Wrong Diagnosis on Homework Help from Parents
Authors find correlation, mistake it for causation
As reviewed by Jay P. Greene
“Don’t help your kids with their homework,” declares a recent Atlantic Magazine article by education reporter phenom Dana Goldstein. “Most measurable forms of parental involvement seem to yield few academic dividends for kids, or even to backfire—regardless of a parent’s race, class, or level of education.”
How does Goldstein know this? Her article relies on a new book, The Broken Compass: Parental Involvement with Children’s Education, by sociologists Keith Robinson and Angel L. Harris of the University of Texas and Duke, respectively. Robinson and Harris examine 63 different measures of parental involvement in their children’s education. They look at things like whether parents say they communicated with teachers or administrators, observed classrooms, helped pick classes and do homework together, established rules and consequences for grades and homework, and volunteered at school. In general, they find that parental involvement does not boost students’ academic achievement. In fact, their analyses usually show that parental involvement is harmful. To “control” for the influence of background characteristics, they report results separately for parents of different levels of income, educational attainment, and race.
Demonstrating that most forms of parental involvement are not only not helpful but sometimes counterproductive for student academic achievement would be a remarkable and counter-intuitive finding. But, on close inspection, Robinson and Harris’s analyses do not demonstrate anything of the sort. They simply observe that students who are struggling to make progress in school also have parents who report being more involved in their children’s education. It is quite possible, perhaps even likely, that Robinson and Harris have confused the direction of causation in their correlational analyses. That is, Robinson and Harris’s book might simply be demonstrating that parents become more involved in their children’s education when those children are failing to make good progress. Lack of academic progress might very well be causing parental involvement, not the other way around.
If it feels as if this point about confusing the direction of causation is just nit-picking, imagine I have a new set of analyses claiming to show that hospitals fail to help improve people’s health. Indeed, my analyses show that people who are hospitalized tend to have worse health outcomes. Without having actually done the analyses, I am quite confident that a regression would show hospitalization having a negative effect on health. I am confident that this negative result would hold even if I reported results broken out by the pre-existence of illnesses. I am sure that among people with heart conditions, for example, those who are hospitalized die at higher rates than those who are not.
The obvious problems with my hypothetical hospital analyses are the same as those afflicting Robinson and Harris’s analyses of parental involvement. People may seek hospital care because they are having more serious health problems, just like parents may become more involved when their children are struggling academically. Hospitals are probably not any more dangerous to your health than parental involvement is to student learning.
The strangest thing about The Broken Compass is that the authors appear entirely unaware of their inability to prove that parental involvement causes flat or negative trends in student achievement. They run one regression after another, never considering, let alone addressing, the fact that these correlational analyses do not establish causation. Taking into account prior student achievement, as well as parental race, income, and educational levels, does not solve this problem. Doing so just shows that, regardless of family background, parental involvement and student academic progress are inversely correlated: the worse the student’s change in achievement, the greater the parent involvement and vice versa. The fact that their research spans three decades of longitudinal data sets with 63 different measures of parental engagement also does not solve their inability to establish causation. A larger and more extensive longitudinal study based on correlations is no more convincing than a small and short-term one. It’s the wrong kind of analysis no matter its scale.
Social science has made enormous progress over the last few decades in finding ways to isolate causation. Researchers are conducting more experiments, as well as using techniques like instrumental variables and regression discontinuity that simulate experiments. Those methods could be used to study the effects of parental involvement. Robinson and Harris could have randomly assigned parents to receive training or incentives to increase their involvement and then compared the academic progress of their children to the progress of children whose parents did not receive that training or those incentives. They could have found some naturally occurring experiment that altered the ability of some parents to be involved. They could have at least acknowledged the problem with establishing causation and properly cautioned readers not to draw overly strong conclusions. But the authors of The Broken Compass do none of these things.
After examining more than 300 pages of The Broken Compass with its dozens of regressions and charts, I know no more about the causal relationship between parental involvement and academic progress than I did before. If the purpose of The Broken Compass were simply to raise questions about this inverse correlation, it might be a fine book. But when the authors and unthinking reporters use it to recommend that parents stop helping kids with homework, they are being irresponsible, no less so than advising sick people to avoid hospitals because they tend to kill you.
Jay P. Greene is professor of education reform at the University of Arkansas.