Lies and Statistics

• It's pretty easy to lie with stats.
• First, you can fudge the data.
• Second, you can do the test over and over again until you get the answer you want.
• So if p~=0.05, if you run it 20 times, you should be able to get the flip answer.
• Third, you can choose the wrong test.
• Obviously, this requires people to be honest and knowledgable.
• Consequently, good science usually requires reproducability.
• If I do it, great. If I do it, then 10 others reproduce it, then it's solid.