From time to time media claimed Trump's poll rate tied to Clinton's or even passed over her. How could they get such result?
One needs to read the polling methodology -- the questions asked, the profile of the individuals asked, how the polled respondents were selected, etc. -- to obtain the answer to that question. With that information, along with an adept understanding of statistical sampling and English grammar/composition, one will know how the pollsters got the result they did and what the result legitimately means and does not mean and what inferences one may reasonably draw from the poll.
Here's one example.
The polls this year seem to be particularly erratic. The last five national polls which included all 4 significant candidates, all taken in approximately the same time frame, show radically different results. Two show Hillary with a 6 point lead and one shows Trump with a 5 point lead. I don't see any significant difference in methodology and I don't have a clue how to explain the variance. I assume that they are somehow sampling different populations but I thought they were all fairly sophisticated in their methods of massaging their results to reflect the actual demographics of the voting population.
RealClearPolitics - Election 2016 - General Election: Trump vs. Clinton vs. Johnson vs. Stein
Let me point out that I'm just assuming the polls initially displayed at the link you provided are the national polls to which you refer in your post. I don't know that they actually are because I haven't clicked on each of them to make sure. When I click to the
Election 2016 Presidential Polls RCP page, I see the five most recent "General Election" polls echo the same general outcome: Clinton ahead in four of the five most recent polls. Presumably, they are the same ones found on the page to which you linked.
I clicked on your link and the "last five" polls, which I took to mean the first five on the list, show Clinton ahead in three of them and Trump ahead in two. At that level, while one can certainly focus on the spread and figure out what gives rise to it, the relevant point that is consistent in four of five of them is that Clinton is ahead nationally.
If I were of a mind to try to figure out why the discrepancies among the spreads in the various polls, what I'd look at first is how the respondents were chosen. It's clear that some sort of random selection method was used for each of them. I suspect, however, that a fair share of the respondents' answers were obtained via online polling methods. That approach to survey participant selection has several pretty well understood biases/weaknesses (not political, necessarily, but bias nonetheless), which is why it's the first place I'd look.
Why does online polling have such a tremendous impact? Because one segment of the population -- older voters -- are not big online users, yet they are among the most likely folks to vote. Because of that, I'd also look at the age stratification of the survey.
Take the Rasmussen survey, for example. It uses census data to derive the targeted age stratas for its survey. That's great and it makes sense; however, if one is going to develop the respondent profile based on census age strata, one needs to assign a greater weighting to the actual answers received from voters who have the attribute "more likely to vote" associated with them and assign a lower weighting value to responses given by folks who have other "likelihood of voting" attributes associated with them. Does the Rasmussen survey team do that? I don't know; I don't see that explained in the methodology. I just know that the "likelihood of voting" attribute is one that needs to be accounted for in weighing the results obtained from survey respondents when the question asked of them is essentially "for whom will you vote."
Lastly, one needs also to consider the dates on which the polling occurred. Current events can have a material impact on poll responses. If ISIS were to execute an attack on Memphis tomorrow, one's answer to a given question may be completely opposite what it was had they not. The events need not even be so dramatic as that example; it all depends on what respondents think is important.
Above I mentioned bias in polling, and I'm referring to polling bias, not political bias. All polls have some element of polling bias in them. The questions that one must answer are:
- What is the nature of the bias in the poll?
- Has it been corrected for and if so, how?
What's an example of polling bias? Well, for the sake of keeping it simple, let's consider the question asked in the Rasmussen poll.
If the 2016 presidential election were held today, would you vote for Republican Donald Trump, Democrat Hillary Clinton, Libertarian Gary Johnson or Green Party candidate Jill Stein?
Consider how the results for that question might differ were the party affiliations of the candidates removed. Consider how the results might differ were the candidate names removed and just the party affiliations indicated. Consider how the results may differ yet again were there no names or parties identified.
The question can be asked in each of those ways, and the results could differ with each way of soliciting the information from respondents. Is the difference significant? I don't know because I have not asked the question each way to find out and then correct for it and/or disclose the difference(s). Have Rasmussen? I don't know; they don't say. Frankly, unless one contacts the organization that conducted Rasmussen's poll and asks, one isn't likely to find out seeing as they've not shared that information on their methodology page.
What are the implications of the discussion above? Well, mostly that the RCP average of polls is likely the best one use to get a general sense of where things stand at a given point in time if looking at national poll is what one is inclined to do. Personally, what I look at most (not that I look at them much to begin with, but if I'm going to look...) are the responses to questions about what issues folks find most compelling and then at polls that attempt to answer the question of which candidate is preferred with regard to those "key" issues. I also look at polls of Independent voters. The other type of poll I look at are "battleground" state election polls.
Based on looking at the stuff I look at, though I don't want Trump to win the election, I think he will.
