Title
betafit postestimation -- Postestimation tools for betafit
Description
This file documents postestimation tools for betafit.
dbetafit displays discrete changes and marginal effects after betafit.
The following standard postestimation commands are also available:
command description ------------------------------------------------------------------------- INCLUDE help post_estat INCLUDE help post_estimates INCLUDE help post_lincom INCLUDE help post_lrtest INCLUDE help post_mfx margins marginal means, predictive margins, marginal effects, and average marginal effects INCLUDE help post_nlcom predict predictions and residuals INCLUDE help post_predictnl INCLUDE help post_suest INCLUDE help post_test INCLUDE help post_testnl -------------------------------------------------------------------------
------------------------------------------------------------------------------- help for dbetafit -------------------------------------------------------------------------------
Syntax for dbetafit
dbetafit [, at(variables_and_values) ]
Description
dbetafit displays changes in predicted dependent variable for three types of discrete changes in explanatory variables and two types of marginal effects:
Discrete changes
Min --> Max shows the change in predicted dependent variable when an explanatory variable changes from its minimum value to its maximum value, while keeping all other variables at their specified values (by default their means). This is the only effect shown for indicator or dummy variables (variables with only two distinct values).
+-SD/2 shows the change in predicted dependent variable when an explanatory variable moves from half a standard deviation below its specified value (by default the mean) to half a standard deviation above its specified value, while keeping all other variables at their specified values. In other words, it shows the effect of a standard deviation change in an explanatory variable, centred on the specified value, on the predicted dependent variable.
+-1/2 shows the change in predicted dependent variable when an explanatory variable moves from half a unit below its specified value to half a unit above its specified value, while keeping all other variables at their specified values. In other words, it shows the effect of a unit change in explanatory variable, centred on the specified value, on the predicted dependent variable.
Marginal effects
MFX at x shows the marginal effect of each (non-indicator or non-dummy) variable, while keeping all variables at their specified values. The marginal effect is the change in predicted dependent variable for a unit change in the explanatory variable, assuming that the effect does not change over that interval.
Max MFX is the maximum marginal effect. Marginal effects change depending on which values of the explanatory variables they are evaluated for. The maximum marginal effect shows the marginal effect you would get if you had chosen those values of the explanatory variables that would maximize the marginal effect. The marginal effect is maximum if the predicted proportion is .5; hence this may lie outside the range of the observed data.
dbetafit is only allowed after betafit in the alternative parameterization, i.e. if one or both muvar() and phivar() is specified or if the alternative option is specified. mfx can be used to get marginal effects after betafit in both the conventional and the alternative parametrization.
Options
at(variables_and_values) is used to specify at which values of the explanatory variables the effects are calculated. variables_and_values is an alternating list of variables and either numeric values or mean, median, min, max, p1, p5, p10, p25, p50, p75, p90, p95, p99. The default is mean for all variables. The statistics p1, p5, ..., p99, are the 1st, 5th, ..., 99th percentiles.
Examples
use http://fmwww.bc.edu/repec/bocode/c/citybudget.dta, clear
betafit governing, mu(minorityleft noleft houseval popdens)
dbetafit, at(minorityleft 0 noleft 0)
(click to run)
Author of dbetafit
Maarten L. Buis, Universitaet Tuebingen maarten.buis@uni-tuebingen.de
------------------------------------------------------------------------------- help for predict -------------------------------------------------------------------------------
Syntax for predict
predict [type] newvar [if] [in] [, statistic var(varname) ]
statistic Description ------------------------------------------------------------------------- proportion proportion (the default) xb xb, fitted values stdp standard error of the prediction sd standard deviation alpha alpha in the conventional parameterization beta beta in the conventional parameterization mu mu in the alternative parameterization, equivalent to proportion phi phi in the alternative parameterization * pearson Pearson residuals * working working residuals * partial partial residuals * scresidual score residuals score first derivative of the log likelihood with respect to the linear predictor. ------------------------------------------------------------------------- INCLUDE help unstarred
Options for predict
proportion (the default) calculates the proportions.
xb calculates the linear prediction.
stdp calculates the standard error of the linear prediction.
sd calculates the estimated standard deviation of the depedent variable.
alpha calculates alpha in the conventional parameterization. This can be specified after estimating a model using betafit, regardless of the parameterization used.
beta calculates beta in the conventional parameterization. This can be specified after estimating a model using betafit, regardless of the parameterization used.
phi calculates phi in the alternative parameterization. This can be specified after estimating a model using betafit, regardless of the parameterization used.
pearson calculates the Pearson residuals, the raw residuals scaled by the estimated standard deviation of the predicted proportion.
working calculates the working residuals as discussed on page 53 of Hardin and Hilbe (2007).
partial calculates the partial residuals as discussed on page 54 of Hardin and Hilbe (2007). This requires that one of the explanatory variables is specified in the var() option.
scresidual calculates the score residuals (see Rocha and Simas (forthcoming)) or "standardized weighted residuals 1" (see Espinheira et al. (2008)). These residuals tend to be more symmetric and normal/Gaussian-like, making them easier to interpret than Pearson residuals.
score calculates the first derivative of the log likelihood with respect to the linear predictions.
References
Cox, Nicholas J. 2005. Speaking Stata: The protean quantile plot. The Stata Journal 5(3): 442-460. http://www.stata-journal.com/article.html?article=gr0018
Espinheira, Patrícia L., Ferrari, Silvia L.P., and Cribari-Neto, Francisco. 2008. On beta regression residuals. Journal of Applied Statistics 35(4): 407-419.
Hardin, James W. and Hilbe, Joseph M. 2007. Generalized Linear Models and Extensions (2nd edition). College Station, TX: Stata Press.
Rocha, Andréa V. and Simas, Alexandre B. (forthcoming). Influence diagnostics in a general class of beta regression models. Mathematics and Statistics.
Winter, Nicholas J.G. 2005. Stata tip 23: Regaining control over axis ranges. The Stata Journal 5(3): 467-468. http://www.stata-journal.com/article.html?article=gr0019
Also see
Online: help for betafit, estimates, lincom, lrtest, mfx, nlcom, predict, predictnl, suest, test, testnl