Multiple Regression Analysis Bernhard Kittel Center for Social Science Methodology University of Oldenburg The Art of Summarizing Relationships The Straight Line 1998 G. Meixner The Straight Line The Art of Summarizing Relationships y Xb ε b X' X X' y 1 Regression Analysis: Issues Measurement Error Model Specification E(b) = b → Case selection Y = a + Xb + e (s.e.) Var(b) → Number of cases Ontological Assumptions The Art of Summarizing Relationships Assumptions Diagnostics Residual structures Modeling Issues Categorical variables Time series Day 1 & 2: The Model and its Assumptions Linearity Identifiability Independent variables exogenous Identically, independently, and normally distributed errors Day 3 & 4: Diagnostics Do the assumptions hold? – Multicollinearity – Residual analysis • Outlying & influential data – Heteroskedasticity Day 5 & 6: Modeling Issues Beyond linear models? – Functional forms • Squares, roots, inverses, logarithms – Categorical factors • Dummy variables – Conditional effects • Interactive models Day 7: Binary response variables How should we deal with dichotomous dependent variables? – Probability models: Logit – Maximum likelihood estimation – Interpretation Day 8 & 9 Longitudinal data How should we deal with repeated observations? – Autocorrelation – Time series analysis – Panel data analysis Day 10: Potentials & Limits of Multiple Regression Equilibrium analysis Statistical sophistication vs. measurement precision Temporality in variables and effects Levels of aggregation The Art of Summarizing Relationships