Earlier today, Public Policy Polling (also known as PPP) released a poll in the Louisiana Senate race that showed three term incumbent Mary Landrieu clinging to a one point lead over Representative Bill Cassidy (R-Baton Rouge). In its write up, PPP noted that this one point lead was a significant tightening of the race since before Obamacare was implemented, when she formerly had a 50/40% lead over Cassidy.
However, in analyzing the internals for the poll, we have concluded that this poll doesn’t tell the complete story about Senator Landrieu’s position in the race against Cassidy (and, for that matter, the other two “major” Republicans in the race).
To put this poll in proper perspective, we would like to (1) discuss various polling methods, (2) discuss the Louisiana electorate, and (3) tie those two topics in with the PPP poll.
Polling is one of the most misunderstood features of a campaign, because it is very subjective. Nevertheless, there are “rules of the game” that can be applied to any poll, particularly the one we are discussing.
At its simplest, a poll can be described as a representative sample of voters at a given point in time. It is the definition of what a “representative sample” is that creates divergent results among pollsters.
Some universal truths must be understood at this point: (1) different segments of the population have different opinions on issues/candidates, and (2) not everyone votes. Therefore, pollsters have to have assumptions as to the representative composition of the jurisdiction being polled – in this case, the state of Louisiana.
To obtain that “representative sample”, the easiest way to poll voters is by dialing a list of phone numbers. However, this method has several flaws: (1) the respondent may not be a registered voter, and (2) the respondent (even if a registered voter) may not be eligible to vote in an upcoming election.
There is another way to get a sample: call a list of registered voters. Of course, this method has its drawbacks: (1) the list may be “stale” if we’re talking about a fast growing suburban area or if the list hasn’t been updated since (to use an example) the 2012 Presidential election, (2) since not everyone votes, getting “unlikely” voters’ opinions doesn’t really add much value to any poll results. While it is true that polling organizations typically ask the voter how likely he/she is to vote, voters typically overestimate their likelihood to vote. To illustrate, you’ll typically see 80-90% of respondents indicating that they plan to vote in a given election (and thus are eligible to participate in the poll). That kind of turnout, however, is generally unheard of – even in a Presidential election year.
There is a more restrictive method, which really is the more technically correct one: letting a voter’s voting history determine who a “likely“ voter is, and only poll those who are “likely.” To illustrate, 68% of eligible Louisiana voters participated in the 2012 Presidential election. Does that mean that 32% of the electorate stayed home? A further review of electoral data reveals that 13% of voters (about 390K) have never cast a ballot, and another 518K (or 18%) last voted BEFORE the Presidential election. In other words, it can be argued that a substantial portion of the non-voters consists of those who are not likely to cast a ballot in a future election. And, more importantly, polling these “non voters” dilutes the quality of the poll sample.
Understanding the Louisiana electorate
To understand the Louisiana electorate, it’s important to realize that Louisiana is a Deep South state. And as such, Louisiana has a substantial percentage of black voters. Given that their vote is overwhelmingly Democratic in partisan contests, the black percentage of the electorate drives the strength of the Democratic vote in a given election.
The February 1, 2014 statewide voter registration statistics show an electorate that is 64-31% white/black, and from reading the poll, JMC has determined that the composition of the poll respondents was 67-31% white/black. From a standpoint of being representative of the voter registration, the PPP poll passes the credibility test. However, black voters typically have a lower electoral participation rate, so that racial composition in a statewide poll is not realistic. To illustrate, here’s the racial composition of the Louisiana electorate in recent elections:
Given that 2014 is a midterm election, the racial composition of prior midterms is a better barometer of the 2014 race. In those midterm elections (2002, 2003, 2007, 2010, and 2011), the black percentage of the electorate was in the 24-27% range – NOT the 31% PPP is assuming will make up the 2014 electorate.
Bottom line: why does this matter?
Given that Louisiana has an electorate that, from a racial standpoint, is fairly polarized, the composition of the poll sample matters a great deal in terms of the reasonableness of the results. That is why we take issue with PPP. Below are the poll results using PPP’s assumptions as to the composition of the Louisiana electorate (basically, “registered voters”):
Given that this is a midterm election, Democratic intensity/enthusiasm will likely be less than it was in 2012. Furthermore, this race is, for the first time, being conducted against the backdrop of some votes (Obamacare, gun control, stimulus, etc.) that can be used to paint Senator Landrieu as not representing Louisiana’s opinion. This is why we prefer to use a midterm electorate similar to the 2002 Senate race (i.e, 71-27% white/black) and recast the PPP results along those lines:
(Revised 2/12 AM) In other words, PPP’s poll is showing a 2-3% bias towards Senator Landrieu. In a contested race like this one, that 2-3% is the margin of victory.
Conclusion (Revised 2/12 AM)
In conclusion, any time a polling organization puts out a poll on a contested race like this one, it is important to independently evaluate the assumptions used to construct their sample. Especially since the PPP poll will be given considerable attention from the local and national news media.