1

There are a number of questions about election polls and margin of error that have been covered in CrossValidated over the years, including one regarding the 2016 US election polls. That one however only addresses validity based on members of a particular poll knowing each other.

In so many cases the state polls showed Clinton beating Trump above the margin of error. Yet states like Wisconsin, Pennsylvania, Florida and Michigan were predicted to be won by Clinton but subject to recount all went to Trump. Of course there are many factors that could affect the accuracy of the polls and the validity/lack of validity of the margin of error. Among those that I can think of are:

  1. difficulty selecting a random sample of likely voters

  2. the election was volatile and hence present opinion may not reflect future opinion

  3. selection bias could exist

Which of these possibilities do you think caused the inaccuracy of the polls or do you have another explanation?

What assumptions about statistical surveys are most commonly violated.

Michael R. Chernick
  • 39,640
  • 28
  • 74
  • 143

1 Answers1

2

  1. It seems highly improbable that sampling variation would cause this problem. How could it affect so many polls then? And wouldn't we have seen this before in other election polls in the past?

  2. It's possible that some voters changed their opinion in the end, almost as they were in the election booth. One possible behavioral effect is that people wanted to "stick it to the system". In that case, given that they had seen Trump being behind in all polls, there would be almost no cost to them voting for him. Maybe they wanted a close election since this would signal to the establishment that while they wanted Hillary, something was really wrong since they were so close to electing Trump. If many felt like this, maybe this explains the last-minute swing. One might conjecture a similar effect in the Brexit vote, where all the polls were also off.

  3. One possible selection bias is that people voting for Trump might be systematically more ashamed of reporting this truthfully than people voting for Hillary. I know for a fact that this is the case in Denmark, where the share voting for the populist anti-immigrant party has for a long time had to have been corrected upwards to adjust for this effect.

Superpronker
  • 722
  • 5
  • 6
  • Superpronker's answer number 3 is one of the factors I thought about as an addition to my question. There has been a lot of talk about Trump voters being ashamed to admit they would vote for him.. Pollsters seem to have shown that this phenomena is real. It raises a key issue in polling. Statistical survey methods assume all replies are truthful. But if some answers are false we can't trust the inference. It is the age old problem of garbage in garbage out. – Michael R. Chernick Nov 28 '16 at 08:16
  • Another factor is complacency. Voters decide not to go to the polls because they think a Clinton victory is a foregone conclusion. This use to be a major issue in National elections when the networks use to declare winners in the Presidential race before the polls closed in places like California. This led to a change that prevent networks from calling the race until all the voting sites are closed. Also no one calls a race in any state until the polls are all closed in that state. – Michael R. Chernick Nov 28 '16 at 08:26
  • This is not a duplicate. I have already explained how it differs from the previous question. Partially answering my question does not make my question a duplicate. I have already edited my question to meet Gung's request. You should remove the duplicate designation. – Michael R. Chernick Nov 28 '16 at 09:24
  • I agree totally. Complacency might be a big thing. Makes you wonder if some voters have regret on the day immediately after. Especially relevant given the low overall turnout. – Superpronker Dec 01 '16 at 12:23