Monday, October 12, 2015

A boring race might be interesting- if the polls are right

Until recently, the Houston mayoral race looked like a boring race not worth caring about. After all, the first poll of the race, back in June, showed two clear frontrunners: State Rep. Sylvester Turner (D) and former Harris County Sheriff Adrian Garcia (D). Months and literally dozens of forums later, a new poll one week ago showed the outcome basically unchanged; Turner and Garcia were headed for a December runoff.

But two new polls have been released since then. The first, from TargetPoint Consulting, showed Bill King (I) surging into second place over Garcia with 18% of the vote. However, the poll was commissioned by a group that endorsed King, making it appear like an unreliable internal poll. However, a few days ago another poll came out showing King tied for second place.

What's more, that poll confirmed the dynamic at play: Garcia is rapidly falling in the polls while King consolidates Republican voters. Finally, we have movement in the polls.

Polling Challenges

Perhaps the race has naturally tightened as actual voting draws near. Candidates are now spending money on advertising, more yard signs (bless their heart) are going up, and local media are throwing increasing coverage towards the race itself. Voters may simply care more as election day draws near. However, polling methodologies may also obfuscate the state of the race.



Pollster
Date
Turner %
Garcia %
King %
Bell %
Costello %
Hall %
Undecided %
10/5 – 10/6
24
14
18
11
8
8
13
9/25 – 10/6
19
9
9
6
5
4
42
9/21 – 9/24
19
19
10
10
9
6
25
5/20 – 6/21
16
12
3
8
2
3
53


Only four polls have been released in the mayoral race so far. Two of them are from the same pollster, a collaboration between centers at the University of Houston and Rice University. Their results are conspicuous  for two reasons. First, the number of undecideds in those two polls dwarf the other two, which were conducted by traditional consulting firms. Second, they were each conducted over a month's time.

Time in the field

A month long survey like this can be useful. The writeup of the poll for KHOU notes that Garcia's numbers plummeted over the month after a barrage of negative stories about his tenure as Harris County Sheriff, something that may not have been picked up in a more typical survey length of two days to a week.

However, such a long survey can also mask what's happening day to day in a race. The sample size may be 567 for the total poll, but breaking that up week by week yields much smaller groups. Now break each weekly group into six major candidates plus undecideds, and you have a lot less confidence in the results. In contrast, a more typical poll conducted over a shorter period would have a stronger snapshot of a race, even if it couldn't pick up change over time as well. Both bring something to the table, and unfortunately there probably isn't enough money floating around for a real tracking poll to better measure change in candidate support.

How many undecided voters?

But the bigger looming issue is the large number of undecided voters in the UH/Rice polls. American Strategies pushed undecided voters on which candidate they were leaning toward. While this pushing only added 1-3% to each candidate, it seems to have brought down the number of undecided voters compared to the university polls (23% versus 53% and 42%). Apparently, that 1-3% per candidate adds up.

This Houston pollster is not the only one facing these methodological issues. One of the better-known cases is the UNH Survey Center, which surveys major races in New Hampshire, one of the most polled states in the nation. Its surveys consistently generate higher than usual undecided voters, a phenomenon that makes elections seem more volatile and candidates seem to have less support than they actually have. Even further, their polls are often out in the field longer than typical, sometimes longer than a week. These issues have real consequences.

Almost half of all voters are probably not still undecided a month before election day. With such small differences in vote share for second place, the UH/Rice poll could miss the actual result, which is kind of the point of asking about elections (elections are one of the few opportunities to test pollster accuracy). The UH/Rice polls have some really interesting questions that make them useful, particularly regarding issues important to Houston voters. But on the horserace, the American Strategies poll appears the most reliable.

Of course, the leading / best funded campaigns have internal poll numbers, the results of which most of us will never see, that could help verify this speculation. My guess is UH/Rice will release one more poll shortly before election day, and hopefully one or two more pollsters drop their own results as well in the next month. With more data, we will be able to make a more educated decision: Is this race really a snoozefest or something worth watching?

For more on this year's Houston mayoral election, see this preview of the race from February. 

No comments:

Post a Comment