Make no mistake, when trying to gain a signal from the onslaught of horserace polls that arrives this time of year, aggregation of surveys as a whole is the best way to go. Averaging the most recent polls or taking their median should give you the clearest picture. However, that’s not to say breaking up these numbers by different survey characteristics won’t return interesting results or speak to important trends (e.g. could a social desirability bias be coming through when you compare live interview vs. online polls?).
I’ve checked polling developments by survey mode, differences by party support for each candidate, and briefly keened in on one particular pollster in the past. Here I’ll expand on that last point: trying to gauge differences by pollster. Overall aggregation is always most important, but often times within-pollster trends can add a new layer of explanation for each candidate’s position in the general election race.
Using data from HuffPost Pollster, I track below the results from five polling outlets that have most often conducted head-to-head matchup polls between Hillary Clinton and Donald Trump. I restrict polls to those that test only these two candidates (i.e. for survey questions in which third party candidate names are not listed), and where the population comprises both registered and likely voters.
Though noisy early on, here’s how the Clinton’s margin over Trump–subtracting Clinton’s percentage support from Trump’s–looks like over the last year.
Only ABC/Post, CNN, and Morning Consult have been asking this trial heat question for nearly the entire year. A mid-May ABC/Post poll that showed Trump ahead affects that pollster’s trend a lot, but for the most part all of the pollsters show Clinton consistently ahead. They don’t all follow similar paths, especially when all of them are consistently sending out polls in the last few months.
On that note, it’s worth restricting the time interval of focus to the start of June onward–the period following Trump clinched the Republican nomination. The trends should be more meaningful during this time, as it captures the process of party reunification/usage of partisanship heuristics (or lack thereof) and how the electorate reacts more broadly to just two competitive candidates in the race. Below, I graph the poll results from these same five pollsters, only starting after Trump secured the nomination (note: I use line graphs rather than the smoothing lines above because of the fewer data points).
For nearly all of the different frequent pollsters, the Clinton margin follows the schedule of the party conventions and their ensuing effects: Trump closes the gap in mid to late July after the Republican convention wraps up–see the sharpest drops for the CNN, Morning Consult, and NBC/SurveyMonkey polls in the graph–but Clinton responds by restoring her lead following her party’s convention. This latter effect has likely proved most durable, as even weeks after the DNC, the trends for all pollsters show Clinton’s margin rising (that likely has as much to do with Trump’s various moments of implosion as well). It seems Ipsos/Reuters was least receptive to this convention-related development, as that firm shows around a five percentage point decline for Clinton starting in early June.
Right up until the last few weeks, the pollster most favorable to Trump has been the Morning Consult, which conducts its surveys online (even at this point, it’s second most). From the five pollsters examined here, that firm held the nadir for Clinton’s margin in the last few months, as well as most often had Clinton with the lowest lead. That’s not the first time Trump has succeeded specifically in Morning Consult polls. If you followed pollster discrepancies during the Republican primary at all, you would recall Morning Consult often yielded the most bullish results for Trump, representing the first poll that consistently had Trump’s primary support at the national level in the 40s percentage-wise.
Ipsos/Reuters and NBC/SurveyMonkey also conduct polls online but haven’t shown Trump as strong during the general election, so it’s hard to ascribe this result for Morning Consult to an effect of online surveys. Maybe it has something to do with question ordering and resulting priming effects for survey-takers. For example, in Morning Consult’s latest poll, it asks about several favorability/approval ratings before getting to presidential candidate matchups. I also thought that if the firm employed a panel, it might have something to do with the characteristics of this group that kept coming back to take each survey–but Morning Consult does not allow a single respondent to take a survey more than once a year. In general, according to its methodology, it seems the firm doesn’t deviate much from the practices of other pollsters (in terms of how it weights, its use of CPS data, etc.). It’s thus hard to pinpoint for certain why Morning Consult has strayed from the polling pack–not in magnitude but in consistency.