As Hillary Clinton has stabilized her general election polling lead over Trump in recent months, recent movement in head-to-head polls both at the national and state level has begun to raise uncertainty about what will happen in the fall. It should be noted that predictive ability really starts to form following the convention periods for both parties, but even at this point, prediction markets and popular statistical models have Clinton at about a 65-70 percent chance of winning the presidency.
In large part, those assessments stem from her polling strength, as despite some widening and narrowing trends in the last months, Clinton has maintained leads in a sizable majority of polls. By one count, since the start of June and through the present day, Clinton has led in 44 of 48 polls, with Trump leading in three.
But that’s what also makes deviations from this trend interesting. Based on the statistical distribution of these samples, we should always expect differences in what the results bring. Deviations from a trend line also validate the statistical methodology; if all poll results were similar or the same, this would indicate “herding,” a detrimental practice that move polls further away from the truth pollsters are after.
At the same time, some polls consistently deviate from overall trends in certain directions. This ought to raise suspicions about biases and methodology employed by the pollster. In the past, Rasmussen Reports has been suspected of a pro-Republican bias, having a “house effect” that consistently leads to polls more favorable to Republican candidates. This comes as likely a result from question wording differences. Over the past months when glancing over Clinton/Trump matchup results or even President Obama’s approval rating numbers, Rasmussen seems once again prone to the same bias.
While a paywall at the site prevents inspection of their survey’s question wording–and thus a chance to really dig into what’s creating a bias, such as priming problem in that case of candidate preferences–comparing the organization’s polling results to those of other groups sheds some light on this pattern. Below, for Clinton/Trump general election margins and Obama approval ratings (approve minus disapprove), I compared the results from Rasmussen’s polls to an average of five polls taken before Rasmussen’s release (with the five excluding Rasmussen). I do this for all polls that are conducted within the month of June and through July.
In the case of the Clinton/Trump vote intention, the average lead Clinton commands is 0.17 percentage points in Rasmussen polls, but 5.3 points in the average of five polls conducted before Rasmussen’s. Thus, Rasmussen polls show a much tighter race than the aggregate does. Interestingly, that discrepancy really begins to form in recent polls (within the selected time frame). In the oldest half shown here, Hillary only does 1.07 percentage points worse in Rasmussen polls.
Moving on to another key political barometer, the incumbent presidential approval rating–often conceived as an influential fundamental in general elections–shows a similar pattern. The average net approval rating for Obama in Rasmussen polls has been -0.1 during this time frame, and the average of five polls done immediately before Rasmussen’s is 3.76. In nine out of the 10 polls here, the approval rating in Rasmussen polls ends up under the average for other polls.
Examining these two polling measures, it becomes fairly clear that a pro-Republican bias has continued in the polling work done by Rasmussen. Coverage of a larger sample of polls and averages would better confirm this point, as there’s always the issue of selecting on the time frame/polls I chose. Nevertheless, starting with June represented the first month in which the candidates for both parties solidified and evaluating these kinds of polling trends could begin in earnest.
Again, it’s worth reiterating that this bias should come as no surprise. According to FiveThirtyEight’s pollster ratings, this pro-Republican tendency has long been in existence, as the most recent ratings–from an analysis of thousands of polls conducted since 1998–had Rasmussen Reports as the firm with the fourth greatest bias toward Republican candidates. Of course, that shouldn’t make it any less of a problem for the pollster.
7/18/16 addendum: As part of the release of FiveThirtyEight’s pollster ratings, the outlet made the data it used–polls released three weeks before various elections since 1998–publicly available. Using this provides another picture of Rasmussen polling going back a few decades.
I reduced the 7,970 total polls included to 6,213 that polled between Republican and Democrat candidates (excluding primary polls). From there, I could also examine 516 total Rasmussen polls during this time frame and with the above restriction. In all polls, Democratic candidates held a lead 51.1 percent of the time, and Republicans did so 44.7 percent of the time. But for Rasmussen polls, 50.8 percent of its polls showed Republican candidates ahead compared with only 43.8 with Democrats leading. Rasmussen polls had Republican candidates ahead 6.1 percentage points more than the average for all polls did.
Agadjanian Politics 