spind {spind} | R Documentation |
spind: Spatial Methods and Indices
Description
The spind
package provides convenient implementation of Generalized
estimating equations (GEEs) and Wavelet-revised models (WRMs)
in the context of spatial models. It also provides tools for
multi-model inference, stepwise model selection, and spatially
corrected model diagnostics. This help section provides brief descriptions
of each function and is organized by the type of model they apply to
or the scenarios in which you might use them. Of course, these are
recommendations - feel free to use them as you see fit. For a more
detailed description of the package and its functions, please see
the vignette Intro to spind (browseVignettes('spind')
).
GEEs
The GEE
function fits spatial models using a generalized
estimating equation and a set of gridded data. The package
also includes S3 methods for summary
and predict
so you can interact with these models in the same way you might
interact with a glm
or lm
.
WRMs
The WRM
function fits spatial models using a wavelet-revised
model and a set of gridded data. The package
also includes S3 methods for summary
and predict
so you can interact with these models in the same way you might
interact with a glm
or lm
. There are also a number
of helper functions that help you fine tune the fitting process
that are specific to WRMs. Please see the documentation for
WRM
for more details on those.
WRM
also has a few other features specific to it. For example, if
you are interested in viewing the variance or covariance of your variables
as a function of level
, covar.plot
is useful. upscale
will plot your matrices as a function of level
so you can examine the
effect of cluster resolution on your results.
Multi-model inference and stepwise model selection
spind
includes a couple of functions to help you find the best fit
for your data. The first two are multimodel inference tools specific to GEEs
and WRMs and are called mmiGEE
and mmiWMRR
. These generate
outputs very similar to those from the MuMIn
package. If you would
like to see how variable importance changes as a function of the scale
of
the WMRR, you can call rvi.plot
. This will generate a model selection
table for each degree of level
(from 1 to maxlevel
) and then
plot the weight of each variable as a function of level
.
spind
also includes a function for stepwise model selection that is
loosely based on step
and stepAIC
. step.spind
differs
from these in that it is specific to classes WRM
and GEE
. It
performs model selection using AIC or AICc for WRMs and QIC for GEEs.
Spatial indices of goodness of fit
Finally, spind
has a number of functions that provide spatially
corrected goodness of fit diagnostics for any type of model (i.e. they
are not specific to classes WRM
or GEE
). These first appeared
in spind v1.0
and have not been updated in this release. The first two are
divided into whether or not they are threshold dependent or not. Threshold
dependent metrics can be calculated using th.dep
and threshold
independent metrics can be calculated using th.indep
.
acfft
calculates spatial autocorrelation of residuals from a model
using Moran's I. You can set the number of distance bins you'd
like to examine using dmax
argument and the size of those bins
using lim1
and lim2
.
Conclusion
The vignette titled Intro to spind provides more information on these functions and some example workflows that will demonstrate them in greater depth than this document. Of course, if you have suggestions on how to improve this document or any of the other ones in here, please don't hesitate to contact us.