Many of us, including me, wait with baited breath for the daily 3 day tracking poll results from Rasmussen and Gallup. However, I suspect that most people do not understand that these polls are notoriously bad at picking up short term trends in support for the candidates. In fact, as I will demonstrate below, they often indicate substantial trends in the opposite direction of the actual trend.
A 3 day tracking poll is one which lists, on a daily basis, the average poll results for each candidate over the last 3 days. Hence, it is often considered more reliable than a poll taken on a single day and, since it is reported on a daily basis, one might conclude that it is the best polling method for spotting daily trends.
However, this conclusion is not correct, and tracking polls are actually notoriously bad at picking up short term trends. The main reason for this is that the results of a tracking poll are as much influenced by the absence from the total 3 day sample of the results from 4 days ago as they are influenced by the results of the day before. Thus, many perceived trends are illusory.
To demonstrate this mathematical fact, below is a hypothetical chart showing possible polling results over a 10 day period. The column entitled "Daily Results" shows the results of each individual day's polling, which information is unfortunately not displayed by Rasmussen or Gallup with the results of their tracking polls. The next column shows the results of the 3 day tracking poll. For comparison purposes, the next column contains the results of a 5 day tracking poll based on the same daily figures and the last column contains the results, day by day, of a poll counting all results from days 1 through 10.
Day | Daily Results | 3 Day Tracking | 5 day Tracking | Count Everything |
---|
Day | 46 | 46 | 46 | 46 |
1 | 44 | 44 | 44 | 44 |
Day | 46 | 46 | 46 | 46 |
2 | 44 | 44 | 44 | 44 |
Day | 46 | 46 | 46 | 46 |
3 | 44 | 44 | 44 | 44 |
Day | 46 | 46 | 46 | 46 |
4 | 44 | 44 | 44 | 44 |
Day | 46 | 46 | 46 | 46 |
5 | 44 | 44 | 44 | 44 |
Day | 46 | 46 | 46 | 46 |
6 | 44 | 44 | 44 | 44 |
Day | 52 | 48 | 47 | 47 |
7 | 38 | 40 | 43 | 43 |
Day | 45 | 48 | 47 | 47 |
8 | 45 | 42 | 43 | 43 |
Day | 46 | 48 | 47 | 47 |
9 | 44 | 43 | 43 | 43 |
Day | 47 | 46 | 47 | 47 |
10 | 43 | 44 | 43 | 43 |
On the first 6 days of the poll, the results were identical -- the first candidate led by 46 to 44 each and every day. However, on day 7 the first candidate had a great day, no doubt at least somewhat aberrational, and led 52 to 38 on that one day. Beginning on day 8, the results moved back to where they had been before day 7, with the first candidate tied on day 8, up by 2 on day 9, and up by 4 on day 10, the candidate's second best showing after day 7. Thus, days 8 through 10 show a small but clear trend in favor of the first candidate.
Note also that the 5 day tracking poll and 10 day poll show a very static race, with the first candidate up 46 to 44 on days 1 through 6 and up 47 to 43 on days 7 through 10.
However, the 3 day tracking poll shows the first candidate leading by 6 points on day 8, by 5 points on day 9, and by only 2 points on the last day, day 10. Thus, contrary to the daily results over the last 3 days, which show a steady trend in favor of the first candidate, and a 4 point gain, the 3 day tracking poll shows a steady trend against the first candidate, and a 4 point loss over the last 3 days. So the tracking poll has changed a seeming 4 point swing in favor of the first candidate into a seeming 4 point swing in favor of the second candidate.
The reason for this is entirely due to the fact that the results of day 7 are no longer reflected in the tracking poll by day 10. That is not an unreasonable result, since they are 4 days old and were probably aberrational. However, the point is that the removal of the day 7 figures results in a highly misleading daily trend indicator, turning a +4 trend into a reverse -4 trend, a startling 8 point negative swing.
In my opinion, all tracking polls should reveal not just the 3 day results but the recent daily results in numerical form. If this were done, right wing pollsters like Rasmussen could not try to influence the campaign narrative by reporting non-existent trends in favor of McCain.