Workshop in Comparative Research Methods: Analysing Cross-National and Multi-level Data in the Social Sciences
In addition to the presentation slides available below, video recordings of the sessions are available via the ELECDEM YouTube channel. To access the videos search 'ELECDEM comparative' on YouTube.
Programme for the ELECDEM workshop in Comparative Research Methods
Delegate list for Comparative Research Methods wksp
Marco Steenbergen, Professor of Political Methodology at University of Zurich
Outline
The course presents an overview of the logic of multilevel analysis and those applications that are particularly relevant to electoral research, namely applications focused on categorical dependent variables. The focus is on model logic an interpretation and less on estimation details.
Part 1: Hierarchical Linear Models
What are multilevel data structures? Statistical complications arising from multilevel data. The hierarchical linear model and its interpretation. Some caveats in multilevel analysis.
Why Multilevel Analysis - Steenbergen (PDF 388KB)
The Hierarchical Linear Model - Steenbergen (PDF 548KB)
A worked example in Stata - Steenbergen (PDF 517KB)
Part 2: Multilevel Logit Models for Binary and Ordinal Dependent Variables
The binary and ordinal logit models. Extensions of the models to multilevel data structures. Interpretation through the odds ratio, predicted probabilities, marginal effects, and elasticities.
Multilevel Logit Model Binary Dependent Variables - Steenbergen (PDF 1636 KB)
Multilevel Logit Model Ordinal Dependent Variables - Steenbergen (PDF 388KB)
Part 3: The Multilevel Multinomial Logit Model
The multinomial logit model. An extension of the model to multilevel data structures. Interpretation through the odds ratio, predicted probabilities, marginal effects, and elasticities.
Multilevel Multinomial Logit Model - Steenbergen (PDF 200KB)
Advanced reading:
Hox, Joop J. Multilevel Analysis: Techniques and Applications.
Leonardo Grilli · Carla Rampichini A multilevel multinomial logit model for the analysis of graduates’ skills
Christopher Wlezien, Professor of Political Science at Temple University
Outline: Pooled Cross Section Time Series Data Analysis
The course provides an introduction to pooled cross section time series data analysis. It will focus on the techniques for analysis of various data sets, from those where the number of time series observations exceeds the number of cross sectional units to those where the number of cross sectional units exceeds the number of time serial observations, the latter of which often is referred to as panel data. There will be an emphasis both on both the logic of analysis and estimation, the latter of which will involve use of Stata software.
Part 1: Analyzing Independent and Dependent Data
This session serves as a general introduction to analysis of cross sectional and time serial data, focusing especially on the latter. In one sense, the session is an introduction to time series data analysis.
Time series and cross section analysis - Wlezien (PDF 289KB)
Part 2: Time Series Issues and the Analysis of Pooled Cross Sections
This session introduces different types of time series models and begins to consider issues in analyzing pooled cross sectional data.
Time series and pooled analysis (a) - Wlezien (PDF 220 KB)
Part 3: Estimating Pooled Cross Section Time Series Models
This session demonstrates the main approaches to estimating pooled models. The focus is primarily on non-binary dependent variables but attention also will be paid to models of binary dependent variables.