As I have written before, my graduate degrees are in Educational Psychology and Research (which involved numerous courses in measurement and statistics.) While I have never conducted a poll, I know how to read one.
Let’s take the latest Survey USA (SUSA) poll (co-sponsored by The Courier-Journal and WHAS-TV) on the Conway vs. Mongiardo race. SUSA is without doubt one of the finest public polling outfits in America. Its accuracy rate has been well established.
As most of you know by now, SUSA claims that this race is a statistical dead heat. Mongiardo is polling at 38% and Conway at 37%. While these data are substantially different from earlier SUSA polls on this race, they are statistically the same as those reported by SUSA a month ago.
Most of you have read that this poll has a plus or minus error rate of 3.9. This error rate, so I have been taught, applies to the reported mean for each candidate. The reported mean for Conway ranges from 41% to 34% and for Mongiardo from 42% to 34%.
However, the given standard error of measurement is less important than the average distance between a poll’s projected margin (1% in the Conway-Mongriado race) and the actual margin. Very few polling corporations report these later data.
Pollster.com has found that Survey USA’s projections have tended to be 4.5% off from the actual outcomes. Daily KOS showed an average difference of 6.11% between SUSA projections for the Obama-Clinton primary cycle and the actual vote.
Because of the cited statistical issues, one cannot claim that Conway has momentum in this campaign. To repeat, the statistical difference between the last two SUSA polls in this contest is meaningless.
Because of the cited statistical issues, one cannot claim that Mongiardo is leading in this race. According to SUSA, Mongiardo or Conway can win this race.
Because of the cited statistical issues, one cannot make any claims about a candidate’s margin of victory. Chances are that it will be razor close but it might not be.