New findings (released March 16) provide important new information on the “medium-run” test score effects of investments in preschool and child care. These new findings are from an ongoing study by Kenneth Dodge, Helen Ladd, and Clara Muschkin at Duke University. The study examines the effects on 3rd grade test scores of North Carolina’s investments in the “Smart Start” program and the “More at Four” program.
“Smart Start” is a state of North Carolina “birth to five” program, begun in pilot counties in 1993. The program provides a wide range of services, including child care, family support, and health care services. Services are provided to children from birth to age five. Services can be provided to children at all income levels. The specific services provided are determined with a great deal of flexibility at the county level. In participating counties, Smart Start spending per child has averaged about $250 per child per year, or $1250 over the five years. These average spending levels per child are calculated using all children in the county, not just those participating in Smart Start services.
“More at Four” is a state-funded preschool program for four-year olds, begun in pilot counties in 2001. Eligibility is limited to “at risk” four-year-olds. “At risk” is determined based on family income and other factors. The state funding goes to a wide variety of preschool providers, including private preschools as well as public schools. The average funding per four-year-old in participating counties is about $1250. This average spending level is also calculated using all four-year-olds in the county, not just four-year-olds participating in the program.
To give another perspective on the size of these North Carolina initiatives, we can calculate what an equivalent effort would cost if implemented nationally. The combined cost of Smart Start and More at Four is about $2,500 per young child per year. The U.S. Census Bureau estimates that there are about 4.3 million four-year-olds in the U.S., with about the same number for each single year of age from birth to age 4. We can multiply $2,500 by 4.3 million to calculate what these two North Carolina programs would cost if implemented at the same scale nationally. The estimate is that a national implementation at the North Carolina level of effort would cost about $11 billion. Given that there are about 311 million people living in the U.S., the estimated cost of such a program package would be about $35 per person. This can be used with your state or local area population to estimate what such a program package might cost in your state or local area.
The Duke University study is worth paying attention to because its methodology is excellent. These two North Carolina programs happened to be implemented in a phased way across North Carolina counties. Only a few counties participated in these programs when the programs were first started, and then the number of counties participating grew over time. The Duke University researchers have information on the 3rd grade test scores of individual children in different North Carolina counties in different years. They estimate how these test scores vary with the county’s spending per child on Smart Start and More at Four, with that spending measured as of the years when the child might have participated in these programs. They restrict this 3rd grade test score analysis to children born in the county, so that the test-takers could have potentially been a participant in the county’s implementation of Smart Start or More at Four. The research controls for school characteristics and characteristics of the child at birth. Finally, the research controls for any fixed effects or time trends in county test scores, as well as for any overall changes in North Carolina test scores over time.
Why is this methodology believable? Essentially, the research is asking the following question: do county test scores tend to jump after the county begins participating in Smart Start or More at Four? And the study seeks to observe whether that jump occurs when we would expect the jump to occur, given the age ranges covered by the program and the typical age of 3rd graders. The implicit comparison group is other time periods and counties for which participation in these programs did not change. Because the study statistically controls for any average differences across counties in the levels or trends of test scores, these comparisons across counties and over time are likely to give a good estimate of the effects of these early childhood investments. This is a good “natural experiment”. It is hard to think of a good reason, other than program effects, why we would see well-timed jumps in many counties’ test scores. Such jumps are unlikely, without program effects, to just happen to occur in the appropriate years after a particular county’s involvement in these programs increased.
As the researchers point out, another good thing about this methodology is that it captures possible community-wide effects of early childhood programs. For example, there are good theoretical and empirical reasons to think that educational achievement may depend on peer effects. The educational achievement gains of one individual student in a class may be positively affected by the average achievement levels of his or her classmates. The Duke University study can capture such peer effects, because the study is looking at the test scores of all 3rd graders born in each county, not just those participating in the program.
The Duke research finds that each of these programs is associated with an increase in test scores that is equivalent to about 2 months of extra achievement. Therefore, the total extra achievement associated with both programs together is 4 months. This extra achievement is averaged over all 3rd graders born in the county. Even with peer effects and other spillovers, obviously we would expect the test score effects for program participants to be greater than 4 months.
What is this extra achievement worth? In chapter 12 of my book Investing in Kids, I address the issue of how much an improvement in early elementary test scores is worth to a state economy, in higher earnings per capita of state residents. Based on that research, I calculate that increasing 3rd grade achievement by 4 months for one student will increase the present value of state per capita earnings by $23,643. (For policy wonks: This is based on converting the test score gains of 4 months into an “effect size” of 0.28, based on figures in Hill, Bloom, Black and Lipsey. I then use the numbers for effects of early elementary test scores in chapter 12 of my book.)
The ratio of state economic development benefits to state program costs for North Carolina’s programs is then calculated to be 8.79. I calculate 8.79 by taking earnings gains for state residents of $23,643 per 3rd grade student divided by state spending of $2,500 per child on the two programs, and then adjusting downward by 7.1% to allow for likely migration out-of-state between age 2 and 3rd grade. This downward adjustment for migration between age 2 and 3rd grade is based on estimates reported in chapter 2 of Investing in Kids and in my previous working paper.
This benefit-cost ratio of 8.79 is quite high. What might explain such a high benefit-cost ratio? First, I suspect that the North Carolina program leverages considerable other local resources. The benefit to cost ratio including such local resources would be somewhat lower, although probably still quite high. Second, the planning and coordination of the Smart Start program may lead to improvements in the quality of the local early childhood system. Such qualitative improvements may have very high benefit-cost ratios. Third, the More at Four preschool program is targeted at at-risk students. Such targeted preschool programs are likely to have high benefit to cost ratios. However, as I have pointed out before, expanding targeted programs to universal programs probably offers both economic and political benefits.
The preliminary findings from the Duke study provide further evidence that state-funded early childhood programs can work at a large scale. These programs can lead to changes in elementary school test scores for an entire county that are large enough to be statistically detectable. These improvements in test scores are large enough to be economically important.