I offer Quantitative Methods (QM) 1, and 2, on a regular basis, and QM 3 on a somehwat irregular basis. All courses are cross-listed between Human Development and Psychology, and are open to graduate students and advanced undergraduate students. In the past I also taught Introduction to Research Methods for undergraduates. Each semester, I offer a reading group (together with my colleague Anthony Ong) that is focused on methods. It is called MSG (Methods Study Group).
QM1 assumes that students may enter the Ph.D. program with diverse backgrounds in statistics. Some students may already have taken one or a whole sequence of graduate statistics courses. Other students may have taken no statistics course at all. During the first couple of weeks, we spend some time on the basics to bring everybody up to speed. We spend a considerable amount of time on foundations of inferential statistics. We will cover in depth what a p-value means, where sampling distributions come from, and how confidence intervals behave in the long run. At the same time, we will also discuss what prior distributions, likelihoods, and posterior distributions are, and how they are used to form Bayes Factors, and Bayesian credible intervals. In short, we will cover both Frequentist and Bayesian statistics. We will often use simulation-based exercises in R to get a more intuitive understanding of these concepts. Once we are comfortable with this foundation, we will discuss the design and analysis of randomized experiments. This leads us naturally to the analysis of variance model, which we will discuss using the more traditional sums-of-squares approach, and the general linear models approach. We will cover several design variants, such as factorial designs, unbalanced designs, fractional factorial designs, and designs that include repeated measures. Throughout the whole course we will weave in aspects of open science, and reproducible research.
QM2 build directly on QM1, and covers the design and analysis of non-randomized studies. We will cover regression models of various flavors. A key aspect of the course is that we will differentiate models that are aimed at description, prediction, and causal inference. In the realm of descriptive models, we will learn to model a few select variables, and how to carefully interpret coefficients from a regression equation. In the domain of predictive models, we will learn about the difference between in-sample and out-of-sample error, and how that relates to the bias-variance tradeoff. Finally, in the domain of causal inference, we will learn about causal assumptions, as encoded in graphical causal models, and how these assumptions inform us what kind of statstical estimate we should seek out to identify our causal effect of interest. As in QM1, we will discuss all models from both a Frequentist and Bayesian perspective.
QM3 is a seminar-stlye advanced methods class. We cover a variety of topics, including propensity score methods, regression-discontinuity designs, causal mediation, factor analysis, and others. Class time is usually spend discussion seminal papers on topics, with heavy emphasis on conducting analyses in R.
MSG in an informal bi-weekly gathering of interested students to talk about research methods. Often, but not always, the papers that we read during a single semester are thematically grouped. In previous years, we have talked about the difference between Bayesian and Frequentist statistic, causal inference, open science, or replicability. Students are encouraged, but not required, to sign up for a single credit of directed readings.
Starting in 2017, I decided to integrate some online learning component to my classes. I am currently using datacamp for this. For the duration of the semester, students can get free access to the video library at datacamp. I use the video lectures on Working with the RStudio IDE and Reporting with R Markdown. In the future I might blog about some experiences with using datacamp as a supplement to my teaching.