A poll can be inaccurate if it draws on a biased sample or asks loaded questions. Regarding the former, an example of a biased sample would be the Truman presidential election in 1948 against Dewey. The Chicago Tribune infamously published a front page story declaring Dewey the victor based on a bad Gallup poll. Those at Gallup had been making phone calls all day, polling people to see if they were voting for Dewey or Truman and their results pointed heavily to Dewey to the point where they were certain he would win. However, since they only used phone calls they were only polling people who owned phones. During the late 1940s, phones were still a luxury for many people. As such, those who owned phones tended to be wealthier and actually represented a specific subset of the population that would be more in favor of Dewey than Truman. Since Gallup missed out on poorer voters, they had a biased sample that ruined their prediction.
Loaded questions can also influence a poll. For example, let's say Tom Smith and Sally Jones are running for the mayor's office in Appletown. Tom is the incumbent. Sally ran for the office a few years ago, but was defeated. Let's say the newspaper wanted to poll people online to see who they were voting for. Compare these possible options:
Who are you voting for in the upcoming mayoral election?
Tom Smith
Sally Jones
Compare that poll to this:
Who are you voting for in the upcoming mayoral election?
Tom Smith, Mayor of Appletown
Sally Jones, Failed Mayoral Candidate
The first poll is relative unbiased and the options are by no means loaded. You are choosing between Tom Smith or Sally Jones. The second poll has loaded options. If you read the choices, you see that the poll is about who should be mayor and the options provided are the current mayor and a failed candidate. People would be more inclined to pick the former over the latter, which would ruin a poll.
