Model Estimation and Data Analysis: Probit and Logit Models (Logistic Regression)

The probit and logit models (logistic regression) for binary choice are the fundamental building blocks of discrete choice modeling of all sorts. LIMDEP and NLOGIT provide many variants and extensions of these models, including panel data models, two part models and a variety of multivariate specifications, as well as all forms of testing and post estimation analyses.

Binary Choice

  • Functional forms: probit, logit, extreme value (complementary log log), Gompertz, Burr, linear, arctangent
  • Proportions data: minimum chi-squared and MLE
  • Marginal effects
    • Standard errors by delta method
    • Effects for dummy variables
    • Effects evaluated at means and specified configurations
    • Marginal effects by strata
  • Numerous fit measures - tabulations of predictions
  • Predicted probabilities: adjustable threshold for predictions
  • Choice based sampling corrections
  • Robust covariance matrices, cluster, sandwich
  • Weights
  • Linear restrictions: impose or test
    • LM tests for specifications
    • Heteroscedasticity
    • Missing variables
    • LR, Wald tests for restrictions
  • Heteroscedastic probit or logit models
    • ML estimation
    • LM tests
    • Partial effects: average, at means
  • Two step estimation: use previous results or pass results to another estimator
  • Semiparametric
    • Klein and Spady
    • Maximum score
  • Nonparametric regression
  • Panel data
    • Random effects - quadrature or simulation
    • Unconditional fixed effects one and two way
    • Random parameters
    • Latent class
    • Conditional logit fixed effects
    • LR and LM tests for effects

Bivariate and Multivariate Probit

Bivariate probit

  • Recursive bivariate probit
  • Individual or proportions data
  • Partial effects
  • Predictions
  • Partial observability models (Abowd, Poirier, Meng/Schmidt)
  • Choice based sampling
  • Restrictions, LM and LR
  • Sample selection model
  • Panel data, random parameters, random effects
  • Simultaneous equations

Multivariate probit

  • Up to 20 equations - GHK simulator
  • Marginal effects
  • Restrictions and tests of restrictions
  • Sample selection model