consider adding basic mapping function

[lmullany]

It would be helpful to add a basic mapping function to the package. This would optional for users, because it would depend on tigris (to get shape file) and ggplot; both of these are already in Suggests. We would condition the usage of the basic mapping functionality on those package. Things to consider:

• what geospatial units should we allow?
• should the user provide their own shape file for non default options? For example, if we can provide county level maps but not other levels, is that useful? It seems that providing a basic shape file would make it more flexible
• what type of labeling etc should we allow? - should be minimal, with sensible defaults
  • should users be able to provide as params their own labels? Or better just to return a ggplot object that can add layers to?

[andrew-farrey]

Follow up question - can the spline methodology handle point-level inputs (e.g., precise incident locations)? That will influence my answer.

I know this is intended for a broad audience, but by default, I'd include ESSENCE's default geometries; counties, zip codes, HHS regions, and states. I'd also allow the user to provide their own shapefile to use custom geometries, if desired.

Labeling - I'd infer from the shapefile (if an sf file is used), and allow the user to pass a column name or named list of values mapped to the centroid names. I doubt most people will use the named list, but if someone wants to create a plot for a presentation or publication and wants to use the built-in plotting function(s), more customizability with sensible default values will make that easier to do.

[lmullany]

@andrew-farrey the approach requires location to location distance. If the locations are single points, then the user has to assign these to larger geospatial units. Mechanically it could work if each case had its own location, but would blow up computationally

Re the map, good points re at least having built in support for essence regions (although states seem too large to be valuable as the unit of clustering)

[andrew-farrey]

@lmullany So point locations are plausible in theory, but operationally ungainly without blowing up computation unless one uses larger centroids. That makes sense.

RE: including states centroids, I was thinking along the lines of CDC-level usage. At the regional/local level you're probably right. At that resolution, a statewide cluster is essentially just a statewide time series anomaly.