Statistics for Sociologists/Chapter 1: Why Statistics? Making Sense of Social Life with Data

Chapter 1: Why Statistics? Making Sense of Social Life with Data Ask a room full of sociology students why they're taking a statistics course, and you will get some version of the same three answers: it's required, they heard it's "not that bad," and they are visibly regretting their choice of major. This chapter exists to argue, gently but persistently, that all three of those students are wrong about at least one thing. Statistics is not just a hoop you have to jump through. It is one of the more interesting tools ever developed for answering a question sociology cares about more than almost any other discipline: what is actually going on with people, as opposed to what we assume is going on with people. That distinction — between what's actually going on and what we assume is going on — is the entire subject of this book. alt=A group of people in a park.|thumb|600x600px|There is a lot going on within any group of people. Statistics are used in Sociology to study groups of people. What Statistics Is (and What It Isn't) Statistics is not magic. It will not tell you what people should do, whether capitalism is good, or why your cousin holds the political views he holds. It cannot generate a fact from nothing, and it cannot rescue a badly designed study. Statistics is, more modestly, a set of tools for describing patterns in data and for deciding how much confidence those patterns deserve. It is also not "just arithmetic," despite what the equation-heavy chapters later in this book might suggest to a skeptical reader flipping ahead. Arithmetic tells you that 2 + 2 = 4. Statistics tells you whether the four-point gap in reported happiness between two groups of people is a real social pattern worth taking seriously, or the kind of noise you'd expect to see even if nothing were going on at all. That second question — is this real, and does it matter — is a much more interesting question than any of the arithmetic required to answer it, and it is the question this entire book is organized around. (If statistics really were just arithmetic, this book would be four pages long, and one of those pages would be a lecture on the order of operations.) Here is a useful mental model: statistics is what you get when you refuse to trust your gut about how the social world works until you've checked. Anecdote vs. Evidence Humans are excellent at building confident, coherent stories out of very small amounts of information. This was probably a useful skill on the savanna. It is a much less useful skill for understanding society, because society does not fit inside any one person's field of view. Consider a genuinely humbling body of research on how badly people estimate basic facts about their own societies. Survey researchers have repeatedly asked people in wealthy democracies to guess things like the percentage of the population that is Muslim, the percentage of teenagers who become pregnant each year, or the share of national wealth held by the top 1%. The results are not close. People in the United States, for instance, have historically guessed the Muslim share of the population at somewhere around 15%; the actual figure is about 1%.Ipsos MORI, "Perils of Perception" surveys, various years. This is not a small rounding error. This is being off by a factor of fifteen while feeling completely confident about it. alt=A dumbbell chart showing perceptions and reality.|center|thumb|800x800px|A chart illustrating the difference between perceptions of social facts and the reality of social facts. The point is not that people are stupid (well... that's mostly not the point!). The point is that human intuition about large-scale social patterns is built from whatever we happen to have personally encountered — the news we watch, the neighborhood we live in, the one loud cousin at Thanksgiving — and none of that adds up to a representative picture of anything. Anecdote feels like evidence. It is not evidence. It is a sample size of one, dressed up in a confident tone of voice. Your uncle at Thanksgiving may be extremely confident about what "everyone" thinks; he is, statistically speaking, an n of one, and no peer-reviewed journal on Earth will publish him, no matter how loudly he clears his throat before the next point. Statistics exists because "I feel like this is true" and "this is true" are different claims, and sociology needs a way to tell them apart. What Sociology Needs Statistics For Sociologists use statistics for a handful of related but distinct jobs: Describing populations. How many people hold a given belief? What does the income distribution actually look like? Description sounds boring until you realize almost no one's intuitive guess about these numbers is correct. Testing hypotheses. Does education predict income? Does religious attendance predict happiness? Sociological theory generates predictions; statistics is how those predictions get checked against reality instead of just sounding plausible in a seminar room. Identifying patterns. Social life is full of relationships between variables that are invisible at the individual level and only show up once you look at enough people at once — this is arguably the founding insight of the entire discipline, going back to Durkheim noticing that suicide, an intensely personal act, follows remarkably stable social patterns.Durkheim, É. (1951). Suicide: A Study in Sociology (J. A. Spaulding & G. Simpson, Trans.). Free Press. (Original work Le Suicide: Étude de sociologie published 1897) Sociologists have, in other words, been doing data science since roughly 1897; we just didn't have nifty computers to help with the analyses back then, not even human computers. Informing policy. Governments and organizations make decisions that affect millions of people based on what the data seem to show. Getting the statistics wrong has consequences that extend well past a bad grade on a problem set. Throughout this book, you will do all four of these things using the General Social Survey, a dataset that includes responses from surveys asking Americans about their lives, beliefs, and behaviors since 1972. You'll meet it properly in Chapter 3. For now, just know that everything from here forward is aimed at letting you ask real questions about real people and get real answers — not perfect answers, but honest ones, with the uncertainty attached. A Field That Occasionally Embarrasses Itself: The Replication Crisis Here is where this book earns some credibility by admitting something uncomfortable: a fair amount of what gets published across the social and behavioral sciences — sociology fully included, not just the neighbors down the hall in the psychology department — does not hold up when someone else checks it. Starting around 2011, researchers began systematically trying to replicate published findings — that is, running the same study again, often with fresh data and more statistical power, to see if it produced the same result. The first wave of this work focused on psychology: a large-scale effort by the Open Science Collaboration attempted to replicate 100 studies from top psychology journals, and fewer than half succeeded.Open Science Collaboration, "Estimating the Reproducibility of Psychological Science," Science, 2015. For years afterward, "the replication crisis" got treated in casual conversation as psychology's problem to solve, which let everyone else in the social sciences feel a little smug about it. That smugness became harder to justify in 2026. A massive replication effort — the DARPA-funded SCORE project — attempted to replicate 274 specific findings drawn from 164 papers published between 2009 and 2018, across six fields: business, economics, education, political science, psychology, and sociology.Tyner, A. H. et al. "Investigating the replicability of the social and behavioural sciences." Nature 652, 143–150 (2026). The replications were not casual efforts, either — they were high-powered (median 99.6% power to detect the original effect), used the original study materials when available, and were peer-reviewed in advance. This was about as fair a test as a field can get. alt=A bar chart illustrating replication rates.|none|thumb|800x800px|This chart, created in R, illustrates the replication rates in the social sciences per Tyner et al. 2026. The results, by discipline, were remarkably consistent — consistently unflattering. Across all six fields, only about half of claims held up: 55.1% of individual claims and 49.3% of papers replicated with a statistically significant result in the original direction. Sociology's rate was 51.1% — essentially indistinguishable from psychology's 49.0%, and statistically no different from any other field in the study (the range across disciplines ran from 42.5% to 63.1%, and the differences between fields were not significant). In plainer terms: if you flip a coin to decide whether a published sociology finding will replicate, you will do about as well as the field's actual track record — which is not a compliment, but it is at least an equitable embarrassment, evenly distributed across the disciplines that study people. It gets more specific, and slightly worse. Among the sociology studies where an effect size (more on these in later chapters) could be directly compared, the original published effect was already fairly modest (median Pearson's r = 0.10) — smaller, on average, than the original effects reported in business, economics, or psychology. When those same claims were replicated, the median effect size shrank further, to r = 0.03. That is not quite zero. It is, however, close enough to see zero from where it's standing. None of this happened because researchers were, as a rule, committing fraud. It happened because of a set of ordinary, human, entirely predictable practices that this book will spend real time on: alt=a bowl of soup|thumb|400x400px|One famous study that failed to replicate was from the lab of now largely discredited researcher Brian Wansink, claiming that people ate more if they were given a bowl of soup that had more soup pumped in from the bottom. Much of Wansink's work was later retracted because the underlying data were manipulated in ways this course will teach you are unethical. Small samples that produce noisy, unstable results that happen to look like a pattern. p-hacking — trying several analyses and reporting only the one that "worked" (more on this in Chapter 18, where it gets a proper dissection). Publication bias — journals overwhelmingly publish studies that find something, so the file drawer fills up with the honest null results nobody ever sees. (The file drawer, it turns out, is a lot roomier than anyone likes to admit — sociology has been contributing to it right alongside everyone else.) No data or code sharing, which meant that for decades, nobody could even check anyone else's work without redoing the entire study from scratch. So: this is not a chapter about psychology's problem. It is a chapter about sociology's problem, documented as recently as 2026, with your discipline's name attached to the data in black and white. The GSS-based examples in this book will use real sample sizes and real reported effects specifically so you get comfortable with what honest, adequately powered social science looks like — which, encouragingly, is not that hard to produce once you know what to watch for, and which is exactly what the rest of this book is trying to teach you to do. Open Science: The Field Grows Up alt=A screenshot of the OSF.io website.|thumb|600x600px|This is the home page for the OSF, which is run by the Center for Open Science. You'll have plenty of chances to work with the website throughout this course. The response to the replication crisis has been a genuine culture shift, usually filed under the heading open science: the practice of doing research in a way that lets other people check it. In practice this means a few concrete habits, all of which you will build over the course of this book: Transparency about what you did — including the analyses that didn't pan out. Data sharing, so other researchers can look at what you looked at. Code sharing, so other researchers can run what you ran, rather than guessing at your methods from a paragraph in a journal article. Pre-registration, which means writing down your hypothesis and analysis plan before you look at the results — the statistical equivalent of calling your shot. Two tools that have become central to this shift are Zenodo, a free platform where researchers post their data and code, and the Open Science Framework (OSF), where researchers can pre-register studies and post pre-prints of their research. Think of these as a public diary for your research — except instead of embarrassing teenage feelings, it holds your regression code, which, depending on how that regression turned out, might be the more embarrassing of the two. You'll set up your own projects in Chapter 2, and you'll see "Open Science Corner" boxes like this one scattered through the rest of the book: Open Science Corner #355C7D #FFFFFF #F8B195 Sharing your code and data is not an act of vulnerability, and it is not something you do only after you've become an established researcher with tenure and nothing left to prove. It is a basic professional habit, on the same level as citing your sources — and after the last decade, the field has run out of excuses not to do it. None of this is presented here as mandatory virtue. It's presented as what careful researchers already do, because the alternative — as an entire replication crisis demonstrated — is occasionally embarrassing itself in front of the entire discipline. Chapter Summary Statistics is a set of tools for figuring out whether a pattern you think you see in social life is actually there, and if it is, whether it's big enough to matter. Human intuition is bad at this job on its own — we build confident stories out of tiny, unrepresentative samples of our own experience, and those stories routinely turn out to be wrong when checked against actual data. Sociology relies on statistics to describe populations, test theories, find patterns invisible at the individual level, and inform decisions that affect real people. The field has recently had to reckon with a replication crisis — and a 2026 large-scale replication effort confirmed that sociology's published findings hold up about as often as psychology's, roughly half the time, with effect sizes that often shrink close to nothing on a second look. This is not one field's embarrassment; it is a shared one, caused by small samples, selective reporting, and a lack of transparency across the social sciences. The discipline-wide response, open science, is less a set of rules than a professional habit of showing your work. This book will build that habit alongside every statistical skill in it. Think About It This chapter has no R code and no practice problems — just some things worth sitting with before Chapter 2 gets you into RStudio. Think of a belief you hold about some group of people (a generation, a political party, a profession — anything) that is based mostly on personal experience or things you've seen online rather than on any data you've actually looked at. What would it take to check whether that belief holds up? What data would you need? The chapter argues that anecdotes "feel like evidence." Why do you think vivid individual stories are so much more persuasive to most people than a table of statistics showing the same pattern across thousands of cases? Is this a flaw we should try to correct, or does it serve some purpose? If a well-known, widely cited finding in your field of interest turned out to fail replication tomorrow, what do you think the honest response should be — from the original researcher, from the field, and from the public that had heard about the finding? Does it matter whether the original researcher did anything wrong? References