Hello everyone.
I have a study design in where I have two groups (control and experimental) with two evaluation periods (pre-test and post-test).
I want to analyse the results using a repeated measures ANCOVA.
In SPSS, as I go to "ANALYZE" -> "GENERAL LINEAR MODEL" -> "REPEATED MEASURES" in the number of levels I have to put "2" (i.e., pre-test and post-test"). After I click "DEFINE", in the "Within-Subjects Variables (Time)" I have to put both pre-test and post test data. In the "Between-Subjects Factor(s)" I have to put the group. However, I also want to put the pre-test result as a "Covariate". This is not possible since the pre-test result is already above in the "Within-Subjects Variables (Time)".
Is my racionale wrong in here? Is it possible to do a repeated measures ANCOVA with the pre-test as a covariate?
Thanks in advance.
ANOVA or Repeated measures of ANOVA is when you have more than 2 groups. In this case (assuming it is not an RCT), if your outcome is a continuous variable you can go for a difference in difference analysis. If it is a binary variable, you can go for logistic regression keeping pre-intervention variable as a covariate.
Dear Vignesh Loganathan , but my question is how do I put the pre-intervention as a covariate if it is already in "Within-Subjects Variables"? Because in the "Within-Subjects Variables" I have to put both moments: pre-intervention and post-intervention.
Thank you.
Analysis of Covariance (ANCOVA) tests whether there is a significant difference between groups after controlling for variance explained by a covariate. In your case pretest can be considered as covariate. However the same variable can’t be considered as ‘Within-Subjects Variable’. That’s why statistician suggests, as mentioned by Jos Feys, go for post = pre + Groups.
Dear Kandasamy Ravichandran , thank you very much for your help. So, as 'Within-Subjects Variable', you're saying that I should put the Post intervention measure. But what is the other variable that I should put in there (since I need two to assess the effect of time)?
The first thing to know is that repeated measure ANOVA is in fact the ANOVA of change scores. If you have two groups and two timepoints, you can calculate the change scores (posttest minus pretest) and run an ANOVA (or t-test), and you will get the same F-statistic as in repeated measures ANOVA.
After clarifying this, we can consider your question as whether you can use the pretest as a covariate in an ANOVA of change scores? Yes. And the results will be mathematically equal to ANCOVA on posttest with pretest as covariates. The only difference is the slope of pre-test (the slope for ANCOVA on change score is 1-B, where B is the slope of pretest in ANCOVA on posttest).
So, when you add the covariate to a repeated measure ANOVA, you are in fact using the common method of analysis for randomized trials, i.e, ANCOVA on posttest with pretest as a covariate. If, using the change scores, make the interpretation simpler (as mentioned by van Breukelen, "clinicians like to think in term of CHANGE", http://dx.doi.org/10.1080/00273171.2013.831743 , p. 903), you can analyze your data this way.
However, whether to use ANCOVA or repeated measure ANOVA without covariate (or ANOVA of change scores) is another issue. It is usually recommended to use ANCOVA when there is randomization or the assignment is based on pretest scores. But, for preexisting groups, analysis of change is probably better, as ANCOVA can be biased. (See the above article by van Breukelen)
Good luck.
In my opinion, it is very important in the analysis of variance with repeated measures of time dimension, and that we can pre-test the covariate when there are at least three time measurements. In repeated measurements, we seek to examine the process of what happened from pre-test to post-test. When the pre-test is covarited, we modify the effect of the pre-test to a philosophy that is not a repeated measurement. But in univariate, we are specifically looking for post-test because there are two times, but we can do it by default, given the group assignment is random.
Best regards,
Just to clarify in relation to "So, as 'Within-Subjects Variable', you're saying that I should put the Post intervention measure. But what is the other variable that I should put in there (since I need two to assess the effect of time)?"
There is no other variable for the ANCOVA in this case. There are two choices of analysis ANCOVA with pre score as covariate and repeated measures ANOVA with time (pre-post) as the within factor. The question is which should be preferred. This depends. In a randomized design both estimate the same treatment effect but the ANCOVA is usually preferred as it has greater precision (depending on the correlation between covariate and outcome Y).
In a non-randomised design there are cases for either analysis, though generally even then ANCOVA is probably more often the preferred model.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size