3
$\begingroup$

I have a scenario where both factors are within-subject. One is time (5 levels) and one is condition (2 levels). The time point of 0 was only measured once in each subject, and is a 'resting value' that is to be used as the time point of 0 for both conditions. I cannot find a way of dealing with this accept for entering in the value twice into the data set as t=0 for both conditions. I feel like this is not a valid approach but I don't know what else there is to do. I could run two separate one way ANOVAs but then I couldn't see an interaction between the two conditions. I use SPSS.

UPDATE: Thanks for the responses. The nature of the methodology/study design includes assuming rest is the same for both conditions (to reduce tissue sampling required). Therefore, entering the rest measures twice in the two-way ANOVA is valid and is the nature of the methodology in this case, albeit a limitation.

$\endgroup$
2
  • 2
    $\begingroup$ Is subtracting the values at t=0 from your remaining conditions (4 time points x 2 conditions) an option? Then, the difference from rest could be the variables you put into your ANOVA. $\endgroup$
    – Amyunimus
    Commented Jun 12, 2012 at 21:11
  • $\begingroup$ @Amyunimus I really like your comment. If you have a binary repeated-measures variable, the ANOVA does nothing else but taking the difference so it clearly is a good idea. Why not post it as an answer? I would upvote it. $\endgroup$
    – Henrik
    Commented Jun 13, 2012 at 0:53

3 Answers 3

3
$\begingroup$

Try subtracting the "resting value" at t=0 from your remaining time points (4x) in both conditions and put the difference from rest as variables into the repeated-measures ANOVA.

$\endgroup$
1
  • $\begingroup$ This may be an option but I have never seen this method used in my field. Often a fold-change from rest is used but I don't agree with not showing the variability which occurs at rest. A problem I see with this is not being able to show significance in the change in the measure from rest as there would be no resting measure, and then again not being able to show the variance at rest. $\endgroup$
    – Danielle
    Commented Jun 13, 2012 at 20:35
1
$\begingroup$

This is a bit of a rough fix, but you might consider recoding your categories into a 9 level variable. So, instead of having two variables, (0,1,2,3,4) and (A,B), you would have one variable (0,1A,1B,2A,2B,3A,3B,4A,4B). Then you could try and run a one-way ANOVA using that variable as your grouping variable. Again, this isn't exactly the best way to handle it, as this won't let you test for either main or interaction effects directly, but you can use post-hoc tests and visual inspection to get a good impression of what is going on in your data.

$\endgroup$
1
  • $\begingroup$ Thanks for the suggestion. I ran this analysis and saw similar effects as the original analysis (an overall effect - which in running the one-way can be indirectly understood as an interaction). $\endgroup$
    – Danielle
    Commented Jun 13, 2012 at 20:38
1
$\begingroup$

I like Amyunimus's answer. But as an alternative, how about fitting the model for t = 1 through 4 and both conditions, then conducting post-hoc tests for each of the fitted values against the value of the response variable at t=0?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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