Hi Maarten,
Thank you for taking the time to respond.
For my data, I am quite certain that the data is not missing at random (NMAR). I have reason to believe that my missing outcome data is related to the outcome data itself. I do have a full set of explanatory variables for all of my observations, however.
Does this mean that I cannot use the typical remedies? What other options are there for analyzing missing data that is non-ignorable?
Does Stata have procedures for the Selection or Pattern-mixture models for NMAR data?
Would it also be acceptable to provide justifications as to why the missing values are more likely to be Domestic cars (if I believed that Domestic car owners were more likely to report it), and then just assign the missing values to Domestic? (and then run sensitivity analyses to compare the results with and without the missing values?)
Thank you!
> Date: Fri, 22 May 2009 09:52:08 +0000
> From: [email protected]
> Subject: Re: st: Missing outcome variables - how to deal with these?
> To: [email protected]
>
>
> --- On Thu, 21/5/09, Tomas M wrote:
>> I have a set of data, with missing outcome variables.
>> I plan to use multinomial logistic regression to
>> analyze the association of my predictor variables with
>> the outcomes. How would I deal with these missing
>> outcome values in Stata?
>>
>> Would I drop the missing values from analyses? Or
>> would I include them as a separate outcome in my multinomial
>> logistic regression models?
>>
>> Would it also be possible to impute the missing outcome
>> values using some function to predict the outcome based on a
>> given set of independent predictor variables (derived based
>> on the set of full and complete data)? If so, how
>> would I go about this in Stata?
>
> Most of the remidies for this type of problem assume that the
> probability of having a missing value only depedends on
> observed characteristics of the respondent (this is the so-
> called Missing At Ramdom or MAR assumption)
>
> If you have only missing observations on the dependent variable
> then just leaving the missing observations out of the analysis
> will lead to unbiased estimates whenever the remedies are
> appropriate (see for example Allison 2002, footnote 1). So, in
> these cases these remedies add very little, except that you use
> information from more observations. However, this is only
> helpful if you have few observations to begin with.
>
> So, in your case I would just declare those missing values as
> missing (=.) and do nothing else, unless you have a very small
> sample size. The missing values will than be automatically
> ignored. If you want to do an imputation I would use Patrick
> Royston's (2004, 2005a, 2005b, 2007) -ice- command.
>
> Hope this helps,
> Maarten
>
> Paul D. Allison (2002) Missing Data, Quantitative Applications
> in the Social Sciences, nr. 136. Thousand Oaks: Sage.
>
> Royston, P. 2004. Multiple imputation of missing values. Stata Journal
> 4(3): 227--241.
>
> Royston, P. 2005a. Multiple imputation of missing values: update. Stata
> Journal 5(2): 188--201.
>
> Royston, P. 2005b. Multiple imputation of missing values: Update of ice.
> Stata Journal 5(4): 527--536.
>
> Royston, P. 2007. Multiple imputation of missing values: further update
> of ice, with an emphasis on interval censoring. Stata Journal 7(4):
> 445--464.
>
> -----------------------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
_________________________________________________________________
Create a cool, new character for your Windows Live™ Messenger.
http://go.microsoft.com/?linkid=9656621
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
