kaylom01@ODJFS.STATE.OH.US wrote: > >Hi all! I have a data set in which I am interested in a small population, >however in this data set there are about 10-20 people missing responses on >each question of interest so I am not including those individuals in the >sample. Is there a suggested way to compare those individuals who are >excluded from the sample because they are missing a response on one of the >12 variables of interest to those who will be included in the sample >because they have completed data in order to ensure that the data is >missing a random and not in some systematic manner (for example only the >low income people did not respond to the education question). Thank you for >your time and help!!

Okay, first I'm really glad to see that you're even *thinking* about MAR vs. MNAR issues. So many people 'conveniently' ignore this.

There are some ways of approaching this kind of problem. I find most of them totally lame. The only way to be *sure* that the data are missing at Random is to go back and find a way to get a legitimate probability sample of the missings and get answers from them. Otherwise, you have no way to be *sure* that they are not different in some fundamental way which would bias the answers to the unanswered questions. (This is one of my problems with tools like the Heckman model applied to this sort of problem.)

HTH, David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

_________________________________________________________________ Mortgage refinance is Hot. *Terms. Get a 5.375%* fix rate. Check savings https://www2.nextag.com/goto.jsp?product=100000035&url=%2fst.jsp&tm=y&search=mortgage_text_links_88_h2bbb&disc=y&vers=925&s=4056&p=5117