Migration, marriage, and mortality: correcting sources of bias in English family reconstitutions.

The impact of excluding migration data from family reconstitutions has usually focused on a comparison of migrant and non-migrant characteristics rather than on the bias occurring in age at marriage or life expectancy. The topic of this discussion is the bias in marriage age and life expectancy in family reconstitution data under preindustrial English demographic conditions. The bias is illustrated with the MOMSIM microsimulation model where migration histories can be assigned at random according to assumptions about age-specific migration. The differences between the general population and the reconstructed subpopulation are attributed to migration effects. Migration is simulation models do not correct the bias. The model is explained and a simulation begun of 50000 married couples who survive to the age of 50 years. Alain Blums model is evaluated and new methods are proposed. The results of the microsimulation show that the mean age at marriage is raised by a little over half a year when mortality censoring is eliminated. When results are subject to both migration and mortality censoring the reconstituted estimates for women understate age at marriage by 2.6 years in the low migration model and understate age at marriage by 4.5 years in the high migration model. In order to achieve unbiased estimates 90% of the cases had to be eliminated in the high migration and high mortality model. For this reason and because English reconstitutions are usually based on 20-30% of marriages that are linked to baptismal records small reconstitutions may not be feasible. The results of Alain Blums minimum mortality procedure show that life expectancies are comparable to life expectancies in the general population. For medium and low migration males life expectancies at age 20 are higher in the general population compared to the expected high life expectancies in the minimum mortality population estimates. An alternative strategy which works well with small populations proposes a new minimum estimate of mortality and is not sensitive to differences in demographic behavior of non-migrants and eventual migrants.