Obviously in a typical multilevel model, there is some nestedness in the dependent variable, but the data I am working with have nesting in the predictor variables instead.
Essentially, we are trying to predict life satisfaction (measured once), from observations of (for example) relatedness, which is measured up to three times. I have treated this as a two-level model with observations nested within individuals, while fixing the intercept (since there is no intra-individual variation in our life satisfaction measure), then added relatedness as an explanatory variable that is allowed to vary.
Sounds like maybe you could alternatively see the 3 time points as a 3-variate observation vector (wide format) in a single-level model. And predict life satisfaction from an intercept growth factor for those 3.
Kane Meissel posted on Wednesday, November 13, 2013 - 3:41 pm
Thanks for your reply. I can see how your suggestion would be a good alternative, but I'm more wondering whether my solution is a reasonable one?
For your approach, it seems like life satisfaction would be a between-level variable (no across-time variation, but instead varying across subjects), so there won't be an issue of fixing the intercept (variance).