The traditional college-level elementary statistics for non-majors course needs a big change. p-value reasoning should be replaced with Bayesian models. The class, call it Stat 101, is the only contact with data analysis methods that most college students will ever have. Stat 101 needs to get it right.
At best, students leave Stat 101 able to define and use p-values. They understand sampling variability and the role of the central limit theorem. They can correctly state confidence interval and hypothesis testing conclusion by the end of the course. They can use a calculator and/or spreadsheet and/or statistical software to do tests and can dutifully interpret p-values. For most, a few months later, all they will remember is, p<.05 is “good.”
Thus, Stat 101 students finish their course having learned an increasingly rejected statistical paradigm which will propagate into their science labs, advanced statistic courses in the social sciences and business, into their capstone papers and beyond into who knows what after they graduate. We are placing the vast majority of our students in a “statistical” hole.
So here is my plea for somebody somewhere to do something. Not that people haven’t tried, and failed by their lights. My interpretation of such efforts is that the authors tried for too much – include too much explanation, too much background, too complicated of models, too complex model checking.
Take a look at an elementary statistics textbook such that it is. The explanations are limited. The types of questions/problems are simple and the subsequent analysis can be written in a sentence. This is appropriate for these students. Why not, instead of a p-value conclusion, have a modeling conclusion based on model sampling. Stat 101 students can just as easily be taught to state conclusion properly in modeling language as in p-value language. Since most students treat calculators or statistical software as a black box anyway, using provided p-value results or model sampling results will make no difference to them. The situations are simple – inferring population parameters from a sample, comparing two populations from samples, using normal or binomial distributions. These simple problems cover most of the cases they will confront in their college career.
So my plea – someone, somewhere write a elementary statistics textbook for non-majors using Bayesian models. Get it underwritten by a major publisher. Most of the content will be like the traditional text. The changes will be just in how the problems are worked out using Bayesian capable software and how the conclusions are stated. this would be good start.
Note: No links in this post. It is just my general sense of the field. I am not an expert in the field of statistics. I have avoided teaching elementary statistics for many years now because I disagreed with the content. I could do this because I am department chair and past retirement age.