LOGSELECT Procedure

OUTPUT Statement

OUTPUT OUT=libref.data-table<options><keyword <=name>>…<keyword <=name>>;

The OUTPUT statement creates a data table that contains observationwise statistics that PROC LOGSELECT computes after fitting the model. In order to avoid data duplication for large data tables, the variables in the input data table are not included in the output data table unless you specify them in the COPYVAR= option.

If your response variable has more than two categories, you can request the "Basic Options" listed in Table 9; the other diagnostic statistics are not available unless you use multinomial-trial syntax. These statistics are computed for every response category, and the automatic variable _LEVEL_ identifies the response category on which the computed values are based. That is, every observation generates several rows in the output data set. If you also specify the OBSCAT option, then the observationwise statistics are computed only for the observed response category, which is indicated by the value of the _LEVEL_ variable. If you specify the PREDPROBS option, then only the PRED and IPRED statistics are output to the data table, and a variable is created to contain the values that correspond to each response category. PROC LOGSELECT names these variables by concatenating the PRED= or IPRED= name with the response category.

If you use multinomial-trial syntax, then the PREDPROBS option is invoked by default, and all the "Diagnostic Options" listed in Table 9 (except for the RESWORK statistic) are available as output to the data table. However, if you also specify the NOPREDPROBS option, then only the RESCHI, RESDEV, and RESRAW statistics are available.

The output statistics are computed based on the final parameter estimates. If the optimization does not converge, then the output data table is not created. If you perform elastic net or LASSO selection, or if you specify the NOXPX or NOSTDERR option, then your analysis does not generate a covariance matrix, and you cannot produce the basic and diagnostic statistics in Table 9 that require a covariance matrix.

For observations in which only the response variable is missing, values of the linear predictor and the predicted probabilities are computed even though these observations do not affect the model fit. This enables, for example, predicted probabilities to be computed for new observations.

You must specify the following option:

OUT=libref.data-table

names the output data table for PROC LOGSELECT to use. You must specify this option before any other options. libref.data-table is a two-level name, where

libref: refers to a collection of information that is defined in the LIBNAME statement and includes the library, which includes a path to where the data table is to be stored, and a session identifier, which defaults to the active session but which can be explicitly defined in the LIBNAME statement. For more information about libref, see the section Using CAS Sessions and CAS Engine Librefs.
data-table: specifies the name of the output data table.

You can also specify the following options:

ALL ALLSTAT

adds all available statistics to the output data table.

ALPHA=number

specifies the significance level for the construction of confidence intervals in the output data table. The confidence level is . The value of number must be between 0 and 1. By default, number is equal to the value of the ALPHA= option in the PROC LOGSELECT statement, or 0.05 if that option is not specified.

COPYVAR=variable COPYVARS=(variables)

transfers one or more variables from the input data table to the output data table.

FIXEDOFFSET=number

specifies a value to use as an offset to the linear predictor instead of obtaining the value from the offset variable. This option is ignored unless you also specify the OFFSET= variable in the MODEL statement.

OBSCAT

requests (for multinomial models) that observationwise statistics be produced only for the observed response level. If you do not specify this option and the response variable has J levels, then the following outputs are created: for cumulative link models, J–1 records are output for every observation in the input data that corresponds to the J–1 lower-ordered response categories; for generalized logit models, J records are output that correspond to all J response categories. This option is not available with multinomial-trial syntax.

PREDPROBS | NOPREDPROBS

creates variables (for multinomial models) for each response category that corresponds to the requested IPRED and PREDICTED keywords. If you use multinomial-trial syntax, then the PREDPROBS option is invoked by default and enables computation of most of the "Diagnostic Options" listed in Table 9. You can override this behavior by specifying the NOPREDPROBS option.

keyword <=name>

specifies a statistic to include in the output data table and optionally names the variable name. If you do not provide a name, the LOGSELECT procedure assigns a default name based on the type of statistic requested. If a statistic is not available for your model then that request is ignored.

Table 9 summarizes the keywords that are available in the OUTPUT statement.

Table 9: OUTPUT Statement Keywords

Keyword	Description	Default Name
Basic Options
INDIVIDUAL	Specifies the individual predicted probabilities	`_IPRED_`
POST	Specifies the posterior predicted probabilities	`_POST_`
PREDICTED	Specifies the predicted probabilities	`_PRED_`
RESRAW	Specifies the raw residual	`_RESRAW_`
XBETA	Specifies the linear predictor	`_XBETA_`
Basic Options Requiring the Covariance
LCL	Specifies the lower confidence limit for the linear predictor	`_LCL_`
LCLM	Specifies the lower confidence limit for the event probability	`_LCLM_`
STDXBETA	Specifies the standard error estimate of the linear predictor	`_STDXBETA_`
UCL	Specifies the upper confidence limit for the linear predictor	`_UCL_`
UCLM	Specifies the upper confidence limit for the event probability	`_UCLM_`
Diagnostic Options
RESCHI	Specifies the Pearson chi-square residual	`_RESCHI_`
RESDEV	Specifies the deviance residual	`_RESDEV_`
RESWORK	Specifies the working residual	`_RESWORK_`
Diagnostic Options Requiring the Covariance
CBAR	Specifies the confidence interval displacement	`_CBAR_`
DIFCHISQ	Specifies the deletion chi-square goodness-of-fit change	`_DIFCHISQUARE_`
DIFDEV	Specifies the deletion deviance change	`_DIFDEVIANCE_`
H	Specifies the leverage	`_HATDIAG_`
RESLIK	Specifies the likelihood residual	`_RESLIK_`
STDRESCHI	Specifies the standardized Pearson chi-square residual	`_STDRESCHI_`
STDRESDEV	Specifies the standardized deviance residual	`_STDRESDEV_`
Option for Generalized Estimating Equations
CLEVERAGE	Names the cluster leverage	`_CLEVERAGE_`
Miscellaneous Options
INTO	Names the level into which the observation is classified	`_INTO_`
LEVEL	Names the response level for a row of the output	`_LEVEL_`
ROLE	Names the role played by the observation in fitting the model	`_ROLE_`

The following list describes these keywords. For more information, see the section Predicted Probabilities and Regression Diagnostics.

CBAR

specifies the confidence interval displacement diagnostic that measures the overall change in the global regression estimates that results from deleting an individual observation. The default name is _CBAR_.

CLEVERAGE CH CLEV

specifies the cluster leverage for GEEs. For more information about diagnostic statistics, see the section Diagnostics for Models Fit by Generalized Estimating Equations (GEEs). The default name is _CLEVERAGE_.

DIFCHISQ

specifies the change in the chi-square goodness-of-fit statistic that results from deleting the individual observation. The default name is _DIFCHISQUARE_.

DIFDEV

specifies the change in the deviance that results from deleting the individual observation. The default name is _DIFDEVIANCE_. This statistic is not supported for GEEs.

H

specifies the diagonal element of the hat matrix (leverage) for detecting extreme points in the design space. For GEEs, the leverage value is the corresponding diagonal element of the cluster leverage matrix. The default name is _HATDIAG_.

If you use multinomial-trial syntax and do not specify the NOPREDPROBS option, then the relevant portion of the hat matrix for observation i is a J–1-dimensional submatrix . Three statistics are computed from this submatrix: the determinant of , the average diagonal element of , and the potential (average diagonal element of ). These statistics are identified in the output table by appending the letters 'D', 'T', and 'P', respectively, to the provided name; by default, their names are _H_D, _H_T, and _H_P.

INDIVIDUAL IPRED IPROB IP

specifies the individual predicted values for the multinomial response levels. For a response variable Y with three levels, 1, 2, and 3, the individual probabilities are , , and . The default name is _IPRED_.

INTO<(cutpoint)>

names the variable that contains the level of the response into which an observation is classified. The default name is _INTO_. Multinomial models classify observations into the level that has the largest model-predicted probability. For binary or binomial response variables, if the predicted probability of an observation equals or exceeds the cutpoint, the observation is classified as an event; otherwise it is classified as a nonevent. You can specify the cutpoint value as a number between 0 and 1. The default value is 0.5.

LCL LOWERXBETA

names the variable that contains the lower confidence limits for the linear predictor. You can set the confidence level by specifying the ALPHA= option. The default name is _LCL_.

LCLM LOWERMEAN LOWER

specifies the lower confidence limits for the probability of the event. You can set the confidence level by specifying the ALPHA= option. The default name is _LCLM_.

LEVEL

names the variable that contains the level of the response for a given row of the output. The default name is _LEVEL_.

POSTERIOR POST

specifies the posterior predicted probability of each observation. If you do not specify the PRIOR= option in the MODEL statement, then the observed proportions from the training data are used as priors, in which case the posteriors from the training data are the same as the individual predicted probabilities. The default name is _POST_.

PREDICTED PRED PROB P

specifies the predicted values (predicted probabilities of events) for binary and nominal response variables and the cumulative predicted probabilities for ordinal response variables. For a response variable Y with three levels, 1, 2, and 3, the cumulative probabilities are and , but by default the last level, , is not output. The default name is _PRED_.

RESCHI PEARSON

specifies the Pearson residual for identifying poorly fitted observations. The default name is _RESCHI_.

RESDEV

specifies the deviance residual for identifying poorly fitted observations. The default name is _RESDEV_. This statistic is not supported for GEEs.

RESLIK

specifies the likelihood residual for identifying poorly fitted observations. The default name is _RESLIK_. This statistic is not supported for GEEs.

RESRAW RESIDUAL R

specifies the raw residual for identifying poorly fitted observations. The default name is _RESRAW_.

RESWORK

specifies the working residual for identifying poorly fitted observations. This statistic is not available for multinomial response models. The default name is _RESWORK_.

ROLE

specifies the numeric variable that indicates the role played by each observation in fitting the model. The default name is _ROLE_. Table 10 shows how this variable is interpreted for each observation.

Table 10: Role Interpretation

Value	Observation Role
0	Not used
1	Training
2	Validation
3	Testing

If you do not partition the input data by specifying a PARTITION statement, then the role variable value is 1 for observations that are used in fitting the model and 0 for observations that have at least one missing or invalid value for the response, regressor, frequency, or weight variables.

STDRESCHI

specifies the standardized Pearson (chi-square) residual for identifying observations that are poorly accounted for by the model. The default name is _STDRESCHI_.

STDRESDEV

specifies the standardized deviance residual for identifying poorly fitted observations. The default name is _STDRESDEV_. This statistic is not supported for GEEs.

STDXBETA

specifies the standard error estimates of XBETA. The default name is _STDXBETA_.

UCL UPPERXBETA

specifies the variable that contains the upper confidence limits for the linear predictor. The default name is _UCL_. You can set the confidence level by specifying the ALPHA= option.

UCLM UPPERMEAN UPPER

specifies the variable that contains the upper confidence limits for the probability of the event response. The default name is _UCLM_. You can set the confidence level by specifying the ALPHA= option.

XBETA LINP

specifies the linear predictor. The default name is _XBETA_.

Last updated: June 22, 2026