Persuasion in Medicine: Experimental Evidence on Sender and Signal Effects

The aim of the study is to identify what sender/signal combinations are most persuasive in encouraging low socioeconomic males living in the U.S. to take-up seasonal flu vaccination. We plan to recruit male participants and randomly assign them to four persuasion treatments: three of which vary dimensions of the sender of a medical recommendation (racial concordance, gender concordance, and authority treatments) and one which varies the signal (standard vs. empathetic). Specifically, we will show participants videos of either Black or white actors providing scripted information on the flu vaccination. We will randomize the race of the sender for all participants and also randomize the authority of the sender for Black participants, with the actor portraying either a doctor or a layperson. We may also randomize the gender of the sender for participants assigned to a racially concordant sender. In addition, we will vary the script used in the experiment between one that acknowledges past injustices and one that does not. We will provide participants a free flu shot coupon and elicit the price at which participants would be willing to trade in the coupon for a cash reward. Lastly, in light of the relevance of vaccination take-up in combating the COVID-19 pandemic, we will assess demand for information about a COVID-19 vaccine, with subjects invited to receive results of a safety and efficacy review from a trusted or standard source. The design requires collection of baseline and endline surveys combined with administrative data from pharmacies about coupon redemption. The primary outcomes of interest are posterior beliefs about seasonal flu, demand and WTP for a free flu shot coupon, redemption of the coupon, and demand for information about a COVID-19 vaccine.

2019-10-31

2021-02-15

Our primary outcomes of interest are prior/posterior beliefs about the risk/benefits of the flu shot, perceptions of the sender, willingness-to-pay (WTP) for a free flu shot coupon, level of attention and recall from the infomercial video, redemption of the flu shot coupon, and demand for information about a COVID-19 vaccine.

We will recruit approximately 3350 African-American or Caucasian adult males with a high-school diploma or less. A pilot study, which ran during the 2019-20 flu season, recruited approximately 850 subjects, and a planned scale-up will recruit an additional 2500 subjects during the 2020-21 flu season. We will oversample African American individuals, emphasizing low-income and minority men because these characteristics are correlated with lower relative take-up of flu vaccination. After informed consent is obtained, subjects will be asked about sociodemographic information (age, education, income and marital status), healthcare experience and past medical history, knowledge and beliefs about flu vaccination, and location of the nearest pharmacy. We will then randomly assign the adult male subjects to videos which contain one of Black or white actors playing the role of either a doctor or a layperson as well as different scripts. The content of the infomercial will be the safety and effectiveness of adult seasonal flu vaccination. After the video, we will obtain information on beliefs regarding vaccination, feedback on the video, attention to the video and willingness-to-pay for a flu shot coupon. We will track coupon redemption, which is redeemable at most pharmacies nationwide.

We will use computer randomization within the online survey.

The unit of randomization is the individual.

No

Not applicable - though we may need to correct for correlation in the error term across actor-script-types.

We plan to recruit approximately 3,350 men (70% African American men and 30% Caucasian men).

Approximately 3,350 men will be recruited. Each subject will be randomly assigned to either racial concordance or discordance group with 50/50 chance. For African American subjects assigned to a racially concordant actor, the subjects will then be assigned to either a doctor or layperson with a 50% chance. For African American subjects assigned to a racially discordant actor, all subjects will be assigned to a doctor actor but will be randomly assigned to either a validation message or normal message with a 50% chance. All Caucasian subjects will be assigned to a doctor actor delivering a normal message. We may also institute a gender concordance treatment in which Black and Caucasian subjects assigned to racial concordance will be assigned to gender concordance or discordance with 50/50 chance. Subjects will be additionally assigned to a COVID-19 vaccine information demand treatment via 50/50 randomization to either a standard or trusted source.

We need a sample size of 2620 (1,720 Black and 900 white respondents) to test our main hypotheses with 80% power and at a significance level of 0.05. Using a sample size of 2620, we can detect a difference in means (in standard deviation units) of at least 0.20 between all our treatment arms. Specific minimum sample sizes needed to detect a 0.20 standard deviation difference in means between treatment arms are provided below. We need 385 subjects for each treatment (i.e. racial concordance) and control (i.e. racial discordance) to test an overall racial concordance effect across both Black and white respondents assigned to an expert figure. To detect a 0.20 standard deviation effect size within each racial group, we need 415 subjects for treatment (i.e. racial concordance) and control (i.e. racial discordance) group within Black respondents, and 370 subjects for treatment (i.e. racial concordance) and control (i.e. racial discordance) group within white respondents. We need 430 subjects for treatment (i.e. concordance) and control (i.e. gender discordance) group within Black respondents assigned to a race-concordant expert delivering a standard message, and 450 subjects for treatment (i.e. gender concordance) and control (i.e. gender discordance) group within white respondents assigned to a race-concordant expert delivering a standard message. We additionally require 430 for each treatment (i.e. empathetic treatment arm) and control (i.e. standard message arm) within Black respondents assigned to a race-discordant expert in order to detect a 0.2 effect size in units of standard deviations. Provided that we recruit 65% Black and 35% white respondents and randomize 50-50 for each treatment, we arrive at an overall minimum sample size of 2620 to test our main hypotheses. We aim to recruit more than 2620 respondents as we plan on recruiting 70/30 Black/white respondents and aim to maximize our power within our budget.

pre_analysis_plan_12oct2020.pdf