The life expectancy of older couples and surviving spouses

Individual life expectancies provide information for individuals making retirement decisions and for policy makers. For couples, analogous measures are the expected years both spouses will be alive (joint life expectancy) and the expected years the surviving spouse will be a widow or widower (survivor life expectancy). Using individual life expectancies to calculate summary measures for couples is intuitively appealing but yield misleading results, overstating joint life expectancy and dramatically understating survivor life expectancies. This implies that standard "individual life cycle models" are misleading for couples and that “couple life cycle models” must be substantially more complex. Using the CDC life tables for 2010, we construct joint and survivor life expectancy measures for randomly formed couples. The couples we form are defined by age, race and ethnicity, and education. Due to assortative marriage, inequalities in individual life expectancies are compounded into inequalities in joint and survivor life expectancies. We also calculate life expectancy measures for randomly formed couples for the 1930–2010 decennial years. Trends over time show how the relative rate of decrease in the mortality rates of men and women affect joint and survivor life expectancies. Because our couple life expectancy measures are based on randomly formed couples, they do not capture the effects of differences in spouses’ premarital characteristics (apart from sex, age, race and ethnicity, and, in some cases, education) or of correlations in spouses’ experiences or behaviors during marriage. However, they provide benchmarks which have been sorely lacking in the public discourse.


General comments
The paper makes a nice descriptive empirical contribution to the literature by documenting the importance of taking into account within-couple mortality correlations, which impact life expectancies of married couples. Although this point seems obvious, this is important knowledge for individual planning and for policy makers, and the authors conclude that tools such as the "Life Expectancy Calculator" on the Social Security website could be augmented to calculate point and survivor life expectancies.
Generally, the empirical analysis seems convincing, although it is not always quite clear what the authors have done in every step, and how the data was used.
My comments are mainly centered around the structure of the paper and the way that the analysis is documented.
1) I think the paper would be more interesting to read if it more clearly explained early on in the paper, perhaps in the Introduction, why we would expect that it is important to take withincouple mortality correlations into account. The background and previous literature on intramarriage correlations and sorting into marriage is not introduced and discussed before 7 pages into the paper (lines 181-195), and the literature survey is very short. The paper would improve by extending the literature discussion on why we should expect intra-spouse correlation in health and mortality, citing papers on 1) assortative mating in health, wealth etc. (e.g. Guner, Kulikova & Llull in European Economic Review, Vol. 104, May 2018) and selection into marriage, 2) intra-marriage correlation in health behavior through habits, smoking, drinking, exercise, diet etc., as well as information exchange within couples (e.g. Fadlon and Nielsen, 2019, American Economic Review, "Family Health Behaviors"), 3) coordination of couples' retirement decisions that may impact health and life expectancy (e.g. Gustman and Steinmeier 2000 in Journal of Labor Economics). This section may also include a short discussion of health effects of bereavement of a spouse.
2) The structure of the Methods section is a bit confusing. It starts out with a very brief mention of the data, then goes on to present the idea of constructing life expectancies based on individual mortality data and showing some results based on this approach in Table 1, then moves to a short literature discussion (mentioned in point 1 above), and then finally presents the methods used. A restructuring of this section would be helpful.
3) Following point 2), I suggest that the authors create a specific Data section explaining in a bit more detail about the 2010 Census and NCHS data for readers who are not quite familiar with these datasets. It was not clear to me whether the authors had access to information at the individual for both partners in a couple? Also, what is the source of the life expectancies from these data? Maybe it is stated in another part of the paper.
4) It is not clear to me what data are used to create Table 1. Moreover, in general, tables should be self-explanatory, without having to read the text in detail. I suggest making a note to Table 1 with instruction about how to read it. Moreover, in line 197: When speaking about "Our calculations, …", I presume that the authors are speaking about the calculations leading to Table  1? 5) Methodological contribution: In order to assess the significance of taking proper account of intra-couple correlation in health and mortality rather than just using individual data, would it be possible e.g. to generate a table like Table 1, but just based on the data for couples? 6) I found the last four lines of the Conclusion confusing, especially point c), as I thought that the point of the whole exercise in the paper is to show the importance of taking marital status into account when calculating life expectancies for couples.

Specific and minor comments
Lines 94-98 in the Introduction more or less repeat the text in lines 65-68.