis, we aggregate this variable to the category/day level and regress the average daily dwell-time onto the number of features (and control variables). Results from this regression are reported in column (7) of Table 9 and show a small and insignificant effect. We note that dwell-time is measured in seconds, and average daily dwell-time has a mean (standard deviation) of 53 (41) seconds. A one-standard-deviation shift in the number of features changes dwell-time by only 0.29 seconds (0.29 = 0.036 * 8).
We note that we would ideally like to measure the time a consumer spent contemplating which product to buy in the category. Total time spent in the vicinity of a given category is likely to be a noisy measure of search time (see Seiler & Pinna (2016) for a detailed discussion of the measurement error associated with path-tracking-based dwell-time measures). We therefore assess robustness of the null effect to using an alternative measure that only captures the amount of time spent near the specific product that was picked up (rather than the entire category). Results from this regression are reported in column (8) of Table 9 and also yield an insignificant result and an effect size that is small in magnitude.