Estimation results

Models

fw_publish, estimates publishes the current estimation results as a model. It writes the same model.json that R and Python publish, so Framework Cloud and the offline reader show a Stata model like any other.

sysuse auto, clear
regress price mpg weight i.foreign
fw_publish, estimates title("Price model") slug(price-model)

The model has a coefficient table with estimates, standard errors, test statistics, p-values, and confidence intervals, plus fit statistics. The mapping runs in Mata. Python is not needed.

Supported commands

Command	Model type	Ratio view	Extras
regress	linear		R², adjusted R², residual SD
logit, logistic	glm, binomial, logit	odds ratio	pseudo R², null log likelihood
probit	glm, binomial, probit		pseudo R²
poisson	glm, poisson, log	rate ratio	pseudo R²
nbreg	glm, negative binomial, log	rate ratio	alpha
glm	glm, family and link from Stata	by link	deviance, dispersion
mixed	mixed		random effects, groups, REML
stcox	survival	hazard ratio	events
ivregress	linear		estimator, instruments

Every model also gets the number of observations and, when Stata has a log likelihood, AIC and BIC.

Other commands that leave e(b) and e(V) publish a generic coefficient table. fw_publish prints a note when it does.

Coefficient table

• Estimates stay on the model scale. logistic results are stored as log odds. The reader offers odds, rate, and hazard ratios.
• _cons becomes (Intercept).
• Base and omitted factor levels are left out.
• Multi-equation models prefix terms with the equation name.
• Models with residual degrees of freedom use t statistics. The rest use z statistics.

Ancillary parameters

Parameters that are not coefficients stay out of the coefficient table:

• nbreg reports alpha, or delta with dispersion(constant), as a fit statistic. R and Python report theta the same way.
• Other free parameters, such as tobit's var(e.y) and ologit's cutpoints, go in an Ancillary parameters table. Variances, covariances, correlations, cutpoints, and tobit's sigma get no test, as in Stata's output. Variances and standard deviations get Stata's log-scale interval, and correlations its hyperbolic-arctangent interval.
• tobit also reports sigma, the residual standard deviation.

level() sets the confidence level. The default is c(level), 95 unless you changed it with set level, as in R and Python. fw_publish does not read the level of the estimation command: after regress, level(90), give level(90) to publish 90% intervals.

logit foreign price mpg
fw_publish, estimates level(90)

Mixed models

mixed adds a random-effects table: a standard deviation for each component, correlations, and the residual. It covers independent, identity, exchangeable, and unstructured covariance. The model also records the number of groups at each level and whether it used REML.

webuse pig, clear
mixed weight week || id: week, covariance(unstructured)
fw_publish, estimates title("Growth model")

Saved copy

saving() keeps a copy of model.json.

fw_publish, estimates saving("$fw_outputs/price-model.json", replace)

Destinations, versions, and stored results work as for files. See Publishing.