predict.selection function

Predict method for fitted sample selection models

Calculate predicted values for sample selection models fitted with function selection.

## S3 method for class 'selection'
predict( object, newdata = NULL,
   part = ifelse( type %in% c( "unconditional", "conditional" ),
      "outcome", "selection" ),
   type = "unconditional", ... )

Arguments

• object: a fitted object of class selection.

• newdata: optionally, a data frame in which to look for variables with which to predict. If omitted, the fitted linear predictors or the fitted response values are returned.

• part: character string indicating for which equation the predicted variables should be calculated: "selection" for the predicted values of the selection equation and "outcome" for the predicted values of the outcome equation; this argument is automatically chosen depending on the value of argument type.

• type: if argument type is "link"

  and argument part is "selection", the linear predictors of the selection equation are returned; if argument type is "response"

  and argument part is "selection", the predicted probabilities of the selection equation are returned; if argument type is "unconditional"

  and argument part is "outcome", the unconditional expectations are returned, i.e. E[y|X] = X %*% beta; if argument type is "conditional"

  and argument part is "outcome", the conditional expectations are returned, i.e. E[y|X,Z,w=1] = X %*% beta + rho * sigma * dnorm( Z %*% gamma ) / pnorm(Z %*% gamma ).

• ...: further arguments (currently ignored).

Returns

In most cases, a numeric vector of the predicted values is returned. However, there are three exceptions: (i) when predicting the unconditional expectations of a Tobit-5 model, a matrix with two columns is returned, where the two columns correspond to the two outcome equations (E[yo1] and E[yo2]); (ii) when predicting the conditional expectations of a Tobit-2 model, a matrix with two columns is returned, where the first column returns the expectations conditional on the observation being not selected (E[yo|ys=0]), while the second column returns the expectations conditional on the observation being selected (E[yo|ys=1]); (iii) when predicting the conditional expectations of a Tobit-5 model, a matrix with four columns is returned, where the first two columns return the conditional expectations of the first outcome equation (E[yo1|ys=0] and E[yo1|ys=1]) and the last two columns return the conditional expectations of the second outcome equation (E[yo2|ys=0] and E[yo2|ys=1]).

Author(s)

Arne Henningsen and fg nu

(the code is partly based on the code posted by fg nu at https://stackoverflow.com/questions/14005788/predict-function-for-heckman-model)

See Also

selection, predict, predict.probit, residuals.selection, and selection-methods.

Examples

## Greene( 2003 ): example 22.8, page 786
data( Mroz87 )
Mroz87$kids  <- ( Mroz87$kids5 + Mroz87$kids618 > 0 )

# ML estimation
m <- selection( lfp ~ age + I( age^2 ) + faminc + kids + educ,
   wage ~ exper + I( exper^2 ) + educ + city, Mroz87 )
   
predict( m )
predict( m, type = "conditional" ) 
predict( m, type = "link" ) 
predict( m, type = "response" ) 
predict( m, newdata = Mroz87[ 3:9, ] )