A pollster house effect is the consistent gap between one polling firm’s results and the average of other polls of the same races at the same time. If a firm’s polls regularly show Republicans 2 points better than everyone else’s, it has a 2-point Republican house effect. House effects usually come from method choices, such as how the firm reaches people, weights its sample or defines likely voters, rather than from any attempt to tilt results.
A house effect isn’t the same as being wrong. A firm that leans away from the pack can turn out closer to the result than the average, if the pack was off.
House effect vs. bias: what’s the difference? #
The two get mixed up constantly.
- House effect compares a pollster with other pollsters during the campaign. It’s measurable before anyone votes.
- Bias compares a pollster with the actual results. It can only be measured after the election.
In 2016, 2020 and 2024, polling averages understated Republican presidential support in many swing states, so firms with a Republican house effect often ended up with less bias than the average. In 2022, several firms that showed stronger Republican numbers than the average overstated Republican performance in some Senate races. House effects point in whatever direction a firm’s methods push, and whether that helps depends on the year.
What causes a house effect? #
How the pollster reaches people #
Live phone, automated phone, text-to-web and online panels each reach different slices of the public, and people answer differently to a person than to a screen. Our explainer on polling mode effects covers the evidence.
The likely-voter screen #
Only a fraction of registered voters turn out, especially in midterms, so pollsters have to decide who counts as a likely voter. Some ask about intent and enthusiasm, some use past voting records, some combine both. A screen that expects higher turnout among younger voters produces different numbers from one that expects an older electorate. Our guide to registered vs. likely voters goes into this.
Weighting choices #
Every pollster adjusts its raw sample to match the population, but on different variables. Weighting by education became standard after 2016. Whether to also weight by past vote or party registration is a live debate, and that single decision can move a poll a few points.
Question wording and order #
Asking the presidential approval question right after a series of questions about prices can pull approval down compared with asking it first. Small differences in how firms word and order questions add up over many polls.
How do polling averages handle house effects? #
Averages take one of two approaches.
Adjust for it. Model-based averages estimate each pollster’s house effect by comparing its polls with others in the same races over time, then shift its results by that amount before averaging. If a firm runs 2 points more Democratic than the consensus, a poll from it showing the Democrat up 5 is entered as roughly a 3-point lead. Silver Bulletin, for example, measures house effects alongside its pollster ratings. Our guide to pollster ratings covers how those grades work.
Limit any one firm’s influence. Simpler averages don’t adjust results, but they stop a single pollster from dominating. Election Tracker’s 30-day average takes this route. If a pollster polls the same race several times in the window, it counts once, at its strongest sample, and newer polls count more than older ones. A firm with a lean still counts, but it can’t flood the average by polling every week.
Both approaches have trade-offs. Adjustment works well for firms with long records, but it assumes a firm’s lean stays steady. Limiting influence is transparent and simple, but a single leaning pollster can still move a race that has few polls.
Why is herding worse than a house effect? #
A steady house effect is useful information: once you know a firm runs 2 points to one side, you can mentally adjust. The bigger danger is herding, when pollsters nudge their results toward the average late in a race so they don’t look wrong. Herding hides real disagreement and can make the whole field miss together. Our explainer on polling herding shows how analysts detect it.
How can you spot a house effect yourself? #
- Compare a pollster with itself. If a firm with a known Republican lean shows the Republican up 2, and its previous poll had the Republican up 6, that’s movement toward the Democrat, even though the Republican still leads.
- Compare it with the average. A new poll sitting several points from the average, on the same side as that firm’s past polls, is probably its house effect at work rather than a sudden shift.
- Check the population. Registered-voter polls and likely-voter polls often differ in a consistent direction, which can look like a house effect.
- Look across several races. A firm with a lean usually shows it in most of the races it polls, not just one.
In Election Tracker, each race’s poll list shows every published poll with its pollster and field dates, so you can scroll to a firm’s earlier polls in the same race and see whether its new number is a change or just its usual position.
Frequently asked questions #
Does a house effect mean a poll is fake or untrustworthy? #
No. Every legitimate poll makes assumptions about who to reach and how to weight, and consistent assumptions produce a consistent lean. As long as a pollster is transparent about its method, a steady house effect is something analysts can account for.
Which pollsters have the biggest house effects? #
It changes as firms update their methods. In recent cycles, firms such as Rasmussen Reports and Trafalgar Group have typically run more Republican than the polling average. Other firms have leaned the other way in particular years. Check a current house-effects table rather than relying on reputation.
How is a house effect different from the margin of error? #
The margin of error describes random variation from sampling, which averages out over many polls. A house effect is systematic: it points the same way poll after poll and doesn’t shrink with a bigger sample.
Should I ignore pollsters with a house effect? #
No. Throwing out every pollster with a lean would leave you with few polls and could remove the ones that end up closest to the result. Keep them, know which way they lean, and give the average more weight than any single poll.