Friday, June 19, 2015

Got Data? Probably not like your econometrics textbook!

Recently there has been a lot of discussion of the Angrist and Pischke piece entitled "Why Econometrics Teaching Needs an Overhaul." (read more...) and I have discussed before the large gap between theoretical and applied econometrics.

But here I plan to discuss another potential gap in teaching and application and this is a topic that often is not introduced at any point in a traditional undergraduate or graduate economics curriculum, and that is hacking skills. This becomes extremely important for economists that someday find themselves doing applied work in a corporate environment, or working in the area of data science. Drew Conway points out the there are three spheres of data science including hacking skills, math and statistics knowledge, and subject matter expertise. For many economists, the hacking sphere might be the weakest (read also Big Data Requires a New Kind of Expert: The Econinformatrician) while their quantitative training otherwise makes them ripe to become very good data scientists.

Drew Conway's Data Science Venn Diagram

In a recent whitepaper, I discuss this issue:

Students of econometrics might often spend their days learning proofs and theorems, and if they are lucky they will get their hands on some data and access to software to actually practice some applied work rather it be for a class project or part of a thesis or dissertation. I have written before about the large gap between theoretical and applied econometrics, but there is another gap to speak of, and it has nothing to do with theoretical properties of estimators or interpreting output from STATA, SAS or R. This has to do with raw coding, hacking, and data manipulation skills; the ability to tease out relevant observations and measures from both large structured transactional databases or unstructured log files or web data like tweet-streams. This gap becomes more of an issue as econometricians move from more academic environments to corporate environments and especially so for those economists that begin to take on roles as data scientists. In these environments, not only is it true that problems don’t fit the standard textbook solutions (see article ‘Applied Econometrics’), but the data doesn't look much like the simple data sets often used in textbooks either.  One cannot always expect their IT people to be able to just dump them a flat file with all the variables and formats that will work for your research project. In fact, the absolute best you might hope for in many environments is a SQL or Oracle data base with hundreds or thousands of tables and the tiny bits of information you need spread across a number of them. How do you bring all of this information together to do an analysis? This can be complicated, but for the uninitiated I will present some ‘toy’ examples to give a feel for executing basic database queries to bring together different pieces of information housed in separate tables in order to produce a ‘toy’ analytics ready data set.

I am certain that many schools actually do teach some of the basics related to joining and cleaning data sets, and if they don't then others might figure this out on the job or through one research project or another. I am not certain that this gap needs to be filled  necessarily as part of any econometrics course. However, it is something students need to be aware of and offering some sort of workshop, lab or formal course (maybe as part of a more comprehensive data science curriculum like this) would be very beneficial.

Read the whole paper here:

Matt Bogard. 2015. "Joining Tables with SQL: The most important econometrics lesson you may ever learn" The SelectedWorks of Matt Bogard
Available at: http://works.bepress.com/matt_bogard/29  

See also: Is Machine Learning Trending with Economists?

Wednesday, June 17, 2015

Farmlink and the Rise of Data Science in Agriculture


At a recent Global Ag Investing Conference Dave Gebhardt (Chief Strategy Officer for FarmLink ) spoke about the rise of data science in agriculture. You can read the story and find a link to the podcast here:

In the podcast he discusses the way data science is revolutionizing agriculture, and how we are at a "tipping point where advances in science, IT, technology, and computing power have put a whole new level of opportunities before us."

This sounds a lot like what I have previously discussed in relation to big data and the internet of things: 

Watch more about how FarmLink is leveraging IoT, big data, and advanced analytics:



Related:

 Big Ag Meets Big Data (Part 1 & Part 2)

Saturday, June 13, 2015

SAS vs R? The right answer to the wrong question?

For a long time I tracked a discussion on LinkedIn that consisted of various opinions about using SAS vs R. Some people can take this very personal.  Recently there was an interesting post at the DataCamp blog addressing this topic. They also provided an interesting infographic making some comparisons between SAS and R as well as SPSS.  Other popular debates also include python in the mix. (By the way, it is possible to integrate all three on the SAS platform and you can also run R via the open source integration node in SAS Enterprise Miner 13.1).

Aside: For older versions of SAS EM-can you drop in a code node and call R via PROC IML?

Anyway, getting back to the article, I tend to agree with this one point:

"While these debates are a good thing for the community and the programming language as a whole, they unfortunately also have a negative effect on those individuals that are just in the beginning of their data analytics career. Biased opinions on all sides of the table make it difficult for new data analysts to see the forest for the trees when choosing a statistical programming language."

While I agree with this notion, I want to reflect for a minute on the concept of a programming language. If you think of SAS as just a programming language, then perhaps these kinds of comparisons and discussions make sense, but for a data scientist, I think one's view of analtyics should transcend just a language. When we think of an overall analytical solution there is a lot to consider, from how the data is generated, how it is captured and warehoused, how it is extracted and cleaned and accessed by whatever programming tool(s), how it is visualized and analyzed, and ultimately, how do we operationalize the solution so that it can be consumed by business users.

So to me the relevant question is not, which programming language is preferred by data scientists, or which program is better for implementing specific machine learning algorithms; but perhaps what is the best analytical solutions platform for solving the problems at hand? 

Friday, June 12, 2015

Linear Literalism & Fundamentalist Econometrics

Your tweet stream may have included the recently trending article by Angrist and Pischke entitled: "Why econometrics teaching needs an overhaul". Here is an excerpt:

“In addition to its more up-to-date contents, our book renews the econometrics canon by abandoning the childish literalism of the legacy approach to econometric instruction. In this spirit, we eschew the notion that regression is tied to a literal linear model. Regression describes differences in averages whether or not these averages fit a linear equation. This is a universal property – one that is reliably true – and we don’t intimidate readers with descriptions of the punishments to be meted out for the failure of classical assumptions. Our regression discussion begins by challenging readers to ask themselves, first, what the target causal effect is, and, second, by asking, ‘what is the regression you want’? In other words, what would you like to hold fixed when trying to regress-out an average causal effect?”

They are referencing their text Mastering Metrics, which I highly recommend.  In their other text, Mostly Harmles Econometrics, they also state:

"In fact, the validity of linear regression as an empirical tool does not turn on linearity either...The statement that regression approximates the CEF lines up with our view of empirical work as an effort to describe the essential features of statistical relationships, without necessarily trying to pin them down exactly."  - Mostly Harmless Econometrics, p. 26 & 29

On their MHE blog a reader asks about their pedagogy which focuses on 'best linear projection' vs the traditional BLUE criteria to which they respond:

"our undergrad econometrics training (like most people’s) focused on the sampling distribution of OLS. Hence you were tortured with the Gauss-Markov Thm, which says that OLS is a Best Linear Unbiased Estimator (BLUE). MHE and MM are largely unconcerned with such things. Rather, we try to give our students a clear understanding of what regression means. To that end, we introduce regression as the best linear approximation to whatever conditional expectation fn. (CEF) motivates your empirical work – this is the BLP property you mention, which is a regression feature unrelated to samples. (MM also emphasises our interpretation of regression as a form of “automated matching”)." read more...

The notion of using regression as a means of making like comparisons has also been echoed by Andrew Gelman:

http://andrewgelman.com/2013/01/understanding-regression-models-and-regression-coefficients/

"It's all about comparisons, nothing about how a variable "responds to change." Why? Because, in its most basic form, regression tells you nothing at all about change. It's a structured way of computing average comparisons in data."

Linear literalism or fundamentalist undergraduate econometrics (being tortured with BLUE as A&P might put it) can have long term consequences for students. I think this has caused harms that I encounter from time to time even among more seasoned practitioners and even graduate degree holders. This isn't too different from what Leo Brieman described as a 'statistical straight jacket' that can arbitrarily limit fruitful empirical work. Overly clinical concerns with linearity, heteroskedasticity, and multicollinearity might crowd out more important concerns around causality and prediction.

Heteroskedasticity

We could simplify this as a notion of non-constant variance. As Angrist and Pischke note:

"Our view of regression as an approximation to the CEF makes heteroskedasticity seem natural. If the CEF is nonlinear....the residuals will be larger, on average, at values of X where the fit is poorer...as an empirical matter, heteroskedasticity may matter little" -Ch 3, p.46-47 MHE

 Of course the concern is correct standard errors and valid inference, which can be addressed via heteroskedasticity corrected standard errors. But I am afraid some students, after taking a traditional econometrics course, may terminate all thought processes after a cookbook test hints of its existence.

Multicollinearity

When covariates are highly correlated, it may be difficult to parse out the independent information about each variable and  lead to inflated standard errors. Again this is a phenomena related to inference, not prediction. Even in professional and academic settings when I have presented or attended other presentations related to forecasting or predictive analtyics you will get the occasional criticism or self aggrandizing questions about multicollinearity being a concern.

"Multicollinearity has a very different impact if your goal is prediction from when your goal is estimation. When predicting, multicollinearity is not really a problem provided the values of your predictors lie within the hyper-​​region of the predictors used when estimating the model."-  Statist. Sci.  Volume 25, Number 3 (2010), 289-310.

Undue criticisms and literalism related to multicollinearity often results from a failure to recognize the differences between goals related to explaining vs. predicting.

Paul Allison offers some additional advice on when not to worry about multicolinearity. I have highlighted a couple points of interest here.

This kind of linear fundamentalist paradigm can lead students and practitioners to adopt more complicated methods than necessary or abandon promising empirical work altogether, become overly critical and dismissive of other important work done by others, or completely miss more important questions related to selection bias and identification and unobserved heterogeneity and endogeneity.

Some of this also is a the result of the huge gap between theoretical and applied econometrics.

See also:
Marc Bellemare discusses a similar vein of literalism that is averse to linear probability models here.

Linear Probability Models

Regression as an empirical tool
Quasi-Experimental Design Roundup