2
$\begingroup$

I have a longitudinal dataset of radiation exposures of an occupational cohort. A percentage of the exposure values are missing and I would like to multiply impute the missing values (it is one option of several we are comparing). The data are recorded in long format (one row for each exposure entry) and there are multiple exposure measurements per worker. However, the data are time-unstructured (different data collection schedules for each worker) and unbalanced.

I want to account for the correlation between repeated measurements on the same worker. However, because of the time-unstructured nature of the dataset, I am unable to convert my dataset into wide format and impute that way. I have begun reading about about using multilevel imputation for such a scenario, but I rather unfamiliar with this approach, including within R. Is this an appropriate method to investigate?

Improve this question
$\endgroup$
4
  • $\begingroup$ When you say "different data collection schedules for each worker" do you mean something like one worker is tested weekly and another only monthly, or something like each worker is tested every 6 weeks but on staggered schedules, or something like each worker is tested based on suspected exposure or based on previous test values? $\endgroup$
    – Wayne
    Commented Feb 18, 2015 at 17:37
  • $\begingroup$ Workers are tested periodically (likely based on suspected exposure) but there is no visible structure to the testing schedule. For example, one worker may be tested one day per month (not necessarily on the same day each month) and the next worker may have only four tests all year. So there is no common unit of time I can assign to all workers. $\endgroup$
    – PamD
    Commented Feb 19, 2015 at 21:53
  • $\begingroup$ Why not just make time a continuous variable? Nothing about panel data models really requires that the time be categorical. $\endgroup$
    – generic_user
    Commented Apr 4, 2019 at 12:07
  • $\begingroup$ For the imputation part, are you looking specifically for a canned package, ir a general approach? $\endgroup$
    – generic_user
    Commented Apr 4, 2019 at 12:08

1 Answer 1

Reset to default
1
$\begingroup$

The mi package might be what you are looking for. It can accommodate longitudinal data and works with lmer and glmer from the lme4 mixed effects models package. The vignette has an example of a longitudinal cohort study.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.