I am conducting an intensive longitudinal analysis with multilevel models.
I have some participants that don't have any variation on the predictor variables. Of course they don't really provide any useful information to the model that can be analyzed.
But I have also heared, that keeping such cases in the dataset could introduce bias to the results. However, I can't find any literature on this topic.
Does anyone know a source that discusses this and recommends excluding these cases from the sample?
Hello Pascal,
I'm a bit unclear about your query. Do you mean that a batch of cases has exactly the same score on each predictor? If so, omitting them would bias estimates of score variance on the variables (and standard errors) and most likely the regression coefficient estimates associated with them.
Aside from the fact that you might find these instances inconvenient, do you have a substantive basis, whether theoretical, or empirical, to warrant jettisoning such data? (For example, do you have after-the-fact information to suggest that such instances were not accurate portrayals of the true values for individuals?) If not, it sounds as if doing so would be ill-advised.
Good luck with your work.
David Morse thanks for your reply!
Maybe I can specify a bit. I am investigating how social control strategies influence affect and physical activity levels on a daily level. Therefore I am interested in the within-person effects (e.g. on days where one partner use a certain strategy, the physical activity of the other partner is higher).
I do have 9 couples which never indicated using any control strategy at all. Therefore their value for the predictors is always 0. Of course, the physical activity levels and affect still fluctuate.
I am not sure myself though, if it is a good idea to exclude them or not. It is just that I have heared from someone, that leaving them in would add no information to the model and could in fact introduce bias. If that is true I don't know, as I haven't been able to find any literature on it yet.
Hello again Pascal,
It's not uncommon among survey researchers to create groupings of cases such as "Did not respond," "Refused to respond," and the like. The reason for this is that, such groups often are seen to differ on other variables in subsequent analyses. Therefore, omitting them would indeed hide potentially useful information.
Ultimately, it's your call as to how to handle this situation, just be clear in your explanation of data handling what you chose to do and why.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS