We recruit subjects on Amazon Mechanical Turk (AMT) to an online experiment which consists of two parts. The first part is a survey in which we elicit demographic information (age, gender, education) as well as measures of cognitive ability, trust, risk preferences, and labor supply on AMT. After the survey, we describe the second part, which consists of an experimental search task.
In the search task, subjects have to buy a fictitious product. They can search for the lowest price for this product in up to 100 shops. The price at each shop (in USD) is drawn from a uniform distribution on the interval [a,b]. This distribution is shown to subjects in the instructions to the search task and on the screen where they conduct their price search. Their payoff from the search task if they purchase the fictitious product at price p equals b – p. Upon entering the search screen they can also push a button to indicate that they do not want to search at all. In this case, their payoff from the search task is zero. Before subjects can enter the search task, they have to respond to a comprehension check. After entering the search task, subjects have three days for searching, i.e., they can take breaks and return to the search task by clicking on the link to the experiment.
To search an online shop, subjects have to record and manually enter a 16-digit code. The code varies between shops and subjects. The copy-and-paste option is disabled so that the task requires some effort. After discovering the price at a shop subjects see an overview page with all prices found so far. They can then purchase the fictitious product at any previously sampled shop or continue their search (the maximal number of searches is 100).
We implement the following treatments.
(1) Piece Rate Treatments: These treatments are identical to the baseline treatment, except that the price interval is given by [za,zb] for some value z < 1 and subjects earn a piece rate g > 0 for each searched shop in addition to the realized price savings. We consider three piece rate treatments with varying values of the piece rate g.
(2) Scale Treatments: These treatments are identical to the baseline treatment, except that the price interval is given by [za,zb]. In one scale treatment we have z = 1 (i.e., this treatment is identical to the baseline treatment), in another scale treatment we have z > 1.
(3) Distribution Treatments: These two treatments are identical to the baseline treatment, except that the distribution over prices is skewed to the right (or to the left). That is, the support of the price distribution is still [a,b], but there is more probability mass on low (or high) prices.
(4) No-information treatments: These treatments are identical to the baseline treatment and the two distribution treatments, except that subjects obtain no information about the price distribution, neither in the instructions nor on the screen where they conduct their search.
In all treatments, subjects earn 1 USD for the completion of the survey in the first part of the experiment (in addition to the earnings in the second part).
We will exclude subjects from the final sample who do not conduct at least one search and also do not indicate that they do not want to search at all (by pushing the corresponding button). Moreover, we exclude subjects who do not purchase the fictitious product at the lowest discovered price.