Fair enough.
If it is due to selection effect, the top end students would have similar access to that kind of job from a wide range of less selective or more selective colleges. If it is due to treatment effect, then the top end students would have better access to that kind of job from some colleges over others.
1 Like
I understand selection and treatment effect and thank you for explaining it so well. What I didn’t understand was the point you were trying to make about the database.
The NYT’s data comes from students who received federal financial aid. These are not full pay students. It shows us salaries for students roughly four years after graduation. If i’m not mistaken we have no way of knowing whether they obtained their job from selection or treatment effect, or a combination of both. We just see the salaries four years out.
I started the thread to call attention to the data in the article. I thought it might be helpful to students and parents in their college search journey.
1 Like
Duke’s listed median is nearly double all other colleges in the United States – far higher than all Ivies, MIT, Stanford, etc. It’s also higher median earnings than occurs for all other majors at all other colleges, including majors that are associated with higher earnings than mathematics. This makes suspect a measurement or recording error.
Consistent with this, when I look at the raw data, I see the following for median Duke mathematics major earnings 4-years out.
- Median for Males – $105k/year
- Median for Females – privacy suppressed due to small sample size
- Median for Both Genders – $297k/year
- Median in previous sample year that includes overlaps 1 year of sample above – $99k/year
The obvious explanation is there was a typo – the actual total is $97k/year, not $297k/year. $97k/year is similar to the median in previous sample year and is similar to the median for males in sample year.
6 Likes
I suppose an objection to a 297 median would be that half the class needs to be above 297. It is not enough for one person to be at 297. The only reason a high median is remotely possible is if the class size is small. Like 3 or 5 people (in which case I think they don’t report). It is harder to have a high median in a class of 300, like there are at MIT for a math major.
1 Like
Thats a big oops if that is a mistake.
I agree the underlying data has a lot of limits and so should not be taken too seriously when making college choices.
That said, presenting it in an easy to use package is nice.