T test problem

#1
Dear all,

Doing a t-test I got this results :

200 400
Mean 944.58333 960
Variance 402.08333 222.727273
Observations 12 12
Pooled Variance 312.4053
Hypothesized Mean Difference 0
df 22
t Stat -2.13652
P(T<=t) one-tail 0.0220045
t Critical one-tail 1.7171444
P(T<=t) two-tail 0.0440091
t Critical two-tail 2.0738731

According with my notes if t Stat < -t Critical two-tail or t Stat > t Critical two-tail, we reject the null hypothesis. This is the case, means are unequal. But statgraphics tells me that means are equal. Someone could help me with this?

Thank you for your support.
 

Dason

Ambassador to the humans
#4
Oh so you have multiple comparisons going on. Have you read up on multiples comparisons corrections?
 
#5
I am working with the first outpost and doing a "re-check" just in case with the other analysis. This is the first time a I had a disagreement.

No, I didn't read about it. But I will look into it.

Thanks
 

hlsmith

Omega Contributor
#6
When you make multiple tests you need to correct for false discovery. You seem to be making familywise comparisons and your software is correcting for this. You need to look up "Tukey HSD" correction to better understand the approach, which deals with adjusting your level of significance needed to declare an effect. An example may be if I give you drug A and then compare drug A to drugs B-Z, eventually I am going to find a spurious association by chance, given the shear number of comparisons. So in order to maintain the specified a priori type I error rate I need to correct the significance level used.