setupSTdataset: Creates the data structure that is then used to build specific models via `create.data.model`.

Description

Creates the data structure that is then used to build specific models via create.data.model. This includes observations, observation locations, smooth temporal trends, geographic covariates and spatio-temporal covariates (in the future).

Usage

setupSTdataset(rawobs, covardat, covarnames, trendf,
               varnames=list(yraw="lac",date="intended_wednesday",idobs="site_id",idcov="site.id",xcoord="lambert.x",ycoord="lambert.y",long="longitude",lat="latitude"), x.to.km=1000,transform=log,scale=TRUE,mesa=TRUE)

Arguments

rawobs

Data frame with the raw observations. Should have vectors whose names are identified by the items ``yraw'', ``date'',``idobs'' in varnames.

covardat

Data frame with the regression predictors. Should have vectors whose names are identified by the items ``idcov'', ``xcoord'', ``ycoord'', ``long'', ``lat'' in varnames.

covarnames

Character vector with the names of variables in covardat that should be included in the modeling dataset. If you are not sure, then just specify all names (i.e., ``names(covardat)'').

trendf

Matrix or data frame with the time trend, on the modeling scale. Its format should be compatible with the output of SVD.smooth: columns represent the smooth time trends, and the row names are in the R

varnames

List of character strings denoting the variable names needed to identify and match the various data components. See above.

x.to.km

Numeric, conversion factor from the coordinate scale to km. Defaults to 1000. Set to 1 to ignore.

transform

Function, the transformation link from the scale in which observations appear in rawobs, to the modeling scale. Defaults to log.

scale

Logical: should the covariates in the LUR component be scaled to each have mean 0 and variance 1? Defaults to TRUE.

mesa

Logical: should we assume that location ID names follow the MESA-Air convention? Defaults to TRUE.

Value

A list with data frames described in mesa.data, except for the ``SpatioTemp'' array.

encoding

latin1

Details

This function is a time-saving aid, for creating the basic data structure for fitting spatiotemporal models, with some built-in error checks. The output of this function can be fed as the mesa.data argument into create.data.model, to build specific models, or to prediction functions. The function separates out the geographic information in covardat into the ``location'' and ``LUR'' components, ensuring that all location IDs with observations have location and covariate data. It also ensures that all observation dates have a trend function value -- if some do not exist, it calculates them using smooth.spline, giving out a warning. If you have a high-dimensional covariate set but plan to use only a small subset, then covarnames helps produce a more compact data structure. Specify only the variables you really think you might experiment with. The scale option is recommended when building and checking the model, because it improves numerical stability of fit.mesa.model, and makes the effect estimate magnitudes comparable. When producing a final model for scientific use, you might want to set scale to FALSE, and manually divide the covariates into units that are scientifically communicated, but not too far in magnitude from unity. Be sure that your variable names are correctly specified in varnames. If you are struggling with this function, you can always construct a mesa.data list manually by yourself, as explained in the package tutorial.