tuning numbers to make them sing

[adularia]

Put together a fairly good statistical analysis for a problem at work today; luckily, it came out in our favor, if I have not done something colossally stupid like set the wrong null hypothesis. I am under the impression that H0 should always assume no intervention, no change, the commonsense default assumption... but doubts creep in, when the no-

Comments

[craigp]

if it came out in your favor, then by definition you did it correctly. isn't that what they pay statisticians for?

*ducks*

[adularia]

I'm working on an extra-persnickety client these days. If I did it wrong (unlikely) and they notice (likely), it won't be fun. :/

[rocket_jockey]

Your null hypothesis can be anything that you wish to examine: it doesn't have to be the "I didn't touch it" condition.

When I have my library out of boxes again, I'll see if I can slide you a copy of one of my stats texts. The most advanced one I have is Statistics for Engineering and the Sciences (I think that's the title, at least) and is a pretty high-level text.