Poll crosstabs are the tables, usually at the back of a pollster’s release, that break the headline result down by groups such as age, gender, education, race, party and region. To read them well, find the number of people interviewed in each group, remember that small groups carry much bigger margins of error than the headline, and compare any surprising subgroup number with other recent polls before believing it.
A headline like “Candidate A leads by 4” is an average of groups that vote very differently. Crosstabs show you which groups make up each candidate’s coalition, and they’re where most misleading poll stories come from.
What are poll crosstabs? #
“Crosstabs” is short for cross-tabulations: tables that show how answers to one question vary across another variable. In political polls, one variable is the question itself (“If the election were held today, who would you vote for?”) and the other is a characteristic of the respondent. Common breakdowns:
- Gender: men and women.
- Age: usually 18 to 29, 30 to 44, 45 to 64 and 65+.
- Race and ethnicity: white, Black, Hispanic, Asian and other groups.
- Education: four-year college degree or not, sometimes split further by race.
- Party: Democrats, Republicans and independents.
- Geography: urban, suburban and rural, or regions within a state.
- Past vote: how respondents say they voted last time.
How do you read a crosstab table? #
Most crosstab tables put the answer choices down the left side and the groups across the top. Read each cell as “what share of this group gave this answer.”
| Vote choice | Total | Men | Women | College degree | No degree |
|---|---|---|---|---|---|
| Candidate A | 48% | 41% | 54% | 53% | 45% |
| Candidate B | 45% | 51% | 40% | 41% | 48% |
| Undecided | 7% | 8% | 6% | 6% | 7% |
| Sample (n) | 1,000 | 480 | 520 | 400 | 600 |
Follow the “Women” column down to Candidate A and you get 54% of women backing A. Columns add to 100% (give or take rounding), rows don’t.
The bottom row is the one to watch. n is the number of people interviewed in that group, and it tells you how much to trust the column above it.
Why is the subgroup margin of error so much bigger? #
A poll’s reported margin of error, say ±3.1 points for 1,000 interviews, applies to the whole sample. Every subgroup has its own, larger margin. A quick estimate for any group is:
MoE ≈ 98 ÷ √n (in percentage points, at 95% confidence)
For a 1,000-person poll:
| Group | n | Approximate margin |
|---|---|---|
| Everyone | 1,000 | ±3.1 |
| Women | 520 | ±4.3 |
| College graduates | 400 | ±4.9 |
| Voters 18 to 29 | 120 | ±8.9 |
| A group of 100 | 100 | ±9.8 |
Those figures are for each candidate’s share. The margin on the gap between two candidates within a subgroup is roughly double. Pew Research Center gives a real example: a Hispanic subgroup of about 15% of a 1,067-person poll had a margin of about ±8 points per candidate and ±16 points on the difference.
So if young voters go from 55% for a candidate in one poll to 47% in the next, you’re probably looking at noise, not a shift. Our guide to margin of error goes deeper on why.
Weighted vs. unweighted: which n should you look at? #
Some releases show both an unweighted n (how many people in the group were actually interviewed) and a weighted n (how many they count as after adjustment).
Pollsters weight because raw samples never match the population. Older, college-educated people tend to answer surveys more readily than younger people without degrees, so the pollster counts under-represented respondents a little more and over-represented ones a little less. Our explainer on how pollsters weight survey data walks through the process.
For judging reliability, the unweighted n is the one that matters. If a state poll interviewed 40 young voters and weighted them up to represent 120, each of those 40 people carries about three times the normal weight. One unusual respondent can swing that column noticeably. If a crosstab shows a wild swing, check the unweighted n first. Below about 100, treat the column as a rough hint.
What are the common mistakes when reading crosstabs? #
Building a story on one subgroup #
“New poll shows Candidate B up 15 with suburban women” makes a good headline. Before you believe it, ask how many suburban women were interviewed, what the last five polls showed for that group, and whether anything happened that would explain a big move. If other polls showed B trailing with that group and nothing changed in the race, the new number is most likely an outlier.
Ignoring the undecideds #
Younger and less engaged voters often have high undecided shares. A 40% to 30% lead among voters under 30 leaves 30% undecided. The honest reading is “this group hasn’t made up its mind,” not “Candidate A has locked it up.”
Applying national crosstabs to a state #
A national poll’s crosstabs don’t describe any particular state. Hispanic voters in Florida, Texas and Arizona have voted quite differently from one another, and white non-college voters in Wisconsin don’t mirror those in Georgia. Use state polls for state questions.
Treating house effects as real shifts #
Pollsters differ in how they weight, whom they count as likely voters and how they contact people. Those choices can produce consistent differences in subgroup numbers from one firm to the next. Compare a pollster’s new crosstabs with its own previous release, not with a different firm’s. Our piece on pollster house effects explains the pattern.
Step-by-step checklist for reading crosstabs #
- Check the population. Adults, registered voters or likely voters. In a midterm, likely-voter results can differ from registered-voter results by several points.
- Find the group’s n. Use the unweighted figure if it’s given.
- Estimate the group’s margin with 98 ÷ √n, then double it for the gap between candidates.
- Look at the undecideds in that group.
- Compare with other polls. A subgroup trend is believable when several pollsters show it.
Step 5 is the one people skip. Election Tracker lists every published poll in a race with its pollster, sample size, population and field dates, and pollster releases open in an in-app browser, so you can jump from a race’s poll list to each release’s crosstabs and check whether the other recent polls agree.
Frequently asked questions #
What is the difference between weighted and unweighted crosstabs? #
Unweighted figures show how many people in each group actually answered. Weighted figures adjust those responses so the sample’s makeup matches the target population. The results you read in the table are weighted; the unweighted n tells you how solid each column is.
Why do some pollsters not publish crosstabs? #
Transparent pollsters publish full crosstabs and methodology, and pollster ratings reward that. A pollster that releases only a topline may be protecting a proprietary method, or the subgroups may be too small to hold up. Either way, you have less to judge the poll by.
Is a sample of 1,000 enough to trust the crosstabs? #
It’s enough for big groups like men, women or college graduates, where margins run around ±4 to ±5 points. It isn’t enough for small groups like young Black voters or rural Hispanic voters, where the n can fall below 100. Pollsters who care about a small group oversample it.
Why do different polls show different results for the same subgroup? #
Partly chance, because small groups carry big margins, and partly method. Pollsters weight differently, screen for likely voters differently and reach people by phone, text or online panels. Two polls can agree on the topline and still differ by 10 points on a small subgroup.