There were a number of press reports on how wrong the public opinion polls were in predicting the outcome of Iowa Caucus for the Republican Party. Going into caucus day, the expectation, with caveats, was that Donald Trump would win by about four percentage points, defeating Senator Ted Cruz. After the votes were counted, Cruz would be victorious, defeating Trump by four points.
Everyone acknowledges that predicting the outcome of the Iowa Caucus is especially difficult due to voters changing their minds before the vote. A caucus is not like a regular primary election where the voter goes to the polling place, casts his vote, and then leaves. A caucus requires voters to attend a polling place, listen to various speeches advocating one candidate of another, after which voters can be persuaded to change allegiances. Traditional polling cannot reflect this phenomenon. Also, there was the issue of turnout. The thinking went that if there was a much higher turnout for this caucus than there was for prior Republican presidential caucuses in Iowa, this would help Trump since many believe that he is expanding the party beyond the traditional Iowa Republican voter and this expansion would be responsible for the increase.
The polls pointed to Trump winning. The significant increase in turnout would also point to a Trump victory.
Cruz won the night.
Were the polls wrong?
As a factual matter, the polls were wrong. However, there may have been clues in the polling data that would have predicted a Cruz victory had someone closely scrutinized the polls.
One of the main sources of political polling is RealClearPolitics (RCP). This is a website that aggregate polls for various polling companies.
In the last four polls prior to the Iowa Caucus were as follows:
Des Moines Register Poll (polling days: 1/26 – 1/29): Trump was up by five points
Quinnipiac Poll (polling days: 1/25 – 1/31): Trump was up by seven
Opinion Savvy Poll (polling days: 1/29 – 1/30): Trump was up by one
Emerson Poll (polling days: 1/29 – 1/31): Trump was up by one
In each of the polls, one would reason to believe Trump would win. This however would miss important information. The poll that matters would be on caucus night. The date of the caucus is 2/1. One should look at the polls to see if there is any trend. Conclusions would not be easily determined given that there are only four polls but it is still worth investigating. The Des Moines poll had roughly an average date of 1/27; Quinnipiac 1/28; Opinion Savvy 1/29; and Emerson 1/30. The two most recent polls, taken two and three days prior to the caucus, suggest that Trump was only up by one percentage point, down from five and seven points only a couple of days earlier. This suggests that there could be a trend, if the quality of the polls is roughly equal. There are events that could suggest such a trend. Trump skipped the most recent debate, which some viewed as unwise and, perhaps, disrespectful. Trump, instead of going to the debate, held a fundraiser for the veterans; a move that many felt exploited the military for political gain. Glenn Beck, who is very popular with conservatives, endorsed Cruz in the days leading up to the caucus.
Below is a link to a graph that shows the polling data of the Cruz support minus the Trump support (shown in blue). Shown in red is a statistical regression line that attempts predict future margins based on prior ones (shown in red).
As mentioned, there are only four data points and they do not lie neatly on a straight line. However, the line that would be implied by the four data points gives us a Cruz win on 2/1 of 3.8 percentage points. The actual margin for Cruz was 3.3 points. One could argue that the polling predicted the Cruz victory.
The regression line does not perfectly describe the relationship between the margin and the dates. As a result, one would not have placed full confidence in the expected margin of 3.8 points. But given the apparent trend in polling data, it should have raised a red flag that Trump was in a bit of trouble prior to the voting.