Instructor: Levente Littvay (TK PTI)
While it feels like regression analysis is not getting much love these days in the social sciences, it is still foundational for survey analysis, the tool causal inferential models utilize for most (if not all their) estimation, and they are the basis of the fancy, advanced techniques receiving most of the attention today in social science outlets. For this reason, it is not only worth learning regression foundations well, but it is essential to understand how advanced techniques build upon regression to analyze more flexibly. At the same time, such studies deal with their own set of challenges and complexities. Ordinary least squares regressions are simple but have many assumptions one must make and understand. Extensions of regression tools through the use of different estimators and generalizations can relax these assumptions and offer immense modeling flexibility.
What can you expect from the course?
Building on our last session, where we reviewed the assumptions of regression models, we will start to explore OLS regression in the presence of interactions, the estimators and link functions necessary to deal with binary, ordered, and unordered (multinomial) outcomes with a specific emphasis on the correct interpretation of such regression results. Second, we venture into the complexities of various data structures that emerge in voting behavior research, cross-country surveys, repeated cross-sectional surveys, panel data, or other within-person analyses. We will approach this complexity through three tools: clustered standard errors, fixed and random effects corrections, also moving into the world of multilevel modeling. In addition, we will also consider the (admittedly frustrating) topic of survey weights and discuss considerations and limitations in their applications.
Any and all of these topics could easily become a week-long workshop of its own. So, do not expect a full treatment of all these topics. (GLM, panel data analysis, multilevel modeling, sampling and weighting, missing data, and courses of their own offered at methods school like MethodsNET). But we can highlight the foundations important to start down any of these paths and glance at (and take home some R code for) examples of such analyses. The workshop will have an applied focus and will not go into the mathematical foundations of these tools.
Pre-requisites
For the applications, I use R, so basic R knowledge is strongly recommended. At least, you should be able to load a dataset and run a simple lm command. If you don’t know how to do this, please get there before the workshop. (This is not much work, you can definitely do it on your own.) Also, once you know what you are doing, you can easily transfer your knowledge to the analysis tools of your preference. If you missed the previous introductory regression workshop, please review Lewis-Beck and Lewis-Beck’s Applied Regression: An Introduction (second edition) or any other coveted introductory regression text and John Fox’s Regression Diagnostics book, both from SAGE’s Quantitative Applications in the Social Sciences (little green book) series.
Who is this workshop for?
If you are comfortable with applied regression in the social sciences and need to take the next step to deepen your knowledge, you have come to the right workshop. If you have never done any regression analysis and would like to receive an introduction, this workshop is probably not for you just yet.
Schedule
Day 1: Review, Interactions, Introduction to Link Functions, Logistic Regression, Understanding Odds
Day 2: More on link functions. General Linear Models. Binary, Ordered, Count and Multinominal Outcomes.
Day 3: Complex data structures. Sampling Weights. Clustered Analysis. Fixed and Random Effects Models. Introduction to Multilevel Data Structures.