Revaluing the German Model

Six years ago, when I first started studying the German economic model (the soziale Marktwirtschaft, or social market economy, also known as the Rhenish model, ordoliberalism, or the stakeholding model), it was in serious disrepute.  Unemployment seemed permanently high, taxes were high, and it was claimed by many pundits that the soziale side of the equation was choking the Markt's efficiencies in a case of terminal Eurosclerosis.

But during the economic crisis of the last couple of years, the German model's virtues have become apparent, and it is back in favor.  Angela Merkel now proclaims from the Davos podium that other developed countries need to follow the German example.

Even The Economist is singing Germany's praise.  The current issues has three pieces on Germany, calling the economy "a machine running smoothly," a success case that "owes more to judgment than to luck," and a model to be emulated.  As The Economist's chart here shows, it has been a splendid decade for Germany.

The Economist, true to their liberal perspective, gives too much weight to the effects of capital and labor market liberalization.  These are certainly important factors, but, as I have argued in previous posts, Germany's current success owes much to the ethos of Mittelstand companies (often family-owned and valuing long relationships with employees) and programs such as the Kurzarbeit system that kept many Germans working, albeit with reduced hours, during the downturn.

Even Merle Hazard is (literally) singing Germany's praises

Ethnographic Science and Bayesian Statistics

Is anthropology a science?  As I suggested below, a lot rests on how one uses the word “science,” and I follow Deirdre McCloskey in an ecumenical, expansive definition.

But, even taking a stricter understanding of science, can ethnography based on participant observation pretend to be as rigorous as random sample surveys and rigid statistical analysis?  I think so, and to understand why we need to look back to Thomas Bayes’ (1702-1761) approach to statistics.  

Bayes was a pioneer of statistics, but in the pre-probablistic era of the field.  So, Bayesian statistics works off fewer and less random data points.

Thomas Griffiths and Joshua Tenenbaum (2006, an article in Psychological Science) have taken a surprising look at how folks actually make predictions and found support for a Bayesian model.

Griffiths and Tenenbaum came up with experiments in which they presented participants with an isolated piece of information and ask them to make a generalized inference from it:

Movie grosses: Imagine you hear about a movie that has taken in 10 million dollars at the box office, but don’t know how long it has been running. What would you predict for the total amount of box office intake for that movie?

Poem lengths: If your friend read you her favorite line of poetry, and told you it was line 5 of a poem, what would you predict for the total length of the poem?

Terms of U.S. representatives: If you heard a member of the House of Representatives had served for 15 years, what would you predict his total term in the House would be?

Baking times for cakes: Imagine you are in somebody’s kitchen and notice that a cake is in the oven. The timer shows that it has been baking for 35 minutes. What would you predict for the total amount of time the cake needs to bake?

Waiting times: If you were calling a telephone box office to book tickets and had been on hold for 3 minutes, what would you predict for the total time you would be on hold?

It turns out that people’s predictions about such things were extremely accurate.  Significantly, in all of these items have well-established probability distributions (normal, Erlang, power-law, poisson, etc.), distributions that people know intuitively and thus are able to place lone pieces of data into these internalized distributions.  This allows accurate predictions from lone pieces of data.

Working with good priors, Bayesian inferences are at the heart of ethnography.

Ethnographers who spend lots of time in a place and a lot of time talking to people there, internalize distributions and probabilities. For certain domains they become almost intuitive. Turns out we are pretty good at estimating those things with just one data point BECAUSE we have a pretty good distribution curve in our minds already.