Issues

Data sources

• Cooperative Election StudyCumulative common content 2006-2024 with the Cumulative Policy Preferences items, policy items 2007-2021Weights: weight (single year), weight_cumulative (pooled years)
• State presidential returnsMEDSL state presidential returns, 2004 to 2024 electionsWeights: none (full counts)

The numbers on this page are read from meta.json and one file per issue in the same folder, all aggregates with no respondent-level rows.

Methods

Data: the Cooperative Election Study (CES) cumulative file, with the Cumulative Policy Preferences items merged onto the common content. Twenty items, each with a declared party direction that the coding check uses. The policy battery runs 2007 to 2021 and ends there; the 2022 to 2024 rows carry none of these items.

Weights: single-year estimates (the trend, the party gap and the latest-year breakdown) use the year weight. Within every year it is proportional to the cumulative weight, so the choice changes no single-year number. Pooled estimates (states, counties, the pooled breakdown and the regressions) use the cumulative weight, which gives every survey year the same total weight, so a pooled estimate averages the waves equally whatever their sizes.

Intervals: 95% intervals from the Kish effective sample size with a normal approximation. They treat the weighted sample as a simple random sample of an opt-in panel, so they are model-based and ignore any design effect beyond the weights.

Party: self-identified Democrat, Republican or Independent from the three-category party question; leaners are not separated, and respondents answering Other or Not sure are in the all-respondent line only.

Floors: 300 respondents for a state estimate and a state regression, 150 respondents for a county estimate. National cells need at least 100 respondents and an effective sample of 100 (150 and 100 for the all-respondent line of a year). Cells below a floor are left out, not shown as zero.

State estimates and county estimates are weighted shares pooled over every year an item was fielded. County estimates are raw means with a standard error, not model-based estimates.

State coefficients: each state gets a weighted linear probability model of support on education, family income, age, female, non-White and born-again or evangelical, all standardized with the national weighted mean and standard deviation of the issue's pooled sample, plus year effects. The scatter against state Democratic margin is a precision-weighted regression of the coefficients on the margin, with robust intervals and a precision-weighted correlation.

State Democratic margin: the two-party Democratic margin in the presidential election before each year fielded (2004, 2008, 2012, 2016 or 2020, the mapping CES uses), from the certified state returns, averaged over the years fielded with every year counting equally.

Each issue is checked before it is published: the sign of the Democratic minus Republican gap in each of 125 item-years is compared with the direction the substance of the item implies, and the 2018 answer counts of 16 items are compared with the 2018 CES guide. The sign check flagged 2 item-years (Raise fuel-efficiency standards, 2018; Repeal the Affordable Care Act, 2014); each is handled as noted under its issue. Cells below 300 respondents (states) or 150 (counties) are not shown.

What these data cannot show

• These are opinions of respondents to an opt-in online panel, weighted to national benchmarks. State and county values are raw weighted means and inherit the panel's sampling error, which the intervals describe only under a simple-random-sample assumption.
• The cumulative file harmonizes items across years, and the wording of an item can differ by wave; the 2018 fuel-efficiency item, for example, asked the opposite proposal, and is excluded. A share supporting a proposal does not measure how much respondents care about it.
• State coefficients describe associations within a state at the ages, incomes and education levels of its respondents. They do not show that a trait changes opinion, and each carries its own sampling error, so differences between states are noisier than the point estimates suggest.
• Nothing here covers 2022 onward: the policy battery ends in 2021.

How to cite

@misc{singh2026exploreissues,
  author = {Singh, Vishal},
  title = {Issues: party gaps, groups, states and counties on 20 CES policy questions},
  year = {2026},
  url = {https://vishalsingh.org/politics/explore/issues},
  note = {In The Public Mind. Published 30 September 2026.}
}