Compared to what

1
Compared to what?

• Jeanine Meyer
• Mathematics/Computer Science New Media
• Natural Sciences 3003

2
4 D's

• Definition
• Denominator (out of what), definition of
  denominator, absolute value
• Difference (compared to what, what's the
  difference)
• Distribution

3
Topics relating to voting

• Youth voting
• Claims for big increase for primaries for 2008
• Colbert Bump
• Claims that appearing on the Comedy show gives
  candidates 'a bump'

4
Comment

• Predictions about voting done in terms of
  demographics
• Stereotypes you are a certain gender, race,
  ethnicity, age and so you will vote .
• Unfortunately, some truth to this but, not
  completely.
• Microtargeting
• http//www.nytimes.com/2008/04/16/dining/16voters.
  html What's for Dinner? The Pollster Wants to
  Know
• Look at You Might Be
• Keep in mind that the consultants may need
  something to say

5
Categories / Populations

• Assignment to categories may be problematic
• Life is more complex
• Example news story from August, 2007Women in
  their 20s out-earning Men in NYC (median 117)
• Example news reports (predictions) on voting in
  places like Florida

6
About women vs men

• Two populations
• 53 of these women were college graduates
  compared to 38 of men!!
• In each category, men still out-earning women
• SIGH
• Butcollege educated women outearning men without
  college degrees AND there are more of them.

7
Florida

• Very complex situation
• Older native residents versus 'snow birds'
• Cubans versus other Latinos
• North west (panhandle) versus specific industries
  (space, Disneyworld,etc.) mid-state versus Miami
  versus
• Western part of State is in different time zone.
  Polls close later!

8
US Time zones
9
How is something actually measured?
Not bad

• Exit polls
• Ask people leaving the polls (so it is pretty
  definite they did vote) how they voted
• Ask or guess demographics, issues
• Extrapolate from voting district
• Note after the results are in, people tend to
  say they voted with the winner.
• Federal Election Commission data
• Legally required and available (200 and up?)

Better
10
Polling

• Pick a random sample of people (in the category
  to be studied)
• Any person in the population equally likely to be
  asked
• Ask them the question
• Can make prediction about general population,
  within
• Margin of error
• Confidence level

11
Mathematics

• The distribution of the fractions of samples is
  more tightly distributed than the views of the
  general population.
• Still.it is possible to get a outlier (extreme)
  sample.
• Typical statement is we are 95 confident that
  the actual result (say percentage of people
  saying they will vote X) is within the margin of
  error (this is a number they calculate, say 3-4
  points) of our finding P.

12
Consider

• Flipping a fair coin many times
• 100.
• Fraction of heads will be close to half
• 50
• But you wouldn't be surprised if it was slightly
  off
• 47, 48, 49, or 51, 52,
• You would be surprised if it was way off
• 80, 90??
• But this could happen!
• Statistical analysis gives the margin of error
  and the chances of it happening

13
Note for PA primary

• The polling results may be within margin of
  error.
• The pollsters also should make assessment of who
  will actually vote.
• Most people don't vote, but many will not tell
  that to the pollster.
• Many new registrations / switches
• Most polls have been accurate, if the full
  information is given. The news media sometimes
  mis-states information before and afterwards.

14
Youth voting

• Youth?
• Typically, 18 to 30, but sometimes 17 to 24, 30
• Historically, very low levels of voting
• NOTE
• historically, voting participation in USA is low
  across the age distribution compared to other
  places
• Primaries much worse than general elections
• Off years much worse than presidential years

15
Gold standard of testing

• Double-blind
• Randomized
• Exact measurements, classifications
• Appropriate, adequate time intervals
• Sometimes not possible!

16
What is the denominator

• Measure by dividing actual voting
• Registered voters
• Voter eligible population (this would not
  include people in jail, felons, illegal
  immigrants, people on the move)
• Voter age population
• http//elections.gmu.edu/Voter_Turnout_2008_Primar
  ies.htm
• Comparison to previous years?

17
Youth voting

• In place of (youth voters) divided by (total
  youth population) or something similar
• Youth vote versus total voting
• Gives comparison (what's the difference) with
  2004)
• http//pewresearch.org/pubs/730/young-voters

18
Interpret results

• Chart does not say anything about turnout.
• These results COULD come about because older
  folks stayed home.
• In fact, 2008 primaries had record turnouts.
• But youth increase was even more..

19
Colbert Bump

• James Fowler, UCSD, The Colbert Bump in Campaign
  Donations More Truthful than Truthy,
• http//jhfowler.ucsd.edu/colbert_bump.pdf
• Appeared to be present for John Hall, successful
  challenger for NY 19.
• Colbert experience was very enjoyable for
  candidate, staff and volunteers
• How to evaluate?
• Can't arrange a double-blind test or cloning the
  candidate and comparing

20
Selection bias

• Maybe candidates who agree to go on the show are
  better candidates
• Maybe candidates that are invited to go on the
  show are better candidates
• Does Colbert want to boost specific candidates?
• Does Colbert want a good show?

21
Fowler approach

• Systematic procedure for identifying a match for
  each person that did appear on the Colbert show
• Party (note more Democrats than Republicans)
• Incumbency compare incumbents with incumbents,
  and challengers with challengers
• Same or similar amount of funds raised at time of
  appearance

22
Fowler approach continued

• Compare average number of donations and dollar
  amounts for time periods before and afterwards
• Note Use FEC data
• Absolutely and also as
• Use statistical tests to evaluate significance of
  results
• This does not remove effects of selection bias,
  but indicates if there is a difference
• Significant difference is one that is unlikely
  (say, less than 1/20 of arising from chance)

23
Results

• Democrats yes!
• Note Democrats who chose to participate
  generally were not doing as well
• Advantage was at significant level
• Republicans no!
• Appears to be opposite effect, but not
  statistically significant
• Note Republicans who chose to participate
  generally were doing well

24
(No Transcript)
25
Recommend

• Read papervery clear, and funny
• Problem required (requires) inventive analysis

26
Predictions?

• Tomorrow in Pennsylvania
• Results vs
• Expectations
• "Buy on the rumor, sell on the news."
• Nomination
• Election

27
Summary

• Definition(s)
• Denominator
• What's the Difference
• Compared to what
• Distribution