Multilevel Multiple Imputation
Posted: Wed Jan 30, 2013 11:13 pm
Hello everyone,
I'm hoping someone can tell me what I'm doing wrong because I've hit a wall with my dissertation. So here goes:
My dissertation (roughly) involves the following steps:
1) Simulate multilevel data (two level model): The model looks like this: Yij=(G00 + G01*Wj + uoj) + (G10 + G11*Wj + u1j)*X1ij + (G20 + G21*Wj + u2j)*X2ij + eij
2) Impose missingness (missing at random and missing completely at random) on two level 1 variables (the outcome Y and a level 1 covariate X1 where missingness on both is dependent on the values of X2): the probability of missingness is: P(Rij)=exp(B0+B1X2ij)/(1+exp(B0+B1X2ij))
3) Impute missing data using a multilevel model
So far, I've tried completing step 3 above using the package "pan" in R and can't seem to get the specify the psi matrix (the variance covariance matrix for the random effects) as something other than an identity matrix. After trying a lot of different things in the pan package, I purchased MLwiN.
I started off by specifying the multivariate model depicted in the attached document (Y_4 is the Y variable with 50% missing data and X1_4 is the X1 variable with 50% missing data) using IGLS.
Then, I tried to use these starting values using MCMC to obtain the multiply imputed datasets and MLwiN gives me an error stating that: MCMC Error: Residual S.E. column is too short or contains missing values(MCMC).
Any thoughts or recommendations would be greatly appreciated!
I'm hoping someone can tell me what I'm doing wrong because I've hit a wall with my dissertation. So here goes:
My dissertation (roughly) involves the following steps:
1) Simulate multilevel data (two level model): The model looks like this: Yij=(G00 + G01*Wj + uoj) + (G10 + G11*Wj + u1j)*X1ij + (G20 + G21*Wj + u2j)*X2ij + eij
2) Impose missingness (missing at random and missing completely at random) on two level 1 variables (the outcome Y and a level 1 covariate X1 where missingness on both is dependent on the values of X2): the probability of missingness is: P(Rij)=exp(B0+B1X2ij)/(1+exp(B0+B1X2ij))
3) Impute missing data using a multilevel model
So far, I've tried completing step 3 above using the package "pan" in R and can't seem to get the specify the psi matrix (the variance covariance matrix for the random effects) as something other than an identity matrix. After trying a lot of different things in the pan package, I purchased MLwiN.
I started off by specifying the multivariate model depicted in the attached document (Y_4 is the Y variable with 50% missing data and X1_4 is the X1 variable with 50% missing data) using IGLS.
Then, I tried to use these starting values using MCMC to obtain the multiply imputed datasets and MLwiN gives me an error stating that: MCMC Error: Residual S.E. column is too short or contains missing values(MCMC).
Any thoughts or recommendations would be greatly appreciated!