← Back to series HSCI 410

Exploratory Data Analysis For Epidemiology

Office Hours — the companion podcast for HSCI 410

Each episode is meant for listening after the lesson. Sarah and Kiffer add perspective and critique, work through the questions students tend to find most difficult, and add further worked examples, including some that are harder than those in the lesson. Every page has the audio with a visual companion and captions, a summary, a list of chapters and a downloadable transcript.

Foundations & descriptive analysis

Lessons 01–02

A Structured Approach to Data AnalysisSarah and Kiffer work through adjustment sets, collider bias, missing-value codes, apparent overdispersion and cut-points, with extra worked examples for the lesson on structured data analysis.

Data Cleaning & Descriptive AnalysesSarah and Kiffer work through outlier fences, missing-data mechanisms, compounding missingness, standardized differences in Table 1, and a debate about medians and means for health care costs.

Regression models

Lessons 03–04

Linear & Logistic RegressionSarah and Kiffer work through odds ratios, non-collapsibility, interaction terms, and calibration versus discrimination, then debate turning blood pressure into a yes-or-no outcome.

Generalized Linear ModelsSarah and Kiffer work through odds ratios, relative risk ratios, offsets and overdispersion with new examples, and debate whether a five-point rating can be treated as a number.

Dependent data

Lesson 05

Modelling Dependent DataSarah and Kiffer work through small ICCs, the limits set by the number of clusters, cluster-specific and population-averaged odds ratios, and missed visits in repeated measures.

Exploratory visualization

Lesson 06

Exploratory Data Analysis and VisualizationSarah and Kiffer work through boxplot outliers, compounding missing data, squared correlations and group composition, and they debate whether exploring data can go too far.

Measurement & path models

Lessons 07–08

Measurement and PsychometricsSarah and Kiffer test what alpha shows, correct correlations for unreliability, trace measurement precision along the trait, and debate how the De Jong Gierveld scale should be scored.

Mediation, Moderation and Path AnalysisSarah and Kiffer discuss mediator-outcome confounding, centred interaction models, interaction scales, inconsistent mediation and equivalent path models, with worked examples and a debate on cross-sectional mediation.