1
$\begingroup$

I would love to get some advice regarding the following please!

I am involved in a study that is looking at cognition across 3 separate groups (1= the clinical group of interest; 2= clinical control group; 3=healthy control group). All participants completed a range of cognitive tests. Preliminary analysis on the entire sample (n=51) showed that there was a general intellect difference between groups (grp 3 was higher than grps 1+2), which meant that any subsequent analysis showing differences on the cognitive tests could be due to intellect alone rather than a true difference. So the sample was specifically matched on IQ, resulting in an adjusted sample size of n=33 (11 per group).

I was originally instructed to perform repeated measures ANOVAs on the refined sample for each of my cognitive tests. As such, the layout of my SPSS datafile shows 11 rows, so there are 3'groups' per variable (e.g., test1_group1; test1_group2; test1_group3).

Now, there were no significant differences for any of these cognitive tests. However, there was a significant difference in terms of mood, and so I would be interested in running a repeated measures ANCOVA to investigate how much mood is impacting on the cognitive results. However, given the layout of my variables, I am becoming confused at how to run the repeated measures ANCOVA using SPSS, as my variable options re: the mood questionnaire produces 3 options (e.g., Mood_group1; mood_group2; mood_group3) - do I choose the mood variable for the group with the significant difference only, or use all 3 variables as covariates (which complicates the output somewhat!)

To add yet another layer of complexity, I consulted with some friends about my analyses. They were confused why I was analysing using repeated measures and suggested I instead alter the layout of my SPSS datasheet so that I would have 33 rows (to match the n=33), and run a series of independent one-way ANOVAs, which would seem to solve my 3 variables per 'person' dilemma re: performing an ANCOVA, but leaves me extremely confused as this seems to make sense, but is not in line with what I was told originally.

I'm sorry this has turned into a longer post than I had intended but hopefully it provides enough background in order for some well-needed advice to be thrown my way!! Any input/ideas would be greatly appreciated!

$\endgroup$
1
$\begingroup$

I think you would be better off using linear regression, which is essentially the same as an AN(C)OVA, but slightly more general. Using a linear regression allows you to control for the influence of group membership and intellect separately, as well as by interaction. I would not recommend throwing away data. Selecting data post hoc introduces sampling bias so that you can no longer safely assume a random sample. This essentially means that your results are less valid and less generalizable.

You should organize your data in a different way. In SPSS the rows contain cases, and the columns contain variables. The best way to organize your data would be to include all 51 cases over 51 rows. Instead of having every measure three times for every group, you will have one variable for every measure plus two dummies to indicate group membership. You need two group dummies because you always need k-1 amount of dummies to discern between k number of groups. How to code dummy variables is explained here.

$\endgroup$
1
  • $\begingroup$ your answer is a good one. The OP should have remarked that going from n=51 to n=33 is a strong signal than an inappropriate analysis has been chosen. There is no justification for matching, especially when it discards valuable data. Covariate adjustment is much less arbitrary and is more efficient. $\endgroup$ Aug 3 '13 at 11:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.