In walkerke/tidycensus: Load US Census Boundary and Attribute Data as 'tidyverse' and 'sf'-Ready Data Frames

knitr::opts_chunk$set(message = FALSE, warning = FALSE, fig.width = 8)

The main intent of the tidycensus package is to return population characteristics of the United States in tidy format allowing for integration with simple feature geometries. Its intent is not, and has never been, to wrap the universe of APIs and datasets available from the US Census Bureau. For datasets not included in tidycensus, I recommend Hannah Recht's censusapi package (https://github.com/hrecht/censusapi), which allows R users to access all Census APIs, and packages such as Jamaal Green's lehdr package (https://github.com/jamgreen/lehdr) which grants R users access to Census Bureau LODES data.

However, tidycensus incorporates a select number of Census Bureau datasets outside the decennial Census and ACS that are aligned with the basic goals of the package. One such dataset is the Population Estimates API, which includes information on a wide variety of population characteristics that is updated annually. Also available through tidycensus is the Migration Flows API, which estimates the number of people that have moved between pairs of places in a given year.

Population estimates

Population estimates are available in tidycensus through the get_estimates() function. Estimates are organized into products, which in tidycensus include "population", "components", "housing", and "characteristics". The population and housing products contain population/density and housing unit estimates, respectively. The components of change and characteristics products, in contrast, include a wider range of possible variables.

The post-2020 Population Estimates are no longer on the Census API, but they are downloadable and usable in tidycensus. Follow the guide below for some examples.

Components of change population estimates

By default, specifying "population", "components", or "housing" (housing is not available for 2020 and later) as the product in get_estimates() returns all variables associated with that component. For example, we can request all components of change variables for US states in 2023:

library(tidycensus)
library(tidyverse)
library(tigris)
options(tigris_use_cache = TRUE)

us_components <- get_estimates(geography = "state", product = "components", vintage = 2023)

us_components

The variables included in the components of change product consist of both estimates of counts and rates. Rates are preceded by an R in the variable name and are calculated per 1000 residents.

unique(us_components$variable)

Available geographies include "us", "state", "county", "metropolitan statistical area/micropolitan statistical area" or "cbsa", "combined statistical area", and "place".

If desired, users can request a specific component or components by supplying a character vector to the variables parameter, as in other tidycensus functions. get_estimates() also supports simple feature geometry integration to facilitate mapping. In the example below, we'll acquire data on the net migration rate between 2022 and 2023 for all counties in the United States. We'll also use the shift_geometry() function from the tigris package to shift and rescale counties outside the continental US for national mapping.

net_migration <- get_estimates(geography = "county",
                               variables = "RNETMIG",
                               vintage = 2023,
                               geometry = TRUE,
                               resolution = "20m") %>%
  shift_geometry()

net_migration

We'll next use tidyverse tools to generate a groups column that bins the net migration rates into comprehensible categories, and plot the result using geom_sf() and ggplot2.

library(showtext)
font_add_google("Roboto")
showtext_auto()

order = c("-15 and below", "-15 to -5", "-5 to +5", "+5 to +15", "+15 and up")

net_migration <- net_migration %>%
  mutate(groups = case_when(
    value > 15 ~ "+15 and up",
    value > 5 ~ "+5 to +15",
    value > -5 ~ "-5 to +5",
    value > -15 ~ "-15 to -5",
    TRUE ~ "-15 and below"
  )) %>%
  mutate(groups = factor(groups, levels = order))

state_overlay <- states(
  cb = TRUE,
  resolution = "20m"
) %>%
  filter(GEOID != "72") %>%
  shift_geometry()

ggplot() +
  geom_sf(data = net_migration, aes(fill = groups, color = groups), size = 0.1) +
  geom_sf(data = state_overlay, fill = NA, color = "black", size = 0.1) +
  scale_fill_brewer(palette = "PuOr", direction = -1) +
  scale_color_brewer(palette = "PuOr", direction = -1, guide = FALSE) +
  coord_sf(datum = NA) +
  theme_minimal(base_family = "Roboto", base_size = 18) +
  labs(title = "Net migration per 1000 residents by county",
       subtitle = "US Census Bureau 2023 Population Estimates",
       fill = "Rate",
       caption = "Data acquired with the R tidycensus package | @kyle_e_walker")

Estimates of population characteristics

The fourth population estimates product available in get_estimates(), "characteristics", is formatted differently than the other three. It returns population estimates broken down by categories of AGEGROUP, SEX, RACE, and HISP, for Hispanic origin. Requested breakdowns should be specified as a character vector supplied to the breakdown parameter when the product is set to "characteristics".

By default, the returned categories are formatted as integers that map onto the Census Bureau definitions explained here: https://www.census.gov/data/developers/data-sets/popest-popproj/popest/popest-vars/2017.html. However, by specifying breakdown_labels = TRUE, the function will return the appropriate labels instead. For example:

la_age_hisp <- get_estimates(geography = "county", 
                             product = "characteristics", 
                             breakdown = c("SEX", "AGEGROUP", "HISP"),  
                             breakdown_labels = TRUE, 
                             state = "CA", 
                             county = "Los Angeles",
                             vintage = 2022)

la_age_hisp

With some additional data wrangling, the returned format facilitates analysis and visualization. For example, we can compare population pyramids for Hispanic and non-Hispanic populations in Los Angeles County:

compare <- filter(la_age_hisp, str_detect(AGEGROUP, "^Age"), 
                  HISP != "Both Hispanic Origins", 
                  SEX != "Both sexes") %>%
  mutate(value = ifelse(SEX == "Male", -value, value))

ggplot(compare, aes(x = AGEGROUP, y = value, fill = SEX)) + 
  geom_bar(stat = "identity", width = 1) + 
  theme_minimal(base_family = "Roboto") + 
  scale_y_continuous(labels = function(y) paste0(abs(y / 1000), "k")) + 
  scale_x_discrete(labels = function(x) gsub("Age | years", "", x)) + 
  scale_fill_manual(values = c("darkred", "navy")) + 
  coord_flip() + 
  facet_wrap(~HISP) + 
  labs(x = "", 
       y = "2022 Census Bureau population estimate", 
       title = "Population structure by Hispanic origin", 
       subtitle = "Los Angeles County, California", 
       fill = "", 
       caption = "Data source: US Census Bureau population estimates & tidycensus R package")

Migration flows

The American Community Survey Migration Flows dataset estimates the number of people that have moved between pairs of places. The estimates are calculated based on where a person lived when surveyed and where they lived one year prior to being surveyed. The data is available at three geographic levels: county, county subdivision (minor civil division), and metropolitan statistical area (MSA). Because the number of movers may be small for some pairs of counties, the data is aggregated over a five-year period. The estimates for each five-year period represent the number of people that moved between places each year during that period.

The data is set up such that for each county, you can find the number of people that moved to that county from each of the other counties in the US as well as the number of people that moved from that county to each of the other counties. The net migration for each pair of counties is also provided (although this is simply the moved to minus moved from).

Using get_flows()

The get_flows() function from tidycensus provides access to these estimates. The only required argument for get_flows() is geography, which can be set to "county", "county subdivision", or "metropolitan statistical area". If geography is set to "county" and no other arguments are set, data for all pairs of counties is pulled from the Census API. This is a large data request as it will get all combinations of counties that had movers. More commonly, you might be interested in knowing the flows in and out of one county, county subdivision, or MSA. In this case, you can specify the state and county or MSA.

Here we get county-to-county flow data for Westchester County, NY:

wch_flows <- get_flows(
  geography = "county",
  state = "NY",
  county = "Westchester",
  year = 2018
  )

wch_flows %>% 
  filter(!is.na(GEOID2)) %>% 
  head()

With the default setting of get_flows(), data is returned in a "tidy" or long format. Notice that for each pair of places, there are three rows returned with one row for each variable (MOVEDIN, MOVEDOUT, and MOVEDNET) and the the estimate and margin of error for these variables are in columns. GEOID1 and FULL1_NAME are the FIPS code and name of the origin county. In this case, it will also be Westchester County since that is the only county we requested. GEOID2 and FULL2_NAME are the FIPS code and name of the destination county.

One simple question we can answer with this data is, to which county did the most people move from Westchester?

wch_flows %>% 
  filter(variable == "MOVEDOUT") %>% 
  arrange(desc(estimate)) %>% 
  head()

The MOVEDOUT variable only estimates the number of people that moved out of Westchester County and doesn't account for the number of people that moved in to Westchester from each county. If you are interested in net migration (moved in - moved out), you can use the MOVEDNET variable.

wch_flows %>% 
  filter(variable == "MOVEDNET") %>% 
  arrange(estimate) %>% 
  head()

You may have noticed that there are some destination geographies that are not other counties. For people that moved into to Westchester from outside the United States, the Migration Flows data reports the region that they moved from, such as Africa or Asia. Since this dataset is based on the American Community Survey, there is no way of knowing how many people moved out of the United States, so for all pairs of US to non-US places, the value of MOVEDOUT and MOVEDNET is NA. The GEOID of non-US places is also NA.

wch_flows %>% 
  filter(is.na(GEOID2)) %>% 
  head()

Demographic characteristics

Datasets between 2006-2010 and 2011-2015 have the ability to cross flow data with selected demographic characteristics such as age, race, employment status. For instance, the following call will get the number of movers to and from the Los Angeles-Long Beach Metro Area by race.

la_flows <- get_flows(
  geography = "metropolitan statistical area",
  breakdown = "RACE",
  breakdown_labels = TRUE,
  msa = 31080,   # los angeles msa fips code
  year = 2015
  )

# net migration between la and san francisco
la_flows %>% 
  filter(str_detect(FULL2_NAME, "San Fran"), variable == "MOVEDNET")

Note that the demographic characteristics must be specified in the breakdown argument of get_flows() (not the variable argument). For each dataset there are three or four demographic characteristics to choose from. For more information and to see which characteristics are available in each year, visit the Census Migration Flows documentation.

Mapping migration flows

An additional feature of get_flows() is an option to return spatial data associated with each place. In contrast to other spatial data available through tidycensus, get_flows() returns point rather than polygon geometry. To get geometry with the flows data, set geometry = TRUE. The return of get_flows() will now be an sf object with the centroids of both origin and destination as sfc_POINT columns. The origin point feature is returned in a column named centroid1 and is the active geometry column in the sf object. The destination point feature is returned in the centroid2 column.

phx_flows <- get_flows(
  geography = "metropolitan statistical area",
  msa = 38060,
  year = 2018,
  geometry = TRUE
  )

phx_flows %>% 
  head()

With the centroids attached to each pair of places, it is straightforward to map the migration flows. Here, we look at the most common origin MSAs for people moving to Phoenix-Mesa-Scottsdale, AZ. To make an interactive map of the flows, we'll use the excellent mapdeck package. To use mapdeck, you'll need a Mapbox account and access token.

library(mapdeck)

top_move_in <- phx_flows %>% 
  filter(!is.na(GEOID2), variable == "MOVEDIN") %>% 
  slice_max(n = 25, order_by = estimate) %>% 
  mutate(
    width = estimate / 500,
    tooltip = paste0(
      scales::comma(estimate * 5, 1),
      " people moved from ", str_remove(FULL2_NAME, "Metro Area"),
      " to ", str_remove(FULL1_NAME, "Metro Area"), " between 2014 and 2018"
      )
    )

top_move_in %>% 
  mapdeck(style = mapdeck_style("dark"), pitch = 45) %>% 
  add_arc(
    origin = "centroid1",
    destination = "centroid2",
    stroke_width = "width",
    auto_highlight = TRUE,
    highlight_colour = "#8c43facc",
    tooltip = "tooltip"
  )

walkerke/tidycensus documentation built on March 28, 2024, 7:04 p.m.