The Political Feasibility of Income Guarantees

Each individual in my experiment will receive a survey. The first half of the survey asks about basic demograhpic information. The second half of the survey asks individuals.

2025-05-12

2025-05-13

The primary outcomes of interest will be the binary decisions of individuals in responding to the hypothetical redistribution outcomes asked about in the survey are desirable.

Each subject will fill out a simple, short survey. Part 1 will collect basic demograhpic information about subjects (Gender, Age, Marital Status, Children, Ethnicity/Race, US Citizenship, State of Residence, Education, Employment Status, Income, President Vote, Political Identity).

In the second part, I will ask them to answer a series of questions of the following form:
"Suppose that the government spent $1,000, and as a result transfers $Y {in cash/in value through currently existing welfare programs} on top of current benefits to a household who 1) currently makes income I, 2) will respond to the transfer by reducing labor supply by D. {In addition, the household in question has a job which is consider to be at high risk of being automated due to AI., In addition, the household in question has a job which is not considered to be at high risk of being automated due to AI.} Do you think this transfer was a net positive or net negative for society?"

From question to question, I vary Y, I, D, as well as the discrete options in braces and record the survey-taker's response to each question.

Randomization is done through the JavaScript code underlying the Qualtrics survey.

Questions will be randomized at the individual-question level.

No

2500 subjects

2500 subjects answering 10 binary choice questions.

The distribution of "treatments" is drawn from a continuum of values, so this question is poorly defined for my setting.

The purpose of this experiment is not to estimate a specific treatment effect but instead to generate enough variation to well-identify the parameters of a structural model of political preferences. This structural model's parameters are ultimately pinned down by conditional choice probabilities of the form Pr[Yes | Y,I,D,characteristics] as defined above. Assuming that Y,I,D are normalized to be i.i.d. Uniform random variables, this involves estimating 4 3-dimensional CCP functions. I give a lower bounds for how precisely this CCP can be nonparmaetrically estimated. Assume a naive kernel-based estimator where to make a prediction for a given conditioning variable, we take an average of all observations within a 0.1 radius of a given point. With 25,000 observations (10 questions x 2,500 participants), this gives roughly 240 observations to estimate this CCP, which therefore has gives us a margin of error of roughly +/- 5% of the true CCP.