News for hackers. Stuff that matters.
Re:Statistics Done Wrong – The woefully complete guide by Paradigma11 on Tuesday November 05, 2013 @03:04PM (#6677810)
Quite a few years ago i devised an ambitious method to achieve significance while sitting through another braindead thesis presentation (psychology):

If you are interested in the difference of a metric scaled quantity between two groups do the following:

1.) Add 4-5 plausible control variables that you do not document in advance (questionaire, sex, age...).

2.) Write a r-script that helps you do the following: Whenever you have tested a person increment your dataset with the persons result and run a:

t-test

u-test

ordinal logistic regression over some possible bucket combinations.

3.) Do this over all permutations of the control variables. Have the script ring a loud bell when significance is achieved so data collection is stopped immediately. An added bonus is that you will likely get a significant result with a small n which enables you to do a reversed power analysis.

Now you can report that your theoretical research implied a strong effect size so you choose an appropriate small n which, as expected, yielded a significant result ;)

Re:Statistics Done Wrong – The woefully complete guide by chrislipa on Tuesday November 05, 2013 @10:46PM (#6680509)
XKCD did it first:

http://xkcd.com/882/