A cross-sectional poll interviews a fresh sample of people once, giving a snapshot of opinion at one moment. A panel study interviews the same people repeatedly, so it can show which individuals changed their minds and why. Most election polls you see in the news are cross-sectional, even when the sample is drawn from an “online panel,” which is a different thing and the source of a lot of confusion.
What is a cross-sectional poll? #
A pollster draws a sample from a population, such as likely voters in Pennsylvania, interviews them once, and reports the results. Next month it draws a brand-new sample. Each wave is independent.
This is the standard design for horse-race polls and approval ratings. When a headline says a candidate leads by 3 points, you’re almost always looking at a cross-sectional poll.
What it’s good at: an unbiased estimate of where the whole population stands right now, without the problems that come from interviewing the same people over and over.
What it can’t do: tell you who changed. If a candidate goes from 48% to 46% between two waves, you can’t tell whether 2% of people switched, whether 10% switched in each direction and it netted out, or whether the two samples just differed by chance.
What is a panel study? #
A panel study, sometimes called a longitudinal study, recruits a group of respondents and re-interviews the same people over weeks, months or years. The American National Election Studies (ANES) have long used panel components, re-interviewing voters before and after presidential elections, and academic projects such as the Cooperative Election Study have run multi-year panels.
What it’s good at: measuring individual change. A panel can show that 6% of voters moved from Candidate A to Candidate B while 4% moved the other way, which a cross-section would report only as a 2-point net shift. It can also measure the effect of an event by interviewing the same people before and after it.
Where it struggles:
- Attrition. People drop out over time, and not at random. If younger or less engaged respondents leave faster, the panel drifts away from the population.
- Conditioning. Being asked about politics repeatedly can change people. Panelists may follow the news more closely or become more consistent in their answers than they otherwise would be.
- Cost. Keeping track of the same people for months and paying them to stay costs more than drawing fresh samples.
The confusing part: “online panels” are usually not panel studies #
Many modern pollsters recruit a large pool of people who agree to take surveys, called a panel, and then draw a fresh sample from that pool for each poll. YouGov’s opt-in panel, Ipsos’s KnowledgePanel and Pew Research Center’s American Trends Panel all work this way. Each individual poll is still a cross-section: a new subset of panel members answers each time.
So there are two separate questions to ask about any poll:
| Question | Options |
|---|---|
| How was the pool of respondents recruited? | Probability-based (random address or phone recruitment) or opt-in (people who sign up) |
| Are the same people re-interviewed and compared individually? | Panel study (yes) or cross-section (no) |
The first question affects quality most. A probability-based panel recruits people at random by mail or phone and, in some cases, provides internet access, which keeps it representative. An opt-in panel relies on volunteers and heavy weighting. We cover the risks in are online opt-in polls reliable?.
Side-by-side comparison #
| Cross-sectional poll | Panel study | |
|---|---|---|
| Who is interviewed | A new sample each time | The same people each wave |
| Shows | Net change in the population | Individual change, who switched and in which direction |
| Main risk | Sampling error between waves | Attrition and conditioning |
| Cost | Lower per wave | Higher over time |
| Typical use | Horse-race polls, approval ratings | Academic research, campaign message testing, studying persuasion |
Why the difference matters when you follow an election #
Suppose a president’s approval drops 4 points in a month in cross-sectional polling. That tells you the national mood moved. It can’t tell you whether independents soured, whether the president’s own supporters became less enthusiastic, or both. A panel that re-interviewed the same people could answer that directly.
Panels also explain something that confuses people every cycle: large amounts of individual churn behind small net changes. Panel research consistently finds that more voters change their stated preference during a campaign than the flat toplines suggest, with movements in both directions largely cancelling out.
Campaigns use this. A campaign testing an ad might interview a group of voters, show some of them the ad, and re-interview everyone to measure how many were persuaded. That’s a panel design, and results like these rarely become public.
What about tracking polls and rolling samples? #
A tracking poll interviews a fresh small sample every day and reports a rolling average of the last few days. It’s cross-sectional, just frequent. A 3-day rolling sample of 300 people per day reports 900 respondents each day, two-thirds of whom were also in yesterday’s number, so day-to-day moves are smaller than they look. Our explainer on tracking vs. benchmark polls covers how campaigns use each.
How to read either kind of poll #
- Check whether “panel” means recruitment or design. Most poll releases that mention a panel are describing how respondents were recruited.
- For cross-sections, compare the same pollster over time. Mixing pollsters mixes methods.
- For panel results, look for attrition figures. A good release says how many people from wave one completed wave two.
- Don’t over-read small net shifts. A 2-point move in a cross-section is often within sampling error. A panel can show whether it reflects real switching.
When you follow individual polls in Election Tracker, our free iPhone app, each card lists the pollster, sample size, population and field dates, which is usually enough to tell a daily tracker from a one-off survey. The 30-day average on top counts each pollster only once per race (using its largest sample), so a frequent tracker can’t dominate the number. Non-response is the other big threat to both designs; see non-response bias.
Frequently asked questions #
Which is more accurate, a panel or a cross-sectional poll? #
Neither is more accurate in general; they answer different questions. A cross-section is better for a clean snapshot of the whole population. A panel is better for measuring how and why individuals change, as long as attrition is handled well.
Is a YouGov poll a panel study? #
Usually not. YouGov recruits people into an online panel, but each poll typically draws a fresh sample from that pool, making the individual poll cross-sectional. Some YouGov and university projects do re-interview the same people, and they say so in the methodology.
What is panel attrition? #
Attrition is when respondents drop out between waves. It matters because the people who leave often differ from those who stay, which can bias later waves. Good panel studies report attrition rates and weight to correct for it.
What is a rolling cross-section? #
It’s a design where a small fresh sample is interviewed every day and results are reported as a rolling average over several days. It combines the fresh samples of a cross-section with frequent updates, at the cost of smoothing over sudden shifts.