Pollster.com

Articles and Analysis

 

Why Does a Weaker Result Move the Trend Up?

Topics: 2008 , Pollster.com , Slate Scorecard , The 2008 Race

As many of you know, our partnership with the online journal Slate resumed last week, as they kicked off their new Election Scorecard feature, which is once again powered and provided by Pollster.com. For now, the Slate feature displays our most recent trend estimate for each candidate in the early primary states as well as handy Flash graphic. The display of trend updates has both the writers at Slate and yours truly watching the daily twitches in the numbers more carefully. Today's movement in the national numbers for the Democratic primary reveals an idiosyncrasy in the way our trend estimates behave that I want to explain.

Some background: The trend lines we plot in our charts are different from the rolling averages we plotted for the races for Senate, Governor and U.S. House in 2006, and from the "polling averages" you see on other web sites. A polling average makes use of data from just the most recent polls included in the average (be it 5 polls or some other number). Our approach - developed by our partner, Professor Charles Franklin - has been to plot line based on a "local regression" that takes into account all available data for the current estimate, not just the most recent 4 or 5 or 8 polls.

The key difference between trend estimates and rolling averages is that an average produces a new estimate for each combination of polls included in the average at any point in time. The regression line produces a trend line - a line, rather than a point - with a particular slope that is either moving up, down or staying level at any point in time.

Another key issue is the level of sensitivity that Professor Franklin built into the regression model that produces the trend line. I'll let him explain:

I've chosen an estimation method and designed the approach we take so that the trend estimator should be resistant to bias due to a single organization or a single poll. While it can be fooled under the right circumstances, those should be both rare and short lived, rather than common and long term.

Franklin explains the mechanics of the estimator in more detail in posts here, here and here.

This brings me to the most recent odd twitch in the national averages for the Democratic presidential trial heat. Late last week, our last update of the national Democratic numbers had Hillary Clinton at 43.2%. Yesterday, we updated the charts with a new national poll conducted and released by the Republican firm, Public Opinion Strategies, that gave Clinton 40% of the Democratic vote. Yet despite showing a result for Clinton that was below her latest trend estimate, the addition of the new poll moved her estimate up higher by nearly a full percentage point (from 43.2% to 44.1%).

10-23-USTopzDems_sml.png

Why? It is all about the what the regression estimate tells us about the trend evident in the last 10 or so polls. The chart makes clear that our most recent estimate of the trend is sharply up for Clinton. As per Franklin's design, the addition of just one new poll did not significantly lessen that upward slope. However, since the end date of the new poll comes a full week since the last poll, the line has moved forward in the upward direction for another week, thus producing a nearly one point increase.

The point is, we're not just adding one new poll and dropping one old poll from a last-five or last-six poll average. We are gradually updating a trend line based on all the data available.

 

Comments
Andrew:

That makes sense.

____________________



Post a comment




Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.

MAP - US, AL, AK, AZ, AR, CA, CO, CT, DE, FL, GA, HI, ID, IL, IN, IA, KS, KY, LA, ME, MD, MA, MI, MN, MS, MO, MT, NE, NV, NH, NJ, NM, NY, NC, ND, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VT, VA, WA, WV, WI, WY, PR