Pollster.com: How to Plot Rolling Average Results?

December 31, 2007

With today's update, we confront the technical issue of how to incorporate rolling-average tracking surveys into our trend charts. What follows is a quick summary of what rolling-average tracking is and how we will incorporate the results of these surveys into our trend estimates. A warning: This post is mostly technical. We just want to make sure we document our procedure for those who sweat such details.

Rolling-average surveys, like the Zogby tracking in Iowa, do a relatively small number of surveys each night but report the rolled-together average of the last several nights of calling (usually three or four). So yesterday's release was based on interviews conducted on Thursday, Friday and Saturday, while today's release is based on interviews conducted on Friday, Saturday and Sunday. We will save the discussion of the pros and cons of rolling-average tracking for another day (though readers are free, as always, to chime in with their comments).

In 2004, Zogby did rolling average tracking in Iowa and six different organizations ran rolling average tracking in New Hampshire. The various media/pollster partnerships have shifted somewhat since, but we assume that the first waves of other New Hampshire rolling average surveys will begin to appear soon.

When including rolling-average results in the data that produce our trend line estimates, we want to avoid including each day's numbers as the base of interviews largely overlaps and the frequency of releases means that rolling averages tend to dominate simpler one-sample surveys (which usually spread interviewing over multiple days). So we will update our charts in a way that ultimately plots only every third release (for a three-day rolling average), but includes the most recent release as the final point for that pollster.

Here is the procedure we will use to avoid including data based on overlapping samples. Suppose a pollster uses a three-day rolling average:

On the first day, we will plot their first release on the chart.
On the second day, we will plot their second release and temporarily remove the first day's data from the chart.
On the third day day, we will plot their third release and remove the second day's data from the chart.
On the fourth day, we will add back the first day's release (and it will remain permanently) and also plot the fourth day's release.
Going forward, we will continue to plot only the most recent release for each three-day window, restoring every third release so the chart ultimately includes no overlapping samples.

The table of results that appears below the standard chart on each poll page will, however, include all poll releases (with links to source data) for every pollster. Those surveys not plotted on the chart will be marked with an asterisk (*).

-- Mark Blumenthal

December 31, 2007 in Pollster.com, The 2008 Race