About | Contact | Advertise | RSS
HomeThe PollsPollster Blogs

November 05, 2006

New: "Pollster Compare" Senate Charts

With less than 36 hours to go before the polls open, and the final poll releases coming at a furious pace, I want to point out a new set of charts we added to Pollster earlier this weekend that should help those puzzling over contradictory results in recent Senate polls.  We call them "Pollster Compare" charts and, as of last night you can see them for the 12 most competitive Senate races.

These charts allow you to compare the trend in the vote margin between the various pollsters in each race and spot the "house effects" that make one set of results consistently different across.   Each race has two charts.  One simply connects the dots for each pollster, adding a regression "local trend line" that estimates the combined trend picked up by all telephone polls.  A good example is the Maryland Senate race (copied below).  Note that the three automated polls by SurveyUSA have all shown the race virtually tied, while other polls (including the automated surveys from Rasmussen Reports) show a narrowing race, with Democrat Ben Cardin typically leading by roughly five percentage points.


Another comparison of interest is the Tennessee Senate race.  We posted an earlier version of the pollster compare chart, which coincided with a lot of speculation about whether Democrat Harold Ford was doing consistently better on the surveys (by Rasmussen and SurveyUSA) that use an automated audio interview rather than a live interviewer.


To help facilitate that sort of comparison, we have also included a second format of chart that plots dots for each poll and three sets of gray "local trend" regression lines to estimate the combined trend for different types of polls:

  • All telephone polls (solid gray)
  • Telephone polls using live interviewers (gray dashed line)
  • Telephone polls using Interactive Voice Response or IVR (gray dotted line)


Looking at the Tennessee chart, the IVR and live interviewer lines have converged since SurveyUSA started conducting surveys in that state in July.  So whatever made Ford perform consistently better in surveys by the other automated pollster -- Rasmussen Reports -- it was probably not the IVR methodology. 

--Mark Blumenthal on November 05, 2006 in Pollster.com, The 2006 Race


TrackBack URL for this entry:

Listed below are links to weblogs that reference New: "Pollster Compare" Senate Charts:


Might be a good idea to put error bars on each data point. (the 95th percentile confidence interval is the standard, right?)

Posted by: meep | November 6, 2006 2:46 PM

Post a comment