Stata 15 help for xt_vce_options

[XT] vce_options -- Variance estimators


estimation_cmd ... [, vce_options ...]

vce_options Description ------------------------------------------------------------------------- vce(oim) observed information matrix (OIM) vce(opg) outer product of the gradient (OPG) vectors vce(robust) Huber/White/sandwich estimator vce(cluster clustvar) clustered sandwich estimator vce(bootstrap [, bootstrap_options]) bootstrap estimation vce(jackknife [, jackknife_options]) jackknife estimation

nmp use divisor N - P instead of the default N scale(x2|dev|phi|#) override the default scale parameter; available only with population-averaged models -------------------------------------------------------------------------


This entry describes the vce_options, which are common to most xt estimation commands. Not all the options documented below work with all xt estimation commands; see the documentation for the particular estimation command. If an option is listed there, it is applicable.

The vce() option specifies how to estimate the variance-covariance matrix (VCE) corresponding to the parameter estimates. The standard errors reported in the table of parameter estimates are the square root of the variances (diagonal elements) of the VCE.


+-----------+ ----+ SE/Robust +--------------------------------------------------------

vce(oim) is usually the default for models fit using maximum likelihood. vce(oim) uses the observed information matrix (OIM); see [R] ml.

vce(opg) uses the sum of the outer product of the gradient (OPG) vectors; see [R] ml. This is the default VCE when the technique(bhhh) option is specified; see [R] maximize.

vce(robust) uses the robust or sandwich estimator of variance. This estimator is robust to some types of misspecification so long as the observations are independent; see [U] 20.22 Obtaining robust variance estimates.

If the command allows pweights and you specify them, vce(robust) is implied; see [U] 20.24.3 Sampling weights.

vce(cluster clustvar) specifies that the standard errors allow for intragroup correlation, relaxing the usual requirement that the observations be independent. That is to say, the observations are independent across groups (clusters) but not necessarily within groups. clustvar specifies to which group each observation belongs, for examples, vce(cluster personid) in data with repeated observations on individuals. vce(cluster clustvar) affects the standard errors and variance-covariance matrix of the estimators but not the estimated coefficients; see [U] 20.22 Obtaining robust variance estimates.

vce(bootstrap [, bootstrap_options]) uses a nonparametric bootstrap; see [R] bootstrap. After estimation with vce(bootstrap), see [R] bootstrap postestimation to obtain percentile-based or bias-corrected confidence intervals.

vce(jackknife [, jackknife_options]) uses the delete-one jackknife; see [R] jackknife.

nmp specifies that the divisor N-P be used instead of the default N, where N is the total number of observations and P is the number of coefficients estimated.

scale(x2|dev|phi|#) overrides the default scale parameter. By default, scale(1) is assumed for the discrete distributions (binomial, negative binomial, and Poisson), and scale(x2) is assumed for the continuous distributions (gamma, Gaussian, and inverse Gaussian).

scale(x2) specifies that the scale parameter be set to the Pearson chi-squared (or generalized chi-squared) statistic divided by the residual degrees of freedom, which is recommended by McCullagh and Nelder (1989) as a good general choice for continuous distributions.

scale(dev) sets the scale parameter to the deviance divided by the residual degrees of freedom. This option provides an alternative to scale(x2) for continuous distributions and for over- or underdispersed discrete distributions.

scale(phi) specifies that the scale parameter be estimated from the data. xtgee's default scaling makes results agree with other estimators and has been recommended by McCullagh and Nelder (1989) in the context of GLM. When comparing results with calculations made by other software, you may find that the other packages do not offer this feature. In such cases, specifying scale(phi) should match their results.

scale(#) sets the scale parameter to #. For example, using scale(1) in family(gamma) models results in exponential-errors regression (if you assume independent correlation structure).


When working with panel-data models, we strongly encourage you to use the vce(bootstrap) or vce(jackknife) options instead of the corresponding prefix command. For example, to obtain jackknife standard errors with xtlogit, type

. webuse clogitid . xtlogit y x1 x2, fe vce(jackknife)

If you wish to specify more options to the bootstrap or jackknife estimation, you can include them within the vce() option. Below we refit our model requesting bootstrap standard errors based on 300 replications, we set the random-number seed so that our results can be reproduced, and we suppress the display of the replication dots.

. xtlogit y x1 x2, fe vce(bootstrap, reps(300) seed(123) nodots)


McCullagh, P., and J. A. Nelder. 1989. Generalized Linear Models. 2nd ed. London: Chapman & Hall/CRC.

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index