Introduction
There is a familiar ritual to the undergraduate statistics course. You’re handed a decision tree, invited to follow its branches, and eventually arrive at a test and a p-value to report. The process is tidy, sequential, and almost entirely thoughtless. At no point are you asked to reflect seriously on what you are doing, or why, or whether the question you are answering bears any resemblance to the question you started with.
My degree was in psychology; a field that relies heavily on statistical inference, and one that has paid a very public price for treating it carelessly. Yet even there, statistics was something to endure rather than be enlightened by. The implicit mantra was that mathematics is hard, the procedures are unpleasant, and the sooner it is over the better. It was a checkbox that needed ticking to get your degree, not a way of thinking that might change how you understood the world.
That has frustrated me for nearly two decades. And I don’t think the frustration is misplaced.
The logic gate curriculum
The procedural presentation gives the impression that statistical analysis is a series of objective logic gates requiring no creativity and very little judgment. Follow the steps, obtain the answer. The methods are presented as a toolkit – seemingly fixed, finite, and authoritative – rather than as a set of imperfect instruments for reasoning under uncertainty.
What this produces is a generation of statistical technicians. They can run the tests they were taught. They know which boxes to check. But they have outsourced their thinking and judgment, and are entirely at the whim of the sampling process as a result. The more difficult and more illuminating methods are withheld to postgraduate courses, leaving undergraduates with a perfunctory understanding of statistical analysis that can take years of subsequent practice to unlearn.
The p-value is perhaps the clearest illustration of the problem. It survives in the curriculum not because it is the most useful or the most interpretable quantity, but because it is procedurally convenient. It collapses a genuinely difficult inferential question into a binary, and students learn to report it because it feels like an answer. The irony is that the p-value is one of the most routinely misunderstood concepts in all of science, yet it remains the centrepiece of undergraduate statistical education.
But the distortion cuts deeper than just the analysis stage. When students are trained on a fixed menu of tests, they learn — implicitly, if not explicitly — to design experiments that those tests can handle. In some cases that is perfectly reasonable. But it can perniciously narrow the scope of scientific enquiry, bending the question to fit the method rather than the other way around. Worse, a design chosen for its statistical convenience may not actually answer the question the researcher started with, yet the results get interpreted as though it did. The curriculum doesn’t just shape how scientists analyse data — it shapes what they think to ask.
A philosophical narrowing
The standard undergraduate curriculum is not just procedurally impoverished, it’s philosophically blinkered in a way that most students never realise.
Almost universally, undergraduates are taught frequentist statistics and nothing else. No alternatives are mentioned, let alone explored. This would be defensible if frequentism were the only coherent framework for statistical inference, but it isn’t. Bayesian statistics – in which probability represents a degree of belief updated in light of evidence – offers a fundamentally different and, to many scientists, more natural way of thinking about what data can and cannot tell us.
The Bayesian framework maps directly onto how scientists actually reason. We have prior knowledge. We collect data. We update our beliefs accordingly. That is science. The frequentist framework, by contrast, asks us to reason about the long-run behaviour of a procedure across hypothetical repeated samples – a construction that is philosophically coherent but considerably more remote from the questions researchers often want answers to. Students trained exclusively in frequentist methods don’t just lack a set of tools, they leave university without knowing that a different way of thinking about evidence exists at all.
What was lost
The cost of this is difficult to quantify, but it is real. The replication crisis — which hit psychology particularly hard — had many causes. But it’s hard to ignore that researchers trained to treat statistical significance as a destination, rather than as one imperfect signal among many, were not well placed to resist its pull. They followed the procedures they had been taught, reported the numbers they had been trained to report, and drew conclusions the framework encouraged them to draw. The curriculum may not have caused these errors, but it did little to equip researchers to avoid them.
A more honest pedagogical approach would ask something harder of students. It would treat inference as a genuinely difficult problem rather than a set of tasks to be executed. It would introduce the idea that there are multiple coherent frameworks for reasoning about evidence, that judgment cannot be fully automated, and that uncertainty is something to be characterised rather than resolved by crossing a threshold.
Statistics, taught well, is not something to endure. It is one of the most powerful tools we have for understanding the world – imperfect, contested, and endlessly interesting. That is what the curriculum could have conveyed. For most students, it didn’t come close.