Saturday, September 24, 2011

About Those Indianapolis Mayoral Polls

A recent campaign poll showed Indianapolis Mayor Greg Ballard up 12 points. Melina Kennedy's campaign though recently released one suggesting she is up 2 points on the Mayor.

What's the difference?  There is a lot of focus on margin of error.  While that's too complicated to explain here, I would note that the larger the sample size the less the margin of error.  Also, the media misreports margin of error when they say a lead is outside or inside a margin of error.  Actually a 4 point margin of error means both candidates numbers can be off 4 points on either side. Thus there a candidate would need a 9 point lead to be outside a 4 point margin of error.

But the bigger error by far in these types of campaigns is the sample selected to poll.  If Democrats want a poll favorable to them, they poll more Democrats.  If Republicans want a poll favorable to them, they poll more Republicans.

WISH-TV reporter Jim Shella claims to have seen the two polls and provides some insight into the sample used in both polls:
 I was permitted to review both polls and found that the major difference, aside from the head to head numbers, is in the breakdown of Republican and Democratic voters in the polling samples. Democrats believe they have a 10 point advantage in terms of party identification by voters in Marion County. Republicans think the Democratic advantage is only 4 points.
So the poll showing Ballard ahead by 12 points only polled 4% more Democrats than Republicans, while the poll that showed Kennedy up by 2 points polled 10% more Democrats than Republicans.

In 2008, the baseline numbers in Marion County favored the Democrats 60-40, i.e. 20 points.  In 2010, the baseline numbers, in the midst of a huge Republican year, favored the Democrats 55-45.

Looking at those baseline numbers, I would have to question any poll that bases its sample on the Democrats only having a 4% advantage on registration in Marion County.  Obviously the edge is substantially bigger than that.  Even a 10% advantage estimate might be too low.

No comments: