|
Field
Sample size (or number of clusters) by treatment arms
|
Before
1250 control condition, 1250 excuse condition, 1250 no excuse condition. To gain statistical precision, we also report some specifications in which we pool with pilot data (N=778, with 400 in the excuse condition and 378 in the no excuse condition).
We ask our survey provider to restrict the survey to respondents who had not taken our pilot. To further ensure that our sample does not include repeat respondents, we also include a post-outcome question asking respondents whether they have taken a previous online survey that discussed the Lott study. We expect this number to be small and plan to exclude respondents who respond in the affirmative from our main specifications (though we will also report results with the entire sample).
Our survey uses Qualtrics' built-in feature to map respondents' IP to their city, which we then mention several times in the survey. This feature occasionally fails (typically when respondents are taking the survey from remote areas), resulting in no city being displayed. We will drop these respondents from all specifications.
|
After
1250 control condition, 1250 excuse condition, 1250 no excuse condition. To gain statistical precision, we also report some specifications in which we pool with pilot data (N=778, with 400 in the excuse condition and 378 in the no excuse condition). We also report the results from the replication experiment (N=701 in the excuse condition and 705 in the no excuse condition).
We ask our survey provider to restrict the survey to respondents who had not taken our pilot. To further ensure that our sample does not include repeat respondents, we also include a post-outcome question asking respondents whether they have taken a previous online survey that discussed the Lott study. We expect this number to be small and plan to exclude respondents who respond in the affirmative from our main specifications (though we will also report results with the entire sample).
Our survey uses Qualtrics' built-in feature to map respondents' IP to their city, which we then mention several times in the survey. This feature occasionally fails (typically when respondents are taking the survey from remote areas), resulting in no city being displayed. We will drop these respondents from all specifications.
|
|
Field
Intervention (Hidden)
|
Before
We randomize respondents into one of three conditions: "excuse", "no excuse", and "control".
We provide respondents in the "excuse group" and the "no excuse group" with information about a recent study (Lott 2018) that finds that undocumented immigrants in Arizona commit crimes at substantially higher rates than comparable US citizens. We tell all participants that they will have the opportunity to authorize a donation to Fund the Wall (www.fundthewall.com), an organization that supports the proposed US-Mexico border wall, and that their individual donation decision will be published on a website that we create (such that others from their city may see their decision). Participants in the "excuse" and "no excuse" groups are shown a screenshot of the website and see that the page contains information about the Lott study, allowing us to hold second-order beliefs (beliefs about what website visitors will know) constant. Participants in the control group see a screenshot of the website, but they do not see any information about the Lott study.
The key treatment varies the availability of an "excuse" for donating: that is, whether the website makes it clear that respondents knew about the Lott study prior to donating. Our main hypothesis is that when respondents know that website visitors will know that they were informed about the Lott study prior to donating (the "excuse group"), they are more likely to donate than when respondents believe that website visitors will believe that they were not informed about the Lott study prior to donating (the "no excuse group"). We also hypothesize that participants in the "excuse" group will be more likely to donate than participants in the control group.
References:
Lott, John R., Undocumented Immigrants, U.S. Citizens, and Convicted Criminals in Arizona (February 10, 2018). Available at SSRN: https://ssrn.com/abstract=3099992 or http://dx.doi.org/10.2139/ssrn.3099992
|
After
We randomize respondents into one of three conditions: "excuse", "no excuse", and "control".
We provide respondents in the "excuse group" and the "no excuse group" with information about a recent study (Lott 2018) that finds that undocumented immigrants in Arizona commit crimes at substantially higher rates than comparable US citizens. We tell all participants that they will have the opportunity to authorize a donation to Fund the Wall (www.fundthewall.com), an organization that supports the proposed US-Mexico border wall, and that their individual donation decision will be published on a website that we create (such that others from their city may see their decision). Participants in the "excuse" and "no excuse" groups are shown a screenshot of the website and see that the page contains information about the Lott study, allowing us to hold second-order beliefs (beliefs about what website visitors will know) constant. Participants in the control group see a screenshot of the website, but they do not see any information about the Lott study.
The key treatment varies the availability of an "excuse" for donating: that is, whether the website makes it clear that respondents knew about the Lott study prior to donating. Our main hypothesis is that when respondents know that website visitors will know that they were informed about the Lott study prior to donating (the "excuse group"), they are more likely to donate than when respondents believe that website visitors will believe that they were not informed about the Lott study prior to donating (the "no excuse group"). We also hypothesize that participants in the "excuse" group will be more likely to donate than participants in the control group.
References:
Lott, John R., Undocumented Immigrants, U.S. Citizens, and Convicted Criminals in Arizona (February 10, 2018). Available at SSRN: https://ssrn.com/abstract=3099992 or http://dx.doi.org/10.2139/ssrn.3099992
In October 2020, we conducted a replication of the two main conditions of the pre-registered experiment (“Excuse” and “No Excuse”) with 1406 respondents recruited through Lucid. A power calculation based on the effect size of our main experiment indicated that we would need a sample of approximately 1400 respondents to have roughly 80 percent power to detect a treatment effect of the same magnitude as in the pre-registered experiment.
|