To get started working with tidycensus, users should load the package along with the tidyverse package, and set their Census API key. A key can be obtained from http://api.census.gov/data/key_signup.html.

library(tidycensus)
library(tidyverse)

census_api_key("YOUR API KEY GOES HERE")

There are two major functions implemented in tidycensus: get_decennial, which grants access to the 1990, 2000, and 2010 decennial US Census APIs, and get_acs, which grants access to the 5-year American Community Survey APIs. In this basic example, let’s look at median gross rent by state in 1990:

m90 <- get_decennial(geography = "state", variables = "H043A001", year = 1990)

head(m90)
## # A tibble: 6 x 4
##   GEOID       NAME variable value
##   <chr>      <chr>    <chr> <dbl>
## 1    01    Alabama H043A001   325
## 2    02     Alaska H043A001   559
## 3    04    Arizona H043A001   438
## 4    05   Arkansas H043A001   328
## 5    06 California H043A001   620
## 6    08   Colorado H043A001   418

The function returns a tibble with four columns by default: GEOID, which is an identifier for the geographical unit associated with the row; NAME, which is a descriptive name of the geographical unit; variable, which is the Census variable represented in the row; and value, which is the value of the variable for that unit. By default, tidycensus functions return tidy data frames in which rows represent unit-variable combinations; for a wide data frame with Census variable names in the columns, set output = "wide" in the function call.

As the function has returned a tidy object, we can visualize it quickly with ggplot2:

m90 %>%
  ggplot(aes(x = value, y = reorder(NAME, value))) + 
  geom_point()

Searching for variables

Getting variables from the Census or ACS requires knowing the variable ID - and there are thousands of these IDs across the different Census files. To rapidly search for variables, use the load_variables function. The function takes two required arguments: the year of the Census or endyear of the ACS sample, and the dataset - one of "sf1", "sf3", or "acs5". For ideal functionality, I recommend assigning the result of this function to a variable, setting cache = TRUE to store the result on your computer for future access, and using the View function in RStudio to interactively browse for variables.

v15 <- load_variables(2015, "acs5", cache = TRUE)

View(v15)

By filtering for “median age” I can quickly view the variable IDs that correspond to my query.

Working with ACS data

American Community Survey data differ from decennial Census data in that ACS data are based on an annual sample of approximately 3 million households, rather than a more complete enumeration of the US population. In turn, ACS data points are estimates characterized by a margin of error. tidycensus will always return the estimate and margin of error together for any requested variables. In turn, when requesting ACS data with tidycensus, it is not necessary to specify the "E" or "M" suffix for a variable name. Let’s fetch median household income data from the 2011-2015 ACS for counties in Vermont; the endyear is not necessary here as the function defaults to 2015.

vt <- get_acs(geography = "county", variables = "B19013_001", state = "VT")

head(vt)
## # A tibble: 6 x 5
##   GEOID                       NAME   variable estimate   moe
##   <chr>                      <chr>      <chr>    <dbl> <dbl>
## 1 50001    Addison County, Vermont B19013_001    59688  2199
## 2 50003 Bennington County, Vermont B19013_001    49573  2146
## 3 50005  Caledonia County, Vermont B19013_001    45323  1559
## 4 50007 Chittenden County, Vermont B19013_001    65350  1634
## 5 50009      Essex County, Vermont B19013_001    36599  1779
## 6 50011   Franklin County, Vermont B19013_001    58199  2034

The output is similar to a call to get_decennial, but instead of a value column, get_acs returns estimate and moe columns for the ACS estimate and margin of error, respectively. moe represents the default 90 percent confidence level around the estimate; this can be changed to 95 or 99 percent with the moe_level parameter in get_acs if desired.

As we have the margin of error, we can visualize the uncertainty around the estimate:

vt %>%
  mutate(NAME = gsub(" County, Vermont", "", NAME)) %>%
  ggplot(aes(x = estimate, y = reorder(NAME, estimate))) +
  geom_errorbarh(aes(xmin = estimate - moe, xmax = estimate + moe)) +
  geom_point(color = "red", size = 3) +
  labs(title = "Household income by county in Vermont",
       subtitle = "2011-2015 American Community Survey",
       y = "",
       x = "ACS estimate (bars represent margin of error)")