I don't mean to introduce politics, but here is an amazing public
opinion poll anomaly, from the Clinton-Obama race:
http://www.gallup.com/poll/104530/Gallup-Daily-Tracking-Election-2008.aspx
The results have remained in a statistical dead heat within 2%,
consistent and unchanged, for 5 days during one of the most hotly
contested and closely watched races in the last 40 years. Public
opinion is rarely this stable.
The people at Gallup are experts at polling. Their results are
usually reliable and repeatable, although that does not mean they
actually predict the outcome of the next vote in this particular
race. That sounds kind of stupid I realize, but I mean that this poll
is probably a good reflection of general public opinion, and if you
doubled the number of respondents the numbers would not change much,
but in a volatile race of this nature, you cannot predict which
segments of the public will vote. For example, this year young people
are voting in larger numbers than usual. Plus many states allow
people to vote for a week or two in advance of the election, so the
results reflect samples at different times during the run-up to the election.
People have criticized public opinion polls but a sophisticated
reader will understand that they have their limitations and results
must be interpreted. Polls resemble instrument readings from complex
instruments such as silicon detectors, where not all control
parameters are under control. You have to take into account
unavoidable sources of bias and noise, and the fact that all public
opinion polls are backward looking, "lagging indicators." Sort of
like running a calorimeter with 24-hour latency.
Professional pollsters are embarrassed about this year's
inaccuracies. John Zogby, who is a smart cookie, was asked about his
incorrect prediction of the California race. His comment: "Let the
record show I'm sitting in the corner in a fetal position right now.
It was not my finest hour and it was embarrassing."
http://seattlepi.nwsource.com/opinion/352352_diazonline24.html
Understanding limitations and interpreting results is also crucial
when you read about cold fusion experiments, or social science
research, or other "fuzzy" research into subjects that are largely unknown.
- Jed
