Stats - Bias

advertisement
Geology 204 – Measures of Variability
Announcements / science in the media
Eyes in Back of Head project grading is still underway.
XC: Posted! First in possibly a series.
Questions from before or readings?
Statistics quotes
He uses statistics as a drunken man uses lamp posts - for support rather than for illumination. ~Andrew Lang
Statistics can be made to prove anything - even the truth. ~Author Unknown
While the individual man is an insoluble puzzle, in the aggregate he becomes a mathematical certainty. You
can, for example, never foretell what any one man will be up to, but you can say with precision what an
average number will be up to. Individuals vary, but percentages remain constant. So says the statistician.
~Arthur Conan Doyle
98% of all statistics are made up. ~Author Unknown
Bias
Q:Think of a way to prove that you are above-average in this class, without
cheating (except statistically)
A:
Find the class average on some type of test
You take a similar test and keep taking them until you happen to do better than the
average
Why should we care?
Could you design a way to show that your drug was safe when it actually was not?
How does drug research work?
Drug companies fund scientific studies of their drugs.
The companies decide whether to publish the results of the studies or not.
Law & Order 10/6/04, “Coming Down Hard”
Polling & Bias
Check out http://www.pollingreport.com/social.htm
Note the style of questions
Note that in CNN poll many say “options rotated”. Why?
Note the style of the ABC News/WP poll questions (esp. #2)
Contrast with the style of the Fox News questions. What’s going on here?
Confidence in action:
NY Times article on SSRIs. Recently, FDA recommended to stop giving these
anti-depressants to kids, because of increased danger of suicide
British Study on Paxil: 20 out of 34 possible suicidal events for Paxil group
compared with 8 out of 17 for placebo group
Assume same definition was used for both
Is this a significant difference?
Let’s express the suicidal event incidence as a percent in each case, so we can
compare them better:
For the 20/34, the incidence is: 58.8%
For the 8/17, the incidence is: 47.1%
Margins of error:
We can find the margin of error from: http://www.raosoft.com/samplesize.html
For the 34 group, the margin of error is +/- 16.8%
This gives a 95% confidence range from 42% to 75.6%
For the 17 group, the margin of error is +/- 23.8%
This gives a 95% confidence range from 23.3% to 70.9%
Download