Two new polls on the New Jersey governor's race this week generated a lot of discussion. When I first saw them Tuesday morning, I thought they presented a great potential (warning: over-used-cliche approaching) "learning moment" regarding our charts. Within minutes, some of you who had been playing with our charts discovered a few quirks -- ok, bugs -- in the way our charts work. Today I want to briefly discuss both.
Let's start with the substance. The new polls from Quinnipiac and Fairleigh Dickinson Universities show Republican Chris Christie leading incumbent Democratic Governor John Corzine by ten and five percentage points respectively. The margins prompted some surprise given the closer results seen on other recent polls. Does that mean, as Josh Marshall put it on Tuesday, that a few weeks of "terrible (or seemingly terrible) news for Christie, hasn't had much of an effect?"
Here are the most recent polls:
Most political junkies tend to look first at the margins separating the top two candidates. In this case, scanning the mostly red numbers in the far right column creates the narrative that Marshall and others saw developing in New Jersey: The double-digit leads favoring Christie earlier in the summer narrowed to single digits, including two surveys that showed the margins narrowing to just two percentage points either way. Then the new polls appear, one (from Quinnipiac) showing Christie ahead by ten points and one (from Fairleigh Dickinson) showing him ahead by five. The numbers produce a narrative: At first Corzine appears to be "closing," then maybe he isn't or, perhaps, Christie is "gaining" once again.
There is another way to look at polling data, however, especially in a race featuring an incumbent. While not as gripping as a narrative it probably gets close to the truth about the status of the race over the last month or so. While the vote for Christie varies widely across the different surveys, the vote for Corzine is much more consistent. In fact, all of the results fall within the range of 39%, plus or minus 5% (which happens to be the margin of sampling error typically reported for surveys of 500 interviews). Look at our chart below, altered (using the choices tool) to display only the Corzine percentages and trend line, and you can see the pattern plainly. Since July, the blue dots appear to be randomly scattered around the blue line. The trend is essentially flat.
Now let's look only at the red Christie results. The lower numbers on the Neighborhood (35%) and the two Democracy Corps polls (40% and 43%) help bend back the red line ever so slightly. In this case, the small-c conservative nature of our loess regression trend line is doing what it is designed to do, taking all the data into account but smoothing out the line and minimizing the impact of the bigger "outliers." In this case, the inflection on Christies trend is small. Christie's percentage of the vote, as estimated by out trend line, has declined by only two percentage points (from 48.9% to 46.8%) since Memorial Day. A trend-line based on a straight averages of recent polls would show more variation.
The next question is, why so much variation in the Christie number? Part of the explanation is that some polls prompt for independent Chris Daggett, some do not. Also, some polls -- especially the Neighborhood survey -- use methods that make it easier for respondents to be "undecided." As always, variations in the way pollsters define "likely voters" probably play a role as well. Finally, the level of Christie's support may be changing.
However, the important pattern -- more consistency for the better known incumbent candidate and much variation around the other choices -- is not unusual. While Corzine's current level of support may change, the fact that he seems stuck at just under 40% should be, at least for now, a continuing source of concern for Democrats.
Now for the recently discovered bugs in our charts. You can, of course, use the "filter" tool to drop particular pollsters, and one obvious thing to examine is what happens if you drop an outlier like, for example, Neighborhood poll. What appears to happen, however, is that dropping just a poll or two will also make a significant change in the smoothing factor used to draw the line, something that should not happen. One alert Pollster reader noticed that by dropping the one-and-only New Jersey poll done by Zogby International -- a poll that is now over a year old -- causes a big shift in the current estimate. Again, something like that should not happen, is a bug in the chart. With more than 40 polls available to plot, the chart should leave the smoothing factor constant and it is not. Having isolated the bug, we are at work on a fix. Apologies for not noticing it until now.
Also, if you use the "choices" tool to select something with a percentages under 20% (such as "undecided" or a candidate like Daggett) strange things happen with the "current estimate" percentage that appears in the box in the upper right corner of the chart. Unfortunately, that box displays the last point of the line that appears on the chart as it is drawn. So, for example, if the entire line his hidden (because the percentage range begins at the default value of 25%), the chart shows "0.0%" as the current value. While this unintended "feature" works consistently, it is not at all intuitive and something we need to repair. Until then, a simple tip: When displaying a choice with small percentages, expand the lower percentage range to 0 (by either using the "Pct Range" tool or by clicking a few times on the "-" sign that appears at the lower end of the percentage axis when you hold your mouse over it).
Have you experienced any other problems or odd behavior with our charts? We are putting together a list of "known issues," so if you have, please drop us a line.
How will Trump’s administration impact you? Learn more