Notes

^1.

See LaRochelle-Côté, Myles, and Picot (2010) for a complete discussion of population coverage issues in the LAD.

^2.

Two variables determine the marital status of individuals in our sample. First, we use a "description of individual" variable named "INDFLI" to identify adult males and females who were married or had common-law partners over the first three years of the sample (at ages 54, 55, and 56). Second, we identify individuals whose individual description changed to "lone parent" or "non-family person," in order to identify marital dissolution. Finally, we distinguish separations from widowhood or widowerhood by using information about the year of death of the married or common-law partner through a variable named "YOD__P". If no year of death is provided at the time of marital dissolution, the individual is considered separated or divorced. Other family status variables in the LAD may not provide accurate results and are not recommended for analytical purposes.

^3.

As a very small portion of widows and widowers do not provide a year of death for their (married or common-law) partner, we use the Survey of Labour and Income Dynamics (SLID) in order to compare divorce and separation rates with our sample of divorcees. Very similar rates are obtained for separated people aged less than 65, but divorce and separation are very rare events in the SLID after age 65. We drop separation cases that happen after age 65, since the majority are likely reporting errors.

^4.

Some individuals identified as singles at age 54 to 56 might be, in fact, people who divorced or separated at a younger age. Since this article is focused on the effect of late-life divorce on replacement rates, it is appropriate to use the married population at age 54 to 56 as a starting point, comparing the replacement rates of a treatment group of divorcees with a control group of always-married people. The impact of past marital status history on the retirement income security of singles is a topic that warrants further research.

^5.

Because we are using a permanent definition of income (income averaged over three years), our sample includes women with an easily identifiable (and consistent) marital status over 1982, 1983, and 1984.

^6.

The analysis does not take into account changes in wealth; rather, it considers changes in income resulting from divorce or widowhood or widowerhood. This may understate the effect of divorce: the widowed will inherit the family home, but the divorced receive only one-half of the assets associated with the home.

^7.

As in LaRochelle-Côté, Myles, and Picot (2008, 2010), before-tax family income is defined as all sources of income at the family level. No adjustment is made to the income for deductions. After-tax family income is defined as before-tax family income minus taxes paid. There is a possibility of double-counting some income between husband and wife after a divorce. In some years, alimony payments were deducted from the income of the sender and reported in the income of the receiver. Deductions are not accounted for in these data. When deductions for alimony payments are removed from the sender's income, the results do not change. Child-support payment deductions, which could be claimed between 1986 and 1997, are not accounted for in our definition; given the age of our sample, few families have children for whom child support payments would be received.

^8.

As mentioned, all incomes and income components are adult-equivalent-adjusted. To arrive at adult-equivalent-adjusted income, all family incomes, or their components, are divided by the square root of family size; this is the most common manner of adjusting family income. The adult-equivalent-adjusted family income can be interpreted as a per capita measure of family income, after one takes into account both family size and economies of scale available to individuals who live in larger families. All individuals in the same family have the same adult-equivalent-adjusted family income. For example, if an individual belongs to a family of four and has an adult-equivalent-adjusted family income of $25,000, his/her family's total unadjusted income would be $50,000. The adult-equivalent-adjusted income is $50,000 divided by the square root of 4. Since results may be sensitive to the choice of equivalence scales, sensitivity tests with alternative equivalence adjustments produced changes in magnitudes in our results but no changes in our substantive conclusions (see the Appendix). To ensure that the results present a more stable "permanent" income picture, income figures are all expressed in three-year moving averages. For example, the family income of an individual in 1983 (for example, an individual aged 55) is actually his or her average family income over 1982, 1983, and 1984, inclusively. Because 2008 is the last year of available data in the LAD, results can be reported between 1983 and 2007.

^9.

Recall that the sample excludes separations occurring after 1995, as many separation cases in the LAD after age 65 are widows not reporting the year of death of their husbands.

^10.

When ln(RR_it) is the dependent variable, the magnitude of the effect of separation or divorce on replacement rates is similar to that of widowhood or widowerhood.

^11.

Age dummy variables allow for the most flexible pattern in the association between age and replacement rates, but, in a model using a cubic in age, coefficients on separation or divorce and widowhood or widowerhood changed little.

^12.

This increase could be related to alimony or related payments, as women in the bottom and middle quintiles may be more likely to receive a lump sum from which they benefit in subsequent years. Other possible factors could include changes in employment earnings, earnings from social programs, or increased income for others in the family.