Comment on "Tests of Certain Types of Ignorable Nonresponse in Surveys Subject to Item Nonresponse or Attrition" (WP-09-10)
Christopher H. RhoadsThe current paper points out some problems with the paper by Sherman (2000) referenced in the title. Misunderstandings about the terms Missing at Random (MAR) and Missing Completely at Random (MCAR) are clarified. A necessary and sufficient condition to justify a complete case analysis of bivariate, binary data when interest is in the conditional distribution of one variable given the other is presented. The non-existence of a test for MAR is noted. The impossibility of testing a condition that is sufficient to ensure unbiased estimates from an analysis of complete cases is also noted. Hence Sherman's proposed tests of ignorable nonresponse are falsified.
Christopher H. Rhoads, Postdoctoral Fellow, Institute for Policy Research, Northwestern University