\name{relrisk.ppp}
\alias{relrisk.ppp}
\title{
  Nonparametric Estimate of Spatially-Varying Relative Risk
}
\description{
  Given a multitype point pattern, this function estimates the
  spatially-varying probability of each type of point, or the ratios of
  such probabilities, using kernel smoothing.
  The default smoothing bandwidth is selected by
  cross-validation.
}
\usage{
\method{relrisk}{ppp}(X, sigma = NULL, ..., varcov = NULL, at = "pixels",
           relative=FALSE,
           se=FALSE,
           casecontrol=TRUE, control=1, case)
}
\arguments{
  \item{X}{
    A multitype point pattern (object of class \code{"ppp"}
    which has factor valued marks).
  }
  \item{sigma}{
    Optional. The numeric value of the smoothing bandwidth
    (the standard deviation of isotropic
    Gaussian smoothing kernel).
    Alternatively \code{sigma} may be a function which can be used
    to select a different bandwidth for each type of point. See Details.
  }
  \item{\dots}{
    Arguments passed to \code{\link{bw.relrisk}} to select the
    bandwidth, or passed to \code{\link{density.ppp}} to control the
    pixel resolution.
  }
  \item{varcov}{
    Optional. Variance-covariance matrix of anisotopic Gaussian
    smoothing kernel. Incompatible with \code{sigma}.
  }
  \item{at}{
    String specifying whether to compute the probability values
    at a grid of pixel locations (\code{at="pixels"}) or
    only at the points of \code{X} (\code{at="points"}).
  }
  \item{relative}{
    Logical.
    If \code{FALSE} (the default) the algorithm
    computes the probabilities of each type of point.
    If \code{TRUE}, it computes the    
    \emph{relative risk}, the ratio of probabilities
    of each type relative to the probability of a control.
  }
  \item{se}{
    Logical value indicating whether to compute standard errors
    as well.
  }
  \item{casecontrol}{
    Logical. Whether to treat a bivariate point pattern
    as consisting of cases and controls, and return only the
    probability or relative risk of a case.
    Ignored if there are more than 2 types of points.
    See Details.
  }
  \item{control}{
    Integer, or character string, identifying which mark value
    corresponds to a control. 
  }
  \item{case}{
    Integer, or character string, identifying which mark value
    corresponds to a case (rather than a control)
    in a bivariate point pattern.
    This is an alternative to the argument \code{control}
    in a bivariate point pattern. 
    Ignored if there are more than 2 types of points.
  }
}
\details{
  The command \code{\link{relrisk}} is generic and can be used to
  estimate relative risk in different ways.
  
  This function \code{relrisk.ppp} is the method for point pattern
  datasets. It computes \emph{nonparametric} estimates of relative risk
  by kernel smoothing.

  If \code{X}  is a bivariate point pattern
  (a multitype point pattern consisting of two types of points)
  then by default,
  the points of the first type (the first level of \code{marks(X)})
  are treated as controls or non-events, and points of the second type
  are treated as cases or events. Then by default this command computes
  the spatially-varying \emph{probability} of a case,
  i.e. the probability \eqn{p(u)}
  that a point at spatial location \eqn{u}
  will be a case. If \code{relative=TRUE}, it computes the
  spatially-varying \emph{relative risk} of a case relative to a
  control, \eqn{r(u) = p(u)/(1- p(u))}.

  If \code{X} is a multitype point pattern with \eqn{m > 2} types,
  or if \code{X} is a bivariate point pattern
  and \code{casecontrol=FALSE},
  then by default this command computes, for each type \eqn{j},
  a nonparametric estimate of
  the spatially-varying \emph{probability} of an event of type \eqn{j}.
  This is the probability \eqn{p_j(u)}{p[j](u)}
  that a point at spatial location \eqn{u}
  will belong to type \eqn{j}.
  If \code{relative=TRUE}, the command computes the
  \emph{relative risk} of an event of type \eqn{j}
  relative to a control,
  \eqn{r_j(u) = p_j(u)/p_k(u)}{r[j](u) = p[j](u)/p[k](u)},
  where events of type \eqn{k} are treated as controls.
  The argument \code{control} determines which type \eqn{k}
  is treated as a control.

  If \code{at = "pixels"} the calculation is performed for
  every spatial location \eqn{u} on a fine pixel grid, and the result
  is a pixel image representing the function \eqn{p(u)}
  or a list of pixel images representing the functions 
  \eqn{p_j(u)}{p[j](u)} or \eqn{r_j(u)}{r[j](u)}
  for \eqn{j = 1,\ldots,m}{j = 1,...,m}.
  An infinite value of relative risk (arising because the
  probability of a control is zero) will be returned as \code{NA}.

  If \code{at = "points"} the calculation is performed
  only at the data points \eqn{x_i}{x[i]}. By default
  the result is a vector of values
  \eqn{p(x_i)}{p(x[i])} giving the estimated probability of a case
  at each data point, or a matrix of values 
  \eqn{p_j(x_i)}{p[j](x[i])} giving the estimated probability of
  each possible type \eqn{j} at each data point.
  If \code{relative=TRUE} then the relative risks
  \eqn{r(x_i)}{r(x[i])} or \eqn{r_j(x_i)}{r[j](x[i])} are
  returned.
  An infinite value of relative risk (arising because the
  probability of a control is zero) will be returned as \code{Inf}.

  Estimation is performed by a simple Nadaraja-Watson type kernel
  smoother (Diggle, 2003).
  The smoothing bandwidth can be specified in any of the following ways:
  \itemize{
    \item \code{sigma} is a single numeric value, giving the standard
    deviation of the isotropic Gaussian kernel.
    \item \code{sigma} is a numeric vector of length 2, giving the
    standard deviations in the \eqn{x} and \eqn{y} directions of
    a Gaussian kernel.
    \item \code{varcov} is a 2 by 2 matrix giving the
    variance-covariance matrix of the Gaussian kernel.
    \item \code{sigma} is a \code{function} which selects
    the bandwidth. 
    Bandwidth selection will be applied
    \bold{separately to each type of point}.
    An example of such a function is \code{\link{bw.diggle}}.
    \item 
    \code{sigma} and \code{varcov}
    are both missing or null. Then a \bold{common}
    smoothing bandwidth \code{sigma}
    will be selected by cross-validation using \code{\link{bw.relrisk}}.
  }

  If \code{se=TRUE} then standard errors will also be computed,
  based on asymptotic theory, \emph{assuming a Poisson process}.
}
\value{
  If \code{se=FALSE} (the default), the format is described below.
  If \code{se=TRUE}, the result is a list of two entries,
  \code{estimate} and \code{SE}, each having the format described below.
  
  If \code{X} consists of only two types of points,
  and if \code{casecontrol=TRUE},
  the result is a pixel image (if \code{at="pixels"})
  or a vector (if \code{at="points"}).
  The pixel values or vector values
  are the probabilities of a case if \code{relative=FALSE},
  or the relative risk of a case (probability of a case divided by the
  probability of a control) if \code{relative=TRUE}.

  If \code{X} consists of more than two types of points,
  or if \code{casecontrol=FALSE}, the result is:
  \itemize{
    \item (if \code{at="pixels"})
    a list of pixel images, with one image for each possible type of point.
    The result also belongs to the class \code{"solist"} so that it can
    be printed and plotted.
    \item
    (if \code{at="points"})
    a matrix of probabilities, with rows corresponding to
    data points \eqn{x_i}{x[i]}, and columns corresponding
    to types \eqn{j}.
  }
  The pixel values or matrix entries
  are the probabilities of each type of point if \code{relative=FALSE},
  or the relative risk of each type (probability of each type divided by the
  probability of a control) if \code{relative=TRUE}.

  If \code{relative=FALSE}, the resulting values always lie between 0
  and 1. If \code{relative=TRUE}, the results are either non-negative
  numbers, or the values \code{Inf} or \code{NA}. 
}
\seealso{
  There is another method \code{\link{relrisk.ppm}} for point process
  models which computes \emph{parametric}
  estimates of relative risk, using the fitted model.

  See also 
 \code{\link{bw.relrisk}},
 \code{\link{density.ppp}},
 \code{\link{Smooth.ppp}},
 \code{\link{eval.im}}
}
\examples{
   p.oak <- relrisk(urkiola, 20)
   if(interactive()) {
      plot(p.oak, main="proportion of oak")
      plot(eval.im(p.oak > 0.3), main="More than 30 percent oak")
      plot(split(lansing), main="Lansing Woods")
      p.lan <- relrisk(lansing, 0.05, se=TRUE)
      plot(p.lan$estimate, main="Lansing Woods species probability")
      plot(p.lan$SE, main="Lansing Woods standard error")
      wh <- im.apply(p.lan$estimate, which.max)
      types <- levels(marks(lansing))
      wh <- eval.im(types[wh])
      plot(wh, main="Most common species")
   }
}
\references{
  Diggle, P.J. (2003)
  \emph{Statistical analysis of spatial point patterns},
  Second edition. Arnold.
}
\author{\adrian
  
  
  and \rolf
  
}
\keyword{spatial}
\keyword{methods}
\keyword{smooth}