Teachers
Chetty et al
After their ground-breaking analysis in 2011 The long-term impacts of teachers was cited by President Obama in his 2012 State of the Union address, Chetty et al. followed up the analysis with some improvements Measuring the impacts of teachers I: Measuring the impacts of teachers II:.
First they showed that if you condition on a student's prior test scores and other observable variables, it creates an unbiased estimate of teachers' causal impact on test scores Measuring the impacts of teachers I:. They then make the assumption that if you use the same set of controls for predicting long-term impacts, the result will likely be unbiased Measuring the impacts of teachers II:.
They find assigning a child to a teacher's with a 1SD higher Value Added for one year
- increases the probability of college attendance by 0.82pp (95% CI = 0.68pp to 0.96pp)
- increases annual earnings at age 28 by $350 (1.65%) (95% CI = $170 to $530) - though they do cause a negative impact prior to the age of 23, consistent with kids going to college. They choose the age of 28 because they can't get sufficient sample sizes larger than this, but they cite Haider as showing that income at age 28 is very predictive of lifetime earnings
- reduces the probability of teen birth by 0.61pp (95% CI = 0.49pp to 0.73pp).
They also find that teachers' effect on test scores fade out over time before stabilizing at roughly 25% of the initial impact.
They then validate their assumption that unobservable variables are irrelevant after controller for prior test scores and observable variables by using a quasi-experimental approach to estimate the effect of teacher quality on college attendance. They find an effect of 0.86pp (95% CI = 0.41 to 1.31). Given the wide confidence intervals, this isn't perfect validation, but its quite suggestive.
They find weak evidence (p~0.10) that teacher quality affects females' incomes ~20% more than males'. They find significant evidence that teacher quality affects children from richer/white households more than poorer/POC ones, even as a percent of their incomes (p~0.000).
They do some back-of-the-envelop calculation and end up supporting the idea of firing a bottom-5% teacher after their first couple years. However, this is premised on the assumption that all these income gains are purely positive-sum. This seems unlikely. Finally, they find paying high-quality teachers more is much less cost-effective.
What Makes a Good Test-Score Teacher?
This meta-analysis finds
A literature review Coenen found
- There is generally no correlation between teach educational attainment and teacher's effect on test scores. The possible exceptions are math and science in high school.
- There is generally no correlation between certification attainment and teacher's effect on test scores, but (again) it does correlate with higher scores in math (and possibly English). They also find a positive correlation for National Board Certified Teachers.
- Teachers' own test scores generally don't correlate with their effect on students' scores except (again) in math.
- Teachers' tenure positively correlates with their effect on test scores.
- Evidence is mixed regarding whether having a teacher with the same gender/race as oneself has a positive effect on test scores.
Most of the above correlations likely stem from selection effects.
todo
- Harris
- Rice
Beyond Test Scores
- Jackson
- Booker
- Deming, D. (2009)
- Deming, D. J. (2011)