Hello all,
I am conducting a randomised controlled trial (control vs. experimental) looking at pre- and post- self-reported measures. I want to test if various demographic variables have an effect on the outcomes as well. To my understanding, for some variables e.g. age, an ANCOVA can be used and the variables would be put in as covariates. However, for other categorical variables that have 2+ possible options, for instance. educational level (undergraduate, postgraduate,doctorate, masters), I would need to create a dummy variable and then put the specific options e.g. undergraduate etc. in as covariates as ANCOVAs. Is this accurate? How is this so?
I hope this is clearly expressed!
Dear Lydia,
I think that is correct. If you suspect the ordinal structure of your other categorical variables with several options could be important then you could consider Jonckheere-Terpsra or Spearman correlation. However, these are both one dimensional nonparametric tests which would not even allow you to include your two experimental groups.
From my understanding, the drawback of ANCOVA is that it assumes the two experimental groups have the same slope. An alternative would be multiple linear regression which would require you to create dummy categorical variables in the same way.
Stay alert with some problems in your approach. At first, if the same person points the pre and post condition, the repeated measure analysis may be more appropriate than the split-plot design. The use of covariates requires some attention because the variables can be correlated, like age with educational level (doctors older than masters, older than postgraduates and so on). This association (colinearity or multicolinearity) can cause problems in the tests.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA