I expect that I'm unable to provide a completely convincing argument to counter the claim that social science isn't really science, but I want to talk about it a bit anyway. Or, more accurately, I want to quote Howard Becker talking about it.

But, before that, I want to note that I think sometimes the topic gets muddled because it's not very clear to people what the boundaries of "social science" are, especially in relation to the Humanities. This confusion probably owes in large part to the "social sciences" and "humanities" influencing each other, whether in the form of postmodernist or feminist philosophy informing debates about positivism within social sciences, or the proliferation of various "cultural studies" (gender studies, African American studies, etc.) programs in the humanities.

I would not argue that literally every publication or work dealing with society and culture is scientific, although I would also not argue that all of it needs to be in order to be worthwhile. Instead, my more modest goal is to try outline some general thoughts on how at least some work in the social sciences can be said to be scientific, with much the same commitments to scientific epistemology as other sciences, although the social sciences also face some unique challenges.

With that in mind, the title of this thread draws on the 2017 book Evidence, by the sociologist Howard Becker, who I recommend very highly. This book is in large part a critique of flaws common to sociological research. But it also provides a useful philosophical outline of the social sciences qua science, and I hope thinking about the relationship between these three concepts of Data, Evidence, and Ideas will be useful. Here's how Becker introduces the subject:

Becker goes on to detail the difficulties Wallin and Waldo ran into when they tried to use the data gathered via this questionnaire, largely tied to an inability to classify nearly 25% of respondents. The example is more broadly useful, though, because it concisely demonstrates so many of the methodological challenges of social science, for example in the selection of methods intended to operationalize complex concepts like "social class" in order to measure them, so as to be able to draw valid inferences from those measurements, to support various theories.

Broadly then, my argument is that "social science" is in fact scientific to the extent that researchers employ methods to ensure the reasonable validity and reliability of the data they collect, in relation to what ideas they wish that data to serve as evidence for. So, in the example of Wallin and Waldo, the problem (and it's a common one) they ran into was a disconnect between their data and the idea of social class. Problems with the data made it unsuitable as evidence of social class. Most of social science methodology is about overcoming similar problems, by standardizing approaches known to work well for producing data, e.g. in the use of surveys, structured interviews, content analysis, and so on.

When Becker writes that we wish to make an argument about our data that "no one could have recorded reality in that form if our argument wasn't correct", he hits on probably the most important objection, again evident in the example of social class: it is too complex of a concept to admit of any simple measure. It's not like trying to measure the mass of an object. Social science data are rarely so conclusive as to admit of only a single explanation. But, I would argue that even if social sciences must sometimes embrace more modest goals than some physical sciences, they are nonetheless science. Here Becker cites the work of the mathematician George Polya:

Essentially, then, the point here is that there is not a difference in kind between so-called "hard" and "soft" sciences, but given the complexity of the social world social scientists certainly need to respect the difficulty involved in making the "plausible conjectures" of social scientific arguments reliable and respectable. Previously I compared measuring social class to measuring mass, suggesting that the former was far more complex. But Becker spends some time in this book pointing out that we have gone through great difficulties and large expenditures of time and money to do things like measure masses of fundamental particles. One explanation for the relatively lower level of success of social science research in comparison to some physical sciences may simply be that we aren't trying hard enough.

In any case, Becker says:

There's other interesting topics related to social science research methodologies that I've love to talk about: debates between quantitative and qualitative approaches, "inductive" vs. "deductive" research plans (hypothesis testing vs. the generation of new hypotheses from raw data), specific arguments in support of the validity of standard methods (e.g. surveys), and so on. But hopefully this is a useful broad apology for social science as science.

Models of Inquiry: Linnaeus and Buffon

In his 2002 book The Politics of Large Numbers: A History of Statistical Reasoning, Alain Desrosières highlights two historical scientists associated with differing models of scientific inquiry: Carl Linnaeus and Georges-Louis Leclerc, Comte de Buffon. I have not read this book, but I'm borrowing from Becker's summary of it in Evidence because I think it points towards a useful way of thinking about arguments between scientists who favor mostly quantitative methods and those who favor more qualitative methods. Plus the Comte de Buffon has a totally sweet name.

Stated this way, the Buffonian model may seem obviously superior; where do the Linnaean categories come from? But there are many cases where a priori classifications seem less arbitrary, i.e. if I begin some research by formulating a hypothesis about differences between men and women, and then go on to classify data according to gender, most of the time that will seem plausible. There are many similarly "common sense" ways of grouping characteristics that are employed in social sciences: age, education level, geographic location, ethnicity, and so on. Desrosières (and Becker) associate this model of inquiry with hypothesis testing, and it's the model most familiar from introductory classes on the scientific method. As Becker puts it:

In other words, hypotheses emerging from prior research are like the Linnaean classification scheme, and the work is to see how new data fit into the scheme. This model of inquiry fits easily with ideas about falsification and hypothesis testing, and is often used in quantitative research, where the goal is to evaluate large, hopefully representative, datasets and analyze along relatively narrow lines to test various hypotheses. A great deal of the value of this model is in the ability to generalize conclusions from statistical significance over meaningful sets of data.

But, this model also highlights in an interesting way the problem of the "theory-ladenness" of observation, e.g. working in this model one will tend to pay attention mostly to the characteristics of phenomena which seem relevant to the hypotheses in question, while backgrounding other characteristics which may also be potentially interesting. In the social sciences -- where the phenomena to be investigated are so rich and varied, and causal analysis so difficult because of the sheer number of causal factors -- this is an important limitation.

The Buffonian model, by contrast, is similar to many qualitative methods, where researchers construct classification schemes in an iterative way by analyzing richer data without necessarily trying to reduce each case to a set of numeric dimensions (although qualitative researchers very often count things). The goal is to develop new insights into patterns in data, which may challenge existing classification schemes. The value of this model is that it may be somewhat less vulnerable to "theory-laden observation", and it can allow some deeper insights into the complexity of causal processes underlying some social phenomena one wants to understand. The drawback is that this model of inquiry is generally time consuming and thus often performed using smaller datasets: interviews, content analysis of texts, participant observation, and so on. It is less easy to assess statistical significance or the representativeness of conclusions. It is also probably less easy to ensure methodological rigor, or at least the considerations may be more subtle than the problems of rigor which appear in statistical analyses, though they are no less important. In sociology, a widely used "buffonian" methodology is called grounded theory, and it's the approach taken in the content analysis research I've participated in.

I hope that the way I've framed the strengths and weaknesses of the two approaches suggests that they ought to be complementary, each helping to address the weaknesses of the other. I would like to argue that a lot of suspicion of social science is rooted in suspicious of the "Buffonian" model of inquiry, and I think by viewing it within the framework of "data, evidence, and ideas" hopefully it may seem less suspicious. Especially if understood as having value in relation to statistical methods.

Here's Becker's take: