New to this - anova test

[Line Andersen]

I am very new to stata so bear with me..
I have conducted an experiment where we measure a hormone in the blood and because of large interpersonal variation, I have defined a new variable, “Change”, which is the difference in hormone concentration in a given person between 2 treatments. I total I have 3 treatments (control, diet 1, diet 2). Some patients received the control and diet 1, others control and diet 2, or diet 1 and diet 2.
Now I want to see if there is significant difference between the different diets, using my variable “Change”. A statistician recommended to find the 2-sided p-value for pair equal to 0, using ANOVA.

Can anyone help with a command for that? I want to know the p-value for each group of “double-treatments”, e.g. control and diet 1, control and diet 2, and diet 1 and diet 2.

I tried:
oneway Difference Difname, tabulate bonferroni
But it doesn’t seem right.
Anyone who can help?

[Phil Bromiley]

Try anova.

[Line Andersen]

Phil Bromiley<-- Really, that was your help? I told you I am new to stata, and that was what you gave me.. Gee, thanks a lot!

[Carlo Lazzaro]

Line:
despite being symphatetic with your need for a fully helpful reply, please consider that no lister is in any way obliged to reply (neither to be interested in how many days/months/years elapsed from the first time the original poster opened Stata from the first time).
Please also note that we are all busy persons (like you) and sometimes, due to time constraints or multitasking, reply with a general advice; it's up to the original poster to investigate whether the provided advice is good or not.
Please also note that you do not provide (via -dataex-) any excerpt/example of your data that allow interested listers to give a more positive reply.
At the top of that, it's good tradition of the list to keep friendly, relaxed and sometimes funny relationships among all members, even if they do not know each other personally and kindness (even if a lister did not give the advice that eases our methodological problem) is highly appreciated.
That said, hoping that you can keep on enjoying this forum, you may want to take a look at the Example 15: Repeated-measures ANOVA, -anova- entry, Stata .pdf manual.
Thanks for your time.

[Joseph Coveney]

I'll second Carlo's suggestions, and recommend further that you look further into your crossover study's design, for example, what the basis of randomization was, and whether the sequence (control diet first and diet 1 second versus diet 1 first and then control diet etc.) of diets was likewise part of the randomization scheme. If there were steps undertaken to identify sequence / carryover effects (you mention consulting a statistician at some point; maybe at the design stage?), then your ANOVA model should take those into account. To get a flavor of what I'm referring to, take a look at the types of models that are fitted with Stata's pharmacokinetics suite: at Stata's command line, type

help pk

for more information.

In addition, with hormone levels in blood, there is the real possibility that the values you measure, even after change scores are computed, are positively skewed. You might want to take a look at Stata's generalized linear models (GLM) commands for an approach to accommodate such characteristics in your data.