# two-way ANOVA on actual glucose data

#### ivn

##### New Member
Hi!
I have some trouble with the interpretation of the post hoc results. For my thesis i have to comapare the effect of a new antidiabetic drug with metformin. This is a classical mixed design, where I have two randomly assigned groups of diabetic patients who will get metformin and a new antidiabetic drug. We compare the blood glucose level before and after six months of the treatment, and I have to see is there any difference between the drugs.

I have used a 2way repeated measure ANOVA and I get these results:

before/after - p=0.00000447254906
between the drugs - p= 0.0763001352
interactions - p=0.120083026

After I found a statistically significant result, I analyzed it further with a Bonferroni post hoc test and I found that both drugs lower the glucose level after six month of therapy:
metformin - 0.0355931195
drug2 - 0.00049297311

And my question is - how can I know, what drug is statistically more effective, when they both significantly lower the blood glucose level?

still using the Bonferroni post hoc test, comparing the base level of glucose, before the treatment, between the drugs, i get a p value = 0.146. but comparing the levels after 6 months, the Bonferroni gives me a p value = 1.

Is the way I am doing it correct? What about other post hoc tests? Is there a nonparametric 2way repeated measure anova and how would you do it if the data is not normally distributed?

Thanks a lot!

#### Karabiner

##### TS Contributor
And my question is - how can I know, what drug is statistically more effective, when they both significantly lower the blood glucose level?
The interaction is not significant. This means,
the pre-post-differences are not significantly
different between groups (or, the difference
between groups does not significantly differ
between time points). Therefore, there is no
evidence that one drug is more effective than
the other.

With kind regards

K.

#### hlsmith

##### Omega Contributor
Wouldn't hurt to graph your data, and if possible, add on confidence intervals.