Two organisations interview people over the same week and publish different leaders. That can be unsettling when each release is presented as the state of the nation, but the gap deserves investigation before anyone concludes that a pollster is incompetent or politically motivated.
Compare what was measured and how it was processed before comparing the headline. Some differences come from sampling variation; others recur because organisations make different choices.
Not every difference is a house effect
A house effect is a tendency for an organisation to put a party higher or lower than other organisations over a series of comparable polls. One unusually high result is not enough to establish it.
Fieldwork can explain an apparent conflict. Interviews completed before a major event and interviews conducted after it need not measure the same opinion. Publication dates can obscure that sequence; use the dates recorded in our polling database.
Even genuinely contemporaneous samples contain different respondents. Sampling uncertainty gives another reason why identical methods would not produce identical results every time.
Questions can differ
A voting question may ask about a general election in broad terms or direct respondents to their own constituency. Parties can appear as named choices, within an “other” option, or in a follow-up. The order and wording of questions also belong in the comparison.
YouGov’s methodology explanation describes its use of a constituency question. That is one identifiable reason to examine the underlying question rather than assume every headline arose from the same exercise.
Prompting a party and reporting a party are different decisions. A party might appear in the detailed answers without being named in the initial question. An absent published figure must not be treated as zero.
Recruitment and processing can differ
Organisations may recruit through different panels or sampling designs. They may use different weighting targets, ask different follow-ups, or make different assumptions about turnout and undecided voters.
Ipsos documented substantial changes when it introduced its 2025 method. Comparisons which span such a change need to distinguish the new series from the old. A pollster’s historic tendency is not a permanent property of its name.
How to measure a persistent gap
A useful analysis compares overlapping periods, rather than averaging an organisation’s whole year against rivals which polled in different months. In a changing political environment, the latter comparison can mistake timing for method.
One possible exploratory calculation compares each poll with an average of other organisations around the same dates. Repeating that across a series reveals whether the deviations tend to point in one direction. The window, comparison set and treatment of repeated polls should be disclosed.
Such a calculation measures difference from a chosen benchmark, not difference from the unknowable true opinion on those dates. A pollster which sits outside the pack may be closer to reality; the pack can share a mistake. Election accuracy requires a different assessment.
How should readers respond?
Follow changes within an organisation as well as the wider picture. A repeated rise across several methods is more persuasive than a movement confined to one series, although common errors can affect several organisations together.
A polling average can reduce the impact of one distinctive series, depending on its design. It should not conceal the disagreement entirely. When the spread between organisations is large, the spread itself is part of the evidence.
Sources and further reading
- YouGov: current voting-intention process
- Ipsos: introducing its revised methodology
- British Polling Council: disclosure of methods
Reviewed 7 October 2026. Worked examples are illustrative unless an election or poll is identified.