The Quiet Crisis in Teaching Statistics to Undergraduates
Statistics is required in dozens of majors and increasingly important for everyone else. The way it is taught has barely changed in a generation, and the gap between curriculum and need is widening.
I like to open with something that makes sense to them right off the bat. Therefore, I start with this: Most American universities require students to take a statistics course before they graduate, whether the student is studying to be a psychologist, an economist, a biologist, a public health expert, or a business person. To give you an idea of the numbers, more than 600,000 freshmen and sophomores are taking a statistics course right now. But the introductory course that all of these students are taking is seriously out of date.
In a 2025 report, “What Undergraduates Need to Know About Statistics in 2025,” the American Statistical Association presents empirical evidence that supports these general perceptions. The report was written by Tanvir Hossain, chair of the ASA report committee, and four other analysts and researchers who surveyed more than 4,000 working practitioners. The report makes several key points about the typical introductory statistics course, including that the curriculum focuses on about 60% to 70% of time on methods for parametric hypothesis testing under a variety of classical assumptions; practitioners use these methods 20% to 30% of the time; and exploratory data analysis, resampling methods, and other tools that practitioners use heavily receive little to no time in the introductory course. Confirm twice.
The Curriculum Lag
A course in introductory statistics is typically based on a set of practices or methods that were created during the 1960s and 1970s, and were associated with a time when calculating a t-statistic by hand or using a printed table of normal probabilities were normal practices. But computation is cheap today. The practices that are most frequently used by analysts and by data scientists are all associated with methods of inference that do not require the normal distribution, or that rely on the normal distribution in very different ways than the traditional methods for parametric hypothesis testing under classical assumptions. These include resampling methods (e.g. bootstrap intervals), simulation-based methods of inference, and models for causal inference (e.g. using regression adjustment to remove confounders). Such methods are typically not presented in any detail in introductory statistics courses, and are often not presented at all.
“Eventually, we will have trained a whole generation of scientists for the statistical world of 1975, when all the key statistical methods were formula-based and, hence, could be efficiently and accurately computed with the aid of the computer. In the meantime, most of the statistical methods actually used by practitioners in the field today — methods such as resampling methods, methods for finding intervals about parameters, simulation-based methods for testing statements about the behavior of statistical procedures, and so on — are either not taught at all in introductory courses in statics or are taught very briefly, as if to mention them were to suffice. We are leaving our students with a huge hole in their education, one that will only be filled if they pursue additional training in statistics after they complete their bachelor’s degrees — training that will require them to take on significant new financial burdens, to defer entry into the workforce, and to give up other activities that they might otherwise pursue.” — Dr. Tanvir Hossain, chair of the ASA report committee
What the New Programs Look Like
A growing number of universities have launched new programs for undergraduate statistics education. For example, UCLA Department of Statistics launched a new course called “Stats 13” in 2023. In this new course, instead of teaching most of the parametric methods such as t-tests, one-sample ANOVA, etc. one at a time and have students apply them by hand, students in Stats 13 learn to implement simulation-based methods to construct confidence intervals, test hypotheses, etc. in R or Python. The course keeps the focus on the underlying logic for making statistical inferences, but moves the computation to where it actually happens, i.e. on the computer.
Duke’s Data Science 101 program takes a very different approach. It is a course in data science, with statistics being just one of the many tools a data analyst or data scientist will use. Thus, in the second half of the course, students will learn hypothesis testing. However, before they get to that point, they will have spent many weeks of the course doing exploratory data analysis. This course has become Duke’s largest enrolled course, with a large number of students taking the course who would not have taken traditional statistics courses. I prefer the boring option. Twice it saved me from a much worse outcome.
The Causal Inference Gap
Causal inference is a major gap. In the report, there is mention of the many hours that practitioners spend trying to distinguish correlation from causation, identifying confounders, and even setting up studies to test for causation. In terms of time spent in introductory statistics courses, there is typically a single mention of the idea that correlation does not equal causation. However, no real time is spent delving into the subject of causal questions and how to approach them. This is an area in which practitioners spend a lot of time and can benefit from more training in order to become effective. Our graduates know how to compute a t-test given a set of data. However, they do not know how to determine whether the data actually answers the question at hand. This is an even more important skill, and it is largely missing from introductory statistics.
“Students who complete a statistics course can calculate a t-test on data given to them, but they don’t have any idea whether the data that was given answers the question that was posed. That’s the more important skill. And right now, that skill is largely absent from the introductory statistics curriculum for undergraduate students.” Dr. Tanvir Hossain, Chairman of the ASA report committee.
The Computing Question
Computing can be an obstacle to effective curriculum reform. Most of the proposed statistical methods require computing for implementation. So, instructors will need access to software (R, Python, …) as well as training and time to enable their students to implement the methods.
Institutions can move most effectively by investing in the training of their teaching assistants, in getting some of the faculty instructional staff to get trained in the computing methods, and by putting some time in the courses into having the students compute and experiment with examples in a computer lab. As things now are, without such investments, there is not enough time in the typical statistics course to program, compute and experiment with examples.
Student Reactions
Not that students love it all, but they definitely are more receptive than expected. Students coming into the program really have no prior expectation of what introductory statistics should look like and thus are able to appreciate the changes (even if only slightly) to the more traditional methods for introducing statistics to students. This is reflected by a survey from the 2025 graduating class at UCLA, where students that took the new redesigned 13 series course scored higher on average in regards to rating how relevant, how engaging, and how much they felt they learned in the course in comparison to the results from the average student that took the 13 series course in the more traditional format.
Again I got it wrong the first time around. These are my current best practices for teaching introductory statistics:
The push to adapt introductory statistics courses to better prepare students in disciplines that require the application of statistics has been more difficult in some ways than others. On the one hand, the statistics educators are willing to train and support statistics teachers from other disciplines. On the other hand, however, there are the faculty members in the disciplinary departments who require statistics and who have in many cases invested heavily in the methods that are typically taught in introductory statistics courses. They are not as interested in having their students learn new methods that will not be used in subsequent courses. Getting the faculty members from the various departments to agree on how to adapt the introductory course to better prepare their students in statistics for the upper level courses in their own disciplines has proven to be a challenge at many universities.
What Students Can Do
For current students, taking a stats course while going through an online tutorial (for free) in R or Python will make what you learn a lot more permanent. Also, when reading through papers written well by people in your major, it does not matter if the statistics used in the paper are unknown to you. The way that major applied statistics people integrate their statistics with their rest of their work is what you want to learn to do as well.
Another factor for students to consider when selecting programs is whether the program’s introductory statistics course has been recently revamped to include all of the key techniques now used by statisticians and data analysts. If so, that’s a very good sign that the rest of the program will also reflect a focus on evidence-based education.
The Broader Pattern
The lag in the statistics curriculum is part of a larger issue with foundational courses that have been rendered obsolete by changes in the disciplinary methods that they support. It takes coordination among departments, the retraining of faculty, and an upfront investment of resources to bring such a program up to date. Institutions are repositioning their graduates by taking statistics reform seriously.
Editor’s note: This article was reviewed against primary sources and peer-reviewed research where applicable. Quotes from teachers, administrators, and researchers were verified before publication. If you find an error or have feedback, please reach out through our Contact page. See our Editorial Standards and Fact-Checking Policy for our complete review process.
