Hi folks,

I intend to run a generalized additive mixed-effects model in R and my sample comprises 31 startup firms surveyed over 12 weeks during the Covid19 crisis (one survey each week). Around 1/3 of the samples missed participation in almost half of the time points. Omitting these can be expected to not only decrease the efficiency a lot but also induces some sort of selection bias (firms with problems don't participate regularly) .

I use the mgcv package in R and as far as I know there is only the possiblity to do a single impution. The same goals for Hyndman's forecasting method with which holes in a time series can be interpolated by forecasting methods (which is also a single imputation).

Has anyone an idea how to specify multiple imputation, that is creating a dozen data sets, run the GAMM and integrating / averaging the results appropriately (according Rubin's recommendations)?

A speciality is that it is a mixed effects model, hence, nesting of time points in startups should be considered....

Best,

Holger