CRITERIA Subcommand (QUANTILE REGRESSION command)

The optional CRITERIA subcommand specifies the model settings.

QUANTILE = value(s)

The optional keyword specifies the quantiles. A single value 0.5 is the default value. Multiple values can be specified with each value separated by a blank space (or blank spaces). A grid of quantiles from value1 to value2 with the increment of value3, connected by the keywords TO and BY, can also be specified. When a grid of quantiles is specified, only one valid set of [value1 TO value2 BY value3] is allowed. It must satisfy that 0 ≤ value1 ≤ value2 ≤ 1. In cases where value1 = value2, it is equivalent to specifying a single value1, regardless of value3. In order to restrict the number of quantiles, 0.01 ≤ value3 ≤ 1. The grid can be mixed with other quantiles, and can be placed anywhere.

Notes:

• All specified values must be in [0,1].
• Only unique values are used in cases where duplicated values are specified (a warning message is generated to indicate that all the duplicated values are excluded from the analysis).

IID = boolean

The optional keyword determines whether or not the error terms are independently and identically distributed. This is an option for the post-estimation of the variance-covariance of the parameter estimates and the confidence intervals for the predicted target values. TRUE is the default setting. When FALSE is specified, the computation time may be significantly increased for large models.

BANDWIDTH = {BOFINGER**} {HALL_SHEATHER}

The optional keyword determines which bandwidth method is used to estimate the variance-covariance matrix of the parameter estimates. The BOFINGER bandwidth method is the default setting.

CILEVEL = value

The optional keyword specifies the significance level. The value must be a single double value in (0,100). 95 as the default value.

METHOD = {AUTO**}{SIMPLEX}{INTERIOR_POINT}

The optional keyword specifies the model estimation method. The SIMPLEX option calls for the simplex algorithm developed by Barrodale and Roberts. The INTERIOR_POINT option calls the Frisch-Newton interior-point non-linear optimization algorithm. The AUTO option is the default setting, which allows the procedure to automatically choose the appropriate estimation method.

TOL = {1E-12**} {value}

The optional keyword specifies the tolerance value for the matrix manipulations in the interior-point method. Must be a single double value in (0,10−3] with 10−12 as the default setting.

CONV = {1E-6**} {value}

The optional keyword specifies the convergence criterion for the numerical method. Must be a single double value in (0,10−3] with 10−6 as the default setting.

MAXITER = {2000**} {integer}

The optional keyword specifies the maximum number of iterations. The value must be a single positive integer with 2000 as the default setting.