23rd December 2013 at 10:10 pm #1305Lindsay GreenleeMember
Hello Everyone! Thank you in advance for your help – I really appreciate it!
I have a mixed 2x2x2x3 design – one repeated measures variables (2 levels), three categorical predictors (2 with two levels and one with three). In addition, I have a continuous measure.
During my dissertation proposal meeting on this topic, there was great discussion on whether I should analyze this data using MANCOVA, with the continuous IV entered in as a covariate (which yields no significant results) or whether a regression with coding is more appropriate. Due to the amount of variables I have, it seems like regression coding is not a good feasible option. Plus, I have always learned that the covariate option in GLM isn’t necessarily a covariate, but this is where continuous predictors should go also as long as you build a custom model to factor in interactions.
Another approach is to take the continuous measure (an individual difference measure) and split that into three groups: high, medium, and low and use it as another factor (which yields some significant results). However, considering I actually have the continuous measure, I am having a hard time justifying why I should group the factors other than making the data analysis more straightforward for the committee. Does anyone know a justification for doing this?
In addition, any ideas why using the continuous measure yields no significant results while using the categorical one does?
Thank you and happy holidays!27th December 2013 at 8:06 pm #1306Dave CollingridgeParticipant
As you’ve discovered, it is possible for a continuous variable which is non-significant in a model to become significant when categorized. One reason for not being significant may be high variability in the continuous predictor. When you collapse into three groups and treat the predictor as a factor, you are in a sense removing the predictor variability. I would say that in, most cases, it is best to go with a continuous variable because it allows for a more sensitive measure. Exceptions are when a researcher is interested in commonly accepted categories and/or the continuous data does not produce significant outcomes due to high variability. Anyway, best to play it safe and go with what your committee suggests as long as it is not an egregious error.
- You must be logged in to reply to this topic.