Providing Consumers with Sale Information: Evidence from a Field Experiment

How to cite

Eliaz, Kfir; Oren-Kolbinger, Orli; Weisburd, Sarit. 2018. "Providing Consumers with Sale Information: Evidence from a Field Experiment." AEA RCT Registry. AEARCTR-0003354. https://doi.org/10.1257/rct.3354-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Kfir Eliaz* , Orli Oren-Kolbinger , Sarit Weisburd

* Primary investigator

Registry ID
AEARCTR-0003354
DOI
https://doi.org/10.1257/rct.3354-1.0
First registered
Last updated
Status
Completed
Trial dates
2016-01-29 – 2016-05-05
Countries
United States of America ()
Registry record
http://www.socialscienceregistry.org/trials/3354

Abstract

Does providing consumers with information about discounts help them realize more savings? We address this question using data from a field experiment on a website for online grocery shopping. Our results illustrate the difficulty in using information provision to steer shoppers towards cheaper alternatives (of equal or higher quality than their substitutes). We find that providing (treatment) shoppers with promotional information on sale categories increases the probability of purchasing within the category. This effect is driven by an increase in purchasing rates for both the reduced priced items and regularly priced substitutes. Our analysis focuses on understanding how item placement, promotional information, and the way promotional information is displayed impact consumer choices in a multi-product environment where prices are changing.

Experimental design

1. The platform. We partnered with a website that offers a purchase and next day delivery service from a large U.S. supermarket in a University city. The website includes roughly 3,000 items that are sold in the supermarket store. These items are divided into several categories to help shoppers perform an intuitive search (e.g. produce, dairy, etc.). Shoppers can also search the website for any item they would like to purchase using a search command. Shoppers need to add the items that they would like to purchase to their basket, and at checkout they pay for the products plus a flat delivery fee of $2.99 for each order. During the period of the experiment there was no option to re-order previous baskets or to add items from previous orders. In addition, all prices were fixed and there were no promotional sales. Shoppers are required to choose a delivery date and a two-hour delivery window. The cutoff time for next day delivery is midnight every day. These shoppers are mainly students (80 percent) with some professors (10 percent). Only 10 percent of shoppers are unaffiliated with the University. The website was interested in encouraging its registered customers to increase the frequency and volume of their purchases, and to learn how different promotional tactics affect shopping behavior. To achieve this goal, it planned to conduct a series of randomized control trials, and agreed to allow us to influence the design of these trials in a way that would also enable us to address our questions. Hence, the experimental design was somewhat constrained by the objectives of the website.

2. Temporary discounts. The experiment was conducted over a period of 13 weeks during which the website offered temporary discounts so that the prices of some select items fluctuated, dropping during the sale and rising when the sale expires. Discounted items were marked on the website with two asterisks (**), and a footnote at the bottom of the screen explained that the marked item was on sale and the original higher price was specified. This method of marking discounts was used because of the following reasoning. First, we did not want discounts to be too salient so there would be an advantage to receiving an email that provided information on which items were discounted. Second, we wanted to allow any shopper who accessed the website to find out about the temporary sale if he or she exerted some effort in noticing fine details. The experiment focused on items in 28 categories that were popular with shoppers in the pre-experiment period. Each of these categories include at least 2 items that could be considered substitutes. Each month a different set of categories were discounted so that a discount on an item was valid for one month. The items whose prices were manipulated during the experiment are defined as target items, and their alternatives are defined as substitute items. During the period with the lowest relative discounts (in percentages) on target items, the highest discount was 25%, while during the period with the highest relative discounts, the maximal discount was 75%. The discounted target items fell into four general categories: (i) organic and conventional items, (ii) same items that are offered in different sizes (e.g. jumbo avocado and regular avocado) or bulk quantities (e.g. apples, that are offered as single units or 3lb bags, or milk that is offered in 0.5gal and 1gal) (iii) brand names vs. generic store brand (e.g. Aunt Millie's breads vs. generic supermarket whole wheat bread), and (iv) two competing brands of the same exact product (e.g., Dasani vs. Ice Mountain mineral water in bottles of the same size).

3. Rebates. In weekly emails, shoppers were offered an immediate rebate (applied at the time of checkout) if they spent at least $20 and also bought at least one unit of an item from a given group of eligible items (which changed every week). During the first three weeks of the study, the rebate was equal to the flat delivery fee of $2.99 (it was presented to shoppers as free delivery), and in the last three weeks it was raised to $10. Between the fourth and the eleventh week the rebate was $2.99 for the control group and $10 for the treatment group (the difference between these two groups is explained below).

4. Treatment and control. The 355 shoppers who had made purchases in the second half of 2015 were randomly divided into two groups -- 178 in treatment and 177 in control. The difference between these groups was that treatment shoppers received additional information on discounted items in the weekly email. In order to separately measure the effect of the email contents from a general salience effect or compliance effect, both groups were sent weekly promotional emails. However, during the entire period of the study the email to the control group did not mention any price discounts. In contrast, the email to the treatment group displayed the following: four product categories (e.g., milk, eggs, fruits, bread) that were on (temporary) sale that month, the biggest discount available in each of the categories (expressed in percentage points) and a link to the relevant page of each category. The treatment group was also informed that discounted items were marked by "**". During the second half of the study (from the sixth week on), shoppers in the treatment group began to receive a more detailed weekly email. For these weeks the email included a line alerting shoppers to the fact that many organic items were now on sale and even cheaper than non-organic items. Additionally, those who had purchased a substitute item in a category that is now on sale received a personalized email alerting them to this fact (e.g. "Don't forget to consider some alternatives to your last purchase of eggs that we have on sale this month").

Design details

Randomization method
The randomization was done in office by a computer.
Randomization unit
Individual.
Clusters
355 shoppers.
Observations
355 shoppers who make purchasing decisions in 28 food categories. For each of these 355 shoppers, we track their decision of whether or not to make a purchase in each category over the duration of the experiment (129,220 observations).
Arms
355 shoppers: 177 shoppers were assigned to control and 178 to treatment.
Primary outcomes
Searching for substitutes is costly, and the likelihood of realizing savings from discounts decreases with search costs. The likelihood of buying a discounted item (of weakly higher quality) versus its substitute increases by 21 percentage points (s.e. 3.2) when the discounted item and its substitute appear adjacent on the screen (the average pre-discount relative purchase rate is roughly 26 percent). The effect of the sale is much smaller for items that are farther apart from their substitutes (an increase of 11 percentage points s.e. 1.8). On average, purchasing a discounted item versus its substitute resulted in savings of 88 cents relative to the non-discount price (which averaged $2.99). This means that displaying substitutes next to each other (which then does not require shoppers to search for the different substitutes of a particular product) can lead to a 10 percent increase in the savings uptake, or 9 cents per item for the average shopper. It is important to keep in mind that the average American supermarket carries close to 40,000 items, so that even if only one percent of these items is on sale, the placement of these items could result in significant changes to consumer surplus. Providing shoppers with information on sales does not necessarily lead to savings. Notifying shoppers of food categories with discounts increases their probability of making a purchase within these categories by roughly 200 percent relative to the control group (an increase of 1.6 percentage points (s.e. 0.6) at an average purchase rate of 0.8 percent (s.d. 9)). During each week of the experiment discounts were offered in roughly 13 food categories, such that shoppers could save a total of $11 if they switched to each discounted item from its more expensive substitute in every category. However, the benefit of the sale averaged $4 as 70 percent of the increase in purchases made by treatment shoppers was due to an increase in the purchase rate of the more expensive substitute item. The framing of information impacts choices. The largest difference between the treatment and control groups is observed for items in the discounted food category that was listed first in the treatment email. During most weeks the top listed category consisted of items with the smallest discounts, and while control shoppers were less likely to purchase discounted items in this category, treatment shoppers were more likely to purchase. Thus, the difference in purchasing rates between the treatment and control group is driven by the first item category with treatment shoppers increasing their purchase rate of the discounted item by 1.8 percentage points (s.e. 0.8) more than the control group.

Cited by

No citation data available for this registration yet.

Keywords