Home > data science, finance, math education, news > A modeled student

A modeled student

February 16, 2012

There’s a recent article from Inside Higher Ed (hat tip David Madigan) which focuses on a new “Predictive Analytics Reporting Framework” that tracks students’ online learning and predicts their outcomes, like whether they will finish the classes they’re taking or drop out. Who’s involved? The University of Phoenix among others:

A broad range of institutions (see factbox) are participating. Six major for-profits, research universities and community colleges — the sort of group that doesn’t always play nice — are sharing the vault of information and tips on how to put the data to work.

I don’t know about you but I’ve read the wikipedia article about for-profit universities and I don’t have a great feeling about their goals. In the “2010 Pell Grant Fraud controversy” section you can find this:

Out of the fifteen sampled, all were found to have engaged in deceptive practices, improperly promising unrealistically high pay for graduating students, and four engaged in outright fraud, per a GAO report released at a hearing of the Health, Education, Labor and Pensions Committee held on August 4, 2010.[28]

Anyhoo, back to the article. They track people online and make suggestions for what classes people may want to take:

The data set has the potential to give institutions sophisticated information about small subsets of students – such as which academic programs are best suited for a 25-year-old male Latino with strength in mathematics, for example. The tool could even become a sort of Match.com for students and online universities, Ice said.

That makes me wonder- what would I have been told to do as a white woman with strength in math, if such a program had existed when I went to college? Maybe I would have been pushed to become something that historical data said I’d be best suited for? Maybe something safe, like actuarial work? What if this had existed when my mother was at MIT in applied math in the early ’60’s? Would they have had a suggestion for her?

Aside from snide remarks, let me make two direct complaints about this idea. First, I despise the idea of funneling people into chutes and ladders-type career projections based on their external attributes rather than their internal motives and desires. This kind of model, which as all models is based on historical data, is potentially a way to formally adopt racist and sexist policies. It codifies discrimination.

The second complaint: this is really all about money. In the article they mention that the model has already helped them decide whether Pell grants are being issued to students “correctly”:

Students can only receive the maximum Pell Grant award when they take 12 credit hours, which “forces people into concurrency,” said Phil Ice, vice president of research and development for the American Public University System and the project’s lead investigator. “So the question becomes, is the current federal financial aid structure actually setting these individuals up for failure?”

In other words, it looks like they are going to try to use the results of this model to persuade the government to change the way Pell Grants are distributed. Now, I’m not saying that the Pell Grant program is perfect; maybe it should be changed. But I am saying that this model is all about money and helping these online universities figure out which students will be most profitable. I’m familiar with constructing such models, because I was a quant at a hedge fund once and I know how these guys think. You can bet this model is proprietary, too- you wouldn’t want people to see into how they are being funneled too much, it might get awkward.

The article doesn’t she away from such comparisons either. From the article:

The project appears to have built support in higher education for the broader use of Wall Street-style slicing and dicing of data. Colleges have resisted those practices in the past, perhaps because some educators have viewed “data snooping” warily. That may be changing, observers said, as the project is showing that big data isn’t just good for hedge funds.

Just to be clear, they are saying it’s also good for for-profit institutions, not necessarily the students in them.

I’d like to see a law passed that forced such models to be open-sourced at the very very least. The Bill and Melinda Gates Foundation is funding this, who know how to reach those guys to make this request?

  1. February 16, 2012 at 9:44 am

    How Khan Academy is using Machine Learning to Assess Student Mastery



  2. February 16, 2012 at 10:16 am

    Monetizing the data acquired from online learning with analytics is being tested:

    Felix Salmon discusses Udacity’s business model:


    “But still, online education is young enough that it’s worth trying many different models to see which ones work. Udacity seems to be built on the standard VC model of get scale first, worry about monetizing it later. And if Udacity does end up with millions of students, I should imagine that there are quite a lot of companies which would pay Udacity to be able to reach those students. Simply charging technology companies to put job opportunities in front of students with given grades and qualifications would probably generate quite hefty fees. So long as the education itself remains free, I don’t think that being a for-profit is in and of itself a bad thing.”


  3. February 25, 2012 at 9:42 pm

    Hi Cathy – thanks for sharing your thoughts on this. I’m equally concerned about student modelling being used incorrectly or to prematurely shuttle students into areas that they really don’t belong (or want to be involved with).

    In terms of your point about these models being open sources – I agree completely!! You might find a concept paper that we posted online last year about “open learning analytics” interesting in this regard…: http://solaresearch.org/OpenLearningAnalytics.pdf


  1. February 21, 2012 at 7:04 am
  2. February 23, 2012 at 8:46 pm
  3. February 24, 2012 at 7:35 am
Comments are closed.
%d bloggers like this: