Seemingly Unrelated Estimation for Model Comparisons
The suest package lets you compare predictions and marginal effects across regression models. It combines two or more separately fitted models into one object with a joint robust covariance matrix. The combined object works with the marginaleffects package to compare predictions, slopes, and average comparisons across models.
Supported models
lm(), binary logit, probit, and complementary-log-log models fitted by glm() or glm2::glm2(), Poisson and other supported GLMs, negative-binomial models fitted by MASS::glm.nb(), ordered logit and probit models fitted by MASS::polr() or restricted ordinal::clm() specifications, multinomial logit models fitted by nnet::multinom(), and parametric survival or censored-regression models fitted by survival::survreg(), and beta regressions fitted by betareg::betareg(). Poisson and negative-binomial zero-inflated models fitted by pscl::zeroinfl() are also supported. Truncated Gaussian regressions fitted by truncreg::truncreg() are supported as well, as are direct Tobit models fitted by censReg::censReg() and unweighted two-stage least squares fitted by fixest::feols() without absorbed fixed effects. Heteroskedastic binary probit and logit fitted by Rchoice::hetprob() are supported. Maximum-likelihood IV probit fitted by Rchoice::ivpml() and bivariate probit fitted by mvProbit::mvProbit() are supported with their documented adapter restrictions. Unweighted individual fixed-effects, between-effects, and Swamy-Arora random-effects linear panel models fitted by plm::plm() are supported; unbalanced random-effects fits can retain small engine-specific differences. Single-level random-intercept Gaussian models fitted by nlme::lme() with method = "ML" are supported, including both variance components. Unweighted individual random-intercept binary logit/probit and gamma random-effects Poisson models fitted by pglm::pglm() are supported under the restrictions documented in suest(). Unweighted binomial-logit, Poisson-log, and negative-binomial NB2 log models fitted by glmmTMB::glmmTMB() with one conditional random intercept are supported under the restrictions documented in suest(). Unweighted GEE fitted by geepack::geeglm() is supported for Gaussian identity, binary logit/probit/cloglog, and Poisson log models with independence or exchangeable correlation. Explicit weight_type = "pweight" support is available for the model families corresponding to Stata suest2’s ordinary pweight route. Joint cluster-robust covariance is available through the cluster argument to suest(). Compatible systems fitted in multiple imputed datasets can be pooled with suest_mi() using Rubin’s rules. The initial MI interface supports joint coefficients, covariance, summaries, and coefficient-level hypotheses.
Reference
Mize, Trenton D., Long Doan, and J. Scott Long. 2019. “A General Framework for Comparing Predictions and Marginal Effects Across Models.” Sociological Methodology 49(1):152–189. doi:10.1177/0081175019852763
See also
Useful links: