Overview
Spring 2025. Taken as a graduate course.
Statistical concepts, methods and data analysis, with emphasis on the assumptions behind each method and the effects of violating them, using statistical software. Topics include statistical models, random sampling, the normal distribution, estimation, confidence intervals, tests and inferences for one and two populations, and analysis of variance.
Statistics is the discipline of drawing conclusions from data that vary. Two patients given the same drug respond differently, two samples from the same population have different averages, and a screening test that is usually right is sometimes wrong. The course took a biostatistical view of the subject, in which the data are usually measurements on people and the questions come from medicine and public health: how common a disease is, how good a diagnostic test is, whether a treatment changes an outcome, and how many patients a study needs. Its material was adapted by the instructor from Pagano, Gauvreau and Mattie's Principles of Biostatistics (3rd ed., 2022), and the units below follow the course's order.
How the articles are written
Each article teaches one method from the beginning, derives the results a student needs with
every symbol defined, and checks the method by simulation where a simulation shows something a
formula cannot. The examples use public health and medical data that anyone can obtain, such as
the National Health and Nutrition Examination Survey (NHANES) of the US Centers for Disease Control
and Prevention, or classic published data such as the sleep trial that Student used in 1908 to
introduce the t distribution, and synthetic data with a fixed random seed where a known
truth is needed. Every number in an article is produced by a script in the site's repository
(tools/experiments/applied-statistics/), and the figures are drawn from the same
scripts' output. Definitions and results are cited to the NIST/SEMATECH e-Handbook of
Statistical Methods and to the original papers. The articles do not work the course's
assignments.
The notation follows the course's formula sheets: $z_{1-\alpha/2}$ and $t_{n-1,\,1-\alpha/2}$ for the quantiles that cut off the upper $\alpha/2$ of the standard normal and t distributions (the sheets sometimes write $z_{\alpha/2}$ for the same point; the articles always use the subscript $1-\alpha/2$), $H_0$ and $H_A$ for the null and alternative hypotheses, $\bar{d}$ and $s_d$ for the mean and standard deviation of paired differences, and $s_p^2$, $s_W^2$ and $s_B^2$ for the pooled, within-groups and between-groups variances.
The course stopped before correlation and regression, which the library's Data Science collection covers, from simple linear regression (including confidence and prediction intervals for a fitted line) to what violated assumptions do to interval coverage. The information retrieval notes' significance testing article already works the paired t-test, sign test, Wilcoxon signed-rank test, randomization test and bootstrap on one example, and the articles here link to it rather than repeat it.
Topics
Each unit leads to a page of detailed articles; units without a link are still being written.
- Descriptive statistics. Types of numerical data, tables and graphs, and measures of center and spread.
- Rates and standardization. Rates in public health, and direct and indirect standardization.
- Probability. The rules of probability, conditional probability, and relative risk and the odds ratio.
- Screening and diagnostic tests. Sensitivity and specificity, predictive values and Bayes' theorem, likelihood ratios, and ROC curves.
- Theoretical probability distributions. The binomial, Poisson and normal distributions.
- Sampling distribution of the mean. The distribution of a sample mean and the central limit theorem.
- Confidence intervals. Intervals for a mean with z and t, coverage, one-sided bounds, and sample size.
- Hypothesis testing. Tests for a mean, the two kinds of error, and power and sample size.
- Comparison of two means. Paired and two-sample t-tests, and sample size for comparing two means.
- Analysis of variance. One-way analysis of variance and multiple comparisons.
- Nonparametric methods. The sign test, the Wilcoxon signed-rank and rank-sum tests, and the Kruskal-Wallis test.
- Inference on proportions. The normal approximation to the binomial, intervals and tests for a proportion, and comparing two proportions.
References
- NIST/SEMATECH. e-Handbook of Statistical Methods. National Institute of Standards and Technology. https://www.itl.nist.gov/div898/handbook/
- Pagano, M., Gauvreau, K., and Mattie, H. (2022). Principles of Biostatistics (3rd ed.). CRC Press. The course text, as adapted by the instructor.
- Student (1908). The probable error of a mean. Biometrika, 6(1), 1–25.