Some Thoughts on Sept. 11
First a moment of silence for those who were killed 9 years ago today. But it is unfortunate that 9/11
has become highly politicized and a day of hatred. The biggest news stories of the past week have been about
people who want to burn Korans and whether some people should be allowed to build a mosque on private property
near the existing mosque and strip joints in the holy New York financial district.
Consider that 9/11 is not that
unique a day. Veterans Day--originally
on 11/11--was never a day to hate people who used poison gas during WWI. Memorial Day, now mostly for picnics,
but originally proclaimed as Decoration Day
by Abraham Lincoln was a day of respect to put flowers on the graves of fallen soldiers.
It generally doesn't bring up the subject of slavery although the "peculiar institution" does have something to do
with it. A closer
analogy is Pearl Harbor Day, when 2403
Americans died (vs. 2671 Americans and 327 from other countries on 9/11). But 9 years afterwards it was hardly a focal
point of anti-Japanese hatred despite President Roosevelt saying it was
"A date that will live in infamy." Perhaps the closest analogy is
4/19. What? You have forgotten already?
April 19, 1995 was the day mass murderer Timothy McVeigh and his co-conspirator set off a bomb that killed 168 people and damaged
324 buildings in Oklahoma City. He bitterly hated the federal government and would now be considered a member of
the tea party movement. He was a white Christian Republican but 4/19 rarely inspires hatred of these groups. Odd.
Nate Silver Predicts GOP has a 2/3 Chance of Taking the House
Statistician Nate Silver has built a
of the 435 House races and by running simulations with the model, has determined that the Republicans have a 2-in-3
chance of getting a majority in the House. For people not familiar with stochastic modeling, a lot depends on the
details of the model. Silver has included every parameter and bit of data he could get a hold of, which is generally
a good thing, but the devil is in the details. For example, current polling shows that when asked: "Will you vote
for the Democrat or the Republican?" (without naming the specific local candidates), the GOP is up by 8%. Such a
factor can be included in the model. The hard part is figuring out how heavy to weigh it. Should 25% of the predictive
power of each district be based on this number? 10%? 2%? The model maker has to assign a weight to this factor and to
all the other factors as well. Different model makers use different input factors and different weights and get different
results. The same is true of economic forecasting, where one model may predict 3% growth next year and a different model
may predict only 1%. Everything depends on the details of the underlying model (which in reality is just
a computer program and if it is complicated enough, may contain programming errors--3 to 10 errors per 1000 lines
of code is the industry average).
We also have a model (or use the
"House races" link to the right of the map) but it is an algorithmic model rather than a stochastic model. It consists
of a series of rules (called predicates), each of which tests some parameters of the district, such as the PVI, and
after all the rules are applied, comes to a prediction. For example, one of our rules is than any incumbent who is running
and who got 55% of the vote or more in 2008 will be re-elected. Why 55%? Why not 53% or 58%? Again, here the model
maker's judgement and intuition play a big role. In this case, the knowledge that the
reelection rate for incumbents has
never dropped below 85% for 40 years and has been above 93% for the past decade suggests that anyone who won big time
in the previous cycle is probably safe.
One way to reduce the uncertainly in looking at models is to compare the results of two completely different
models, in this case a probabalistic model (Silver's) and a deterministic model (ours). To make the comparison
possible, any district Silver rated as 51% or more likely to flip parties is considered a takeover.
He gives 55 districts that are likely to flip. Clicking on any of them brings up a description of the race or a map.
AL-02 AR-01 AR-02 AZ-01 AZ-05 CA-11 CO-03 CA-04 DE-AL FL-08 FL-24
GA-08 HI-01 IA-03 IL-10 IL-11 IL-14 IN-08 IN-09 KS-03 KY-06 LA-02
LA-03 MD-01 MI-01 MI-07 MO-04 MS-01 ND-AL NH-01 NH-02 NM-02 NV-03
NY-24 NY-29 OH-01 OH-15 OH-16 PA-03 PA-07 PA-08 PA-10 PA-11 PA-12
SC-05 SD-AL TN-06 TN-08 TX-17 VA-02 VA-05 WA-03 WI-07 WI-08 WV-01
In contrast, our model predicts these 33 seats will flip:
AR-01 AR-02 AR-04 AZ-05 CA-18 DE-AL FL-08 HI-01 IL-10 IL-17 IN-08 KS-03 LA-02
MI-01 MI-07 MI-09 MI-11 MN-03 MS-01 NH-01 NJ-03 NV-03 OH-16 PA-03 PA-06 PA-07
PA-12 TN-06 TN-08 VA-02 WA-03 WI-08 WV-01
The overlap here consists of 25 districts, as follows:
AR-01 AR-02 AZ-05 DE-AL FL-08 HI-01 IL-10 IN-08 KS-03 LA-02 MI-01
MI-07 MS-01 NH-01 NV-03 OH-16 PA-03 PA-07 PA-12 TN-06 TN-08 VA-02
WA-03 WI-08 WV-01
All of these are Democratic-held seats except DE-AL, HI-01, IL-10, and LA-02. For completeness, though,
Silver has only two districts that are tied (NC-08) and (TX-23), whereas our model has 17 districts that
are too close to call:
AL-02 CA-04 CA-44 ID-01 LA-04 MD-01 MN-06 NH-02 NY-20 NY-23 NY-24
NY-29 OH-01 OH-15 VA-05 WA-08 WI-07
This simply means our model is more cautious. When there is no clear evidence of how a seat will go, it
is rated a tossup rather than venturing a probably-not-very-good guess. However it is not wise to assume
each party will win half the tossup seats. In wave elections, as 2010 will probably be, one party wins most
of them, in this year probably the Republicans.
Note that these are just predictions from the model. They do not take the polling data into account.
For a table with the polling data, the predictions, and the 2008 election results
Models aren't everything (including ours) and it is important to run a sanity check on the output.
For example, Silver's model predicts a 51% chance that the Democrats will pick up LA-02. This is a majority
black district in New Orleans with a PVI of D+25. It is one of the most reliably Democratic districts in the country,
not having elected a Republican to Congress from 1891 to 2007, a span of over 100 years. In 2008, a Republican,
Ahn "Joseph" Cao won--because the incumbent, "Dollar Bill" Jefferson, was under 16 counts of indictment for
bribery and related crimes. He, as you may remember, was the owner of the freezer in which the FBI found $90,000
in cold hard cash. Despite this, Jefferson lost only by 2.7%. The Democratic candidate this year, is a black man,
Cedric Richmond, who represents part of the CD in the Louisiana House. He got 73% of the vote in 2007.
To say that Vietnamese Republican is an even match for a popular black state legislator in an overwhelmingly black
Democratic district defies all reason.
Another example, Silver says the chance of the Democrats taking back HI-01 is 54%. The current congressman,
Charles Djou, won a special election on May 22 with under 40% of the vote because there were two Democrats in
the race who split the Democratic vote evenly and Hawaii does not have runoffs. Prior to Djou, Democrats occupied
the seat for years. In November Djou will face only one Democrat, whoever wins the Sept. 18 primary. A more reasonable
prediction would say the odds on the Democrats taking back this seat are north of 80%.
Finally, Silver says that Democratic Rep. Stephanie Herseth Sandlin's goose is cooked. He predicts an 82% chance that
she will be beaten by Republican Kristi Noem. But just today (see below) we have a (Rasmussen) poll putting Herseth Sandlin
slightly in the lead. What no model can take into account is a factor like name recognition. In her case, not only
her own, but that of her grandfather who used to be governor. In a Republican year, she's not going to get the
68% she got in 2008, but to say she is a dead duck (to use another poultry analogy for this agricultural state) is
a bridge too far to our way of thinking.
New Definition of Pickup
Under the map is a list of pickups by state. The definition has been changed starting today.
The old definition of a pickup is when a party is predicted to win a seat that it lost in 2004.
The new definition is when a party is predicted to win a seat it currently does not have.
For example, if Pat Toomey (R) wins Pennsylvania, under the old definition, that would not be
a pickup because it was won by a Republican, Arlen Specter (R-PA), in 2004. Under the new
definition it would be a pickup because it is currently occupied by a Democrat, Arlen Specter (D-PA).
The new definition is probably more intuitive for most people.
Today's Polls: CT NC SD-AL
|| Richard Blumenthal
|| Linda McMahon
|| Sep 09
|| Sep 09
| North Carolina
|| Elaine Marshall
|| Richard Burr*
|| Sep 08
|| Sep 08
|| Stephanie Herseth Sandlin*
|| Kristi Noem
|| Sep 08
|| Sep 08
If you like this Website, tell your friends. You can also share by clicking this button
-- The Votemaster
Your donation is greatly appreciated. It will buy ads to publicize the site.