informative_plot {growfunctions}

Plot credible intervals for parameters to compare ignoring with weighting an informative sample

Description

Uses as input the output object from the gpdpgrow() and gmrfdpgrow() functions.

Usage

informative_plot(
  objects = NULL,
  objects_labels = c("ignore", "weight"),
  map = NULL,
  units_name = NULL,
  model = "gp",
  true_star = NULL,
  map_true = NULL
)

Arguments

objects

A list of objects, either all outputs from gpdpgrow(), or all from gmrfdpgrow(). objects includes a model estimated under ignoring the informativeness of the sampling design and another that employs weighting to account for the informativeness. An additional object may be added to represent a separate "iid" (or non-informative) sample from the same population, which will typically be available if the dataset was generated as synthetic data using function, "gen_informative_sample()".

objects_labels

A character vector of length equal to objects that provides labels for each entry in objects. Allowed entries in objects_labels are c("ignore","weight","iid"), where "ignore" denotes a model that ignores the informativeness, while "weight" denotes a model that employs sampling weights and "iid" denotes a model run on a non-informative, iid sample from the same population. Defaults to objects_labels = c("ignore","weight").

map

A list matrices, where each entry is produced from cluster_plot(object)$map. It is comprised of unit labels and cluster assignments for each object in objects. The length of map must be equal to the length of objects.

units_name

The label in each "map" matrix for the observation units. Will be the same as the units_name entry for the previously run, cluster_plot() function.

model

A scalar character input indicating the estimation model for all of the entries in objects. Allowable values for model are c("gp","gmrf"). Defaults to model = "gp".

true_star

An optional, P x M matrix, of true parameter location values, where P denotes the number of parameters per cluster and M denotes the number of clusters. For example, in model = "gp" with a single, rational quadratic covariance, P = 3 and if there are 3 clusters, then M = 3. For a model = "gmrf", with a single covariance, P = 1.

map_true

An optional data.frame object with n rows, the size of the informative sample used for c("ignore","weight") objects that maps the units_name to a true cluster. map_true must have 2 columns (and the rest are ignored), one must be named the same value as input for units_name. The second column must be named, cluster. If the true values derive from running gen_informative_sample() as the source of the true values, one may just input the map_obs data.frame that is listed in the object returned by gen_informative_sample().

Value

A list object containing the plot of estimated functions, faceted by cluster, and the associated data.frame object.

p.compare

A ggplot2 plot object

dat.compare

A data.frame object used to generate p.compare.

Author(s)

Terrance Savitsky [e-mail]

See Also

gpdpgrow, gmrfdpgrow

Examples

## Not run: 
library(growfunctions)
## use gen_informative_sample() to generate an 
## N X T population drawn from a dependent GP
## By default, 3 clusters are used to generate 
## the population.
## A single stage stratified random sample of size n 
## is drawn from the population using I = 4 strata. 
## The resulting sample is informative in that the 
## distribution for this sample is
## different from the population from which 
## it was drawn because the strata inclusion
## probabilities are proportional to a feature 
## of the response, y (in the case, the variance.
## The stratified random sample over-samples 
## large variance strata).
## (The user may also select a 2-stage 
## sample with the first stage
## sampling "blocks" of the population and 
## the second stage sampling strata within blocks). 
dat_sim        <- gen_informative_sample(N= 10000, 
                                n = 500, T = 5,
                                noise_to_signal = 0.1)

y_obs                       <- dat_sim$y_obs
T                           <- ncol(y_obs)

an informative sampling design that inputs inclusion
probabilities, ipr
res_gp_w            <- gpdpgrow(y = y_obs, 
                               ipr = dat_sim$map_obs$incl_prob, 
                               n.iter = 5, n.burn = 2,  
                               n.thin = 1, n.tune = 0)
and fit vs. data for experimental units
fit_plots_w         <- cluster_plot( object = res_gp_w,  
                           units_name = "establishment", 
                           units_label = dat_sim$map_obs$establishment, 
                           single_unit = FALSE, credible = TRUE )

## estimate parameters ignoring sampling design
res_gp_i            <- gpdpgrow(y = y_obs, 
                               n.iter = 5, n.burn = 2, 
                               n.thin = 1, n.tune = 0)
## plots of estimated functions, faceted by cluster and fit vs. 
## data for experimental units
fit_plots_i         <- cluster_plot( object = res_gp_i,  
                                    units_name = "establishment", 
                                    units_label = dat_sim$map_obs$establishment, 
                                    single_unit = FALSE, credible = TRUE )

## We also draw an iid (non-informative, exchangeable) 
## sample from the same population to 
## compare estimation results to our weighted 
## (w) and unweighted/ignoring (i) models

## estimate parameters under an iid sampling design
res_gp_iid          <- gpdpgrow(y = dat_sim$y_iid, 
                               n.iter = 5, n.burn = 2,   
                               n.thin = 1, n.tune = 0)
## plots of estimated functions, faceted by cluster and 
## fit vs. data for experimental units
fit_plots_iid       <- cluster_plot( object = res_gp_iid,  
                           units_name = "establishment", 
                           units_label = dat_sim$map_iid$establishment, 
                           single_unit = FALSE, credible = TRUE )

## compare estimations of covariance parameter credible 
## intervals when ignoring informativeness vs.
## weighting to account for informativeness
objects                  <- map <- vector("list",3)
objects[[1]]             <- res_gp_i
objects[[2]]             <- res_gp_iid
objects[[3]]             <- res_gp_w
map[[1]]                 <- fit_plots_i$map
map[[2]]                 <- fit_plots_iid$map
map[[3]]                 <- fit_plots_w$map
objects_labels           <- c("ignore","iid","weight")

parms_plots_compare      <- informative_plot( objects = objects, 
                                objects_labels = objects_labels,
                                map = map, units_name = "establishment", 
                                model = "gp",
                                true_star = dat_sim$theta_star, 
                                map_true = dat_sim$map_obs)


## End(Not run)