Opinion polls are used across many fields of study.
Sometimes they’re used as a component of another type of research (subjects might be asked their opinions on a range of topics, but the actual study is assessing the negative or positive valence of the words they use when they’re asked questions by pollsters wearing different sorts of clothing, for instance), but most often they’re used to gauge the general feeling about something within a given demographic.
This flavor of large-scale temperature-taking has been in use since at least 1824, when local newspapers in the eastern US polled readers on who they intended to vote for that election year. The resulting numbers generally mirrored the outcome of the popular vote in the region, and that led to more such efforts, mostly at a local level, in subsequent decades.
National surveys (funded by nationally distributed publications) arose in the early 1900s, and by the 1930s these surveys were beginning to show their limitations, in several highly publicized cases leading to incorrect predictions about elections, which led to some interesting research into what became known as participation bias: the realization that folks willing to fill out these sorts of surveys only represent a slice of the overall electorate, not a cohesive cross-section of it.
In the decades since, pollsters have gotten better at broadening and de-biasing their data, but even using modern approaches that help make their samples more representative of the total electorate (often by ensuring the number of participants is large enough to cancel-out hijinks, while including enough members of smaller demographic groups that they’re unlikely to reach wildly unrepresentative individuals), these polls are still laden with issues.
Among said issues is that, even at their best and most optimized, polls can only ever give us a sense of what a generalized swathe of a population is thinking at the moment in which they’re surveyed; and in modern politics, that can sometimes mean the data is outdated even before it’s aggregated and reported upon.
Within this larger collection of unintentional issues related to polls, there are also variations of the concept that introduce intentional bias into the resulting survey data, while others aim to bias the people being polled.
A “push poll” is a type of survey that intentionally nudges a voter’s opinions by asking loaded questions, providing negative information about a candidate or cause (in the guise of a question), and in some cases by using language and innuendo to guide respondents toward a preferred answer.
Many push polls aren’t even gathering data, though they pretend to, and in some cases they show up in the budget of the political party, candidate, or nonprofit that funds them as that kind of expense.
The real goal, though, is to get opposition research findings—negative stuff about one’s opponent—into casual conversation, and to do so in a seemingly unbiased context.
So rather than hearing such information or claims from a clearly paid-for advertisement, folks will be introduced to these rumors or speculations by an ostensibly objective source: someone who’s just asking questions and trying to get the respondent’s opinion about the things they’re spreading.
The lightest-weight version of push polling is basically meant to just keep a particular topic in the ether so that voters are more inclined to vote, and/or are thinking about it when they do vote. In such cases, getting folks to think about the issue from a particular angle is just an added bonus.
The harder-core version of this is a lot more overt and loaded with slanted language that, despite being borderline propagandistic, allows those asking the questions (and those funding them) to maintain a veil of deniability and objectivity.
Variations on this theme have been criticized by legitimate polling organizations, and in a few cases have been criminalized, though barely anywhere, and even then, only regionally.
This is just one sub-species of politically motivated misinformation, but it’s especially relevant in the run-up to elections (and especially at a time in which convincing-sounding chatbots are being used for such purposes), as that’s when these types of polls become more common, and when civically engaged people are most prone to subtle (or not-so-subtle) nudging by folks they have every reason to believe are objective data-gatherers, and nothing more than that.