Hello everyone,
I have data that was collected in 20 sites across NYC (ie level 3 - site). Each site, approximately 10 participants were recruited and baseline characteristic data was gathered (level 2- participant). Further, each participant was asked to wear a device for a week (Level 1- days). Level 1 and 2 have a chunk of missing data. So we decided to perform a hierarchical level multiple imputations (HLMI)
To my understanding, this is a 3-level dataset. In this dataset, we didn't collect any site-level measures, only the individual's perception of their site (level 2). In other words- Level 2 has around 80 variables and level 1 has around 7 variables in my dataset.
1) Any advice on the package for conducting hierarchical level multiple imputations in R?
2) In this case, the cluster will be site or participants?
3) Do I have to define the method of imputation for each and every variable during writing my codes?
Thank you for your time and considerations.
Fine not having any variables at the level, at least for the computation (for substantive issues you might want to what is associated with level - 3 variance). Several R packages for multilevel modeling. https://www.jstatsoft.org/article/view/v067i01 is widely used, but there are others.
Yes it is a three level structure. I am a bit perplexed by “80variables” when you have only 20 by 10 by 7 observations. It would seem to be very underpowered. There could be a three level variable which is some summary measure of participants at each site.
Daniel Wright Appreciates your input and thanks for the paper. Do you have any recommendations on the R package to perform the multilevel data multiple imputations?
Kelvyn Jones Thank you for the response.
the dimension of my dataset it 450 X 94. I don't have any summary variables for each site. Everything is collected at the participant level with his/her 3 or more days wear data.
Sir, maybe you have a good point, maybe 3 levels will be underpowered. would it be wise to conduct a 2 level imputation?
Can you explain the 450 by 94?. And you can calculate/estimate the summary for the site level from the individual data. And can you tell us the nature of the missing ness, if it is MAR and in the response, then the standard model can handle imbalance.
Kelvyn Jones Sir, my survey has 450 individuals enrolled.
Dimensions of the dataset-
1) Number of rows- 1994
2) Number of columns- 94
Nature of missingness- MAR
I am using the MICE and MICEADDS package in R. But hit a block with coding.
Which software/package do you use for imputation?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R Statistical Package