Extrapolating out of range: education edition
This post is by Joseph from West Coast Stat Views (on Observational Epidemiology and more)
This is Joseph.
Matt Yglesias had a post on the effects of closing schools on learning:
I think that if at any time pre-Covid someone had suggested that regular, in-person school attendance was not that important and kids would be okay just watching video lessons and doing online work, that would have been understood as a kind of right-wing techno-libertarian crank viewpoint. Thanks to the pandemic, though, we got to find out if the techno-libertarian cranks are right about school.It turns out that they are not. In Virginia, for example, student test scores plummeted and the racial gap in scores exploded
And we’ve seen this basically everywhere. McKinsey and NWEA found huge learning losses concentrated in poorer kids nationwide. Texas and Indiana reported big early test score declines. A study from the Netherlands indicated that during an eight-week period of virtual schooling, students learned basically nothing on average.
For years, study after study has shown that the effect sizes of education interventions tend to be really small. And when they don’t look small, they tend to be very difficult to scale up. That led some people to infer that schooling is largely pointless. But we learned during the pandemic that if you try something out-of-sample like not having school at all, the effects are actually very large.
I think that this example illustrates two major themes.
One, which we've been discussing for ages, is that it is very difficult to extrapolate data out of the range of (Read more...)