Why Count: The Case for Quantification

Some sociological questions can be answered without counting anything. What does it feel like to work the night shift at a nursing home? What goes on inside a trading floor during a panic? What does a particular religious community understand by the word 'sin'? These are the questions for which the ethnographer reaches for her notebook, and no table of numbers would do the work any better. The previous topic introduced the main qualitative approaches; this one turns to their companion family of methods, which do different work for different questions.

The questions that require quantification are of a specific shape. They are questions about populations, rates, gradients, differences, and changes over time — questions whose answer is not 'yes' or 'no' but 'how much,' 'for whom,' and 'compared to what.' Has the gender wage gap narrowed in the United States since 1980? By how much? Is the narrowing larger for college graduates than for those with only a high school education? Did the reduction in the European Union's youth unemployment rate between 2013 and 2019 occur faster in countries that adopted active labor-market policies? No amount of close ethnographic attention will answer any of these questions; they are claims about distributions across millions of cases, and claims about distributions require systematic measurement.

The founding text of sociological quantification is Émile Durkheim's Le Suicide (1897), the first major work in the discipline to take seriously the idea that a pattern visible only in aggregate data could have genuine sociological content (Durkheim 1897). Durkheim did not invent statistics — the civil statisticians of the nineteenth century, including Quetelet, had already established the field of 'moral statistics' — but he did something consequential: he argued that the suicide rate, which appears to be the most irreducibly personal of outcomes, varies systematically across groups (Protestants versus Catholics, married versus unmarried, soldiers versus civilians) in ways that no aggregation of individual biographies can explain. The rate is the explanandum. It is a social fact (Durkheim 1895/1982, pp. 50-54). If the quantitative program of sociology has a founding move, it is this one.

It is worth being clear about what quantification is not. It is not a claim that numbers are the only kind of knowledge that counts. It is not an ideology of reducing human experience to data points. It is not a commitment to any particular political stance. Goldthorpe, in a careful defense of quantitative sociology against various opponents, puts the matter plainly: quantification is a tool for answering a certain kind of question about the social world, and its epistemic authority comes from the fact that, when used well, it does answer those questions (Goldthorpe 2000, pp. 34-40). It is no more and no less than that.

Three specific things quantification does that other methods cannot. First, it produces population-level claims. When the US Census reports that the median household income in 2022 was $74,580, that number is a claim about the entire population, produced through sampling methods that would not be available to a case-study researcher (US Census Bureau 2023). Second, it produces effect sizes — not just 'does X affect Y' but 'how much.' The answer 'a lot' or 'a little' matters for policy, theory, and scientific progress (Firebaugh 2008, pp. 102-108). Third, it produces trend measurement. Whether a pattern is stable, rising, or falling is itself a finding; rates of violent crime, rates of religious attendance, rates of intermarriage, rates of home ownership — the shape of their change over time is a central sociological fact that only measurement can reveal.

This topic takes those three uses as its spine. The next five pages address, in order: the survey as the workhorse of quantitative sociology; experiments, rarer but powerful; the basics of statistical analysis; common pitfalls in reading quantitative work; and the interpretation of effect sizes. The goal is not to train you as a statistician but to let you read a regression table, recognize what the numbers are doing, and know when to trust them.