November 25, 2014 2014 NFL Prediction Accuracy
Microsoft's Cortana
and Nate Silver's fivethirtyeight.com Elo
are predicting the outcomes of NFL games each week during the
2014 season. The
tables below use Brier
scores to determine the accuracy of the probabilistic
predictions from both models.
Brier scores range from 0 to 1, with 0 meaning the probabilities
in the models perfectly match
the outcomes of the games and 1 meaning no matches were made
(so the closer to 0.000, the better
calibration of the model). If a 50% (0.500) probability were
assigned to each team for each game, the
Brier score would be 0.250. Through the twelfth week of the
season, the Elo model has a Brier score of 0.207 and the Cortana
model has a Brier score of
0.213.

Brier
Scores 
2014 NFL Season Summary 
538 Elo 
Cortana 

Week 1 
0.212 
0.222 
Week 2 
0.256 
0.258 
Week 3 
0.187 
0.209 
Week 4 
0.242 
0.288 
Week 5 
0.149 
0.145 
Week 6 
0.194 
0.194 
Week 7 
0.211 
0.200 
Week 8 
0.217 
0.197 
Week 9 
0.199 
0.201 
Week 10 
0.189 
0.192 
Week 11 
0.260 
0.287 
Week 12 
0.167 
0.167 

Season totals 
0.207 
0.213 
For the season through
week 12,
the Elo model is performing about 17% better than assigning
all teams a 50% win probability
and the Cortana model is performing about 15% better than assigning
all teams a 50% win probability. The Brier scores for the season
reflect that the average win
probabilities among winning teams are 58% for the Elo model
and 57% for the Cortana model.
On a weekly basis, the models
have failed to perform better than assigning each team a 50%
win probability eleven out of twelve
weeks. In week twelve of the season,
the Elo model had a Brier score of 0.167 and the Cortana model
had a Brier score of 0.167. Details for each game are here.
November 21, 2014
2014 Christmas Spending
Shoppers around the country say they are
planning to spend an average of $861 for gifts this holiday
season, up from $801 last year according
to our 30th annual survey on holiday
spending. Planned gift spending
will slightly exceed spending in 2007, a first since the recession.
Year 
Average Spending 
Percent Change 

2014 
$861 
+
8% 
2013 
$801 
 6% 
2012 
$854 
+
32% 
2011 
$646 

2% 
2010 
$658 
+
58% 
2009 
$417 
 3% 
2008 
$431 

50% 
2007 
$859 

5% 
2006 
$907 

4% 
2005 
$942 

6% 
2004 
$1,004 
+
3% 
Average planned spending peaked in 2001 at $1,052.
Average planned spending has recovered from $417 in 2009, with
average planned spending for
2014 slightly above average planned spending of $859 in
2007.
When
average planned spending is adjusted for inflation beginning
with the first survey in 1985,
the average planned spending of $861 in 2014 would be worth
$390 in 1985.
Details from the nationwide survey conducted
November 1720 are here.
November 21, 2014 Obama Job Approvals and the National Economy
Obama Job Approval Ratings 
11/20/14 
Approve 
Disapprove 
Undecided 

Overall 
44% 
51% 
5% 
Economy 
42% 
54% 
4% 
A total of 44% of Americans say they approve of the way Barack
Obama is handling
his
job as president and 51% say they disapprove of the way Obama is handling his job.
In October, 43% of Americans approved
of the job Obama was doing and 52% disapproved. When it comes to Obama's handling
of the economy, 42% of Americans approve and 54% disapprove.
In October, 41% approved and 55% disapproved. Among Americans registered to
vote,
46% approve of the way Obama is handling his job
as president and 50% disapprove. On Obama's handling of the economy, 43% of registered
voters
approve and 53% disapprove. Details from the nationwide survey conducted November
1720 are available at The National
Economy.
November 13, 2014
Pollster Ratings 2014 Colorado US Senate Election
The following table lists the accuracy of pollsters in the November 4, 2014 Colorado
US Senate election
using the measure of polling accuracy proposed by Martin, Traugott, and Kennedy. An accuracy measure of 0.00 indicates that the odds ratio of the survey results matches
the odds ratio of the actual results (the closer to 0.00, the better).
2014 Colorado US Senate Election 

Gardner 
Udall 
Odds Ratio 
Accuracy 

Ipsos/Reuters 
47% 
45% 
1.00 
0.00 
Quinnipiac 
45% 
43% 
1.00 
0.00 
SurveyUSA 
46% 
44% 
1.00 
0.00 
YouGov 
43% 
42% 
0.98 
0.02 
PPP 
48% 
45% 
1.02 
0.02 
Marist/NBC 
46% 
45% 
0.98 
0.02 
Vox Populi 
46% 
43% 
1.02 
0.02 
Monmouth 
47% 
46% 
0.98 
0.02 
ORC/CNN 
50% 
46% 
1.04 
0.04 
Gravis 
48% 
43% 
1.07 
0.06 
YouGov/CBS/NYT* 
46% 
47% 
0.93 
0.07 
Harstad 
43% 
44% 
0.93 
0.07 
Keating* 
44% 
45% 
0.93 
0.07 
Strategies 360* 
44% 
45% 
0.93 
0.07 
Rasmussen 
51% 
45% 
1.08 
0.08 
Mellman* 
37% 
39% 
0.91 
0.10 
Benenson* 
44% 
47% 
0.89 
0.11 
Suffolk 
46% 
39% 
1.13 
0.12 

Actual 
48.32% 
46.16% 


There are no differences between all 18
polls at the 95% confidence level. *The YouGov/CBS/NYT, Keating, Strategies 360,
Mellman, and Benenson polls had Udall
leading.
November 13, 2014
Pollster Ratings 2014 Colorado Gubernatorial Election
The following table lists the accuracy of pollsters in the November 4, 2014 Colorado
gubernatorial
election
using the measure of polling accuracy proposed by Martin, Traugott, and Kennedy. An accuracy measure of 0.00 indicates that the odds ratio of the survey results matches
the odds ratio of the actual results (the closer to 0.00, the better).
2014 Colorado Gubernatorial Election 

Hickenlooper 
Beauprez 
Odds Ratio 
Accuracy 

Strategies 360 
46% 
43% 
1.00 
0.00 
YouGov/CBS/NYT 
48% 
44% 
1.02 
0.02 
YouGov 
44% 
42% 
0.98 
0.02 
Vox Populi 
49% 
44% 
1.04 
0.04 
ORC/CNN 
49% 
48% 
0.96 
0.05 
Marist/NBC 
46% 
41% 
1.05 
0.05 
Ipsos/Reuters 
46% 
46% 
0.94 
0.07 
PPP 
46% 
46% 
0.94 
0.07 
SurveyUSA 
46% 
46% 
0.94 
0.07 
PPP/LCV 
47% 
47% 
0.94 
0.07 
Monmouth 
50% 
43% 
1.09 
0.08 
Rasmussen* 
47% 
49% 
0.90 
0.11 
Quinnipiac* 
43% 
45% 
0.89 
0.11 
Suffolk* 
43% 
45% 
0.89 
0.11 
Gravis* 
44% 
48% 
0.86 
0.15 

Actual 
49.21% 
46.07% 


There are no differences between all 18
polls at the 95% confidence level. *The Rasmussen, Quinnipiac, Suffolk, and Gravis
polls had Beauprez
leading.
Novmeber 13, 2014
Pollster Ratings 2014 Alaska US Senate Election
The following table lists the accuracy of pollsters in the November 4, 2014 Alaska
US Senate election
using the measure of polling accuracy proposed by Martin, Traugott, and Kennedy. An accuracy measure of 0.00 indicates that the odds ratio of the survey results matches
the odds ratio of the actual results (the closer to 0.00, the better).
2014 Alsaka US Senate Election 

Sullivan 
Begich 
Odds Ratio 
Accuracy 

YouGov/CBS/NYT 
48% 
44% 
1.02 
0.02 
Moore 
42% 
38% 
1.03 
0.03 
Rasmussen 
47% 
42% 
1.05 
0.04 
PPP 
46% 
45% 
0.95 
0.05 
Harstad 
44% 
44% 
0.93 
0.07 
Hellenthal** 
39% 
49% 
0.74 
0.30* 

Actual 
48.58% 
45.37% 


There are no differences between all 6
polls at the 95% confidence level. *The accuracy measure for the Hellenthal poll
falls outside of the 95% confidence interval. **The Hellenthal poll had Begich
leading.
November 13, 2014
Pollster Ratings 2014 Alaska Gubernatorial Election
The following table lists the accuracy of pollsters in the November 4, 2014 Alaska gubernatorial
election
using the measure of polling accuracy proposed by Martin, Traugott, and Kennedy. An accuracy measure of 0.00 indicates that the odds ratio of the survey results matches
the odds ratio of the actual results (the closer to 0.00, the better).
2014 Alsaka Gubernatorial Election 

Walker 
Parnell 
Odds Ratio 
Accuracy 

PPP 
46% 
45% 
0.99 
0.01 
Hellenthal* 
43% 
44% 
0.94 
0.06 
YouGov/CBS/NYT* 
39% 
42% 
0.90 
0.11 
Rasmussen 
50% 
43% 
1.12 
0.12 

Actual 
47.94% 
46.30% 


There are no differences between all 4
polls at the 95% confidence level. *The Hellenthal and YouGov/CBS/NYT polls had
Parnell
leading.
November 7, 2014
Pollster Ratings 2014 Iowa US Senate Election The following table lists the accuracy of pollsters in the November 4, 2014 Iowa
US Senate election
using the measure of polling accuracy proposed by Martin, Traugott, and Kennedy. An accuracy measure of 0.00 indicates that the odds ratio of the survey results matches
the odds ratio of the actual results (the closer to 0.00, the better).
2014 Iowa US Senate Election 

Ernst 
Braley 
Odds Ratio 
Accuracy 

Selzer/DMR 
51% 
44% 
0.97 
0.03 
Suffolk 
47% 
43% 
0.92 
0.09 
PPP 
48% 
45% 
0.89 
0.11 
Marist/NBC 
49% 
46% 
0.89 
0.11 
ORC/CNN 
49% 
47% 
0.87 
0.14 
Optimus 
45% 
44% 
0.86 
0.16* 
Rasmussen 
48% 
47% 
0.85 
0.16* 
ARS/Fox News 
45% 
44% 
0.86 
0.16* 
Monmouth 
47% 
46% 
0.86 
0.16 
YouGov/CBS/NYT 
44% 
44% 
0.84 
0.18* 
Ipsos/Reuters 
45% 
45% 
0.84 
0.18* 
Quinnipiac 
47% 
47% 
0.84 
0.18* 
GarinHartYang 
47% 
47% 
0.84 
0.18* 
Loras College** 
44% 
45% 
0.82 
0.20* 
YouGov** 
42% 
43% 
0.82 
0.20* 
PPP/LCV** 
47% 
48% 
0.82 
0.20* 

Actual 
52.2% 
43.7% 


There are no differences in accuracy between the Selzer, Suffolk, PPP, Marist/NBC, ORC/CNN, and
Monmouth
polls at the 95% confidence level. *The accuracy measures for the Optimus, Rasmussen,
Anderson Robbins/Shaw/Fox News, YouGov/CBS/New York Times,
Ipsos/Reuters, Quinnipiac, GarinHartYang,
Loras College, YouGov, and Public Policy Polling/League of
Conservation Voters polls fall outside
of the 95% confidence intervals. **The Loras College, YouGov, and the PPP/LCV polls had Braley
leading.
November 7, 2014
Pollster Ratings 2014 Iowa Gubernatorial Election The following table lists the accuracy of pollsters in the November 4, 2014 Iowa gubernatorial
election
using the measure of polling accuracy proposed by Martin, Traugott, and Kennedy. An accuracy measure of 0.00 indicates that the odds ratio of the survey results matches
the odds ratio of the actual results (the closer to 0.00, the better).
2014 Iowa Gubernatorial Election 

Branstad 
Hatch 
Odds Ratio 
Accuracy 

Monmouth 
58% 
37% 
0.99 
0.01 
Loras College 
55% 
34% 
1.02 
0.02 
Marist/NBC 
59% 
36% 
1.03 
0.03 
Rasmussen 
55% 
37% 
0.94 
0.06 
Selzer/DMR 
59% 
35% 
1.06 
0.06 
ARS/Fox News 
53% 
36% 
0.93 
0.07 
Suffolk 
54% 
37% 
0.92 
0.08 
YouGov 
49% 
35% 
0.88 
0.12 
YouGov/CBS/NYT 
50% 
37% 
0.85 
0.16* 
PPP 
51% 
40% 
0.80 
0.22* 
Quinnipiac 
52% 
41% 
0.80 
0.22* 

Actual 
59.0% 
37.2% 


There are no differences in accuracy between the Monmouth, Loras College, Marist/NBC, Rasmussen
Selzer/DMR,
ARS/Fox News, Suffolk, and YouGov
polls at the 95% confidence levels. *The accuracy measures for the YouGov/CBS/NYT,
PPP, and Quinnipiac polls fall outside
of the 95% confidence intervals.
Why doesn't most advertising work? Take a look at 10 Rules for More Effective Advertising.
