F test significant but simple contrasts not

[brett_a_baxter]

Hi All,

I have conducted a repeated measures ANOVA, examining one of my dependent variables over 5 time points. I have a significant main effect from my F test but not when i am examining my simple contrasts (see images below). Could anyone offer any advice on why this is and what should be done?

Thank you in advance,

Brett

 

[hlsmith]

Not an ANOVA person, but you have the 3 v 1 pretty close. perhaps that is the one triggering the F-test. But I don't recall all of those 'sources'?? Also, a correction for pairwise comparisons could be getting applied in the second table.

 

[Miner]

That is one possibility. A second possibility is that there is another, stronger contrast that is driving the main effect.

 

[hlsmith]

Do you mean something like 2 v 3?

 

[Miner]

Yes. Assuming t0 = 1, t1 = 2, ..., Brett is testing the logical contrasts. However, it is possible that the change over time was not linear, an unaccounted covariate affected the measurements for some time periods, etc. resulting in a different contrast having a larger impact.

 

[hlsmith]

Or something like a j-distribution (0-1.5 servings of wine possibly beneficial to cardiovascular health; 50 servings you are dead)!

 

[brett_a_baxter]

Or something like a j-distribution (0-1.5 servings of wine possibly beneficial to cardiovascular health; 50 servings you are dead)!

Yes. Assuming t0 = 1, t1 = 2, ..., Brett is testing the logical contrasts. However, it is possible that the change over time was not linear, an unaccounted covariate affected the measurements for some time periods, etc. resulting in a different contrast having a larger impact.

Thank you both for your input. I conducted post-hoc analyses just to see if there was a difference elsewhere (i am not interested in this) but it did flag a difference between level 2 and level 3 so I think that may be why. Excuse my naivety here but I am wondering why there is an F statistic for a comparison between 2 levels as this appears to be a pairwise comparison?

 

[Miner]

An ANOVA is an omnibus test. It does not actually test all possible contrasts, but effectively tests the largest effect. You only know that one of the contrasts is significant. That is why you need Post Hoc tests to determine which contrast(s) are the cause.