I read a comment the other day that the US election polls are probably being skewed slightly in Romney's favour by the Rasmussen and Gallup polls (http://www.guardian.co.uk/commentisfree/2012/sep/25/gallup-rasmussen-polling-outliers-lean-republican); there have been a few similar comments about the past performance of Rasmussen especially (http://www.washingtonpost.com/blogs/the-fix/wp/2012/09/17/rasmussen-the-gops-cure-for-the-common-poll/), and lately (2010 midterms, and early this cycle) on Gallup). This could be the result of their methodology or their weighting; I wanted to see if this had much of an impact on the rolling average to warrant concern.
The data was from the RealClearPolitics polling dataset. RCP use the data to create a running average of polls over time, by averaging the most recent polls, while favouring, with weightings, polls which are more recent above those from a little further back.
I used the data to create a straight average of all polls within a given time frame, and conducted that average for each piece of data (i.e. averaged x number of polls prior to each individual poll). This creates a more lagging indicator of actual opinion than the RCP measure, but allowed me to easily adjust the data to take into account a given number of days prior. That is, if a poll ended on day t, I would average that poll and all polls which had a start date of surveying conducted z number of days prior to day t. I then adjusted z to create a smoothed graph.
The Graph below shows the data over time, and then the data with the Rasmussen and Gallup polls removed; below (Chart 2) shows the same data, focused in on the later stages of the polling.
The data was from the RealClearPolitics polling dataset. RCP use the data to create a running average of polls over time, by averaging the most recent polls, while favouring, with weightings, polls which are more recent above those from a little further back.
I used the data to create a straight average of all polls within a given time frame, and conducted that average for each piece of data (i.e. averaged x number of polls prior to each individual poll). This creates a more lagging indicator of actual opinion than the RCP measure, but allowed me to easily adjust the data to take into account a given number of days prior. That is, if a poll ended on day t, I would average that poll and all polls which had a start date of surveying conducted z number of days prior to day t. I then adjusted z to create a smoothed graph.
The Graph below shows the data over time, and then the data with the Rasmussen and Gallup polls removed; below (Chart 2) shows the same data, focused in on the later stages of the polling.
Chart 1
Chart 2
As can be seen, the polling does not seem to indicate that Rasmussen and Gallup have a very large effect on the position. A recent Gallup or Rasmussen poll may have some effect given the
The alternative methodology was similar to the RCP, in that it used a phased average - however I did not make this date dependent. I merely averaged the preceding 5 polls to each given poll, and weighted the 5th poll less than the 4th poll and so on. I then took a sample of the rolling averages to display the pattern of data. This shows similarly, the Rasmussen/Gallup removal makes limited difference to the pattern.
Chart 3
No comments:
Post a Comment