Unbiasedness

1
Unbiasedness

• Methods and Statistics in Political Science

2
Introduction

• What is Unbiasedness?
• Types of Bias
• Examples of Bias

3
What is Unbiasedness

• Specific term, not just a preference.
• Unbiased research is correct on average
• Involves using a procedure that will not
  systematically tilt the outcome in one direction
  or another.
• Application of the same procedure across many
  data sets will provide an average that is correct
  even if individual data sets are incorrect.

4
Types of Bias

• Selection Bias
• Most well known type of bias.
• Involves collecting data from a group whose
  responses will not be representative of the
  average.
• Can be intentional, in the case of people trying
  to skew the response to an issue.
• Can also be unintentional.

5
Bias Present Not Only in Data Collection

• Theories can bias results as well.
• If one picks a theory, then narrows the number of
  cases to only those which support his theory, the
  research is biased and suspect.
• Focusing too hard on the theory may lead one to
  consider less than all the data, which can bias
  the reported inferences.

6
Examples of Bias

• Intentional Selection Bias
• Unintentional Selection Bias
• Event Count Bias
• Perception Bias

7
Intentional Selection Bias

• If someone were polling to guage support for a
  new waste dump site to supplement the overly
  filled sites already in existence, and they
  collected responses in the neighborhood of the
  proposed site, their research would be biased.
• One would expect that support for the new dump
  site would be low in the neighborhood in which it
  would be placed.
• However, a random sampling of the same number of
  respondents from the White Pages for example,
  would probably yield a very different level of
  support.

8
Unintentional Selection Bias

• Literary Digest Poll for President FDR vs.
  Landon
• Poll taken based on a mailing list comprised of
  the phone books and vehicle registrations across
  the country. LANDON WINS!
• NO, FDR won by a landslide. During the Great
  Depression an overwhelming majority of those who
  owned telephones and cars were Republicans.
  20/20 Hindsight from the pollsters at the
  Literary Digest.

9
Event Count Bias

• A study by Gary King of Harvard University and
  published in the American Journal of Political
  Science found that
• Event count analyses are flawed by their nature.
• Event count analysis involves simply tallying the
  number of times something happens, and then
  correlating the data with other data sets.
• Example If one were to analyze the number of
  bills proposed by members of the House of
  Representatives and compare that to the publicity
  received by the member, one would end up with a
  biased result.
• Members who had more publicity to begin with
  would probably receive publicity at a higher than
  linear rate compared to a member who had
  significantly less publicity to begin with.

10
Perception Bias

• When polling groups, one should know about
  predispositions within that group.
• Pami Dua and David Smyth researched economic
  polls and reported in Public Opinion Quarterly
  that
• Over 21 years, the public is neither overly
  optimistic or pessimistic about inflation levels.
• Over that same period, the public was
  considerably more pessimistic about unemployment
  rates.

11
Conclusion

• Unbiasedness is an important goal in political
  science research.
• Bias is not always intentional, so look carefully
  at the data you collect to see if you have
  unintentionally biased the collection.
• Make sure when you limit your cases that you
  arent just excluding data that disagrees with
  your theory.

12
Sources Cited

• Dua, Pami Smyth, David J. Survey Evidence on
  Excessive Public Pessimism About the Future
  Behavior of Unemployment. Public Opinion
  Quarterly, Vol. 57, No. 4. (Winter, 1993), pp.
  566-574. Available at http//links.jstor.org/sici
  ?sici0033-362X2819932429573A43C5663ASEOEPP3
  E2.0.CO3B2-J
• King, Gary Keohane, Robert O. Verba, Sidney.
  Designing Social Inquiry.
• King, Gary. Statistical Models for Political
  Science Event Counts Bias in Conventional
  Procedures and Evidence for the Exponential
  Poisson Regression Model. American Journal of
  Political Science, Vol. 32, No. 3. (Aug., 1988),
  pp. 838-863. Available at http//links.jstor.org/
  sici?sici0092-58532819880829323A33C8383ASMFP
  SE3E2.0.CO3B2-3
• TWO STORIES OF STATISTICAL BIAS
  members.fortunecity.com/jonhays/stories.htm