Some argue there has been no decline in polling support for Gov. Rick Perry among GOP primary voters. The chart above, and I, disagree. That looks like a bit of a dip at the end, from just barely over 30.1% to 27.1%.
But don’t take my word for it. Let’s check a completely independent source, RealClearPolitics.com:
The blue line is Perry. Looks to be down from about 32 to 28.4, according to RCP’s table.
So my trend estimator has taken a dip, RealClearPolitics’ moving average has also dipped.
So why then does The New York Times’ Nate Silver characterize my previous post as
One misconception is that Mr. Perry’s standing had been declining in primary polls prior to Thursday evening. This simply isn’t the case, at least not to any degree of statistical significance.
At first, I feared I had made an embarrassing mistake, but checking the arithmetic shows that the different conclusions between Silver and RCP and I have to do with the time perspective one brings to the data. And that turns out to be just a little more interesting than a mere statistical pissing match.
Silver comes to his conclusion by averaging the polls from Aug 13-Sept 6, and from Sept 7-Sept 19. Happily we both agree that RealClearPolitics.com’s list of national polls of GOP primary voters are the right polls to use, so our different conclusions don’t result from different sets of data. Silver finds the average Perry support before the first debate was 29.0 and after is 28.4 (also RCP’s current moving average.) That’s a change of just -0.6 percentage points, and we agree that is not statistically significant. (In point of fact, Silver uses a simple average, which ignores the differing sample sizes across the polls. If we correctly weight the polls to account for variations in sample size, the pre-debate average is 27.4 and the post-debate(s) average is 27.7, a gain of +0.3 and Silver would seem righter than he knew.)
So why do we differ in conclusions? Silver’s assumption is that the pre-debate polls are all homogeneous, with no time trend across them, and likewise for the post debate polls. If that were true, his approach to pre-post comparisons would be fine.
In contrast, my approach to estimating trends in polling, and the different but philosophically similar moving average approach of RealClearPolitics is that public opinion is dynamic and constantly moving. I and RealClearPolitics estimate trends that respond to these short term fluctuations in polling, rather than compare static pre-post groups. And that does lead to different conclusions about how Governor Perry has been doing recently.
Let’s look at a closeup of Perry’s polling since June 1 (there are few polls that include him before that date in any case.)
The gray line is my trend estimate based on all the polls. It is the same trend as the first chart of this post, except zoomed in to show more recent detail. The recent decline is numerically the same but looks far milder because of the different time scale, 4 months vs 30. The red lines are also trend estimates but calculated separately for the pre- and post-Perry entry data. The sharp jump after Perry enters is hard to capture instantly with any trend estimator, so the red lines provide an alternative view. As you can see, the red and gray trends nevertheless follow each other closely except right at the point Perry enters. The conclusion they offer about recent change is the same.
The critical point we see is that Perry support was rising when he entered and his support continued to rise up to September 1. The red line shows a clear peak there, and the gray line reaches and then maintains a plateau. The red line peak is a shade higher at 31.4%, but it falls off to 28.5% at the end, a decline of 2.9 points. The gray line peaks at 30.1% and declines to 27.1%, a three point decline.
And there lies the difference between Silver’s analysis and mine. Perry entered the race on an upswing and continued to gain before his trend reversed direction just about September 1. Since then, he’s come down about 2.9 to 3.0 points. That isn’t a huge loss. He’s not in free fall. But he isn’t holding steady either.
What drives this are the four polls just before September 1. Two are quite high, above 35%. Another at 29% is as high as any previous poll. And one is rather low. But average them (or fit them with my trend estimate) and together they argue for a peak in Perry support. (The sample-size weighted average for the four is 31.78%, close to the peak of the red line at 31.4%.) One could debate the two highest polls as outliers. Removing them would certainly change the trend. Sometimes outliers are indeed influential and a more conservative approach to the data would consider this but that issue wasn’t part of the difference of opinion that started this exchange since both Silver and I include these two high polls.
If we compare the last four polls before the first debate with the most recent four polls (three entirely after the 2nd debate and one that straddles the 2nd debate) we get an estimate of 27.06 (accounting for differing sample sizes), again quite close to my trend estimate of 27.1, and a decline between the two sets of four polls of 4.7 percentage points, a bit higher than my trend estimate.
My conclusion is that there has been a modest but clear decline in Perry’s support since the first debate. Whether you believe that or not depends on how you view the dynamics of polling. Accounting for short term change, as I do, clarifies the rising support for Perry before September 1 followed by an equal decline since then. If you prefer to treat the earlier polls as static and homogeneous, and likewise those after debates began, then you will agree with Silver’s conclusion that Perry is as popular with GOP voters now as he was before the debates began.
Horse races take at least two horses, so before ending let’s see Romney’s performance over the same period:
It is clear Romney has also enjoyed some dynamics, falling over 5 points but recently reversing that decline. From his low of 17.8% he has now climbed to 22.6%. During some of that time Perry’s rise was outpacing Romney’s but as of polling through September 19, the gap between Perry and Romney has closed to about 4.5 points. Perry remains the front runner.
At the end of my original post on this I suggested that the continued media focus on Gov. Perry’s debate troubles has helped mask continuing potential trouble for Gov. Romney. One reason Perry zoomed up so rapidly is that a sizable share of GOP primary voters are not satisfied with Romney and were happy to embrace a strongly conservative alternative. Perry’s problems don’t mean that lack of enthusiasm for Romney has changed. My analysis here still shows Perry in the lead and if no other strongly conservative candidate enters the race this may well remain the horse race of the year. Romney has yet to demonstrate he can win over those who were reluctant to support him four years ago even when paired against a candidate having some trouble adjusting to presidential debates.