In article <[EMAIL PROTECTED]>, Brian Sandle <[EMAIL PROTECTED]> wrote:
>I need some help with statistical education. >There is always the warning against doing lots of correlations and taking the >ones you think you like. If there is 5% chance of any particular result >happening by chance then the more you take the more likely your result is by >chance. >So you are told to think out a theory and test it by stats. >I feel that that is now rather hit and miss. Your theory and the stats still >have a 5% chance of being wrong. What is needed is an analysis which deals >with that. That usually means wait until 20 more people have repeated your >work, doesn't it? Your theory will have a 100% chance of being wrong; you do not understand hypothesis testing. The 5% chance of rejection is if the null hypothesis is EXACTLY correct, and this is in practice never the case. It may be close to correct, but this is a different matter. When looked at from a logical perspective, all that data can do is to help choose from a set of theories. In the cases which occurred in simple physics, the theories are so simple that "everyone" thought of them, and so they may have appeared to have come from the data, but htis is not so. >My approach would be to look for all correlations with knowledge that 5% are >by chance, to acknowledge that but to try to build a pattern of what is >happening using the partial correlations to help, and to decide on further >directions, not to wait the years for people to check the one correlation you >have done. >I would tend to develop reasoning after seeing a pattern. This is called "spurious correlation". Statistics cannot do your thinking for you. -- This address is for information only. I do not claim that these views are those of the Statistics Department or of Purdue University. Herman Rubin, Department of Statistics, Purdue University [EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
