Social Science & Machine Learning

For those interested, I thought it might me nice to start a conversation about how machine learning can be applied to social issues. Specifically I would love to chat about how this can be applied outside the digital world, but also want to know what you all are interested in!


Sounds great! @dsavage maybe you can tell us a little about your background, interests, and thoughts in this area?

This is indeed an interesting topic. Coming from an economic consulting background, the following link is particularly true I believe, and a good starting point:

Most prominent economic theories are always constrained by a lack of clean, observable data. This constraint suited classical economic theory just fine, when we were predominantly interested in how economic systems operate (efficient market hypothesis, Solow growth model, price signals, monetary policy), but present a real challenge to econometrics and behavioral economics. The latter two are data hungry beasts, by design - the more data you have, the better you can identify unintuitive connections and causality in people’s and systems’ behavior, and the better you can quantify these effects.

The computational power of ML, and general data science practices are already being explored by econ researchers worldwide. Someone will hit gold with the combined approach sometime soon, and econometrics will change the way it approaches data as a whole. This is significant to society because econometricians work at everywhere from Amazon to the World Bank, from the White House to Greenpeace; they are usually the ones tasked with quantifying the effects of suggested policy changes.

There are economists who are skeptical, but once the first breakthrough is made it will be easy to convince them. After all, economists are supposed to be rational players, right? :wink:

@jeremy has a good story about economists being interested in his work.


@dsavage great topic!

Economists rely heavily on statistical learning do study various economic phenomena. There is a whole branch in economics - econometrics - which develops the statistical apparatus to study economic problems empirically (i.e. using data). One of the most commonly cited tools by economists are linear and logistic regressions, plus a whole bunch of various statistical tools applicable in design studies (e.g., regression discontinuity design, difference-in-difference approach).

I think the key difference in econometrics and machine learning are not just the tools per se, but the difference in approaches and goals. Economists (and econometricians) always look for causal relationships, they need to test their theoretical hypothesis using data (empirically), rather than get predictions. Also, from my point of view (which might change as I get to know machine learning more :)), econometricians start solving the problem with theoretical assumptions (remember Gauss-Markov? If you wake up an economist in the middle of the night and ask, you’ll get a precise answer :)) before even looking at the data, whereas machine learning starts analysis first looking at the data.

For a long time, I believe, economists have been satisfied with the apparatus that econometrics offers, however, as data gets bigger, economists pretty quickly turn to using the machine learning tools in their analyses. For example, dynamic pricing at Uber or Airbnb is a great problem for an economist, or studying online auctions - all these problems, however, involve dealing with massive sets of data.
One of my friends, who is getting her PhD in economics in Michigan, is currently applying deep learning techniques in labor economics.In particular, using thousands of photos from people’s profiles on a job search site she studies the effects of photo attractiveness on employment outcome (so, take seriously your photo on LinkedIn ;)).

Even in public economic policy design machine learning could be effectively applied (see here.

So, I believe, there is a great intersection between economics, econometrics and machine learning, and machine learning could greatly expand economists’ toolkit and even open opportunities to better understand and predict economic matters. It’s just a matter of time for ML to be used extensively in economics academic and practitioners’ circles.


I love that we have outed some of the Econ majors. I too come from econometrics background, where as @mvasilenko stated there tends to be a leaning towards causal relationships and explanatory power rather than strict predictive power. What I am starting to understand with machine learning however @jeremy correct me if I am wrong is that the two no longer have to separate approaches?

I majored in both Political Science and Economics during my undergrad and always found myself doing traditional statistical analysis when trying to answer the questions I had about human behavior or social trends. In many ways it never felt like I was doing it well enough (might be a personal problem lol). So maybe Machine Learning is the next step in answering these questions more effectively?

It is just hard for me to believe that the big questions surrounding mass incarceration or the economics behind environmental conservation efforts and other issues have to be answered in these traditional manners. Thoughts?

I wonder if this would be a good resource for folks interested in this intersection?:


Here are some interesting case studies by a startup ‘Social Cops’ in India about how they are using data in social sector by collaborating with public and private counterparts.

If you are interested in Kaggle-ish competitions on social challenges, do check out

1 Like