variables in linear mixed model


I'm performing a longitudinal study, that compares computer use duration measured by 2 different methods: by questionnaire and computer software. During the research, subjects have estimated the duration of their computer use 3 times with a questionnaire, and I paired this with the duration in the according period as measured by software. Furthermore, I asked other questions in the questionnaires. I measured the difference between the 2 methods with a contrast measure: ((duration acc. to questionnaire - duration acc. to software) / (duration acc to questionnaire + duration acc to software)).

My research question is: which factors are influencing this contrast measure (= over- or underestimation when estimating duration)? Because of the fact that my dataset is not well balanced (some subjects answered only 1 questionnaire, others 2 or 3), I use a linear mixed model for the analysis.

My question to you is: should I enter the 2 variables for the 2 separate methods (duration acc to questionnaire and software) to my model as well, and in what way? I'm thinking about covariates that are entered standardly (instead of stepwise entering when statistically important for the model, like I want to do with my fixed factors), but since a covariate should be a continuous variable, this excludes my variable 'duration acc to questionnaire', since it has only 3 categories.

Thanks in advance for brainstorming with me!