4 ms·

I find the creation of physics-inspired sociological models distasteful. It's often a way of manufacturing certainty and authority through mathematical obsfucat

by knzhou 7y ago

I find the creation of physics-inspired sociological models distasteful. It's often a way of manufacturing certainty and authority through mathematical obsfucation. It always runs the same way. First, you think of some political opinion you want to support. Then you come up with a mathematically trivial toy model that throws away 99% of the structure of the problem, whose few dynamical rules are artificially chosen so that they obviously will produce the result you want. (Though this part may look intimidating, it's not once you know how the game is played; I can easily cook up a toy model that "proves" any given political opinion upon request.) Then you put a lot of energy into solving the model in all kinds of cases, showing off your mathematical skills. The end result of all such social models is a set of predictions that are only tenuously linked to reality. A common example of such a claim is "X follows a power law". But the models are usually too weak to actually calculate the power in the power law, and the data too tenuous to even show you have a power law at all. (Most of the time this is done by showing a fuzzy cloud of points on a log-log plot with a tiny range, drawing a line through it (ignoring the fact that literally anything can be fit with a line if you zoom in enough) and declaring victory.) The point of mathematics is to make your postulates and predictions precise. If your postulates are obviously unrealistic, and your predictions are too vague to be meaningful, then the mathematics isn't serving any purpose -- except for giving your work a false authority.

anotherone15 7y ago
Would you rather sociological models not contain any physics or math? What do you propose?
- knzhou 7y ago
  I would prefer an empirical approach, as some take in macroeconomics. Just parametrize things you care about in terms of measureable quantities, and go out and measure them. Once you have that, there's no need to slap on some model that says, like people are a frustrated antiferromagnetic spin lattice on a random small world graph. It doesn't add anything. If you want to explain the results of the measurements, you can do it just as well with words.
  - anotherone15 7y ago
    What about parameters that are hard to measure? Take for instances infectious disease models, often a very simple set of ODEs that are capable of capturing the wide ranging dynamics that we observe. One of the crucial parameters in these models is the transmission parameter, which determines the rate at which susceptibles become infected upon contact with an infected. Transmission is a function of a long string of physical, chemical, biological, social and economic variables. I am not sure how feasible it is to determine from first principles this parameter and so it is obtained from fitting the model to data. So in this case, one does not directly measure the parameter, but infer it from the data conditional on the model.

by knzhou 7y ago

anotherone15 7y ago

Would you rather sociological models not contain any physics or math? What do you propose?

knzhou 7y ago
I would prefer an empirical approach, as some take in macroeconomics. Just parametrize things you care about in terms of measureable quantities, and go out and measure them. Once you have that, there's no need to slap on some model that says, like people are a frustrated antiferromagnetic spin lattice on a random small world graph. It doesn't add anything. If you want to explain the results of the measurements, you can do it just as well with words.
- anotherone15 7y ago
  What about parameters that are hard to measure? Take for instances infectious disease models, often a very simple set of ODEs that are capable of capturing the wide ranging dynamics that we observe. One of the crucial parameters in these models is the transmission parameter, which determines the rate at which susceptibles become infected upon contact with an infected. Transmission is a function of a long string of physical, chemical, biological, social and economic variables. I am not sure how feasible it is to determine from first principles this parameter and so it is obtained from fitting the model to data. So in this case, one does not directly measure the parameter, but infer it from the data conditional on the model.