Analyzing the data from a randomized crossover study (pre-test and post-test) with a placebo and two interventions

#1
Hi all, I'm making my first steps in statistical analysis and hopefully somebody more experienced will be able to help get me started :)

Firstly, I have a randomized crossover study to analyse and my first question is does the nature of a crossover study (n= 10 or n=30 depending on how you look at it) mean I should do anything different form a normal RCT?

Secondly, the only tutorial I have been given told me to analyse normality for the whole column (e.g. pre-test bodyweight. However a textbook I have (by Andy Field) says you should split the data into groups then test normality. Which way is correct?

Thirdly, if my pre-test data is normally distributed but the corresponding post-test column is non-normal, when I want to test repeated measures, should I transform the data, use a non-parametric test, or do something else?

fourthly, my text book 'the spss survival guide' says ANOVA is robust enough to mean you can ignore violations of normality if the sample size is over 30. my sample size is exactly 30 (though it is a crossover with 10 subjects undergoing all 3 conditions so 10 X 3 = 30). should i just ignore that i have non-normal data and use a mixed between-within ANOVA to test for interaction effects?

Please help!!!!!o_O
 

fed2

Active Member
#2
spss is for social "scientists". use SAS like a normal person.

proc ttest has a crossover option, its almost fool-proof. almost.
 

Karabiner

TS Contributor
#4
my sample size is exactly 30 (though it is a crossover with 10 subjects undergoing all 3 conditions so 10 X 3 = 30).
Your sample size is n=10, with repeated measures. But the description is a bit confusing
(at least for me). You have a 1 pre-post test, but there were 3 conditions? Could you tell
us your precise research question, and could you describe your experimental design in
more detail?

With kind regards

Karabiner
 
#5
thank you for engaging with me Karabiner,

my hypothesis is that a when compared with placebo a diet supplement (beetroot juice) will improve performance in a number of facets of rugby-related fitness. I further hypothesize that 6 weeks of the diet supplement will improve fitness more than 2 weeks of the diet supplement.

Therefore, I had 10 subjects undergo the placebo, 2 weeks of supplementation and 6 weeks of supplementation with a pre-test and post-test for each condition.

I hope this helps,

Regards

ProfessorP
 

Karabiner

TS Contributor
#6
So normally you would perform a repeated-measures analysis of variance with the 2 factors
"type of intervention" (3 levels) and "time of measurement" (pre and post). n=10 might
indeed constitute a problem with regard to assumptions. Alternatively, you could calculate
the 3 pre-post difference scores, and use them as dependent variable in a Friedman test.

With kind regards

Karabiner
 
#9
So normally you would perform a repeated-measures analysis of variance with the 2 factors
"type of intervention" (3 levels) and "time of measurement" (pre and post). n=10 might
indeed constitute a problem with regard to assumptions. Alternatively, you could calculate
the 3 pre-post difference scores, and use them as dependent variable in a Friedman test.

With kind regards

Karabiner
Thank you Karabiner

do you have any advice for my second question -

"Secondly, the only tutorial I have been given told me to analyse normality for the whole column (e.g. pre-test bodyweight. However a textbook I have (by Andy Field) says you should split the data into groups then test normality. Which way is correct?"


Thank you again,

Regards,

ProfessorP
 

Karabiner

TS Contributor
#11
a textbook I have (by Andy Field) says you should split the data into groups then test normality. Which way is correct?"
You only have 1 group. You can theoretically assess the distribution at each of the 6 time points.

With kind regards

Karabiner