Avoiding Dissonant Information

2022-08-16

2023-02-16

The proportion of subjects who are willing to pay to avoid dissonant information,
The mean amount of money subjects are willing to pay to avoid dissonant information

Subjects indicate their willingness to pay to avoid dissonant information by completing a task based on a news article which is against their prior beliefs about an issue (women's abortion rights). I ask subjects if they are willing to pay to switch the article that is assigned to them which is against their prior beliefs about abortion with another one which is in line with their prior beliefs about abortion and how much they are willing to pay to switch them.
Subjects who choose to pay any positive amount of money to switch the article which is against their prior beliefs with an article which is in line with their prior beliefs to complete the incentivized task will give us the proportion of people who are willing to pay to avoid dissonant information / avoid completing an effort task which includes dissonant information.
The maximum amount of money subjects are willing to pay to switch articles will be used to create the mean value in each treatment group.
More details about the comparison of the primary outcomes among treatments can be found in the analysis plan.

The performance of subjects in tasks with opposite beliefs, attention avoidance, gender, age (generation), state of residence (USA), political view and extremism, information avoidance scale, belief updating, media consumption, the extremism of the belief on abortion rights.

Tertiary: other demographics

The information preference scale will be measured using the scale developed by Ho et al (2021).

Attention avoidance will be measured by the number of correct statements written in Q1 of each effort task and time spent.

More details on secondary outcomes can be found in the Analysis Plan.

The experiment seeks to examine how varying beliefs about abortion rights and prior exposure to dissonant information affect information and attention avoidance. The experiment starts by asking participants' opinions on abortion rights: Pro-choice or Pro-life. I filter only Pro-choice participants for Session 1 and only Pro-life participants for Session 2 of the experiment. Then, I provide subjects with an effort task which includes reading a short article and answering some questions based on the article. Subjects are told that they will receive 0.10 $ per correct answer they give to these questions. Half of the subjects are randomly allocated to a consonant group (i.e. the article is in line with their beliefs on abortion rights) and the other half are randomly allocated to a dissonant group (i.e. the article is against their beliefs on abortion rights). It means that half of the subjects in each session complete the effort task on an article which is in line with their beliefs on abortion rights and the other half of the subjects in each session complete the effort task on an article which is against their beliefs on abortion rights. The content and which side of the argument the article supports are made clear by giving participants the title of the article and a sentence-long summary of its content. Once they answer the questions based on the article they are randomly allocated to in the first stage, subjects are presented with a different article in the same way as before (i.e first title and a sentence-long summary). However, this time before moving on to reading the article, they are asked if they would like to switch the article assigned to them with a different one which is in line with their prior beliefs on abortion. If they decide to switch articles, they are told that they can use the separate pot of 100 cents given to them at the beginning of the experiment. Any unused amount is added on top of their bonus payments. Once they indicate their preference to switch articles and their willingness to pay, a random number between 0 and 100 is drawn. If their maximum willingness to pay to switch articles is greater than or equal to the random number, the article that is assigned to them (against their beliefs on abortion) is replaced by another one (in line with their beliefs on abortion) to complete the task. If their maximum willingness to pay is less than the random number, then the article which is against their beliefs is not replaced by the one which is in line with their beliefs. Depending on the result of the lottery and their willingness to pay, subjects complete the effort task in the final article.
I also ask some questions about participants' beliefs and background characteristics such as posterior beliefs on abortion rights, political beliefs, media consumption, demographic information, risk preference, and information preference scale (IPS) (Ho et al, 2021).

See the attached documents (the transcripts and the experimental overview and analysis plan), which include more detail regarding the questions asked, treatments, subject recruitment and exclusion criteria.

Randomization is done by Qualtrics

Individual level

No

NA

1000 subjects Any small deviations from these exact figures will be due to sampling discrepancies generated by the Prolific platform.

250 subjects in each treatment. ( 500 participants from the Pro-life group and 500 from the Pro-choice group. 50% of each group will be assigned to a dissonant group and the remaining 50% will be assigned to a consonant group.)

My primary interest is to gain an estimate of the proportion of subjects who are willing to pay any positive amount of money to avoid information. This does not require a statistical test. I base my power calculation on comparing the proportion of people who avoid dissonant information among different treatment groups. The statistical test for this is a two-sided difference in proportions test under the null hypothesis that the proportions of people who avoid dissonant information in our treatments regarding prior beliefs and initial exposure to dissonant information are the same. The relevant assumptions, the Stata command used and its result are below. Power: 0.8 Alpha: 0.05 P1: 0.70 P2: 0.58 Stata command: "power twoproportions 0.7 0.58" Result: N = 250 per treatment In total, I need 1000 observations. In order to account for the observations which might be lost after the data collection period because of exclusion criteria mentioned in the Analysis plan, I will collect 25% more observations [A previous experiment run in Prolific suggested that about 25% of the time participants fail the attention checks, implying N=1000*1.25=1250 observations. To be confident in securing adequate numbers, I will collect N=1500 (350 per treatment).] The second power analysis is based on the second primary outcome variable: the mean amount of money subjects are willing to pay to avoid dissonant information. The statistical test for this is a two-sided difference-in-means test under the null hypothesis that the mean amount of money subjects are willing to pay to avoid dissonant information in our treatments regarding initial exposure to dissonant information and prior beliefs are the same. As the power analysis for this comparison requires a smaller sample size, the binding criterion for the power calculations is the one based on the proportion of people who are willing to pay any positive amount of money to avoid dissonant information. The required sample size is 1500. Note: Effect sizes used in the power analysis are based on the pilot study.

Experimental Design and Analysis Plan