Tuesday, February 16, 2016

How to (mis)use econometrics

Apparently these two pieces of writing (in Indonesian) on LGBT and economic growth hit the nerves of those who have taken at least elementary econometrics course - and/or have been associated with the School. I could not blame them, they have all the reasons to get really annoyed by such sloppy regression analysis.

Take this bit directly from the second article above:
Korelasi menunjukkan arah, sementara regresi menunjukkan hubungan fungsional antar variabel.
Correlation shows the direction [RS: of changes between two variables, perhaps], while regression shows functional relationship between variables.
Wait. Simply running regression would not tell you why the 1 percent increase in defunct economists' access to Facebook leads to 2 percent increase in the ignorance of practical men who believe themselves to be quite exempt from any intellectual influence (HT: J.M. Keynes, circa 1936).

Regression per se will not provide you with a meaningful functional relationship. For that, you need a convincing and logically consistent story line - a theory, on which your empirical model is based. Isn't it just stating the obvious, you say? Yes, of course. First week of Method 101.

But maybe the author thinks that he has a story on how government inclination or support to LGBT leads to slower economic growth [RS: not sure if it's a level or growth effect, but never mind]. Here is what I found in the first writing:
... kecenderungan LGBT yang semakin besar di sebuah negara akan berdampak kepada kondisi kependudukan yang memburuk. Hal ini dapat dijelaskan dari fakta terang benderang bahwa pasangan LGBT tidak dapat menghasilkan keturunan .... Kondisi kependudukan yang memburuk tersebut pada gilirannya akan menghambat ekonomi untuk terus tumbuh.
... growing LGBT population in a country would deteriorate its demographic situation. This can be explained by the crystal-clear fact that LGBT couple can not reproduce... This demographic deterioration, in turn, will prevent the economy to grow.
So I WhatsApp our certified labor economist in the house, barista Ujang, to send me the main literature on the determinants of fertility.

He replied right away and emailed me two major references in the profession: Hotz et.al, 1997,  The Economics of Fertility in Developed Countries (.pdf), and Schultz, 2007, Fertility in Developing Countries (.pdf).

My reading: first, theories, models, and empirical strategies to identify and estimate the determinants of fertility are anything but simple nor easy. Second, there is no discussion on LGBT factor as determinant of fertility rate - if you want to put LGBT into the equation, you have a lot to explain.

OK, fine. With a controversial finding, maybe we can look at the data now?

Unfortunately, it is available either only in a certain office room in Depok and/or if you are willing to co-(re)write the paper -- despite, somehow, the regression results are already widely circulated.