Questions concerning the analysis of my Data

[Kalu]

Hello everyone,


I received some old data that I am now supposed to analyze, however, deciding for an appropriate statistical test turned out to be a little hard for me, so I could Need your help!

The data are basically from a 2x2 factorial design with Pre- and Posttest measures. It is an Intervention study (intervention vs placebo) that was conducted in a group of patients as well as in a group of healthy controls, so in theory there are 4 different groups (patient+intervention, patient+placebo, control+intervention, control+placebo). Moreover, I have results from various questionnaires and neuropsychological tests from two different time points (pre- and post- intervention).
The Problem is, that the group sizes are rather small (around 15) and that most of the measures have different distributions for patients and controls (e.g. skewed for patients but normal for controls).

I am using SPSS, so I would prefer doing analyses with SPSS. My Questions are basically if there is any non-parametric equivalent to a Mixed Design? Or what test(s) do you suggest me to do? Any help is appreciated!

Thanks,
Kalu

 

[hlsmith]

Can you provide a couple more pieces of information? Was the intervention randomized? What is the overall sample size and how many people are in each of the 4 subgroups?

Thanks and welcome to the forum!

 

[Karabiner]

The Problem is, that the group sizes are rather small (around 15)

Since this is in part a within-subject design, error varianvce is reduced and power is larger than in a pure between-subjects design with n=15 per group.

and that most of the measures have different distributions for patients and controls (e.g. skewed for patients but normal for controls).

You say that you have only n=15 per group, and you say that this is sparse, but you make
claims about the population distributions from which the samples were drawn. On which
analyses are these claims based?

With kind regards

Karabiner

 

[hlsmith]

I would have to replicate @Karabiner's comment on assumptions related to distributions. I can create a n=15 simulation of a random normal distribution and it can look skewed given the finiteness of the sample size.

 

[hlsmith]

set.seed(1234)
y1 = rnorm(15, 0, 1)
y2 = exp(y1)
par(mfrow=c(1,2))
hist(y1, main="normal, n=15")
hist(y2, main="lognormal, n=15")

 

[Kalu]

Thank you so much for trying to help!

The complete sample size is N=54 with Group sizes of 14,14,15,11.
And yes, the intervention was randomized.

And my bad for making claims about the population distributions, what I meant is, that the histograms of the groups look very different and not as coming from a normal distribution. However, it seems that based on these sample sizes I cannot make any claims about that. But still, what would be the best way to deal with these issues?

Best, Kalu

 

[Karabiner]

You have n > 50, and as far as i know, with such a total sample
size, you do not need to worry about the distributions within
your groups (i.e. the residuals from the analysis of the group
effect). So it would rather be important to check the remaining
assumption for repetaed-measures/mixed analysis of variance.

With kind regards

Karabiner