Sunday, September 26, 2010

About that 8th District Poll ...

The 8th congressional district poll released last week caused a stir. It showed Patrick Murphy trailing Mike Fitzpatrick by a noticeable margin. The race is likely to be close and both campaigns have released polls showing that. When Murphy defeated Fitzpatrick in 2006 it was a squeaker. The rematch looks to be a similar roller coaster. But the latest polls showed a double digit lead for Fitzpatrick. That seemed a little strange to me.

Granted I don't know much (anything) about polling. Granted, also, that I have been following Murphy's campaigns since 2005, and so could not be called completely objective. All the same, I wanted to dig a little into the numbers and see if anything could account for the numbers.

The Franklin & Marshall poll provides demographic data on those polled. The age distribution seemed a little off to me. Those under 45 years old represent only 17% of the poll respondents. The US Census Bureau provides some basic demographic data on Bucks County (2006-2008). The 8th congressional district includes a sliver of Montgomery County and a bit of Philadelphia as well. In fact in 2006 Murphy lost Bucks County, but the other parts of the district put him over. Of poll respondents 95% live in Bucks, 3% in Montco and 3% in Philly (the narrative states that percentages won't always add up to 100).

Looking at the age distribution in Bucks as listed with the census, 31% of the county's residents are under 45. That's nearly twice the percentage reflected in the poll. The poll report listed 35% of the respondents as 65 and older although the census reports that only 13.6 of the population was over 65. Finding numbers for those who actually vote is tricky. I couldn't find anything by congressional district. Professional politicos can probably come up with that data. The closest I could come was statewide statistics. This also comes from the census bureau (excel files available for download). I found voter registration numbers and percentages that actually voted. It is true that those over 45 register and vote in greater percentages than those under 45. To get even wonkier I compared age level voting numbers for 2004 and 2008, though this was only available by region. In the northeast (including Pennsylvania) the only group to vote in greater numbers in 2008 was the 18-24 age group. For the 25-44, 45-64, and 65+ groups voter participation went down between 2004 and 2008. (Project Vote recently released an interesting report on what they call the "rising American electorate" which includes younger voters.)

The poll's emphasis on older voters may not be representative of the county's population. This could skew the results even further if, following the 2004-2008 trend, more younger people come out to vote, though mid-terms usually don't bring out as many voters as presidential years.

The gender representation in the poll seems to match that of the county. The income groupings in the poll deviate in one category from that of the census. The greatest deviation is in the upper income levels. the poll lists 30% of respondents earning over $200K. The county census lists 44%. Of course it is possible that with the economic upheaval of the past year or so that households might have lost income. Plus, 7% of the poll respondents don't know (or, more likely, won't say, what their household income is). Otherwise there is just a difference of a percentage point of two. Generally speaking those with higher incomes and higher education levels vote more frequently than those with less income and education. How they vote probably depends on where their income is from. The excel census files show that nationwide veterans vote in greater numbers than non-veterans (and female veterans in greater number than male; this surprised me). The poll respondents show 14% are veterans, but the county generally has only 9.6% veterans.

I don't know if the deviations between poll respondents and county residents generally would impact the accuracy of the poll. The age breakdown is the one that I think could change the outcome the most, depending on how the younger people vote.

At any rate, I wanted to dig a little deeper into the poll findings and am thus sharing with readers what I found. You are encouraged to review the numbers yourself. Let me also say that I have the highest respect for the team at Franklin & Marshall and am sure the poll reflects the responses from those polled. I'm just not sure those polled reflect the voters in the district in representative proportions.


DLeopold said...

Here's what the Murphy campaign says:

Pollster Showing Murphy Down Has Uneven Track Record in PA-08

(Bristol, PA) - Yesterday, Franklin and Marshall released a poll showing Patrick Murphy down 10% against his opponent, former Congressman Mike Fitzpatrick, in Pennsylvania's 8th Congressional District race. But their polling has been very inaccurate in PA-08 before. When Murphy and Fitzpatrick first faced off in 2006, Franklin and Marshall put Murphy behind by 9% (39%-48%) on October 23, 2006. Of course, Murphy didn't lose to Fitzpatrick by 9%. He ended up winning the election less than two weeks after that poll was released. The poll's inaccurate numbers were well beyond the claimed margin of error.

The pollster used by Patrick Murphy for Congress in 2006 and presently, Global Strategy Group, had Murphy up by 2% at about the same point that Franklin and Marshall had Murphy down 9% in 2006. Murphy's final margin of victory was about half a percent, exactly in line with Global Strategy's results. Global Strategy Group's recent poll has Murphy with a narrow lead of 47%-43% (polling memo attached here). As expected, this will be a close race and Murphy is taking nothing for granted.

phillydem said...

Jane, F&M is the old Keystone Poll. Terry Madonna still runs the poll and has been following and polling state politics for decades. The poll does try to weight itself based on demographics. Howvere, F&M is still done as a live person-to-person telephone survey. It's not an automated telephone survey poll like Rasmussen, PPP, SUSA, etc. I know we've read a lot about pollsters "missing" younger voters who have cell phones only and not land lines. IIRC, cell phone numbers aren't usually part of telephone polling. The poll's oversampling of older voters may be a result of that group being more likely to have land lines than younger voters.

The positive is that F&M always puts all its data out for review and is one of the most transparent polls out there.