Worker Preferences Over Monitoring

We inform workers of a German crowdworking platform who are about to work on a task that their activities on the website will be subject to close monitoring. We compare the willingess to pay to avoid being monitored to a control group of workers who do receive any information with respect to monotoring of workers' activities.

2018-11-23

2019-06-30

Willingness to pay (WTP) to avoid monitoring. The total effect will be identified using as an outcome the stated WTP to switch to the alternative version of the task. The extensive margin response will be identified using as an outcome an indicator for a strictly positive WTP.

WTP to avoid monitoring. The total effect will be identified using as an outcome the stated WTP to switch to the alternative version of the task. The extensive margin response will be identified using as an outcome an indicator for a strictly positive WTP.

In order to test if the treatment shifts the perceived difficulty of the tasks, we consider as secondary outcomes survey responses on the perceived difficulty of the default and the alternative version of the task. The survey is implemented twice, before and after the subjects state their WTP. In order to test if the treatment shifts subjects’ behaviour once they work on the task, we consider as secondary outcomes measures for the extent to which subjects successfully complete the task (clicking dice showing a given number). Such measures will include an indicator for subjects successfully completing all rounds of the tasks and the share of rounds solved correctly.

We run the experiment on a German crowdworking platform. The subjects are registered workers and are not informed about the fact that they participate in an experiment. The experiment is framed as a regular task concerned with the framing and presentation of online minijobs. We inform workers that they can choose between two different versions of a task, a default version and an alternative. In the treatment group, workers are informed that if they perform the default version of the task, they will be subject to close monitoring. In the control group, no information regarding monitoring is provided. Applying the BDM method, we then elicit workers’ WTP for switching to the alternative version of the task. Workers then work on the task selected by the mechanism.

We run the experiment on a German crowdworking platform. The subjects are registered workers and are not informed about the fact that they participate in an experiment. The experiment is framed as a regular task concerned with the framing and presentation of online minijobs. In the job post, we inform workers that they can choose between two different versions of a task and that they can earn a payoff between €1.50 and €2.50. The entire experiment lasts about 8 minutes and we notify all workers that no data is shared with the crowdworking platform or any other party.

All workers earn a fixed participation fee of €1.50 conditional on completing the HIT. The welcome page informs workers about potential payoffs and about the fact that they will not be paid if they do not perform the task as requested. After the welcome page, workers have to pass an attention check. The following instructions inform workers that they can choose between two versions of a task, a default version and an alternative. The instructions do not provide any details about the versions of the task, but workers are informed about the fact that both versions of the task consist of a clocked activity that has to be performed for 1:30 minutes, that in both versions workers cannot pause during the task, and that irrespective of the version performed, the clocked activity requires the same effort.

After the instructions page, workers are redirected to a survey page consisting of a question regarding the perceived relative difficulty of the two versions of the task. Workers state their perceived relative difficulty on a scale consisting of five items (version A much easier than version B, version A somewhat easier than version B, version A and version B equally difficult, version A somewhat more difficult than version B, version A much more difficult than version B).

Next, workers are redirected to a page that informs them that the next step will be the decision which version of the task they will perform. In the treatment group, workers are informed that if they perform the default version of the task, they will be subject to close monitoring. The statement reads: „All your activities and entries on the website are subject to close monitoring”. After being redirected to the next page, workers are informed about the mechanism that selects the version of the task to be performed. Workers are informed that they receive an endowment of €1.00 that they can use to pay for switching to the alternative version of the task. Applying the BDM method, we then elicit workers’ WTP for switching to the alternative version of the task. To state their WTP, workers use a slider ranging from €0.00 to €1.00 with €0.05 intervals. The price for switching is randomly drawn from a uniform distribution. Before learning about the outcome of the mechanism, workers are again asked to state their perceived relative difficulty of the tasks. The survey is implemented using the same wording and the same response items as the first survey on the perceived relative difficulty. The next page informs workers about the outcome of the mechanism. If the drawn price is larger than the stated WTP, workers are informed that they will perform the default version of the task. If the drawn price is smaller or equal to the stated WTP, workers are informed that they will perform the alternative version.

Given the evidence from previous literature that some subjects (typically the less experienced ones) misconceive the BDM mechanism, resulting in a stated WTP that does not reflect the true WTP, we might also switch to a version of the mechanism where subjects are not asked to state their WTP, but to indicate if they are willing to pay for switching to the alternative version of the task for a complete menu of possible prices. The mechanisms would then randomly draw a price, and the pre-determined choice of the subjects for the drawn price would be implemented.

After being redirected to the next page, workers are informed that they are about to start working on the task. The page shows an example of the task. The task is the same, irrespective of whether workers perform what has been labelled the default version or what has been labelled the alternative version. The task consists of a panel of 25 dice showing different numbers. Workers perform the task for 15 rounds. In each round, they are asked to click a die showing a given number, with the number changing between rounds. In each round, the panel of dice is visible for six seconds. If a workers does not click any die in more than one round, she is informed that she did not perform the task as requested and as a consequence, her participation has been terminated. Clicking a die that does not show the correct number does not lead to exclusion from the experiment. Workers who drop out at this stage (or at any other stage) do not receive any payoff.

After completing the task, workers are asked to fill in a short demographics survey (gender, age group, type of school completed, region (state) where schooling was completed), before being redirected to the final page providing the redemption key for claiming the payoff earned during the experiment.

On-the-fly randomization using python. Randomization into treatment and control takes place once workers arrive at the web page hosting the experiment.

Individual worker

No

We do not cluster the randomization, implying that the number of clusters is equal to the number of observations (workers).

We ran pilots to test the functionality of our webpage and to estimate the parameters for a power calculation. Based on these pilots, we estimate the required sample size for a two-sample means test (power 0.8, alpha 0.05) to lie in the range of 150 to 500 subjects. We note that the power calculations based on the pilots are quite imprecise in terms of required sample sizes. This is due to the high share of workers stating a WTP of zero (up to 70% in some subgroups), leading to imprecise estimations of the standard deviation of the stated WTP.

Symmetric sample sizes between treatment and control, meaning 75 to 250 subjects (workers) in treatment and control.

From our pilots (with limit sample size), we estimate the mean WTP in the control group to lie between 0.13 and 0.2. The SD of the WTP is estimated to lie between 0.24 and 0.3. Assuming a maximum sample size of 500 workers and symmetric treatment groups and based on a two-means test, we estimate the minimum detectable effect size from a linear model for the WTP to be about 0.07 Euro.