Seminar in Advanced Political Science Methodology (2011) [Syllabus]
Topic 1: Binary Models
Predicted Probabilities from Logistic Regression Models
These handouts identify methods in Stata and R to calculate predicted probabilities.
Predicted Effects Graphs for Continuous Variables
Here, you can see how to construct a predicted probability graph over the values of a continuous variable holding the others constant at some typical value.
Predicted Effects Graphs for Continuous Variables
Here, you can see how to construct a predicted probability graph over the values of a continuous variable holding the others constant at some typical value. For the Stata version of this, I modified, slightly the praccum command that comes with SPost. This may have since been changed, but at the time praccum would not save the information about the confidence bounds. The praccum2.do command will save the confidence bounds of the predicted probabilities.
Predicted Effects Graphs for Continuous Variable Interactions
There is some controversy about what exactly is necessary/sufficient to conclude that there is an interaction effect in non-linear models like this (see Berry, DeMeritt and Esarey, AJPS 2010). FOr now, we will treat this essentially like interactions in linear models.
Predicted Effects Graphs for Interactions (2)
This is a graph that plots the interaction of a continuous and categorical variable.
Confidence Bounds on Predicted Probabilities
Here we are using simulation to get confidence bounds for predicted probabilities. This is accomplished easily with Clarify in Stata and the R handout shows how to do Clarify-type simulations "by hand" in R.
Average Effects
Here, we consider a different strategy for figuring out effects. The average effect, as discussed by Hanmer and Kalkan in "Behind the Curve", uses small deviations from every different potential starting-point in the data. The figures below show average effects by other variables, which may be useful.
Average Effects (Dummy Variable)
Topic 2: Model Fit
Scalar Measures of Fit
Proportional Reduction in Error
Proportional Reduction in Error is one way of figuring out how well the data fit the model.
ROC Curve Analysis
Likelihood Ratio Tests
Model Testing
Dealing with Model Selection Uncertainty
Topic 3: Ordinal Data Models
Predicted Probabilities I
Predicted Probabilities II
Parallel Regressions Assumption
Model Fit
Topic 4: Nominal Data Models
Estimation
Predicted Probabilities I
Predicted Probabilities II (polynomials)
Model Fit and Testing
Conditional Logit
Topic 5: Count Data Models
Poisson Estimation
Predicted Rates
Negative Binomial Estimation
Negative Binomial Predicted Rates
Zero-Inflated Negative Binomial Estimation
Zero-Inflated Negative Binomial Effects
Topic 6: Duration Models
Exponential Models
Weibull Models
Cox Proportional Hazards Models
Binary Time-Series Cross-Section Models
Topic 7: Missing Data and Multiple Imputation
|
|
|
|
|
