Structural Analysis of Xenophobia

We use a two-layered experiment design. We have two interventions, so our sample will be divided into four treatment arms.

The first intervention is the manipulation of publicity. The treatment group will receive a message telling them that their actions will be public while the control group will be told that their actions are anonymous.

The second intervention randomly provides information. The treatment group will see one animation about rising anti-Chinese sentiment in the US. The control group will not see the animation.

Besides, we ask List randomization questions to assess potential social desirability bias in anti-Asian sentiment and perception about racism against Asian Americans.

2020-12-16

2020-12-23

We have three primary sets of outcome variables.
The first set of outcome variables is about anti-Asian sentiments. The second set of outcome variables is about the perception of racism in the US. The third set of outcome variables is three measurements of discriminatory behaviors.

We collect multiple measurements for each set of outcome variables, not to cherry-pick the most responsive survey items later, but to apply nonparametric estimation techniques to estimate measurement errors. Having multiple measurements is one of the essential identifying assumptions.
The first set of outcome variables on anti-Asian sentiment is a shorter version of the realistic threat measure, intergroup anxiety, and prejudice measures used in Stephan et al. (1999).
The second set of outcome variables on perception about racism in the US is measured by asking how much respondents agree to five statements. The statements are (1) whether a person who publicly says he/she dislikes Asian Americans can maintain good social relationships with most people, (2) if a person refers to the novel coronavirus as “China virus,” the person will face severe criticism, (3) whether people will stand by Asian Americans and will fight for their rights, (4) whether people will support if the US government reduces the number of Asian immigrants because they threaten this country, (5) whether people will support if the US government deports Chinese immigrants if they have any Chinese government connections.
The third set of outcome variables is discriminatory actions. We will ask three questions: a petition question, a dictator game, and a donation question. The dictator game is an incentivized survey item with real money at stake, and the other two questions ask about intentions to donate and to sign a petition.
In the petition question, we present two different petition forms, with one urging to protect the United States’ interests from Chinese threats and another urging to protect the safety and rights of Asian Americans in the US. We ask which petition survey participants would like to sign. If they choose to sign a petition to protect the US from Chinese threats, the racist action is coded as 1.
In a dictator game, we randomly match respondents with two other survey participants, who gave different answers in the previous petition question, respectively. Next, we ask respondents to split $1 between themselves and their matched partners. To remove deception, we will randomly select 10% of the survey sample and will make actual payment to survey participants based on responses. We will explain that their answers will not affect the probability of winning the lottery. We repeat dictator games twice with two different partners to remove individual fixed effects, which includes generosity in sharing money with someone else. We randomize the order of two dictator games to remove any order effect. If survey respondents share more money with a partner who signed a petition to protect the country from Chinese threats than with a partner who signed a petition to protect Asian Americans’ safety and rights, the racist action is coded as 1.
In a donation question, we present descriptions about two organizations with opposing attitudes to Asians: the Center for Immigration Studies (CIS) and the Asian Pacific Policy & Planning Council (A3PCON). Next, we ask which organization respondents are willing to donate if they are given $1. If they choose to donate to CIS, the racist action is coded as 1.

Reference
Stephan, Walter G., Oscar Ybarra, and Guy Bachman. "Prejudice toward immigrants 1." Journal of Applied Social Psychology 29.11 (1999): 2221-2237.

To screen out survey participants who pay little attention to the survey, we have inserted two attention check questions before the two interventions. In the first attention check question, we ask respondents how they are currently feeling, but we tell them to check only the “None of the above” option in the middle of the question. Survey participants who fail to pay attention to our message and choose any other choices will be screened out. The second attention check question uses the same trick except that we ask the respondents which device they are using to participate in the survey. We ask them to check “Other” as a correct answer.

To remove any order effect, we randomize the order of multiple choices when we measure racist actions.

To screen out survey participants who pay little attention to the survey, we have inserted two attention check questions before the two interventions. In the first attention check question, we ask respondents how they are currently feeling, but we tell them to check only the “None of the above” option in the middle of the question. Survey participants who fail to pay attention to our message and choose any other choices will be screened out. The second attention check question uses the same trick except that we ask the respondents which device they are using to participate in the survey. We ask them to check “Other” as a correct answer.

To remove any order effect, we randomize the order of multiple choices when we measure racist actions.

We randomize using a Qualtrics randomizer.

We do individual level randomization.

No

100 in the pilot survey and 3000 in the main survey.

100 in the pilot survey and 3000 in the main survey.

50% of the sample is in the treatment arm and 50% of the sample is in the control arm.

We will compute MDE using the pilot data, which is yet to be collected.