Where are all the polls for the U.S. Senate Race in Louisiana?
[This is the first of two posts this week on the state of polling in Louisiana]
Seventeen in New Hampshire. Twenty-six in Ohio. Twenty-seven in Pennsylvania. Twenty-eight each in Florida and North Carolina. These are the numbers of publicly released polls in the U.S. Senate races in these states since January 1, 2016, as collected by the Huffington Post.
Here in Louisiana? As of today it is nine, or at least that is the number of polls I could track down. That's far less than what comes out in a tossup race like those in New Hampshire or Pennsylvania, but does that mean it is a low number?
Typical for Louisiana
This is about the same number as we had at this point in the 2014 U.S. Senate race. Using the counts of polls collected by Real Clear Politics in each election since 2004, the number of polls in this year's race looks similar to each of the state's last three U.S. Senate races.
Different poll averaging sites have different standards for which polls to include, so the numbers in the table above should not be taken as the definitive count of polls. Still, using just one site to compare across elections in just one state gives a sense of whether the amount of polling this year differs from the past. (Real Clear Politics is the only site I could find that collects polls throughout the Election Year in all the U.S. Senate races in Louisiana since at least 2004.)
You have to go all the way back to 2004, Louisiana's last open seat race, to see any real difference in the amount of polling. If anything, there is more polling today, not less, than there was back then.
One other thing worth noting: There haven't been many more polls released from late September to Election Day. Just five or fewer polls came out after this point in the campaign in three of these races. The 2014 election stands out as an exception. In that race, the number of polls more than doubled over the final weeks before the primary.
How do we compare to other states?
Another poll averaging site, FiveThirtyEight, has collected all publicly released polls during the final three weeks of each U.S. Senate race since 1998. Looking at just this late period of the campaign, the amount of polling in these races has increased nationally over the pat 18 years. The total number of polls more than doubled.
This is mostly because more races are polled today than in earlier years. Less than 75 percent of contested U.S. Senate races had any polling in 1998; in 2014, every contested race had at least one poll in the final weeks. It is also because the amount of polling in these races has increased slightly. Looking just at the races that had any polls, the average number of polls rose from fewer than four in 1998 to more than eight in 2006 and 2010. In 2014, the number was just over six. So more states races are polled and the number of polls in a race is on the rise.
Except for the 2014 primary, Louisiana's races for the U.S. Senate have seen fewer than the average number of polls. The number of publicly released polls from the final three weeks of Louisiana's races have fairly consistently fallen in the three to five range, about 1.5 to 5.5 fewer polls than the national average depending on the year.
Lack of competition is partly to blame.
The amount of polling in an election is closely related to how tightly contested the election is. The relationship is captured in the graph below. The red line shows the expected number of polls in a race given how close the winning margin turns out to be. The bigger the margin, the smaller the number of polls.
Most of Louisiana's recent U.S Senate races have not been particularly close. In 1998, Breaux won by 32 points. Six years later when the seat opened up, David Vitter won in the primary phase leading his closest opponent by 22 points. Vitter secured re-election over challenger Charlie Melancon by 19 points in 2010. Louisiana's other senate seat switched hands in 2014 when Bill Cassidy toppled Mary Landrieu by 12 points. Landrieu had previously won a third term by six points over John Kennedy. The only close U.S. Senate election in Louisiana during this period was Landrieu's 2002 win by just three points.
But there is something else going on. Notice the blue points on the graph marking Louisiana's races. They all fall below the red line, meaning that in each case the election had less polling than should be expected based on its competitiveness (or lack of competitiveness) alone.
Does it matter?
Yes. Fewer polls makes it harder to know where the race stands. It is always better to average across a number of polls than to rely on just one. Generally speaking, the more polls available to average across the better the chance that their average comes close to where the candidates actually stand. This is harder when you have fewer polls, especially when those polls are scattered across a longer span of time. Ideally, you have a large number of polls spread over a short window of time. In Louisiana, we have a relatively small number of polls spread over a very long span of time.
Fewer polls also means any questionable surveys in the bunch wield relatively greater influence. In this year's U.S. Senate race in Louisiana, most of the polls are either sponsored by a partisan entity or conducted with lower quality methods such as automated polling of landline numbers and under-representation of individuals who only use cellphones (44 percent of the state's adult population). This does not mean any one of those polls are wrong, but they are more questionable than live caller polls. For example, in runoff polling for last year's gubernatorial contest, automated polls were about twice as far off from the actual victory margin than were live-caller polls.
So, we have to be more careful with the polls when we have less of them. Without a bunch of additional polls for comparison, we cannot make too much of any difference in a new poll because we won't know if it reflects a true change in the race (as the two most recent polls would indicate). Yet, at the same time, by averaging a new poll with polls from weeks and months ago we may drown out actual shifts in the electorate.