In total, 1,845,956 married individuals aged 50 and older are included in the data set; 958,997 of them are male, 886,959 female. The distribution of all persons in the data set by age gap to the spouse is presented in Figure 2 . It shows that most men are between two and three years older than their wives, while most women are two years younger than their husbands.

Approximately 75% of all married men aged 50 and over are married to women who are more than one year younger than themselves; only 10% of all men are at least one year younger than their wives. In contrast, the majority of married women (65%) aged 50 years and over are married to men who are older than themselves, and only 15% have a spouse who is more than one year younger.

Table 2 gives descriptive information on all covariates. It shows the distribution of time at risk measured in days for men and women. In total, I observed 3,271 million person-days for men and 2,907 million person-days for women. The proportion of missing information is highest for duration of marriage. This is because the date of marriage is unknown for all couples who married before January 1, 1990. These couples were assigned to the category unknown for 1,000 days when entering the study population and to the category ? 1,000 days thereafter. A large number of missing values is also found for the variables highest achieved education and highest achieved education of the spouse, with the proportion missing data increasing for older cohorts. I find no indication that this effect influenced the outcome of the regression models.

Table 2.

In the following paragraphs, I present the results of four estimated hazard regression models. For men, the relative risk of dying by the age gap to the spouse and the standard errors of the fourth model are shown in Figure 3 . The corresponding results for women are shown in Figure 4 . Both figures consist of four separate curves showing the relative risk of dying by age gap to the spouse. The reference category, represented by a dotted vertical line, includes all persons who are less than one year younger or older than their spouses. The part of each curve to the left of the reference category relates to individuals with older spouses, the right part relates to individuals with younger spouses. I show only the standard errors of the fourth model because they were virtually the same for all four models.

Table 3.

Figure 4 shows that, similar to that of men, female mortality is higher if the wife is younger than her husband. Women who are more than 7 years but less than 17 years younger have an excess mortality of about 10%. In contrast to the pattern for men, women also have an elevated risk of dying when they are older than their spousespared with the reference category, an excess mortality of 40% is observed in women who are more than 15 years but less than 17 years older than their spouses. The lowest risk of dying is found in women who are about the same age as their husbands, which is the reference category.

These first results provide strong evidence that the age difference between the spouses online hookup Savannah affects individual survival chances. It also shows that the effects are substantially different between the sexes. Next, in Models 2, 3, and 4, I examine the impact of the age gap to the spouse in the presence of additional covariates.