This
week does bring us a slightly closer margin than even last week did. In a two-way
race Clinton now leads by 3: 45.9% to 42.9% which is down from the four point
spread, 46.1% to 42.1% of a week ago. In
the four-way variation Clinton’s lead is down also to 2.2%: 41.8% to 39.6% from
41.3% to 37.9%. However while the margin this week is closer, the trend toward
Trump seems to have stopped. If we focus
on what should be considered the two highest quality surveys (CNN/ORC and the
Washington Post/ABC polls), we see Trump having pulled slightly ahead in the
September 1-4th poll done by CNN, but Clinton up by much more in the
September 5th-8th Washington Post Poll. Updated polls from more recent sources
confirm the point; Trump’s gains have stalled.
Another interesting
tidbit in these polls is that in both cases, the likely voter screen benefited
Trump by about 4 points. Results
filtered through this screen find their way into the polling averages we use,
which creates grounds for some skepticism as they create another chance for a
pollster to make a mistake.
Pollsters have two jobs in figuring
out who is going to win an election. First they have to figure out who a representative
group of people are going to support, and then they also have to figure out who
is going to vote. One of the striking
things about Presidential elections is that amongst registered voters turnout
is usually quite high. Something on the
order of 75% of registered voters vote. The reality that in the end a
registered voter is likely to vote needs to be built into the model of what
constitutes a likely voter. We are going
to do a brief illustration of why a likely voter screen can go somewhat
haywire. Let’s assume candidate A has
the support of 55 registered voters and candidate B has the support of 45
registered voters. However suppose we
have a likely voter screen that eliminates
10 voters, 7 A’ and 3 B’s. A used
to have 55% of the vote but even after
losing 4 more of his voters , he still
only drops the lead from 10% to close to 7%. If you drop 20 voters at the same rate
candidate A still leads. Losing around 4 points off the lead would
require this kind of asymmetry in support.
This sort of asymmetry when it
comes to whose supporters will or will not turnout is also something of a
dangerous thing to predict. This is
particularly true in our particular case when the candidate in the lead seems
to have a state of the art, sophisticated system for getting their voters to
the polls, and the other campaign does not.
This is also something where the number of states with early or mail voting
becomes even more crucial. The early voting window provides a lot of time to
convert a registered voter into one who casts a ballot. Therefore a strategy which depends in large
part on a much higher percentage of your supporters showing up than your opponents
when you did not build an organization for it is a bad assumption to make. It
is also somewhat dangerous to rely only on the RCP average for this very
reason. Building a model that takes into account both the RV and LV sample from
the same poll is dangerous, as is simply ignoring the pollster’s efforts to
screen entirely. But simply trusting the
screen entirely is not without its draw backs. We will closely monitor as we
go.
0 comments:
Post a Comment