BART Procedure

GAUSSPRIOR Statement

GAUSSPRIOR prior-options </option > ;

The GAUSSPRIOR statement enables you to specify the prior parameters for a model that is fitted to a normally distributed response variable. Bayesian additive regression trees (BART) models for normal data use an inverse chi square prior for the variance parameter. For more information about the BART prior for a normal data, see the section Variance Prior. You can use the following prior-options to specify the prior for the variance parameter and the initial variance estimate:

SIGMADF=number

specifies the degrees of freedom of the scaled inverse chi-square prior for the variance parameter. The value of number must be greater than or equal to 1.

By default, SIGMADF=3.

SIGMALAMBDA=number

specifies the scale parameter of the scaled inverse chi-square prior for the variance parameter. The value of number must be positive. You cannot specify both this option and the SIGMAQUANTILE= option.

By default, the scale parameter value is determined by the value of the SIGMAQUANTILE= option.

SIGMAQUANTILE=number

specifies the quantile level to be used to determine the scale parameter of the scaled inverse chi-square prior for the variance parameter. The value of number must be between 0 and 1, exclusive. You cannot specify both this option and the SIGMALAMBDA= option.

By default, SIGMAQUANTILE=0.9.

VAREST=number

specifies the initial value for the variance. The value of number must be positive.

By default, the initial variance value is determined by a least squares regression of the response variable on the predictor variables.

You can also specify the following option:

VARAUTOCORR=numeric-list

specifies the autocorrelation lags for the variance parameter.

By default, autocorrelations are computed at lags of 1, 5, 10, and 50.

Last updated: September 22, 2026