Relocating Alaska and Hawaii on thematic map of the USA with ggplot2

Question

I am trying to create a thematic map showing all 50 US states, but I am having trouble relocating Alaska and Hawaii in a reliable way. I have a couple of ideas but none of them work well. I will demonstrate them now.

First we need to import the data; using the data in the maps package is not enough because it does not include Hawaii and Alaska.

setwd(tempdir())
download.file("https://dl.dropbox.com/s/wl0z5rpygtowqbf/states_21basic.zip?dl=1", 
              "usmapdata.zip", 
              method = "curl")
# This is a mirror of http://www.arcgis.com/home/item.html?
# id=f7f805eb65eb4ab787a0a3e1116ca7e5
unzip("usmapdata.zip")

require(rgdal)
all_states <- readOGR("states_21basic/", "states")

require(ggplot2); require(maptools); require(rgeos); require(mapproj);
all_states <- fortify(all_states, region = "STATE_NAME")

Now we define some plot aesthetics:

p <- ggplot() + geom_polygon( 
  aes(x=long, y=lat, group = group, fill = as.numeric(as.factor(id))), 
  colour="white", size = 0.25
) + coord_map(projection="azequalarea") + 
scale_fill_gradient(limits = c(1,50))

Now we remove all background etc so they don't clash when we overlap the non-contiguous states:

p <-   p + theme(axis.line=element_blank(),
            axis.text.x=element_blank(),
            axis.text.y=element_blank(),
            axis.ticks=element_blank(),
            axis.title.x=element_blank(),
            axis.title.y=element_blank(),
            panel.background=element_blank(),
            panel.border=element_blank(),
            panel.grid.major=element_blank(),
            panel.grid.minor=element_blank(),
            plot.background=element_blank())

Using viewports

The first idea I had was to use viewports:

AK <- p %+% subset(all_states, id == "Alaska") + theme(legend.position = "none")
HI <- p %+% subset(all_states, id == "Hawaii") + theme(legend.position = "none")
contiguous <- p %+% subset(all_states, id != "Alaska" & id != "Hawaii")

grid.newpage()
vp <- viewport(width = 1, height = 1)
print(contiguous, vp = vp)
subvp1 <- viewport(width = 0.25, height = 0.25, x = 0.18, y = 0.33)
print(AK, vp = subvp1)
subvp2 <- viewport(width = 0.12, height = 0.12, x = 0.32, y = 0.27)
print(HI, vp = subvp2)

This looks nice but it is not satisfactory because it is very sensitive to slight changes in the figure, for example resizing or changes in the size and shape of the legend.

Manually moving Alaska and Hawaii

all_states_AKHImoved <- within(all_states, {
  lat[id == "Alaska"] <- lat[id == "Alaska"] - 45
  long[id == "Alaska"] <- long[id == "Alaska"] + 40
  lat[id == "Hawaii"] <- lat[id == "Hawaii"] + 0
  long[id == "Hawaii"] <- long[id == "Hawaii"] + 70
})
p %+% all_states_AKHImoved

This is not satisfactory because Alaska is usually not to scale on most US maps so it looks very big. Also relocating Alaska and Hawaii changes the distortion introduced by the map projection.

Question

Does anyone have any better approaches?

Answer 1

I was looking for another solution as the packages maptools and rgdal were removed from CRAN (and I had some issues with the fiftystater package). I thought I'd share it here because it is surprisingly easy to move Alaska and Hawaii (and also Puerto Rico, if necessary) using shift_geometry() from the tigris package.

Example

To create an (unshifted) map of 50 US states + Washington D.C.:

library(tigris)
library(ggplot2)

# Download US county geo data to reproduce this example but filter out US territories  
tigcounties <- counties()
tigcounties <- tigcounties[as.numeric(tigcounties$STATEFP) < 60, ]

# This creates the original map
ggplot(tigcounties) + 
  geom_sf() + 
  ggtitle ("Unshifted")

Using shift_geometry() on the geo data used above, this can easily be adjusted. position lets you specify where to shift the states to (default is "below" the rest of the US, "outside" to shift it rather to the side). preserve_area defines whether the states may be scaled to fit more smoothly on the map. (Don't forget to note somewhere on or below your plot if states are shifted and area is not to scale!):

geo_shifted <- shift_geometry(
  tigcounties,
  position = "below", # other option: "outside"
  preserve_area = FALSE
  )

ggplot(geo_shifted) +
  geom_sf() + 
  ggtitle("Shifted - below & area size not preserved")

I would prefer to have a box drawn around the moved states as fiftystater does, but I haven't looked into that yet. If anyone knows or finds a solution, I'd appreciate it!

One last note: the documentation of tigris states

"shift_geometry(), while designed for use with objects from the tigris package, will work with any US dataset. If aligning datasets from multiple sources, you must take care to ensure that your options specified in preserve_area and position are identical across layers. Otherwise your layers will not align correctly.

The function is also designed to work exclusively with features in the continental United States, Alaska, Hawaii, and Puerto Rico. If your dataset includes features outside these areas (e.g. other US territories or other countries), you may get unworkable results. It is advisable to filter out those features before using shift_geometry()."

Answer 2

I've published the fiftystater R package on GitHub (devtools::install_github("wmurphyrd/fiftystater")) to provide a simple solution. It is based on the steps from Spacedman's answer (would link but insufficient rep) and is published as a ggplot2::geom_map ready shape data frame named fifty_states to remove the need to install dependencies, track down a source shape file, or tweak elide values.

library(ggplot2)
library(mapproj)
library(fiftystater)

crimes <- data.frame(state = tolower(rownames(USArrests)), USArrests)

p <- ggplot(crimes, aes(map_id = state)) + 
  geom_map(aes(fill = Assault), map = fifty_states) + 
  expand_limits(x = fifty_states$long, y = fifty_states$lat) +
  coord_map()
p

^{fifty states map}

Plot noise can be cleaned up in the usual way, and there is also the fifty_states_inset_boxes function in the package to add inset boxes:

p + scale_x_continuous(breaks = NULL) + 
  scale_y_continuous(breaks = NULL) +
  labs(x = "", y = "") +
  theme(panel.background = element_blank()) +
  fifty_states_inset_boxes()

^{fifty states with inset boxes}

Answer 3

Here's how to do it by projecting and transforming. You will need:

require(maptools)
require(rgdal)

fixup <- function(usa,alaskaFix,hawaiiFix){

  alaska=usa[usa$STATE_NAME=="Alaska",]
  alaska = fix1(alaska,alaskaFix)
  proj4string(alaska) <- proj4string(usa)

  hawaii = usa[usa$STATE_NAME=="Hawaii",]
  hawaii = fix1(hawaii,hawaiiFix)
  proj4string(hawaii) <- proj4string(usa)

  usa = usa[! usa$STATE_NAME %in% c("Alaska","Hawaii"),]
  usa = rbind(usa,alaska,hawaii)

  return(usa)

}

fix1 <- function(object,params){
  r=params[1];scale=params[2];shift=params[3:4]
  object = elide(object,rotate=r)
  size = max(apply(bbox(object),1,diff))/scale
  object = elide(object,scale=size)
  object = elide(object,shift=shift)
  object
}

Then read in your shapefile. Use rgdal:

us = readOGR(dsn = "states_21basic",layer="states")

Now transform to equal-area, and run the fixup function:

usAEA = spTransform(us,CRS("+init=epsg:2163"))
usfix = fixup(usAEA,c(-35,1.5,-2800000,-2600000),c(-35,1,6800000,-1600000))
plot(usfix)

The parameters are rotations, scaling, x and y shift for Alaska and Hawaii respectively, and were obtained by trial and error. Tweak them carefully. Even changing Hawaii's scale parameter to 0.99999 sent it off the planet because of the large numbers involved.

If you want to turn this back to lat-long:

usfixLL = spTransform(usfix,CRS("+init=epsg:4326"))
plot(usfixLL)

But I'm not sure if you need to use the transformations in ggplot since we've done that with spTransform.

You can now jump through the ggplot2 fortify business. I'm not sure if it matters for you but note that the order of the states is different in the usfix version - Alaska and Hawaii are now the last two states.

Answer 4

Once you start shifting things around like this you may as well just represent Alaska and Hawaii as square boxes somewhere in the Gulf of Mexico. It would have the added advantage of making it possible to tell what colour Hawaii was.

You could probably then just go all the way and use a distorted system where each state has equal area, then you could see Rhode Island.

USA Cartogram examples on google images shows the kind of thing. Don't know how many of them have shapefiles or data with them though.

Are you really interested in the relative sizes of states, or do you want a representation that lets people see what the value is for a state?