Friday, November 21, 2008

The Impact of Employment during College on Academic Performance

Jeffrey S. DeSimone (NBER Working Paper 14006; May 2008) looks at the impact of employment during college on academic performance. DeSimone uses the Harvard College Alcohol Study; waves: 1993-2001, and considers the problem of unobserved heterogeneity. Using two-stage GMM regressions, he instruments work hours using paternal schooling and being raised Jewish, which are hypothesized to "reflect parental preferences towards education manifested in additional student financial support but not influence achievement conditional on maternal schooling, college and class."

DeSimone finds evidence for the identifying assumptions of instrument strength and orthogonality. The GMM results show that an additional weekly work hour reduces current year GPA by 0.011 points, roughly five times more than the OLS coefficient. The effects are stable across specifications, time, gender, class and age, but vary by health status, maternal schooling, religious background and race/ethnicity.


Kevin Denny said...

I haven't read the paper but the instruments sound implausible. Being Jewish influences how much paid work you do but not your educational attainment?
And he uses paternal schooling as another instrument with maternal schooling as a covariate. Oy vey!!!

Martin Ryan said...

It is quite different. I'm replicating the paper for an assessment exercise so I'll comment when I'm finished. The data is quite interesting - the Harvard College Alcohol Study.

Kevin Denny said...

What is quite different from what? The question is: is the identification strategy credible? Its hard to think of a small group of people that has contributed more to scholarship and intellectual life than the Jews. And their food's not bad either.

Martin Ryan said...

Well it may have been better to quote the author: "the main innovation of this study is its identification strategy."

De Simone specifies as instruments variables representing human capital
accumulation and preferences of the respondents’ fathers. The
assumption is that paternal schooling attainment is strongly related to student labor supply, yet otherwise unrelated to academic performance.

To quote the author some more:

"It seems reasonable to expect that paternal schooling has a negative impact on student
labor supply. Fathers with higher attainment likely place a greater value on education, and in
turn might provide more financial support to their college-enrolled children to allow them to
spend less time earning money for tuition and living expenses and more time studying."

"The usefulness of this study, therefore, hinges critically on whether paternal schooling
truly is exogenous with respect to student achievement. This assumption is supported by the
traditional view of the intergenerational human capital transmission literature, which is that child
schooling is much more closely related to maternal schooling than paternal schooling (e.g.
Haveman and Wolfe, 1995; Chevalier et al., 2005). Presumably this stems from children
spending more time with their mothers than their fathers (Black et al., 2005)."

So I guess we have to be believe that father's schooling doesn't affect students' academic achievement. Not a short order...

Martin Ryan said...

And the other instrument is of course being raised Jewish. According to the author:

"students raised Jewish will spend fewer hours working for pay in response to greater financial support from their fathers, who have better
means of providing such support and also emphasize schooling and the eventual attainment of skilled jobs. The logic for assuming that being raised Jewish has no separate correlation with
academic achievement also parallels that for paternal schooling."

Kevin Denny said...

Well I have found more than once that paternal schooling better predicts schooling than maternal schooling since it picks up the permannent income of the household better I expect.But even if maternal is better it doesnt follow that you can use paternal education as an instrument. Its not clear how the logic also applies to the other instrument. These instruments are just not credible.

Martin Ryan said...

I think you're right Kevin. It has been useful to think this through as I may have to discuss this paper for 15 mins. in front of Paul Devereux's Ph.D. class.