Home > data journalism, economics, education > How to think like a microeconomist

How to think like a microeconomist

July 23, 2014

Yesterday we were pleased to have Suresh Naidu guest lecture in The Platform. He came in and explained, very efficiently because he was leaving at 11am for a flight at noon at LGA (which he made!!) how to think like an economist. Or at least an applied microeconomist. Here are his notes:

Applied microeconomics is basically organized a few simple metaphors.

  1. People respond to incentives.
  2. A lot of data can be understood through the lens of supply and demand.
  3. Causality is more important than prediction.

There was actually more on the schedule, but Suresh got into really amazing examples to explain the above points and we ran out of time. At some point, when he was describing itinerant laborers in the United Arab Emirates, and looking at pay records and even visiting a itinerant labor camp, I was thinking that Suresh is possibly an undercover hardcore data journalist as well as an amazing economist.

As far as the “big data” revolution goes, we got the impression from Suresh that microeconomists have been largely unmoved by its fervor. For one, they’ve been doing huge analyses with large data sets for quite a while. But the real reason they’re unmoved, as I infer from his talk yesterday, is that big data almost always focuses on descriptions of human behavior, and sometimes predictions, and almost never causality, which is what economists care about.

A side question: why is it that economists only care about causality? Well they do, and let’s take that as a given.

So, now that we know how to think like an economist, let’s read this “Room For Debate” about overseas child labor with our new perspective. Basically the writers, or at least three out of four of them, are economists. So that means they care about “why”. Why is there so much child labor overseas? How can the US help?

The first guy says that strong unions and clear signals from American companies works, so the US should do its best to encourage the influence of labor unions.

The lady economist says that bans on child labor are generally counterproductive, so we should give people cash money so they won’t have to send their kids to work in the first place.

The last guy says that we didn’t even stop having child labor in our country until wage workers were worried about competition from children. So he wants the U.S. to essentially ignore child labor in other countries, which he claims will set the stage for other countries to have that same worry and come to the same conclusion by themselves. Time will help, as well as good money from the US companies.

So the economists don’t agree, but they all share one goal: to figure out how to tweak a tweakable variable to improve a system. And hopefully each hypothesis can be proven with randomized experiments and with data, or at least evidence can be gathered for or against.

One more thing, which I was relieved to hear Suresh say. There’s a spectrum of how much people “believe” in economics, and for that matter believe in data that seems to support a theory or experiment, and that spectrum is something that most economists run across on a daily basis. Even so, it’s not clear there’s a better way to learn things about the world than doing your best to run randomized experiments, or find close-to-randomized experiments and see how what they tell you.

  1. July 23, 2014 at 7:35 am

    Point 3 reminds me a little bit of point 3 in this talk Charlie Munger gave at Harvard in the 90s



  2. Hernan
    July 23, 2014 at 7:49 am

    I have been thinking recently about these two perspectives: (1) using historical data to make a prediction about future realisations of some variable and (2) using historical data to understand whether variable X causes Y and to which extent. Let’s ignore randomised experiments and imagine we only have economic, social, business data. To me those two perspectives were very clear. In my professional life, I have worked with both kinds of models, but mostly type (2).

    If I want to predict (type 1), I optimise the model according to some prediction criteria in out of sample data to make sure I am not overfitting. Often these models are good at out-of-sample prediction, but they are not robust if there is a structural change in the system. For instance, if there is a tax increase, or some firms go bankrupt, or a new law comes into place, the past calibration of the model will not work.

    By contrast, the economic models that attempt to untangle cause and effect, are often concerned with estimating the so-called economic primitives (e.g. the cost curve of firms, the preferences of consumers making product choice, the choice of jobs, etc.) based on the observed outcome of some assumed economic process (supply/demand, maximise utility, etc.). These models are not great at predicting the next out of sample observation. But because they unearth the economic primitives, the researcher can use those estimates to “solve” the new economic model under the new structure. So these models are useful for evaluating policy decisions of governments or organisations.

    My view right now is that prediction is all we care about. What changes when we move from type (1) to type (2) is under which conditions we must make those predictions. The only reason we want to understand causality and the laws of nature is to be able to predict the outcome of future events (yeah, ok, there is also the intrinsic curiosity). In type (2) we want to use data to make predictions under conditions that we have not yet observed. This involves adding plenty of economic and statistical assumptions and giving up predictive ability if things don’t change. Example: if you want to predict next year’s unemployment, you would use a type (1) model (e.g. VARs, etc.). If you want to predict next year’s unemployment in case the minimum wage goes up by 20%, then you would better off with type (2) model because the structural change will influence how many people look for work, how many people companies want to hire, etc.


    • July 23, 2014 at 8:06 am

      I think you have a convincing case for why economists are so concerned with causality. Big data people often focus on nudging people towards buying or clicking, and they don’t care about making big policy changes and guessing what their customers will do.


  3. July 23, 2014 at 7:19 pm

    I would ask the economist: What incentives can be given to put an end to female genital mutilation?


    • Guest2
      July 24, 2014 at 9:53 am

      Something so deeply tied to notions of the sacred are also highly institutionalized, making it resistant to micro-forces. A good example of the limitations of the micro approach.


  4. Min
    July 24, 2014 at 4:22 am

    Some quick responses:

    “People respond to incentives.”

    That pretty much true by definition. What do economists mean by that? IMX, when people with some training in economics use that phrase — not necessarily economics Ph. D.’s, who may use it differently than those with less training –, they mean that people respond to monetary incentives, or that the economics view of human nature is correct, that people maximize their utility in terms of narrow self interest. I somewhat hazily recall a story about a couple of economists eating dinner at a restaurant with a couple of anthropologists, and the question of picking up the tab came up. The economists valued not picking up the tab because they would get the meal for free, while the anthropologists valued picking up the tab because that meant that they had higher status. (The story may be apocryphal, OC. :)) Aside from the question of narrow self interest vs. enlightened self interest, my impression is that economists’ understanding of human motivation is too simple, and not well grounded in observation.

    “Causality is more important than prediction.”

    What is causality? In a deterministic universe, every event that we observe is both cause and effect, and every event has multiple causes and effects. Furthermore, it makes little difference whether the past causes the future or vice versa. Everything just is, in an indicative rather than temporal sense. Causality is how we give reasons for things, how we explain data. To put it another way, we start with a factual description of the data and end up with a condensed description utilizing certain concepts. Now, my understanding of causality is intrinsically related to prediction. That is, if I know some data and I know relevant causal concepts, I can predict other data. And I can test different causal theories by how well their predictions fit the data. To say that causality is more important than prediction is like saying that a good story is more important than the facts. I don’t think that is what Naidu meant. It would have been interesting to hear what he had to say.

    As for child labor, I have some thoughts based upon what I read years ago. I am not going back to check anything now. Our modern view of child labor, indeed, our view of the child, is the result of the industrial revolution. In an agrarian society, there was no concept of child labor. Everybody pitched in when there was work to do; children had chores. Adolescence as we know it did not exist. There was a smoother transition between child and adult through apprenticeship. Adolescence is the result of treating teenagers as more dependent and less responsible than they are capable of being. It is not just “hormones”. Rather, as Ruth Benedict, pointed out in general, adolescence is the result of having a discontinuity between childhood and adulthood.

    As the industrial revolution made farm hands into factory workers, the labor of men and animals was devalued in general. As machines did the heavy work, children were often used to tend the machines. If machines were packed into the space available, young children might be the only ones who could fit under and between them. Besides, children were cheaper to hire and more docile than adults. They were often worked 12 or more hours per day. Yes, some children were maimed and killed, but that is the price of progress. Meanwhile, the fathers of the children may have been without jobs.

    So yes, the third guy is correct that competition was a factor in ending child labor. But that is not all that was happening. Both children and adults were suffering in a way and at a scale that was unheard of before the industrial revolution. The social changes that ended child labor were profound. Again, the view of economics strikes me as simplistic. And the idea that other societies will have the same worries as we do and abolish child labor on their own in the relatively near future seems either naive or disingenuous. As is the idea that sending our money to the factory owners will help produce that change. What incentive do the factory owners have to abolish child labor?

    Now, we may not have the right to impose our values on other societies, even if we regard child labor and sweatshops with revulsion. But that does not excuse our responsibility for those conditions if we buy products made under them. After all, we would be providing monetary incentives for those conditions to continue. We would participate in causing suffering that we would not tolerate in our own societies. Causality is responsibility.


  5. Zathras
    July 24, 2014 at 9:13 am

    For the 3 elements given above, the qualification should be given that these apply to academic microeconomists. There is a whole separate sphere of economists who work in the private sector for who, these rules do not apply, especially about caring about causality over predictions.

    It is in fact a difficult task who transition from the academic to the business world to understand what is needed for predictive microeconomists. They usually start with an idea like Hernan’s above “If I want to predict (type 1), I optimise the model according to some prediction criteria in out of sample data to make sure I am not overfitting.” However, even absent structural change, the out-of sample testing for a predictive microeconomics model is a very small part of the analysis of whether the model makes good predictions. There are fundamental issues with data in the time domain that do not come up either in academic microeconomics or in traditional predictive analytics.


  6. Michael L
    July 24, 2014 at 10:28 am


    As a person who does mathematical modeling, I wonder what you think of the “understanding a lot through the lens of supply and demand” idea. I’m not a professional economist but I’ve studied a fair amount of it. As I understand it, the idea of prices being determined by supply and demand is derived from the assumption of a perfectly competitive market. Yet many markets in the world aren’t perfectly competitive so one wonders to what extent the interaction of supply and demand determining prices idea even applies to them, let along things like whom people choose to marry or whether a person decides to have kids (other things that economists often treat as market like).

    Take the minimum wage debate. Labor markets aren’t perfectly competitive. Economists know this because they’ve developed more sophisticated models of labor markets that are based on the assumption that the labor market isn’t perfectly competitive. Yet when it comes to predicting the effect of a minimum wage, many economists still rely on the model of supply and demand determining prices, a model based on the assumption of perfect competition. It’s like a physicist who states that Newton’s laws only apply when things are moving at a small fraction of the speed of light but then uses those laws to make predictions about something traveling near light speed.


  7. Michael L
    July 24, 2014 at 9:32 pm

    Apology for typing “Kathy” instead of “Cathy”.


  8. July 25, 2014 at 12:24 pm

    TIL 2 out of 3 economists are paid stooges.

    Why say, this guy, this lady, and that guy, when you can take the microeconomists’ approach and note the financial incentives the researchers themselves face. Here are their CVs: http://leahlakdawala.weebly.com/cv.html and http://faculty.wcas.northwestern.edu/~mdo738/vita.pdf


  1. July 30, 2014 at 4:21 pm
Comments are closed.
%d bloggers like this: