180 likes | 283 Views
Centre for Market and Public Organisation. Using difference-in-difference methods to evaluate the effect of policy reform on fertility: The Working Families Tax Credit Mike Brewer (Institute for Fiscal Studies) Anita Ratcliffe (CMPO, University of Bristol) Sarah Smith (CMPO and IFS).
E N D
Centre for Market and Public Organisation Using difference-in-difference methods to evaluate the effect of policy reform on fertility: The Working Families Tax Credit Mike Brewer (Institute for Fiscal Studies) Anita Ratcliffe (CMPO, University of Bristol) Sarah Smith (CMPO and IFS)
The impact of welfare reform on fertility • Government spending per child rose by 50% (in real terms), 1999 – 2003 • Working Families Tax Credit (WFTC) • Means-tested benefits • Child benefit • Biggest increases for low-income families – equivalent to 10% income
Change in child-contingent benefits, 1998 – 2002 Couples, one child
Impact on fertility • The policies were aimed at incentivising work and tackling child poverty, but might they also have affected childbearing? Effect of reforms: Economic model of childbearing • Higher incomes will increase demand for quantity of children OR quality • Lower income volatility will increase demand for quantity of children • Higher benefits reduce the “price” of children (increase fertility) • Employment effect – if gain to work rises (falls) then opportunity cost rises (falls) and fertility falls (rises) • Employment effect positive for lone mothers, but mainly negative for women in couples
Empirical strategy: Differences-in-differences • How can we tell whether childbearing has been affected by the reform (and by how much)? • “Before” versus “after” may be misleading because of other changes over time • Missing data problem – what would childbearing have been in the absence of the reform? • Solution – use a control group (not affected by the reform) to proxy for the change that would otherwise have taken place
Empirical strategy: Differences-in-differences • Differences-in-differences, also known as “natural experiment” • Compare change in childbearing among a treatment group (affected by the reform) with change in childbearing among a control group (not affected by the reform) ∆ childbearingT – ∆childbearingC = effect of the reform
Empirical strategy: Differences-in-differences • Strengths • Clear, simple, intuitive • Potential weaknesses • Plausibility of control • Black box – estimate combined effect of a bundle of changes; little insight into mechanism
Choosing the treatment and control groups • The treatment group must be affected by the reform and the control group must be unaffected (including spillover effects) • The composition of the groups must be the same over time. Otherwise changes that are driven by selection effects will be wrongly attributed to the reform • Cannot split by income, instead split by education • Both groups must be affected by time-varying factors in the same way (or differential changes must be controlled for)
Treatment and control groups • Education • Treatment: Both male and female partner left school at/before compulsory school leaving age • Control: Both male and female partner left school at 18+
Entitlement to child-contingent benefits, couples with children
Data • Family Resources Survey 1995 – 2003 • Large sample, extensive information on education, income and other socio-demographic characteristics • Derive the probability that a woman had a birth in the previous 12 months • Step 1: Allocate children in household to natural mothers • Step 2: Assign randomly-generated date of birth to children (based on their age) if none available. • Step 3: Infer probability that a woman had a birth in previous 12 months based on date of interview and date of birth of child • Use information on number and ages of children to derive (approximate) fertility histories
Comparison of estimated TFR with official measure Annual total fertility rate = number of children a woman would have if she had the age-specific birth rates in that year
Identification • Identifying the effect of the reform relies on successfully controlling for everything else that might affect fertility in the treatment group • Rich set of demographic controls • Age, education, kids in household and age of kids in household, and interactions; region, housing tenure, ethnicity • Average wages for treatment and control groups • Control group intended to capture other (unobservable) time-varying characteristics, but control group has different fertility, and possibly different fertility trends • Control explicitly for differential trends
Regression analysis Dependent variable = birth in last 12 months ** indicates statistically significant at 5% level Controls include age, education, numbers and ages of children, region, housing tenure, ethnicity and wages
Regression analysis Dependent variable = birth in last 12 months ** indicates statistically significant at 5% level, * at 10% level Controls include age, education, numbers and ages of children, region, housing tenure, ethnicity and wages
Some robustness checks • Use longer time period (1990-2004) to control for quadratic trends • Estimate effects of spurious reforms in 1995 and 1996 • Allow for reform to take effect from announcement as well as implementation
Conclusions • The evidence suggests a significant increase in births (particularly first births) among women in couples affected by the reforms • 1.4 percentage point increase in probability of a birth very roughly translates into 20,000 extra births (total births 670,000) • Implied elasticity around 0.25; within range estimated by previous studies • Is it plausible that such a large increase in child-contingent benefits would not affect fertility?