Here we go again - more quantitative self-hatred

So, here we go again. Discover Society has decided to publish another piece of quantitative self-hatred, this time by the American sociologist Brian Castellani. It covers similar ground to David Byrne's piece in Sociology that I wrote about here, which is only to be expected as Catellani and Byrne seem to be close associates. It repeats the same old contentious claims  about the alleged deficiencies of the standard toolbox of quantitative methods and combines them with a complete lack of understanding of the actual state of affairs in British sociology departments. On the plus side Castellani's piece is less intemperate than Byrne's ridiculous polemic so every cloud has a silver lining I suppose. 

Still, we should probably feel a bit of sympathy for Castellani because as a student he seems to have suffered from the intellectual equivalent of child-abuse. He tells us that: "...my quantitative professors argued, statistics (and pretty much it alone) made 'sense' of the complexity of social reality." If they really forced that kind of garbage down his throat then I genuinely feel sorry for him. What can I say? He was clearly taught by people who didn't know what they were talking about and it has soured his whole understanding of statistics. That would explain sentences like: "And being a good quantitative social scientist, you would develop as simplistic a causal model as possible, what Capra and Luisi (2014) call mechanistic or reductionist social science." We're not pulling our punches then. People doing conventional  quantitative social science prefer simplistic models - I assume Castellani is aware of the difference in nuance between simple and simplistic - and what they are doing is mechanistic and reductionist (cries of boo and hiss from les enfants du paradis). Jesus, those quant guys are so primitive it's amazing they can stand on two legs (not like us super evolved complexity guys because complex is always better than simplistic right?).

Well, maybe, but it doesn't seem to imply an ability to stick to the point or to follow a coherent line of argument. What for example is the point of the figure, culled from Wikipedia,  showing various Gaussian probability density functions? It has no clear relationship to anything that Castellani is discussing at the point it is presented to us. Is the idea just to serve up to the home crowd some totemic figures attributable to the enemy  that they can focus their hatred on? Who knows? Who cares? Well, in a complex world seemingly anything goes,  after all Castellani tells us that students, instead of being force fed that quant methods hogwash should have been following a curriculum stuffed with post-positivism, post-structuralism, eco-feminism, deconstruction, constructionism, constructivism, qualitative method and post-modernism. But hang on, isn't that what most British undergraduate sociology courses actually consist of? Come on Brian, pay attention.

There are however, apparently, beacons of hope. Though Oxford, Harvard, Stanford and Cambridge feed their undergraduates a diet of "...mechanistic and reductionist experimental or quasi-experimental design" - Oh the brutes! - the School of Political and Social Science at the University of Melbourne, Science Po in Paris and, the BSc in Sociology at the LSE are held up as examples to us all. In these enlightened institutions, we are told, "...undergraduates are given courses in critical thinking, applied research and interdisciplinary and mixed methods." The gods be praised!

Well, fair dinkum to Melbourne, I can't gainsay 'em because I don't know what they do. Science Po I should know something about because my own institution has an exchange programme with them and I've supervised one or two students on it. I must say, they seemed perfectly normal to me. Not especially enlightened nor especially disdainful of the thin gruel they are obviously being served up in Oxford. And then the BSc in Sociology at the LSE. Well, there's something I do know about. Having been an undergraduate on it 35 years ago, I actually then taught on the degree for ten years - albeit over a decade ago - and am still in intimate contact with sources currently involved in delivering it. Let's say I have a pretty good idea of what the reality looks like on the ground. The rules of good manners and the UK's libel laws prevent me from saying at this point all of what I actually believe to be true. But let me put it this way: the department that hosts the degree is also home to that monument to absurdity resting on a  pedestal of  incompetence called the Great British Class Survey. If the future of quantitative methods  in the UK relies on the skills and understanding of the people associated with that, then God help us, we're doomed.

What we're doomed to is however not entirely clear from Castellani's account. Of course he has his diagram that explains everything. It's a fantastic green and white web of names that starts with Issac Newton in the top left-hand corner who apparently was working in the 1940s  and 50s (you might have to brush up the periodization of your  history of science a little there Brian) and ends up with Emma Uprichard in the bottom right hand corner. In between is more or less everything and everybody plus the kitchen sink, serious thinkers and doers along with notorious bullshit merchants. 

This actually tells me nothing and there's the rub. If you are going to persuade me that I need to learn a new set of tools - let's call them complexity science - to answer the questions I'm interested in then I think I'm entitled to say, OK, but first show me the money. Please show me, by producing convincing empirical evidence, that these tools give better answers to the sorts of questions I and other social scientists are interested in. I don't mean by this demonstrations that the GLM is an inappropriate model for understanding weather systems or the swarming behaviour of birds and fish. I know that. I also don't want a philosophical disquisition about ontology. I want to see the pay-off for standard social science questions. I also don't want demonstrations that RCTs and linear regressions are not appropriate for all questions . Nobody in their right mind thinks they are. In other words cut out the endless - in some cases book length - chat and demonstrate the empirical value of complexity tools when it comes to the sort of questions that most quantitatively oriented sociologists are interested in.

If the answer is that they add little or no value for the conventional questions, then that's fine. If your question is about networks then you need network models - nobody is going to dispute this. I went to my first course on social network analysis in 1989 and very good it was too. Since then I've not worked on problems that involved the analysis of network data so there was no particular pay-off other than gaining an appreciation of what can be done. I also have colleagues whose substantive work involves the use of network models, who develop the statistical theory to make inferences about networks and write the software to do it. None of them feel the need to disparage conventional quantitative methods or feel the need to shroud what they do in a sea of verbiage. I could say much the same sort of thing about colleagues whose work involves the use of agent-based models or system dynamics models. They have these things in their tool box and they take them out when they are the right tool for the job and when they can be shown to work.

Though he doesn't present any in his article probably Castellani has a whole bunch of convincing applications that he can pluck from his CV.  I don't know, I haven't looked at it. I'm less sure that Byrne and associates could do that. Let's face it, this is a guy who tries to tell us that whatever it is we're doing is all wrong yet is  unable to work out correctly the size of his state space (he's out by a factor of 10). In all serious disciplines that would earn you a one way ticket to Palookaville. In sociology it get's you a chair. It also seems that in practice complexity science for Byrne comes down to QCA and cluster analysis. That's odd, for when I look at what seems to me to be the serious stuff on complexity in the social sciences  - for instance at the Santa Fe Institute -  it's all differential equations and random matrices. I think it would be great if we taught our sociology undergraduates that. But then again given that most have given up maths at 16 and break out in a cold sweat if you show them a summation sign this would be a tall order. Probably better to go for rubbishing the modest but tangible achievements of conventional quantitative methods and give them a bit of waffle from Luhmann, Latour, Castells, Urry, Thrift and so on. Then they'll really understand what is going on in the world.