tobit now accepts censoring limits and constraints
Some people think of tobit as being censored at zero. Stata’s tobit estimation command allows you to specify the lower value of the censoring point and specify an upper censoring point. All that is unchanged. You can now specify censoring points—upper, lower, or both—that vary observation by observation. The censoring points can be stored in variables.
tobit now allows constraints.
tobit now has the other standard features that it always should have had, but this is just for completeness. You can, for instance, specify initial values.
tpoisson, ul()
The existing estimation command tpoisson fits truncated Poisson models. It previously fit only left-truncated models. It now fits left-, right-, and both-truncated models. New option ul() specifies the upper truncation limit.
One- and two-sample mean tests with clustered data
Existing command ztest has new option cluster() and other new options to account for clustering.
One- and two-sample proportion tests with clustered data
Existing command prtest has new option cluster() and other new options to account for clustering.
gsem now fits truncated Poisson models
gsem, whether used to fit the new LCA models or the existing generalized SEM models, now fits truncated Poisson models if you specify option family(poisson, ltruncated(…)).
Standard deviations and correlations instead of variances and covariances for multilevel models and generalized SEM
For multilevel models, estat sd displays random effects and within-group error parameter estimates as standard deviations and correlations instead of the variances and covariances reported in the estimation output.
Similarly after gsem, estat sd reports the estimated variance components as standard deviations and correlations.
bayesmh has new options for displaying results
bayesmh now allows the eform and eform(string) options for reporting exponentiated coefficients such as odds ratios, incidence-rate ratios, and the like.
bayesmh now allows new option show(paramlist) to specify which model parameters should be presented in the output. Option show() joins existing option noshow(). Specify one, the other, or neither.
bayesmh now allows new option showreffects to specify that all random-effects estimates be presented in the output. They are not displayed by default.
Postestimation supports new bayes: prefix command
If you use the new bayes: prefix command with multilevel models such as mixed or meglm, then bayesgraph, bayesstats ess, and bayesstats summary have new options.
New option showreffects displays the results for all random-effects parameters.
New option showreffects() displays specified random-effects parameters.
By default, results are displayed for all model parameters except the random-effects parameters.
These new estimation commands may be used with the svy: prefix:
| Command | Purpose |
|---|---|
| svy: asmixlogit | Alternative-specific mixed logit regression |
| svy: heckpoisson | Poisson regression with sample selection |
| svy: hetregress | Heteroskedastic linear regression |
| svy: stintreg | Parametric interval-censored survival regression |
| svy: zioprobit | Zero-inflated ordered probit |
| svy: metobit | Multilevel tobit regression |
| svy: meintreg | Multilevel interval regression |
| svy: eregress | Extended linear regression |
| svy: eintreg | Extended interval regression |
| svy: eprobit | Extended probit regression |
| svy: eoprobit | Extended ordered probit regression |
| svy: gsem | For latent class analysis |
The following existing estimation commands support combined use of svy: and fmm: to fit survey-adjusted finite mixture models:
| Command | Purpose |
|---|---|
| svy: fmm: regress | Linear regression |
| svy: fmm: tobit | Tobit regression |
| svy: fmm: intreg | Interval regression |
| svy: fmm: truncreg | Truncated regression |
| svy: fmm: ivregress | Instrumental-variable regression |
| svy: fmm: logit | Logistic regression |
| svy: fmm: probit | Probit regression |
| svy: fmm: cloglog | Conditional log-log regression |
| svy: fmm: ologit | Ordered logistic regression |
| svy: fmm: oprobit | Ordered probit regression |
| svy: fmm: mlogit | Multinomial logistic regression |
| svy: fmm: poisson | Poisson regression |
| svy: fmm: nbreg | Negative binomial regression |
| svy: fmm: tpoisson | Truncated Poisson regression |
| svy: fmm: betareg | Beta regression |
| svy: fmm: glm | Generalized linear model |
| svy: fmm: streg | Parametric survival regression |
Cauchy distribution
A new family of Cauchy distribution functions—cauchyden(), cauchy(), cauchytail(), invcauchy(), invcauchytail(), and lncauchyden()—compute the density, cumulative distribution, reverse cumulative distribution, inverse cumulative distribution, and natural logarithm of the density.
rcauchy is a Cauchy random-number generator.
Laplace distribution
A new family of Laplace distribution functions—laplaceden(), laplace(), laplacetail(), invlaplace(), invlaplacetail(), and lnlaplaceden()—compute the density, cumulative distribution, reverse cumulative distribution, inverse cumulative distribution, inverse reverse cumulative distribution, and natural logarithm of the density.
rlaplace() is a Laplace random number generator.
Multivariate normal distribution
Mata functions are now available for calculating values and derivatives of the multivariate normal distribution.
