National and state polls can produce different results without contradicting each other. They survey different electorates and may ask about different choices.
A national poll can describe broad congressional party preference. A state poll can measure a named statewide race. A House district poll narrows the electorate further.
The geographic unit matters
Imagine a hypothetical national poll in which Party A leads by four points. A statewide poll might still show Party B's Senate candidate ahead.
The national population includes voters from many states. The statewide race contains one state's electorate, two or more actual candidates, and local circumstances. There is no mathematical rule requiring those estimates to match.
Before interpreting a difference, compare the populations and questions rather than assuming one survey must be defective.
House control is built district by district
The national House vote is not a single contest awarding all seats. Congressional district results determine the members elected.
A party's support can be concentrated in a smaller number of strongly favorable districts or spread across many competitive ones. The same national preference could therefore be consistent with different seat outcomes.
Historical relationships between national voting and seats may inform analysis, but they are not a substitute for the current map and candidate field. Pew's discussion of national midterm polls illustrates the importance of distinguishing national preference from congressional outcomes. Source: Pew Research Center.
Compare compatible evidence
Use a national survey to understand the broad environment. Use statewide evidence to assess a Senate contest. Use district evidence and boundaries to assess a House race.
For all three, record the interview dates and whether respondents are adults, registered voters, or likely voters.
A better headline test
If a headline says a national poll “proves” who will win a particular Senate seat, inspect the link. Does the original source actually survey that state and race?
If not, the headline is making an inference. It should explain the additional assumptions rather than present the national number as local measurement.
Related reading: the generic ballot and competitive House districts.