• GEE;
  • linear regression;
  • logistic regression;
  • probands;
  • twin data


Twin pairs are sometimes included in studies because at least one of them is a proband, and conventionally the analysis of the data is based on the conditional distribution of the cotwin given the proband. In the case of more than one proband in each pair, an often used “ad hoc” method of analysis is to allow each twin to act as proband as well as cotwin. An example of this is the probandwise concordance, which is used as an estimate of the casewise concordance under incomplete ascertainment. In this paper, we show that the method of double entry under a regularity condition provides consistent estimates of parameters, but that conventional standard errors as well as a correction suggested by Stevenson et al. ([1993] J. Child Psychol. Psychiatry 34:1137–1152) will be wrong. Instead, we recommend the use of an easily calculated sandwich estimator. We illustrate the method with twin data from a study of the genetic component in hand eczema. Genet Epidemiol 25:225–235, 2003. © 2003 Wiley-Liss, Inc.