xThe AEA RCT Registry will be down for maintenance on Tuesday, April 20th, from 9pm EDT to 11pm EDT to perform necessary upgrades. The site will be in maintenance mode during that time. We apologize for any inconvenience this may cause.
The main outcome is the effort exerted by workers in each session, measured as the number of envelopes stuffed (in each 20 minute session).
Effort exerted by workers in each session, measured as the number of envelopes stuffed (in each 20 minute session).
Randomization of treatment ordering by computer before start of experiment. Assignment to positive, positive in-kind, negative and neutral gift treatments within each daily session happens by stratifying on worker performance in the first half of the day, as described below. In addition, we have variation within subjects in the fixed wage, piece rate and return to employer as they work in 10 different work periods over the course of the day.
Randomization of treatment ordering by computer before start of experiment. More details in attachment, which becomes available after the end of the experiment
The experiment uses two types of between-subject variation:
1. First, the order of the experimental sessions is randomized. There are 12 types of sessions, which vary (a) the session order (UP or DOWN), (b) the charity order (CHARITY ORDER 1, 2 or 3) in which one of the charities (B, RN or RIC) corresponds to Charity No 1 in the experimental design, and (c) the match order in the final two sessions (MATCH FIRST or MATCH LAST). This produces 2x3x2=12 treatment session types. The order of the 12 treatments was randomly drawn at the beginning of the study. We plan to hold at least 48 experimental sessions (of which 24 have already been completed as of the time of this posting) and ideally 72 total sessions if we can find enough subjects to recruit. In total, there will be at least 4 full loops through the complete set of 12 treatment sessions, and ideally 6 full loops. In any case, we will aim to run a completed loop though all the 12 sessions constituting one full run. On each day that the experiment is run, either one or sometimes two experimental sessions will be run (depending on how many individuals respond to the posted advertisement and show up).
2. The second randomization is at the individual level, within a given experimental session. Randomization occurs during lunch within an experiment session and is used to determine whether a subject is put into the positive gift, in-kind gift, negative gift, or control (neutral) gift treatment. The total number of envelopes each subject created in the previous sessions is totaled and their rank determined. The highest and lowest rank were put into one treatment, the second highest and second lowest were put into another treatment, and the middle two were put into the third treatment. In the very first experimental session, the highest and lowest rank were randomly placed in the negative gift. Following this, in the second group, the highest and lowest rank were place in the the neutral gift. In the third session, they were placed in the positive gift. This pattern was then repeated. Thus, the randomization was one-time and affected which treatment the first pair of "highest and lowest" workers were assigned to. After that, assignment was deterministic, although of course it depended on worker performance on a given day. The goal was the have ability be balanced over time in the different gift treatments combining over all the sessions, and as similar as possible on average even within a session.
NOTE: This experiment registration, including the full specification of the model and planned structural estimation, was written on 21 Nov 2014. This was after one full phase of data collection, with data on 131 subjects collected and two full rounds of treatment orders (that is, 24 sessions) completed. We plan to gather data on approximately 200 more subjects, making for a total of about 330 workers. But note that exact sample sizes will depend on (unpredictable) show-up of the invited workers. This will require collecting at least another two full rounds of treatment order sessions (T1-T12 twice over, in randomized order) and ideally four more full rounds (pending enough subjects to recruit). Notice that the only difference between the sessions run after Nov. 21, 2014 is the addition of an "in-kind gift" treatment, where workers are provided with an unanticipated gift (a thermos with the logo of the employing charity) in the gift periods, so the subjects are split four-ways in rounds 9-10. This treatment was not included in the first 24 sessions, since we thought of including this treatment after the initial 24 sessions.
In addition, we also utilize within-subject variation:
Each worker works for 10 work sessions in day, during which they face varying incentives. This provides a rich source of within subject variation.
The first of the ten work sessions for each subject is a training session, where the worker is paid to stuff practice envelopes, but the envelopes are discarded and not mailed by the employer (as always, there is no deception in the experiment, and the envelopes are truly discarded as announced to the workers). The fifth session, which is immediately after lunch, is a second training session. The last two sessions -- 9 and 10 -- are always the gift sessions, where workers receive unanticipated positive, negative, in-kind gifts or no gift. The remaining six sessions (other than the two training and two gift sessions) vary in order in one of two ways. Each subject is assigned to either UP or DOWN order, each of which specifies a particular sequence of the work sessions. Importantly, the two orders are reversed or mirror images of each other (excluding the position of the training and gift sessions, which are fixed). Particular pay scheme + employer combinations are thus observed (in different subjects) in two different positions -- once early and once late in the day. By averaging across these two occurrences, we can partially deal with issues of learning and tiredness over the course of the day.
Individual level randomization and random ordering of treatments (i.e. randomization at the session level). Full details in the attachment, which becomes publicly available after the experiment is conducted.
Planned Number of Clusters
Approximately 320 subjects, which might vary slightly due to unpredictable variation in no-shows by the invited potential workers. Of course, workers do not know their treatment status before showing up. Note that while treatment varies within each experimental session (since workers are assigned to different gift treatments for the last two work periods), we can allow the clustered standard errors at the experimental session level in the analysis.
Approximately 320 subjects, which might vary slightly due to unpredictable variation in no-shows by the invited potential workers. Of course, workers do not know their treatment status before showing up. While gift treatment varies at individual level, we can cluster standard errors at the experimental session level in the analysis.
Sample size (or number of clusters) by treatment arms
Approx 90 workers in each of positive gift, control and negative gift treatments, and approximately 50 in the positive in-kind gift session. Approximately 300/12=25 workers in each of the 12 experimental treatments. Total sample size will depend on the number of job seekers who actually show up for each session. We intend to have a total of 48 experimental sessions. The number of experimental sessions might increase if worker show up per session is lower than anticipated.
Approx 90 workers in each of treatment arm. Full details in attachment.