Multiple Imputation

[BJ]

Apparently something went wrong when I tried posting this the first
time...apologies if you have already received this...
 
Hello all, this is my first posting, but it's a good one.
I'm analyzing a dataset which is both longitudinal and follows a complex
sampling methodology. For cross-sectional descriptions the complex
sampling module is used, while any analyses are carried using the GENLIN
procedure.  I'm currently deciding on whether to use Multiple Imputation
(probably via AMELIA) to correct for the large numbers of missing
responses.  My questions are:
(1) Is this even a good idea, perhaps I should use some other missing
value analysis;
(2) Once I have multiple imputed datasets, (a) how would I combine the
estimates derived from the complex sampling frequency procedure for each
imputed dataset and (b) similarly for the estimates from the GENLIN
procedure.  i.e. are Rubins rules appropriate (particularly for nominal
data)
 
More detail if required:
833 Households nested within 161 Primary Sampling Units (PSU), which are
in turn nested within 51 Strata, and across 4 sampling waves (taken
every 3 years).
 

weight =

 

Number of PSUs in strata   x

Number of PSUs selected    

 

Number of eligible dwellings in PSU    _ x

Number of dwellings selected from PSU

 

Enumerated dwellings  x  

Screened dwellings          

 

Eligible dwellings selected     _

Eligible dwellings responding

 

 

Thanks in advance,

Brendan Stevenson

[Bauer, John H.]

Something likewise went wrong with my post on June 22, which did not appear until June 23, with the subject "Re: Ahhh...Missing Data Nightmare".  It began:

A Python module rubin.py is available at SPSS Developer Central which uses Rubin's Rules to combine results:

http://www.spss.com/devcentral/index.cfm?pg=downloadDet&dId=55

I won't repost the whole thing, please check the list archives instead.


John Bauer

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Stevenson, Brendan
Sent: Monday, June 25, 2007 8:35 PM
To: [hidden email]
Subject: Multiple Imputation

Apparently something went wrong when I tried posting this the first time...apologies if you have already received this...

Hello all, this is my first posting, but it's a good one.
I'm analyzing a dataset which is both longitudinal and follows a complex sampling methodology. For cross-sectional descriptions the complex sampling module is used, while any analyses are carried using the GENLIN procedure.  I'm currently deciding on whether to use Multiple Imputation (probably via AMELIA) to correct for the large numbers of missing responses.  My questions are:
(1) Is this even a good idea, perhaps I should use some other missing value analysis;
(2) Once I have multiple imputed datasets, (a) how would I combine the estimates derived from the complex sampling frequency procedure for each imputed dataset and (b) similarly for the estimates from the GENLIN procedure.  i.e. are Rubins rules appropriate (particularly for nominal
data)
 ...

Thanks in advance,

Brendan Stevenson