02-checking

02-checking | model checking #

Overview #

A suite of graphical and tabular displays for model checking.

FileDescription
op/Observed vs. predicted plots
ppc/Graphical displays comparing the observed data to data simulated from the posterior predictive distribution. For mixed models, there will be files for each of the distributions. If the model fits well, we should be able to use it to generate data that looks a lot like the data we have in-hand. Subfolders for outputs by stratum and unit or other grouping variables specified by ppc facets in the analysis config file (see HERE for more).
variography/Checks for spatial autocorrelation

Subdirectories #

op/ | observed vs. predicted plots #

FileDescription
op-y-rep.jpgA scatterplot of observed ( \(y\) -axis) vs. the median predicted ( \(x\) ) values. The horizontal lines / whiskers around each point correspond to the 95% HDI for the \(y^{rep}\) ’s associated with each observation.

ppc/ | posterior predictive checks #

Graphical displays comparing the observed data to data simulated from the posterior predictive distribution. For mixed models, there will be files for each of the distributions. If the model fits well, we should be able to use it to generate data that looks a lot like the data we have in-hand. Subfolders for outputs by stratum and unit and any other faceting variables specified by ppc facets, as described here.

FileDescription
y-rep-bayes-p-by-stratum_id[unit_code].csvBayesian p values for mu and sigma by stratum or unit code
y-rep-by-stratum_id[unit_code]-9-draws.jpgSeparate histograms of \(y\) and some ( \(n=9\) ) of the \(y^{rep}\) datasets
y-rep- by-stratum_id[unit_code]-stats.jpgThe distribution of test statistics. The solid, vertical line is the value of the test statistic computed from the observed data, \(y\) , while the underlying bars represents the distribution of the test statistic in the \(y^{rep}\) simulations.

variography/ | spatial autocorrelation #

FileDescription
residuals-df.rdsResiduals for each observation. Used as input to variograms.
variograms.pngFigure shows degree of spatial autocorrelation in data, semi-variance by distance by direction or by any direction. Note that the default is to evaluate up to 1/3 of the total distance represented in the data.