As I've mentionned before, a teacher of mine is trying out a new class this semester, and it basically goes like this:

Read published paper as homework => teacher R1's the paper => next paper.

This is going to be the first in a series of posts outlining the badeconomics in published academic literature. I can't take credit for these posts, I'm laying out the arguments of someone with decades of experience. Take these as just incredibly distilled class notes that happen to fit /r/BE's model perfectly.

The article we're going to look at today is "Expansion and Ascription: Trends in Educational Opportunity in Canada, 1920–1994" by Wanner (1999).

RI: A sociologist wrote it

That's it for this week! Next week we'll be showing why empirical results are too feeble to effectively disprove a priori theorems derived from undeniably true axioms.

...

...

...

...

...

...You're still here?

Well shit, I guess I'll have to give it a serious go. In his article, Wanner pores over a large dataset from Canadian surveys to test if parent's socioeconomic status, gender, first language spoken (remember this is Canada) are predictive of "how deep" into the education system a child will go (using OLS and logit models).

The paper might be in a Sociologists' review, but it's effectively econometrics and the findings are important because they act as evidence of the educational system's effectiveness at helping economic mobility.

Wanner follows the sociologist Mare's (1981) model where each level of education is seen as a "stage", and students will decide using a cost benefit analysis at each level if they should continue with their education.

The paper finds that gender and language inequality in education have been practically eliminated in the 20th century, but that economic class inequality in education persisted, despite massive investment in the Canadian education infrastructure.

Importantly, the paper finds results that agree on all counts with an important study by Shavit and Blossfeld in 1993, which find there is no trend in the effect of socio-economic background on higher education.

As levels of education increase in the model, there's a steep decline in the "mother's status" and "father's status" predictors' values in the regression.

While rich parents' kids will pass high school more often, they're not going to grad school much more than poor kids. Not to repeat myself, but these findings are important as far as policy recommendations go, because they obliquely imply that the investment in higher education infrastructure is inefficient.

RI: This paper might have been given a pass, except for the fact that it was written in 1999 and James Heckman and Stephen Cameron wrote this masterpiece of a paper in 1998. The Heckman paper is about unobserved differences in the Mare model we talked about above.

To explain this, lets assume the Mare sequential model where at each level of education you're faced with the decision of working your way to the next level or not. You make a cost benefit analysis taking into account not only the value you will get from subsequent education, but also the likelihood you will pass the next level of education. This likelihood we can refer to as a student's ability (or motivation), which a student can observe but the researcher obviously can't. What Cameron and Heckman show is that not accounting for this unobservable variable causes omitted variable bias and invalidates the results of your regression.

In simple terms, the deeper level of education we look into, the more we get a sample that self selected for higher ability. People who know they don't have the ability to go for a certain level of education likely won't attempt it. So the result that Wanner(1999) and Shavit and Blossfeld's (1993) found where the predictors decline in value as the level of education increase is entirely explainable by the unobserved ability variable (which increases with level of education and corrects for self selection bias).

As such, Wanner's article above gets the official stamp of BadEconomics.