Using a Multiple Imputation Data set in AMOS
Locked
 
|
Hello,
I have two data sets that I am working with. Set 1 - N = 375, Set 2 N = 175. Both sets of data have 15 variables in them. I have removed all cases with lots of missing data but about 20% of the cases in both sets have data missing from just one or occasionally two variables. There is very little missing data. Most variables only have data missing from 5 or so cases - but different cases for each variable. I was using paririwise deletion in SPSS but can't do that in AMOS. I am doing a path analysis in AMOS and am trying to get a complete data set to work with to avoid using estimating means and intercepts. I am also so far not getting a good model fit so I am hoping a complete data set might help. I did a multiple imputation in SPSS and split the files. However I can't seem to figure out how to use that data as a complete data set in AMOS. In SPSS the analysis include a pooled value but there isn't actually a pooled data set to work from. Any ideas? |
Re: Using a Multiple Imputation Data set in AMOS
|
|
>I have two data sets that I am working with. Set 1 - N = 375, Set 2 N = 175. Both sets of data have 15 variables in them. I have removed all cases with lots of missing data but about 20% of the cases in both sets have data missing from just one or occasionally two variables. There is very little missing data. Most variables only have data missing from 5 or so cases - but different cases for each variable.
Pending the results of a detailed missing data analysis, I believe you are wasting data. I urge you to read up on missing data theory. There are a number of books. John Graham is an very accessible early writer but there are others equally good. A little database search would be useful or a little internet search. In brief, you need to decide whether people refused to answer a question because of what they would have said. The classic example is income--the 1% don't want to say. > I was using pairwise deletion in SPSS but can't do that in AMOS. I can't comment on this. I do not know what options Amos offers. >I am doing a path analysis in AMOS and am trying to get a complete data set to work with to avoid using estimating means and intercepts. I don't use Amos but I believe pretty strongly that Amos uses FIML (full information maximum likelihood) estimation. FIML has been shown to yield the same results as multiple imputation (MI) if the data are either 'missing completely at random' or, simply, 'at random' and if the missing data covariates are included in the model. How Amos estimates models with categorical dependent variables with or without missing data is unknown to me. The documentation or the promotional materials should describe this. >I am also so far not getting a good model fit so I am hoping a complete data set might help. It might but I doubt it because of the use of FIML estimation. >I did a multiple imputation in SPSS and split the files. However I can't seem to figure out how to use that data as a complete data set in AMOS. In SPSS the analysis include a pooled value but there isn't actually a pooled data set to work from. You may remember that there was a question about MI a few days ago on this list. One of the repliers was Jon who described how MI functions in spss. I have not used MI and I do not know how MI integrates with Amos. Based on what he said, I suspect that you will need to save a separate dataset for each imputation and then run each dataset through your model, tabulate the results and apply Rubin's rule for calculating the SEs. I urge you to read the parts of the Amos documentation that relate to its integration with spss. If you want to save separate files, use Xsave within a Do if-else if structure for each value of the imputation dataset numbering variable, which Jon explicitly named and is probably documented in the MI documentation. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Jenna Sent: Wednesday, July 29, 2015 3:06 PM To: [hidden email] Subject: Using a Multiple Imputation Data set in AMOS Hello, I have two data sets that I am working with. Set 1 - N = 375, Set 2 N = 175. Both sets of data have 15 variables in them. I have removed all cases with lots of missing data but about 20% of the cases in both sets have data missing from just one or occasionally two variables. There is very little missing data. Most variables only have data missing from 5 or so cases - but different cases for each variable. I was using paririwise deletion in SPSS but can't do that in AMOS. I am doing a path analysis in AMOS and am trying to get a complete data set to work with to avoid using estimating means and intercepts. I am also so far not getting a good model fit so I am hoping a complete data set might help. I did a multiple imputation in SPSS and split the files. However I can't seem to figure out how to use that data as a complete data set in AMOS. In SPSS the analysis include a pooled value but there isn't actually a pooled data set to work from. Any ideas? -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Using-a-Multiple-Imputation-Data-set-in-AMOS-tp5730322.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by Jenna
2. This is not an endorsement of what I'm about to say, but one workaround one would be to impute the missing data in your original dataset using one of several methods (e.g., regression, stochastic regression, Bayesian) offered in AMOS, and use that single dataset with the AMOS imputed values to perform your path analysis. By performing the analysis on your original dataset with no missing data, you can examine modification indices (MIs) to make *post-hoc* changes to assist in improving model fit. Of course, cross-validation and other issues related to single data imputation should be considered. Ryan On Wed, Jul 29, 2015 at 3:06 PM, Jenna <[hidden email]> wrote: Hello, |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |