Synthetic Sample of US population

A statistically representative synthetic sample of 20,000 Americans. Each record is a simulated survey respondent.

people

Format

A tibble with 20,000 rows and 40 variables:

id: Sequential unique ID
fname: Random first name, see details
lname: Random last name, see details
gender: Biological sex
age: Age capped at 85
race: Race and Ethnicity
edu: Educational attainment
div: Census regional division
married: Marital status
house_size: Household size
children: Has children
us_citizen: Is a US citizen
us_born: Was born in the Us
house_income: Family income
emp_status: Employment status
emp_sector: Employment sector
hours_work: Hours worked per week
hours_vary: Hours vary week to week
mil: Has served in the military
house_own: Home ownership
metro: Lives in metropolitan area
internet: Household has internet access
foodstamp: Receives food stamps
house_moved: Moved in the last year
pub_contact: Contacted or visited a public official
boycott
hood_group: Participated in a community association
hood_talks: Talked with neighbors
hood_trust: Trusts neighbors
tablet: Uses a tablet or e-reader
texting: Uses text messaging
social: Uses social media
volunteer: Volunteered
register: Is registered to vote
vote: Voted in the 2014 midterm elections
party: Political party
religion: Religious (evangelical) affiliation
ideology: Political ideology
govt: Follows government and public affairs
guns: Owns a gun

Source

“For Weighting Online Opt-In Samples, What Matters Most?” Pew Research Center, Washington, D.C. (January 26, 2018) https://www.pewresearch.org/methods/2018/01/26/for-weighting-online-opt-in-samples-what-matters-most/

Details

This dataset was originally produced by the Pew Research center for their paper entitled For Weighting Online Opt-In Samples, What Matters Most? The synthetic population dataset was created to serve as a reference for making online opt-in surveys more representative of the overall population.

See Appendix B: Synthetic population dataset for a more detailed description of the method for and rationale behind creating this dataset.

In short, the dataset was created to overcome the limitations of using large, federal benchmark survey datasets such as the American Community Survey (ACS) or Current Population Survey (CPS). These surveys often do not contain the exact questions asked in online-opt in surveys, keeping them from being used for proper adjustment.

This synthetic dataset was created by combining nine separate benchmark datasets. Each had a set of common demographic variables but many added unique variables such as gun ownership or voter registration. The surveys were combined, stratified, sampled, combined, and imputed to fill missing values from each. From this large dataset, the original 20,000 surveys from the ACS were kept to ensure accurate demographic distribution.

The names were RANDOMLY assigned to respondents to better simulate a synthetic sample of the population. First names were taken from the babynames dataset which contains the Social Security Administration's record of baby names from 1880 to 2017 along with gender and proportion. First names were proportionally randomly assigned by birth year and sex. Last names were taken from the Census Bureau, who provides the 162,254 most common last names in the 2010 Census, covering over 90% of the population. For a given surname, the proportion of that name belonging to members of each race and ethnicity is provided. The last names were proportionally randomly assigned by race.