Ancova when covariate is of theoretical interest My experimental design contains a repeated-measures, categorical variable (three levels, type of sentence) and two continuous predictor variables (two subscales scores derived from a questionnaire). The design is balanced and meets assumptions of normality. Also, I do not expect missing data. I want to know the effect the predictor variables have on the time to read a sentence for sentence type (the dv).
One suggestion was that I analyze the data using ANCOVA. I had thought about doing an ANCOVA, but I wondered about the propriety of it given that the continuous variables are of theoretical interest to me (i.e., I predict it will influence reading time). So, is an ANCOVA appropriate if there are continuous variables of interest?
 A: Yes, ANCOVA and regression are the same under the hood. With ANCOVA you will get an F-test for the statistical significance of the covariate of interest. If you run the model as a regression model, the t-test for the regression coefficient for this covariate would have the same significance value and $t^2 = F$. What you don't get with the ANCOVA is the regression coefficient providing you with information on the nature of the relationship between this covariate and the outcome. An advantage of a repeated measures ANCOVA is that it is easy to implement the repeated measures aspect of this design. Another advantage of ANCOVA is that you get an omnibus test for the significance in the variability in the means across the three conditions of your independent variable, rather than two indicators variables comparing two groups to the third (although other indicator variable coding methods are also possible). Depending on the software you are using, you may be able to get the linear model (regression model) associated with your ANCOVA (in Stata and R, this is straightforward).
