One-way anova and two tailed t-test question

#1
Hello,

I have 5 datasets, the first dataset is a baseline measurement and the other 4 datasets are post-treatment measurements at different time points . I graphed the mean and standard deviation (error bars) and I see that there is a decreasing trend in the data.

The one-way anova showed no significant difference when comparing the 5 datasets. However, I performed a two-tailed t-test to compare the baseline (dataset 1) with each post-treatment datasets ( datasets: 2,3,4 & 5) and I saw a significant difference between some datasets.

Why is that? I thought that the one-way anova results will be the same with the two-tailed t-test.

Thank you in advance for your input :)
 

Karabiner

TS Contributor
#2
I guess by oneway ANOVA you mean repeated-measures ANOVA with 5 levels, and t-test is t-test for dependent groups?

How large was your sample size? And are there missing values during follow-up? If yes, how many?

What do you mean by "significant"/"not significant"? A difference between 0.499 and 0.500? Between
0.00001 and 0.27?

With kind regards

Karabiner
 

Karabiner

TS Contributor
#4
Well, I don't see a decreasing trend, it looks more U-shaped. For the t-tests,
you picked out those 4 comparisons with the largest differences. A p=0.006
with only n=7 is a bit surprising, though.
In the ANOVA, a global test is performed which takes into account that several
levels are included simultanously. So one could consider it as a means against
producing false-positive results.

I honestly do not quite know what to make of all this. The power is low, so a
non-significant result is hardly interpretable. Both t-tests and ANOVA seem
inadequate, because it cannot be shown that their assumptions are fulfilled.
You used one analysis, and after it turned out non-significant you used another
one which produced a statistical significance at one point.

With kind regards

Karabiner