Multiple imputation

[kingston1]

Hi experts,

I am trying to use multiple imputation for left censored bio marker data. Due to the sensitivity on the assay, many smaller values are set as missing because they were undetected.

This is my current syntax (SAS 9.4) but I am unable to impute values outside my observed values range. How do I bypass this and impute values outside of my observed range of values? Thank you

proc mi data=new1 nimpute=10 seed=12345 outimputed

minimum=. . . . .

maximum=0.09 0.78 18.6 6.3 26.8

minmaxiter=100;

mcmc;

var cortpgml crp il6ra ifny stnfr1;

run;

ERROR: An imputed variable is not in the specified range after 100 tries.

[SAS_Rob]

You could try the suggestions in this usage note.
http://support.sas.com/techsup/notes/v8/24/475.html

[DWilson · Accepted solution]

---

@kingston1 wrote:

Hi experts,

I am trying to use multiple imputation for left censored bio marker data. Due to the sensitivity on the assay, many smaller values are set as missing because they were undetected.

This is my current syntax (SAS 9.4) but I am unable to impute values outside my observed values range. How do I bypass this and impute values outside of my observed range of values? Thank you

proc mi data=new1 nimpute=10 seed=12345 outimputed

minimum=. . . . .

maximum=0.09 0.78 18.6 6.3 26.8

minmaxiter=100;

mcmc;

var cortpgml crp il6ra ifny stnfr1;

run;

ERROR: An imputed variable is not in the specified range after 100 tries.

---

I don't think it's prudent to try and impute data outside the observed data values. MI won't help you here because it uses the observed data to estimate the distribution of the data. 

If you want to move forward with this, I think you'd have to have other variables that you could use to predict the bio marker data. You'd develop a model that allows you to predict the missing bio marker data given the values of the other variables.

[kingston1]

Ok, I see what you're saying. So right now I am planning to run a tobit regression model to predict my censored data. Then I can run a multiple imputation and specify a prior distribution based on my tobit model.