# SPSS ANOVA / Multiple Testing

#### DrK89

##### New Member
Dear all,

I have a database with measurements at t = 1 (baseline) and t = 2 (after treatment), and I want to assess whether the treatment causes a difference in the measurement. See the attached jpeg.file for a simplified hypothetical example: 15 participants, 5 are 80 years old, 5 are 60 years old, and 5 are 40 years old. They have scores on measurement 1 and 2; a paired t-test should be the best option (assuming normally distributed data, which is the case in the original database I am working on). The difference is not statistically significant. However, based on some literature, the treatment may be more or less effective depending on the age of the participants. So, if I select only those cases that are 80 years old (Data --> Select cases --> If condition is satisfied: age=80), and I repeat the T-test, I get a significant score, if I repeat the same with particpants that are 60 and 40 years old, the differences are not significant. So I am tempted to say that indeed the treatment works best in older people.

However, based on my limited statistical knowledge, I think this approach is not allowed because of 'multiple testing'. I cannot find online how I am supposed to do this in a statistically correct matter. I have tried a "one-way ANOVA" with measurements 1 and 2 in the dependent list and 'age' as a factor, but I think this is not correct, at least I cannot interpret the outcome. Can someone enlighten me? Let me know if I need to provide more information.

Thanks in advance!

Best
Martin

#### Attachments

• 129.2 KB Views: 2

#### katxt

##### Active Member
One straightforward approach would be to find the differences and do a one way anova on those. If there is no significant difference between the groups then that's the end. If there is a significant difference, ask for post hoc tests.
Age is an odd one. Do you think that there may be a linear age effect with the difference between 80 and 60 being the same as between 60 and 40? In which case perhaps a regression may be better.

#### DrK89

##### New Member
Hi katxt,

Thank you for your answer. Sorry, I forgot to reply earlier.

Yes this was of course the answer - stupid that I did not think of that myself.

Thanks again!
Best
Martin