Thursday, June 10, 2010

But what do the data say? Female labor force participation...

In my last post I asserted that female labor force participation (FLFP) in Turkey was low given their level of economic development and that most Turks are Muslim. But what do the data say?

The graph below (click it to enlarge) shows income per person on the horizontal axis (log scale, PPP), while the vertical axis shows FLFP (women working at home are not counted as participating). The curve shown is the predicted level of FLFP - high in very poor countries, lower in middle income countries, and slightly higher in rich countries.

Turkey is the red dot, below the curve, meaning fewer women work than expected - so something other than the income level explains their low FLFP. Note that the dot size corresponds to population (China and India were omitted because their enormous populations allowed them to bully the results).

Graph 1: Female labor force participation vs. income per capita


The next graph shows the difference between the actual and predicted values - it simply takes the data from the previous graph and displays the vertical distance between each point and the curve. Imagine taking the above graph, bending the curve until it is straight, then placing it on the horizontal axis. Points right on the horizontal axis were predicted perfectly - so clearly there are large prediction errors for many countries - the income level does not explain everything!

Graph 2: Difference in actual and expected FLFP vs. income


A casual look at the above graph shows that many of the countries that have lower FLFP than predicted are predominantly Muslim. The graph below controls for the percentage of Muslims in each country, and yields much better predictions. The religion data is from worldmapper, which was the most comprehensive source I could find, but I don't know its accuracy. Note that once % Muslims is controlled for, Turkey actually has a higher FLFP than predicted (it is above the x-axis).

Graph 3: Difference in actual and expected FLFP vs. income, controlling for Muslim population


Finally, the graph below controls for percentage of Sunnis and Shias in each country, instead of lumping them together as Muslims. I also controlled for Catholics, Protestants, rural population, male labor force participation, and the age distribution. Now the model fits the data quite closely (R2=0.69).

Graph 4: Difference in actual and expected FLFP vs. income, controlling for Sunni and Shia population


So in the final analysis, Turkey actually has a greater female labor force participation rate than expected! Their level is 7.4 percentage points higher than the model predicted (25.1% vs. predicted 17.7%). Is this the Atatürk effect?

This means my prior suspicions were wrong!

But why did I come all the way to Turkey to crunch a bunch of numbers that I could access from the US? I did not - this background analysis is merely a starting point to understand this issue in Turkey's economic and cultural context. More to come.

Now the questions are - why is Turkey's FLFP higher than expected? But more importantly - what can Turkish policy makers and employers do to increase female labor force participation? Better education and vocational training? More creative use of existing talent? And, what do the women say? Do they want to work? Time for some qualitative analysis?