| Message/Author |
|
|
|
|
Dear Drs Muthen,
Is it possible to run multiple imputation on a dataset with count variables? I'm using version 6.11. Also is there any way to specify maximum and minimum values for the imputed values as I'm ending up with minus values?
Many thanks,
Jaclyn |
|
|
|
|
| DATA IMPUTATION is available for continuous and categorical outcomes only. |
|
|
|
|
Thank you for your quick response.
Jaclyn |
|
|
| EFried posted on Tuesday, June 19, 2012 - 12:56 pm
|
|
|
Following up on the mininum and maximum questions - are there ways to limit the imputed values? We work on average workload per week, and we would like to limit the upper imputed values to the highest observed value in the population.
I checked technote7 (MI) and the 6.x handbook (chapter 11) but was unable to find information.
Thank you |
|
|
|
|
| Take a look at the "Values =" option in the data imputation command. |
|
|
| EFried posted on Tuesday, June 19, 2012 - 3:33 pm
|
|
|
I seem to be blind. I checked twice - in two different pdf readers (not being able to copy & paste from the UG seems to sometimes bug the search function in some readers) - for "values" in the UG6, but could not find anything related to imputation. I also searched for "data imputation", but there is nothing about an option "values" either.
Sorry |
|
|
|
|
See page 750 of the UG index, saying
VALUES, 464 |
|
|
|
|
Dear Prof Muthen,
From the UG I understand that you can only specify integer values for continous variables. However, we have multiple continous variables that can also have decimals (so e.g. a variable that can have any value between 1-7). Is it possible to specify a minimum and maximm for those variables?
If this is not possible: Is there any way we can check what values have been imputed, to see if they do not deviate to much from the observed or theoretically possible values?
Thank you! |
|
|
|
|
| Continuous variables can have decimals in Mplus. I don't know what you mean about specifying a min/max for a variable. |
|
|
|
|
Sorry, my message wasn't very complete. I meant during multiple imputation. I want to impute the missing values in my dataset using the data imputation command. However, I would want to specify the minimum and maximum for the continuous values to be imputed. However the VALUES command seems only to specify the range of integers that can be imputed. Is there a way to specify a minimum and maximum, allowing for decimals?
And, if not, is there a way to check what values have been imputed?
I hope this makes any more sense.
Thank you! |
|
|
|
|
| Check the Values and Rounding options of Multiple Imputation described on page 579 of the V8 UG. |
|
|
|
|
Dear Dr Muthen,
Thank you for the quick response. I had read this page. It says that for continues variables you can specify the values to be imputed, but only integers. Therefore, my understanding is that it is not possible to restrict the range of non-integer values when imputing. Is that correct?
Three other questions:
1. Is there a way to ask for the descriptives (incl. min and max) of all imputed datasets?
2. In the output of the imputation analysis, using twolevel basic, the within means are 0. Why is that? Is this an indication of an error?
3. Also, it states that the results are based on 39 imputed datasets whereas 40 imputed datasets have been created. Again, is this an indication of an error?
Thank you! |
|
|
|
|
This is not correct. You can specify non-integer values in the values command
values=x(1.1 2.2 3.3);
1. Look for a section called
"UNIVARIATE SAMPLE STATISTICS"
that contains min/max
2. The means are on the between level. Only one mean can be estimated for each variable and it is estimated as a between level parameter
3. Yes the errors are reported in output:tech9; at the end of the output file |
|
|
|
|
Thanks for your reply.
1. These sample statistics are the descriptives of the non-imputed original data, aren't they?
3. Indeed, in tech9 I have 1 error message stating:
THE H1 MODEL ESTIMATION DID NOT CONVERGE. THE SAMPLE STATISTICS COULD NOT BE COMPUTED. INCREASE THE NUMBER OF H1ITERATIONS.
a) How do I know for which dataset this holds as I seem to have data in all 40 imputed datasets.
b) Can I still run analyses using TYPE = imputation? Or are these analyses no longer valid as there was an error for one of the imputed datasets?
Thank you for your help! |
|
|
|
|
1. You should run type=basic on the data that you want these for
3a. it should be possible to see that in tech9 - if not send your example to support@statmodel.com
3b. Since only one didn't converge I would say that can ignore it.
You can also use this option
OUTPUT: NOCHISQUARE;
since that will skip the H1 computation which you probably don't need anyway.
Or you can follow the instructions in the message to try to resolve the convergence issue. |
|
|
| Back to top |
