Fig 1.
1914 French birth certificate.
The birth certificate gives the date of birth, sex and parental occupations at the time of the birth (zone a). The date of death, if after March 29th 1945, is available in the margin of the birth certificate (zone c). The dates of other life events (zone b) such as marriages and divorces are also notified by law in the margin (since 1897 for marriages, 1939 for divorces) and are used to classify those without a date of death. When no life event is notified after March 29th 1945, the individual is considered dead before this date. Conversely, individuals without a date of death but known to be alive on March 29th 1945 because of a life event after this date are considered alive at age 99. The birth certificate shown was registered on January 20th 1915 and is held by the city hall of the 15th arrondissement of Paris.
Table 1.
Baseline characteristics of subjects.
Fig 2.
Survival curves by sex and occupation of the father.
The survival functions are based on the pooled 10 completed datasets created by multiple imputation. In each plot, grey lines are survival functions for the other sex. Estimates of median ages at death based on B = 500 bootstrap samples are drawn in the lower part of the plots. The survival functions of the French 1915 male and female cohorts as given by the Human Mortality Database (HMD) are also drawn. Child mortality was known to be higher in the “departement” (district) of Paris than nationwide at the beginning of the XXth century. Also drawn are HMD survival functions corrected for this excess child mortality (cor. HMD), based on the quotients of mortality 1q0 and 4q1 as given by ref. 20, which relate to the 1901–1905 female cohort. 1915 HMD 1q0 and 4q1 were corrected assuming that the percent excess Parisian mortality for both sexes was the same in 1915 as in 1901–1905.
Table 2.
Parametric terms from model (1) and model (2).
Fig 3.
Adjusted hazard ratio according to sex and paternal occupation.
The effect of paternal occupation on the hazard ratio on [31; 99 y] is estimated for each completed dataset thanks to a Cox proportional hazard model, and then combined following Rubin’s rules. Also given are estimates on legitimate children alone, for whom no imputation of paternal occupations is needed. For both sexes, the reference group is that of craftsmen’ children. Estimated effects are given ± standard error.