The Effect of Atheism Primes on Self-reported and and Implicit Religiosity

Manipulations
The primes used for the study consist of short video presentations (in English) and a writing task in which subjects are asked to summarize the main point of the primes. There are three conditions: emotional anti-religious argument, rational anti-religious argument, and control. Each prime is approximately 4 minutes long and consists of a speaker making various points about religion, or in the case of the control, about vegetables. The speaker is the same Kenyan narrator in all three videos. Bullet points summarizing the main arguments appear on the screen, accompanying the auditory presentation by the speaker. In the emotional condition, the speaker argues that any God must be very cruel if he allows so many people in the world to suffer, concluding that it is unlikely that God exists. In the rational condition, the speaker argues that the existence of God is unnecessary to explain the universe given our scientific knowledge. In the control condition, the same speaker instead argues that it is important to eat vegetables every day. After viewing the video presentations, participants are asked to summarize the main arguments in their own words by typing into a text box on their screen and then asked a series of questions to ensure comprehension. we will include analyses restricted based on comprehension as described below. The primes were randomly assigned within each session, with approximately 1/3 of participants in each condition. The setup of the lab included dividers between computers so that subjects were unaware of the images shown on the screens of others. Additionally, all subjects wore headphones, so they were unaware of what other participants were hearing. All images used in the implicit religiosity measure (described below) were vetted for comprehension by the experimenters and were further validated through debriefing after several pilot sessions.

2015-04-25

2015-06-15

1. Self-reported religiosity
Self-reported religiosity is one of our two main outcome variables of interest. Following Shariff, et al (2008), we ask a series of 6 questions intended to gauge religiosity. Individuals are asked to rate the extent they agree with each of the following statements on a scale of 1 (strongly disagree) to 5
(strongly agree):
 My personal religious beliefs are very important to me;
 My religion or faith is an important part of my identity;
 If someone wanted to understand who I am as a person, my religion or faith would be very important to that;
 I believe strongly in the teachings of my religion or faith;
 I believe in God;
 I consider myself a religious person

2. Implicit religiosity
Implicit religiosity is one of our two main outcome variables of interest. We employ a single-target Implicit Association Test (ST-IAT) adapted from (but not identical to) the design used by Shariff et al.(2008). The ST-IAT is a computer-based sorting task that uses response time to measure unconscious associations with a target concept. In each block of the ST-IAT, subjects sort three categories of words to the left- and right-hand sides of the screen: synonyms of “real”, synonyms of “imaginary”, and words associated with religion. All language used in the task was vetted for comprehension with the sample population during several pilot sessions. In one block, the subjects sort the religious words and synonyms of “real” to the same side. In a second block, the subjects sort religious words and synonyms of “imaginary” to the same side. The order in which these two blocks occur is randomized. The ST-IAT design assumes that subjects will more quickly sort target words to the side that represents their implicit association with the target. As a manipulation check for the ST-IAT, we also administer ST-IATs using alternatively the names of cartoon characters familiar to most Kenyans and the names of natural landmarks in Kenya. We hypothesize that individuals should be relatively faster to associate cartoons with the concept imaginary and faster to associate natural landmarks with the concept real.

3. Cantril Ladder
To assess how religion may influence an individual’s sense of status, we present participants with a Cantril ladder, which asks participants where they would rate their current life compared to the best possible life they can imagine for themselves, both now and in five years by indicating where they stand (will stand) on a picture of a ladder with steps labeled 0 to 10.

4. Tolerance
We adapt a series of questions from the World Values Survey asking participants “Of the following groups of people, which would you not like to have as neighbors?”
 Drug addicts
 People of a different race
 People who have AIDS
 Immigrants / foreign workers
 Homosexuals
 People of a different religion
 Atheists/people with no religion
 Heavy drinkers
 Unmarried couples living together
 People who speak a different language

5. Positive and Negative Affect Schedule (PANAS)
To assess the possible effect of the primes on affect, individuals complete questions from the PANAS evaluating negative affect after viewing the primes. On a scale of 1 (not at all) to 100 (very much), subjects are asked “How much do you feel...”
 “Distressed, at this moment?”
 “Upset, at this moment?”
 “Guilty, at this moment?”
 “Ashamed, at this moment?’
 “Hostile, at this moment?”
 “Irritable, at this moment?”
 “Nervous, at this moment?”
 “Jittery, at this moment?”
 “Scared, at this moment?”
 “Afraid, at this moment?”
 “Frustrated, at this moment?”
 “Stressed, at this moment?”

6. Demographics
We ask individuals to report their gender, age and religious affiliation.

Self-reported Religiosity:
We evaluate the impact of priming on each of the questions individually, and test for joint significance using Seemingly Unrelated Regression (SUR). In addition, we will analyze a weighted-average index based on the methodology in Anderson (2008). Finally, we will also examine impact of priming on any latent factors identified using factor analysis restricting the analysis to factors with eigenvalues greater than 1 and using a varimax rotation.

Implicit Religiosity:
Using the result of the ST-IAT described above, for each individual we will calculate a D-score as the difference between the mean latency when associating the target concept with the concept "real" and the mean latency when associating the target concept with the concept "imaginary" divided by the pooled standard deviation. Latency (response time) is recorded in milliseconds and shorter latency indicates a stronger implicit association. A lower D-score represents a stronger belief in the validity of religious concepts.

Following the recommendations in Greenwald, et al (2003), we will exclude participants for whom more than 10% of responses are below 300 ms, as well as all responses over 10,000 ms. Individuals who initially respond incorrectly in a trial are required to press the correct response before proceeding to the next trial. We measure latency as the total time from beginning the trial to the entry of a correct response, effectively penalizing incorrect responses with longer latencies (Greenwald, Nosek, and Banaji 2003). To check the robustness of our results, we will also analyze the reports using the approach of Lowes et al. (2015). Instead of dropping responses over 10,000 ms, we will winsorize reponse times to 3,000 ms and replace response times for trials that were initially incorrect with the block mean response time plus the block standard deviation.

Cantril Ladder: We will calculate z-scores for responses to both the present-date and 5-year Cantril ladder question.

WVS Tolerance: We will analyze individual responses and test joint significant with SUR regression; in addition, we will calculate both aggregate and weighted average indexes across all responses.

PANAS: We will evaluate whether priming impacted affect by analyzing individual responses and testing for joint significance with SUR regression; in addition, we will analyze the summary indices of this measure. We will also use PANAS results as a dimension of heterogeneity.

1. Sampling
The study was conducted at the Busara Center for Behavioral Economics (Busara) in Nairobi, Kenya, a facility specially designed for social science studies. Busara maintains an active subject pool of more than 7,000 Nairobi residents. For the present study, 318 subjects who had previously signed-up to be part of the Busara subjet pool were recruited from local universities using SMS and phone calls and were informed that they would be paid KES 300 (approx. USD 7.20 PPP)1. Subjects were told that they were invited to participate in a study about their behavior and preferences. Recruitment was limited to university students to ensure comprehension of the arguments contained in primes. Restricting participation to University students also ensures comprehension English. Although Universities do not typically require an official test of English proficiency, matriculating students are expected to be proficient in written and spoken English, and much of the instruction is in English. Additionally, Busara has confirmed through previous studies that the vast majority of Kenyan university students are highly proficient in English. The sample includes 183 males and 133 females. All subjects were over 18 years of age, with a mean age of 22 and a maximum age of 26.

To control for heterogeneity in religious background, we restricted the sample to Christians by omitting individuals identified as ethnically Nubian from recruitment. Since most Muslims in the Busara subject pool belong to this ethnicity, we believed this to be the best way to identify Christians while avoiding any issues of self-selection that might arise by asking individuals their religion before participation. As described below, 24 individuals in the study either failed to report a religion identified as “no religion” or identified as “other,” and 2 individuals identified as Muslim. We will control for this fact by restricting our analysis as described below.

2. Experimental Procedure
After receiving a text message inviting them to participate in the study, individuals came to the Busara lab in Nairobi, Kenya, for experimental sessions lasting approximately one and half hours. Each session included up to 25 participants. Sessions were administered by two female Kenyan research assistants, who spoke English and Swahili fluently and were trained in helping subjects with comprehension. The experiment was conducted in English.

Subjects were randomly assigned to one of 25 computer workstations with partitions on three sides, so that they were unable to see or speak with the other subjects. Within each session, individuals were randomly assigned to one of the three conditions. All of the treatments and measures were implemented on HP TouchSmart 310 desktop computers running Windows 7. Each participant wore headphones and watched the video prime on his or her own computer. Subjects used the touch screen exclusively to mitigate effects of individual differences in experiences using a mouse and keyboard. All treatments were implemented using z-Tree software (Fischbacher 2007).

Each session progressed as follows:
1. Practice IAT
2. Cartoon ST-IAT
3. Nature ST-IAT
4. Video Prime (4 minutes)
5. Writing task and comprehension questions
6. Self-reported religiosity questionnaire
7. Practice IAT
8. Religion ST-IAT
9. Cantril ladder
10. WVS tolerance questionnaire
11. PANAS questionnaire
12. Demographics survey
At the conclusion of the final questionnaire, participants were debriefed and paid KES 300 in cash.

Computer

Individual

No

None

300 individuals

100 control; 100 science priming condition; 100 emotional priming condition