need help asap plz!

#1
Hi!
I'm a psyc university student, and need help with this stats question that's due tomorrow morning! I'm behind with my reading so I have no clue how to answer it! PLEASE HELP!!! :(

"If we had more than J=2 groups and we used a t-test to compare the largest and smallest means, will the actual probability of a Type I Error be larger or smaller than the (nominal) level we choose?"

THANK YOU!!!
 
#2
ok, so I know that the probability increases if you do multiple t-tests, but this is only asking for the comparison of the larges and smallest means, which would bring it down to two again. Does that mean that the probability will then stay the same?
 

mp83

TS Contributor
#3
Not sure , but I think thet it's not like you say. You don't know the ordering exactly, in fact you estimate it, so it's a multiple comparison still, as it underlies an estimation procedure to order the different means of the groups

I mean , you're not sure thet mu1 > mu2 > mu3 so you can t-test mu1,mu3and get an 100*(1-a)% test
 
#4
I think the problem is that by doing a t-test, you are assuming that you can compare other groups within that range without fear of increased type 1 error.

I know that when comparing such groups, researchers almost always do an ANOVA with some post-hoc test. As such, the t-test would not be the best idea.