Online questionnaires will be sent to a representative sample of United States labor force participants aged 25-54 years old. The target sample is 8,000 respondents. We will use quota-based sampling to ensure representativeness of the sample across broad age categories, regions in the United States, gender, occupational groups, and educational attainment. Respondents will answer a range of survey questions detailing their background (i.e. individual characteristics/personality traits, education, work experience, location etc.) and their beliefs on the likelihood of their job being automated/taken over by robots. The respondents then will be grouped into two treatment and one control group. Randomization will be performed at the individual level, and study participants will have a roughly 33% chance of being assigned to the control group or either one of the two treatment groups. Participants assigned to the two treatment groups will be given truthful information on the average expectation of job automation in their occupation obtained from a previous survey, and how this compares with their own expectations, which we will elicit before treatment assignment. The material from the information treatment comes from surveys conducted over March – May 2020 as part of the “Covid Inequality Project”, with representative samples of members of the labor force in the United States. The two treatments will differ in whether the information will be phrased as coming from other people from the labor force in the United States (`Treatment 1’) or from expert economists (`Treatment 2’). The control group will not receive any information. Then, the respondents will be asked various questions on their political preferences, views on populism, and support for redistribution. This methodology enables the analysis on how a change in perception effect views on populism and political preferences by controlling for gender, age, occupation, industry, municipality, education, income, and personality traits.
Experimental details (hidden): We will estimate the average treatment effect of the treatments on each outcome by running OLS regressions of the outcome on the treatment dummies. To increase precision of our estimates, we will control for baseline characteristics (gender, age, occupation, industry, municipality, education, income, and personality traits). In addition, we will look at several dimensions of heterogeneity. More specifically, we are interested in whether the treatments are differentially effective for (i) men vs women, (ii) old vs young workers, (iii) people in different occupational groups, (iv) people from different socio-economic backgrounds, and (v) people whose expectations about job automation are below / above the numbers from the previous representative study.