Today’s Study of the Week, via SlateStarCodex, considers the impact of intervention programs to help ameliorate the impact of lead exposure on children. Exposure to lead, even at relatively low doses, has a long-established set of negative consequences, particular pertaining to cognitive functioning and behavioral control. This dynamic has long been hypothesized as a source of a great deal of social problems, perhaps even explaining the dramatic rise and fall in crime rates in America in the 20th century, given the rise and fall of leaded gasoline. Those broader questions are persistently controversial and will take years to answer. In the meantime, we have interventions designed to ameliorate the negative impacts of lead exposure, but little in the way of large-scale responsible research to measure their impact. This study is a step in closing that gap.
In the study, written by Stephen B. Billings and Kevin T. Schnepel, a set of observational data is analyzed to see how children eligible for inclusion in a program of interventions for lead exposure compared to a control group that did not receive the intervention. The data, taken from North Carolina programs in the 1990s, is robust and full-featured, allowing the researchers to consider behavioral outcomes for children, later-in-life criminal behavior, educational outcomes, and some other metrics of overall quality of life.
For obvious reasons, the study is not a true experiment – you can’t expose children to lead as an experimental treatment and note the difference. But they are able to approximate an experimental design, first thanks to the number of statistical controls, and second thanks to a trick of the screening process. Lead testing is notoriously finicky, so children are usually tested twice in early childhood. If children were tested once and found to have lead levels higher than the threshold, they would then be tested again several months later. If they were found to have again exceeded that threshold, they would be assigned to the intervention protocol. This provided researchers with the opportunity to examine children who tested above the threshold the first time but not the second and compare them to those who tested above the threshold both times. Because only those who were above threshold twice were subject to interventions, these formed natural “control” and “test” groups, subject to quality and robustness checks. Because those in the intervention groups had higher lead exposure overall, their outcomes were statistically corrected for comparison to the control group.
As discussed in the last Study of the Week, this research uses an intent-to-treat model (“once randomized, always analyzed”) because it was not possible to tell what portion of test subjects actually completed the interventions, and because there will certainly be noncompliers in other real-world populations as well, helping to avoid overly strong estimates of intervention effects.
The interventions included education and awareness campaigns, general medical screenings for overall childhood health, nutritional interventions which are believed (but not proven) to be effective at mitigating the effects of lead exposure, educational interventions, and for higher levels of exposure, efforts to physically locate and remove the sources of contamination, usually lead paint. These efforts can be quite expensive, with an estimated average cost of intervention for in-study participants of $5,288. To my mind this is precisely the kind of thing a healthy society should ensure is paid for.
I want to note that this study strikes me as a monumental task. The sheer amount of types of data they pulled – birth records, housing records, educational data, criminal justice data, and others – must have taken great effort, and wrangling that amount of data from that many different sources is no mean feat. They even investigate which of their research subjects may have lived in the same house. And the sheer number of controls and quality tests employed here are remarkable. It’s admirable work, which will serve as a good model for replication going forward.
Unsurprisingly, lead exposure has a serious impact on educational outcomes:
This is consistent with a large body of research, as suggested previously. The behavioral outcomes are even more pronounced, which you can investigate in the paper. Bear in mind that in the raw numbers there are many confounds – poor people and people of color are disproportionately likely to live in lead-tainted environments, and they are also more likely to suffer from educational disadvantage in general, thanks to many social factors. But these trends are true within identified demographic groups as well.
Luckily, the intervention protocol does have an impact. To estimate it, the researchers combine this data (math and reading at 3rd and 8th grade and grade retention from grades 1-9) into an educational intervention index. They find an overall effect of .117 SD improvement relative to the control group in this index, though with a p-value only significant to .10, not typically considered significant in many contexts. This is perhaps explained in part by the n of 301 and may be improved with larger-n replications. There is a great deal of difference in the metrics that make up the index, listed in Table 4, so I urge you to investigate their individual effect sizes and p-values.
This overall effect size of .117 for the educational index is somewhat discouraging, even though they suggest the intervention does have a positive impact. The biggest positive educational interventions achievable by policy, such as high-intensity tutoring, tend to have a .4-.5 SD effect on quantitative outcomes; the black-white achievement gap in many metrics is around 1 SD. So we’re talking about modest gains that don’t close the educational disadvantages associated with lead exposure. This can perhaps intuitive given that these efforts are largely aimed at preventing more exposure, rather than counteracting the impact of past behaviors. Still, in a world where we’re grasping for the positive impacts we can, and given our clear moral responsibility to help children grow in lead-free environments regardless of the educational impacts, it’s an encouraging sign.
What’s more, the behavioral indices were more encouraging, The researchers assembled an antisocial behavior index including metrics related to school discipline and criminality. Here the effect size was .184, significant to an alpha of .05. These non-cognitive skills make a big impact on the quality of life of students, parents, teachers, and peers. Still fairly modest in impact, but more than worth the costs.
Seems pretty clear to me that we need robust efforts to clean up lead in our environment and to mitigate the damage done to people already exposed. This is an important study and I’m eager to see replication attempts.