Caveats and important information about polls and their analyses.
Necessary caveats. If you are just a “bottom line” person, look at the title of this post and move on to the current poll analysis. If you want to understand the “why” for those conclusions, and to at least understand why polls can and do differ, read on.
In the general election, partisan voters, who historically make up between eighty and ninety percent of the electorate, have their party’s endorsement to help them in deciding whom to support. This is not the case in primary elections. Early in the election cycle, which this surely is, polls overemphasize the importance of name recognition. This often results in wild swings as sizable numbers of voters simply haven’t made up their minds or haven’t heard of many of the candidates. Surely a sizable amount of Hillary Clinton’s and Donald Trump’s early support comes from the fact that nearly all of the voters have heard of them. The same cannot be said for Carson, Fiorina, Rubio, and Cruz on the Republican side, or Sanders on the Democratic side. Should Joe Biden enter the race, he will do so with relatively high name recognition. The Republican field is quite large, but, with the cost of running a campaign so high, we can almost certainly expect withdrawals by some of the candidates whose funds and poll numbers remain low. This makes voters’ second choice an important data source. It is essential to remember that not all polls are created equal. None employs truly random samples, due to the cost and time required to do that. Therefore, polling organizations make assumptions about the make-up of the “likely voter” such as age and race. Since the polling that we consider, at least this early in the contest, are telephone polls, seemingly unimportant variables such as the percentage of cell phones in a sample can affect the accuracy of the results. Because polling organizations aggresively guard their models of the likely voter, as outsiders we have to rely on polls with reputations that are impeccable and that have a history of accuracy. Size of the sample matters–up to a point. National polls generally have a sample size of between 1,000 and 1,500. uch smaller sample sizes are subject to a higher margin of error. Online polls, which by definition use non-probability based samples, are virtually useless. The Drudge Report’s poll, to cite just one example, measures only a sample drawn from their viewers–a population that differs significantly from the electorate. Because of the different models used to derive their samples, it is impossible to compare them exactly. Thus, for example, when one poll shows a 7 percent support for candidIn such polls, the make up of their model of the likely voter often affects their accuracy more than using a larger sample. State polls and national polls that use mate X on a Sunday, and a different poll two days later shows that candidate with 10 percent, one cannot conclude that the candidate’s popularity is rising. For that reason, many analysts use a running average of several polls to demonstrate changes in support. We try to use as many recent polls as possible, and note when we are just using the latest one, specifying it’s source.