help for svrest

Replication-based estimates of covariances for complex survey data

svrest "command" "exp_list" [if exp] [in range] [, [matrices(list) noisily or liststats diparm(eq type name [ci] [ \ ...]) ]


svrest runs the user-specified command, and calculates covariance estimates for the statistics specified in exp_list, using replication methods for complex survey data. svrest is based on bs.

The expressions in exp_list must be separated by spaces and there must be no spaces within each expression. Note that command and exp_list must both be enclosed in double quotes.

See survwgt and svr for details on these replication methods.

svrest estimates the statistics jointly, so it calculates both the variances of each statistic, and the covariances among them. The results are stored as estimates.

See svresttest for testing hypotheses involving the estimated statistics generated by this command.


matrices indicates one entire matrices of results to be included. This allows the easy display of the entire coefficient vector when using svrest for an estimation command not supported by brrmodel. For example, the following would compute replication-based standard errors for the full coefficient vector from a heckman model:

. svrest "heckman income age race weight, select(education)" , mat(e(b))

Important note: It is vital that the estimation in each replicate produce the same coefficient vector. Thus, for example, when running mlogit, be sure to include the base() option, so that the same category is omitted from the estimation for every replicate.

dots requests a dot be placed on the screen at the beginning of each replication, thus providing entertainment of a sort when long, slow commands are run.

noisily requests that output from each replication be displayed.

or requests that statistics be displayed as odds ratios.

liststats requests that a key be displayed, indicating what each statistic is.

diparm requests that single equation statistics be redisplayed in the natural metric. You must indicate the name of the equation containing the statistic, the type of transformation, the name for the statistic, and, optionally, ci to indicate that the confidence interval should be appropriately transformed. When specified, this option simply calls _diparm to redisplay statistics that are not estimated in their natural metric. For example, for heckman, you might do the follwowing:

. svrest "heckman income age, select(educ)" , mat(e(b)) diparm(athrho exp rho ci \ lnsigma exp sigma di)


. svrest "reg mpg weight foreign" "_b[weight] _b[foreign]"

. svrest "kappa cat1-cat3" "r(z)"


I would like to thank members of the Stata technical team, whose suggestions led to the creation of this command.


Nick Winter Cornell University nw53@cornell.edu