predict.iqr: Prediction After Quantile Regression Coefficients Modeling

In qrcm: Quantile Regression Coefficients Modeling

Prediction After Quantile Regression Coefficients Modeling

Description

Predictions from an object of class “iqr”.

Usage

## S3 method for class 'iqr'
predict(object, type = c("beta", "CDF", "QF", "sim"), newdata, p, se = TRUE, ...)

Arguments

object	an object of class “iqr”, the result of a call to iqr.
type	a character string specifying the type of prediction. See ‘Details’.
newdata	an optional data frame in which to look for variables with which to predict. If omitted, the data are used. For type = "CDF", it must include the response variable. Ignored if type = "beta".
p	a numeric vector indicating the order(s) of the quantile to predict. Only used if type = "beta" or type = "QF".
se	logical. If TRUE (the default), standard errors of the prediction will be computed. Only used if type = "beta" or type = "QF".
...	for future methods.

Details

Using iqr, quantile regression coefficients \beta(p) are modeled as parametric functions of p, the order of the quantile. This implies that the model parameter is not \beta(p) itself. The function predict.iqr permits computing \beta(p) and other quantities of interest, as detailed below.

• if type = "beta" (the default), \beta(p) is returned at the supplied value(s) of p. If p is missing, a default p = (0.01, ..., 0.99) is used.

• if type = "CDF", the value of the fitted CDF (cumulative distribution function) and PDF (probability density function) are computed.

• if type = "QF", the fitted values x'\beta(p), corresponding to the conditional quantile function, are computed at the supplied values of p.

• if type = "sim", data are simulated from the fitted model. To simulate the data, the fitted conditional quantile function is computed at randomly generated p following a Uniform(0,1) distribution.

Value

• if type = "beta" a list with one item for each covariate in the model. Each element of the list is a data frame with columns (p, beta, se, low, up) reporting \beta(p), its estimated standard error, and the corresponding 95% confidence interval. If se = FALSE, the last three columns are not computed.

• if type = "CDF", a two-columns data frame (CDF,PDF).

• if type = "QF" and se = FALSE, a data frame with one row for each observation, and one column for each value of p. If se = TRUE, a list of two data frames, fit (predictions) and se.fit (standard errors).

• if type = "sim", a vector of simulated data.

Note

Prediction may generate quantile crossing if the support of the new covariates values supplied in newdata is different from that of the observed data.

Author(s)

Paolo Frumento [e-mail]

See Also

iqr, for model fitting; summary.iqr and plot.iqr, for summarizing and plotting iqr objects.

Examples

  # using simulated data
  
  n <- 1000
  x <- runif(n)
  y <- rlogis(n, 1 + x, 1 + x)
  # true quantile function: Q(p | x) = beta0(p) + beta1(p)*x, with
    # beta0(p) = beta1(p) = 1 + log(p/(1 - p))
  
  model <- iqr(y ~ x, formula.p = ~ I(log(p)) + I(log(1 - p))) 
  # (fit asymmetric logistic distribution)
  
  
  # predict beta(0.25), beta(0.5), beta(0.75)
  predict(model, type = "beta", p = c(0.25,0.5, 0.75))
  
  # predict the CDF and the PDF at new values of x and y
  predict(model, type = "CDF", newdata = data.frame(x = c(.1,.2,.3), y = c(1,2,3)))
  
  # computes the quantile function at new x, for p = (0.25,0.5,0.75)
  predict(model, type = "QF", p = c(0.25,0.5,0.75), newdata = data.frame(x = c(.1,.2,.3)))

  # simulate data from the fitted model
  ysim <- predict(model, type = "sim") # 'newdata' can be supplied

  # if the model is correct, the distribution of y and that of ysim should be similar
  qy <- quantile(y, prob = seq(.1,.9,.1))
  qsim <- quantile(ysim, prob = seq(.1,.9,.1))
  plot(qy, qsim); abline(0,1)

qrcm documentation built on May 29, 2024, 12:09 p.m.