3.11 Miscellaneous & Conclusion

• We can combine multiple approaches we have learned throughout the class for flexible model specifications
  • E.g., Can add polynomial terms (e.g., age²) to a logit/probit model to allow the curve to change directions in a way other than the normal S-curve
• See Appendix 3E for several examples and advice for flexible IV functional forms in logit/probit models

Moving forward

• The binary regression model is a building block for most all categorical models:
  • Ordinal and nominal models are straightforward extensions
  • Zero-inflated count models combine count and binary models
• Many latent variable models (for categorical items and/or categorical latent variables) build directly on binary regression models