Experimental Design Details
We design a simple transcription task on MTurk. We ask workers to assess whether the receipt is legible and, if so, transcribe each item and the price. In addition, we ask them to add up all numbers as a voluntary task. Participants are randomly assigned the name of either a black of a white supervisor. After the fourth receipt, 60% of participants are randomly assigned to receive feedback on their previous performance. Specifically, the assigned manager informs them about whether they are above or below average. Participants then continue transcribing three additional receipts. The group that does not receive any feedback serves as the control group. Afterwards, we collect attitude outcomes such as whether they are interested in working for us again in the future and how much they agree with statements such as “The feedback I received was accurate.” We then debrief participants about the specifics of the research question (the HIT mentioned performance will be recorded for research purposes) and invite them to complete a short survey in which we collect demographic characteristics as well as a test that assesses the implicit bias that participants hold with regard to race. A note on the choice of manager names: One concern with the design is that names are correlated with other attributes than just race. For example, research shows that names are correlated with demographic characteristics such as age, education and income. To account for these potentially confounding facts, we identify both balanced and unbalanced name pairs along these characteristics through an out-of-sample survey in our study population.
Addendum (Oct. 10): In addition the the balanced and imbalanced group, we also include a set of names for which there is an "average" difference in education levels between blacks and whites. These include Christine and Jake (white) and Jamal and Lakisha (black). The analysis will look at three groups: balanced, average imbalanced (new group) and very imbalanced (previously called imbalanced).
We also replicate results for a new list of balanced names, including Tiffany and Jimmy (white) and Jalen and Tyra (black).
Addendum (Nov 12): In order to test the robustness of result and ensure that results are not just driven by particular name parings, we will also add additional name pairings for the:
1) unmatched sample: Dandre, Precious, Nathan, Hannah
2) average matched sample: Deshawn, Lakisha, Chloe, Ethan
As a result we have a total 4 name black-white name pairs for each of the 3 categories (matched, unmatched, average matched).