be held back a grade, to be in special education, and to qualify for remedial serv- ices, although we do not have hard data on how much of these effects are due to unobserved selection bias and how much are “caused” by lack of marriage.
If marriage were to reduce the percentage of children receiving special or remedial services, then family fragmentation would create significant taxpayer costs for public education, as federal and state funding formulas tend to provide large amounts of extra funding for children receiving these services. (These costs may be offset, however, by more teens dropping out of school as a result of family fragmentation, which reduces the direct taxpayer costs of public education.49)
The lack of evidence of exogenous changes in family structure on the likelihood of receiving special education or remedial services or staying in school, and the lack of comparable cost data on remedial and special education services across states, makes it impossible to estimate these costs with confidence. But the lack of data does not mean that family fragmentation has no impact on educational expenditures. Finally, we exclude the approximately 71 percent of Medicaid expenditures devoted to the disabled and the elderly from the analysis, thereby making the cautious assumption that family fragmentation has no impact on these expenditures. Most people do not think of elderly unmarried adults or middle-aged disabled singles as belonging to “fragmented families.” Nonetheless, there is considerable evidence that older adults who are unmarried are more likely to become disabled, to manage chronic diseases less successfully, and to need nursing home care as they age.50 Excluding these large public costs thus likely significantly underestimates the actual costs to taxpayers from the decline in marriage.
2. Ignoring the direct impact of family fragmentation on crime (independent of poverty) underestimates the taxpayer costs.
Estimates of the potential impact of family structure on crime, even those that do not control for selection bias, are large and arguably should not be ignored. As dis- cussed in the section on methodology on page 12, it appears that family fragmen- tation has large effects on crime, both in terms of increasing the likelihood that a child raised outside of marriage will commit crimes51 and the likelihood that adult men will leave criminal activity after they are married.52 While Harper and McLanahan use a large number of control variables to help isolate the effect of fam- ily structure on youth crime, they do not control for unobserved selection effects. Nonetheless, their estimated effects of family structure on crime are extremely large—typically children reared in single-mother households are more than twice as likely to engage in criminal activities as children reared in a married household. For example, they report that children living with a single mother are 2.168 times more likely to be incarcerated than children living with both parents, all else being equal.53 Suppose we had assumed that over half their result was due to selection bias—that the single mothers in their sample possessed such poor parenting skills that even if they got married most of the estimated effect Harper and McLanahan reported was due to selection bias. Specifically, suppose that children reared with