Hi,
I have 2 grouping variables Gender and Exercise Group (Control, Exercise).
Participants visit on 3 occasions and 2 measurements (call them A and B)
are carried out on each occasion.
The number of participants is as follows:
Gender: (12, 12)
Exercise Group: (11, 13)
I carry out a test of normality first on (A1,A2,A3) and (B1,B2,B3) which
indicates B1 in particular violates normality (p-value=.000) according to
both Shapiro-Wilks and Kolmogorov Smirnov. Q1) Is there a rule for which
one of these tests should I use?
Q2) A test of normality needs to be carried out on the dependents only
(not by subgroups) in repeated measures?
Box's Test is passed however in the repeated measures I carry out on the
above.
As normality was violated for some of the dependent variables, my options
are either a) doing transformations or b) splitting file according to
Gender and Exercise Group and then carrying out Friedman's Test for
measures A and B seperately.
In the transformation option, the transformation which results in the best
fit for normality is reciprocal (1/A1, 1/A2, 1/A3) and (1/B1, 1/B2, 1/B3)
Doing a test of normality on the transformed variables results in (1/B1)
only marginally being non-normal now according to Kolmogorov Smirnov (p-
value=.046) whereas it is passed for Shapiro-Wilks (p-value=.071)
Using the reciprocal transformation, I do a repeated measures on the
transformed variables and Box's Test is violated this time (indicating I
need to look at the univariate tests in the output).
My interpretation of the results is now reversed though (as I am using the
reciprocal) so previously where someone had high scores for A1, A2 and A3,
they will now have lower scores on 1/A1, 1/A2 and 1/A3 in comparison to
someone who originally had lower scores on A1, A2 and A3.
Also, similarly whereas previously I had a decreasing trend for A
measurement, I now have an increasing trend for 1/A.
Does the above approach sound OK, use the reciprocal transformation,
report the univariate results in repeated measures and inform the reader
of the different result interpretation (based on the reciprocal
transformation)?
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
