04/14/2008 02:42 pm ET Updated Dec 06, 2017

Comparing Pollsters in Pennsylvania

I know the question just about everyone following the Pennsylvania primary contest is asking today is what polls are showing regarding the impact of the "bitterness" flap involving comments made at a fundraiser by Barack Obama last week. We have two new surveys released so far today, one from ARG conducted as the "bitter" story broke over the weekend (4/11-13) and one from Susequehanna Polling and Research University completed earlier (4/6-10). We will have more clarity on this story's impact as more polls are released, but in the meantime -- thanks to Charles Franklin -- here is a chart that can help put these new results into context (click on the image to pop-up a full-size version).

The chart plots the Clinton margin (Clinton minus Obama) separately for each pollster that has released two or more Pennsylvania surveys since February 1 (except Susquehanna -- Charles ran this chart just before we received that release). We have also plotted our "standard estimate," the margin based on the standard regression trend lines that we plot in our Pennsylvania chart.

A few notes about the data, although as always, your interpretation may differ:

First, every pollster showed some decline in Clinton's margin between mid- and late March, as reflected in the black trend line which shows that margin shrinking from roughly 13 to roughly 7 points.

Second, several pollsters (SurveyUSA, PPP, Insider Advantage and ARG) have shown upticks in the Clinton margin since April 1. These findings affect the trend line by flattening out its slope. But the margins reported by the four pollsters are scattered from well below to well above the trend line, making their effect on the overall estimate less powerful than it would be if they all agreed on the level of support for each candidate.

[Update: Also notice that surveys by Quinnipiac, Strategic Vision and the last ARG showed a down tick, and Rasmussen was flat, just a day or so before the three others reported an up-tick up].

Third, the black trend line continues to show Clinton's margin declining because the regression model that generates it gives greater weight to the larger number of polls conducted earlier showing a decline. The new ARG survey is so different from other polls that, with a more sensitive estimator, the line would move quite a bit in its direction, producing an upward spike. Our more conservative estimator is pulled up by ARG, but it is designed to resist the influence of single poll that falls far from the rest. If other new polls are consistent with ARG, then the trend estimator will move sharply in that direction. But if new results look like other recent polls, then the trend will remain about where it is.

[Charles Franklin updates with much more on the sensitivity of the Pennsylvania estimates].

All of which is a long way of saying that for the moment, your judgement about where the trend is headed depends on what you make of the ARG poll.

Fourth, the Pennsylvania surveys from PPP have been something of an outlier in terms of their sheer volatility. They produced the biggest Clinton margin (by far) in mid-March (in a survey conducted entirely on March 16, on the eve of Obama's speech on race) and the one result showing a slight Obama lead on April 1. Again, only time will tell whether ARG has produced a similarly outlying trend or is the harbinger of comparable results yet to come.

Beyond that, we can only stare at the chart and speculate. So far at least, the two national tracking surveys are not showing evidence in a significant shift since Friday in Obama-Clinton vote preference. As Josh Marshall noted, however, "coverage of this story in Pennsylvania has likely been as intense as anywhere in the country." So we will have to wait and see what other Pennsylvania polls have to say.

On that note: Quinnipiac University has distributed a media advisory that will be releasing their latest survey tomorrow morning. So stay tuned.