new function obs_exp is a diagnostic function to compare observed and…

… expected counts at different covariate aggregations

.Rbuildignore

@@ -1,6 +1,5 @@
 .travis.yml
 LICENSE
-R/obs_exp.R
 ^\.travis\.yml$
 ^appveyor\.yml$
 ^tic\.R$

NAMESPACE

@@ -21,6 +21,7 @@ export(generate.ds.uncertainty)
 export(generate.mb.sample)
 export(latlong2km)
 export(make.soapgrid)
+export(obs_exp)
 export(plot_pred_by_term)
 export(rqgam.check)
 export(trim.var)
@@ -47,6 +48,7 @@ importFrom(graphics,segments)
 importFrom(mgcv,uniquecombs)
 importFrom(mrds,DeltaMethod)
 importFrom(numDeriv,grad)
+importFrom(plyr,ddply)
 importFrom(statmod,qres.binom)
 importFrom(statmod,qres.nbinom)
 importFrom(statmod,qres.tweedie)

NEWS

@@ -5,6 +5,7 @@ dsm news!
  * gamma option removed, since it now is not ignored by gam(). If using method="GCV.Cp" then gamma=1.4 should still be used (but generally we advise using REML rather than GCV for smoothing parameter selection)
  * dsm now requires mgcv version 1.8-23 or higher.
  * dsm.var.prop (and dsm_varprop) now throw an error when there are no covariates in the detection function as it can't do anything useful
+ * new function obs_exp is a diagnostic function to compare observed and expected counts at different covariate aggregations
 
 
 # 2.2.15

R/obs_exp.R

@@ -0,0 +1,48 @@
+#' Observed versus expected diagnostics for fitted DSMs
+#'
+#' Given a covariate, calculate the observed and expected counts for each unique value of the covariate. This can be a useful goodness of fit check for DSMs.
+#'
+#' One strategy for model checking is to calculate observed and expected counts at different aggregations of the variable. If these match well then the model fit is good.
+#'
+#' @param model a fitted \code{dsm} model object
+#' @param covar covariate to aggregate by (character)
+#' @param cut vector of cut points to aggregate at. If not supplied, the unique values of \code{covar} are used.
+#' @importFrom plyr ddply
+#' @export
+#' @author David L Miller, on the suggestion of Mark Bravington.
+#' @export
+#' @return \code{data.frame} with values of observed and expected counts.
+#' @examples
+#' library(Distance)
+#' library(dsm)
+#'
+#' # example with the Gulf of Mexico dolphin data
+#' data(mexdolphins)
+#' hr.model <- ds(distdata, max(distdata$distance),
+#'                key = "hr", adjustment = NULL)
+#' mod1 <- dsm(count~s(x,y), hr.model, segdata, obsdata)
+obs_exp <- function(model, covar, cut=NULL){
+
+  # get data
+  oe <- model$data
+  # add in predictions
+  oe$N <- predict(model)
+
+  # do the aggregation if necessary
+  if(!is.null(cut)){
+    oe[[covar]] <- cut(oe[[covar]], breaks=cut)
+  }
+
+  # for each unique value of term, sum observed and expected
+  oe <- plyr::ddply(oe, covar, function(x){
+    data.frame(Observed = sum(x$count),
+               Expected = sum(x$N))
+  })
+
+  # format the table
+  cn <- oe[,1]
+  oe <- t(oe[,2:3])
+  colnames(oe) <- cn
+
+  return(oe)
+}