Extract Decennial Census and American Community Survey Data
Download summary files from Census Bureau and extract data of decennial censuses and American Community Surveys from your local computer.
2/7/2019: Version 0.6.1 is on CRAN.
- Added GEOID to decennial census for selected summary levels so that it will be easier to compare the same geographic areas in decennial census and American Community Survey.
read_acs5year()run faster in the new version.
totalcensus v0.5.1 is on CRAN. It includes
- Decennial census 2000 and 2010
- ACS 1 year: 2005 - 2017
- ACS 5 year: 2009 - 2017
Installation and setup
# from CRAN install.packages("totalcensus") # development version devtools::install_github("GL-Li/totalcensus")
This package requires downloading census data and you need to create a folder to store the downloaded data. Let's call the folder
my_census_data and assume the full path to this folder is
xxxxx/my_census_data. Run the function below to set the path for the package.
Why another R census package
The census API offers most data in decennial censuses and ACS estimates for download and API-based packages such as
acs make the downloading very convenient in R. So why we need another package?
One advantage is that once you downloaded the summary files, you do not need internet anymore and everything is on your own computer. You do not need to worry about internet interruption or government shutdown. You have total control of the data.
Another benefit of using package
totalcensus is that it makes census data extraction more flexible. It is particularly convenient to extract high resolution data at census tract, block group, and block level for a large area.
Here is an example of how we extract the median home values in all block groups in the United States from 2011-2015 ACS 5-year survey with this package. You simply need to call the function
read_acs5year(). It takes 15 seconds for my 4-years old laptop to return the data of all 217,739 block groups. In addition to the table contents we request, we also get the population and coordinate of each block group.
library(totalcensus) home_national <- read_acs5year( year = 2015, states = states_DC, # all 50 states plus DC table_contents = "home_value = B25077_001", summary_level = "block group" )
With the coordinates, we can visualize the data on US map with
ggmap. Each data point in the figure below corresponds to a a block group, colored by median home value and sized by population. This plot not only displays the median home values, but also tells population densities on the map.
There are additional benefits of using this package:
- You can get detailed urban/rural data from Census 2010. This package use summary file 1 with urban/rural update, while the census API only provide data in summary file 1 before urban/rural update.
- You can get all block groups that belong or partially belong to a city. Original census data do not provide city information for a block group as a block group may not uniquely belong to a city. However, large cities have most block groups within their boundaries and only a small number of block groups run across the borders. The block group level data provide valuable spatial information of a city. This is particularly helpful for ACS 5-year surveys which cover data down to the level of block groups.
- It provides longitude and latitude of the internal point of a geographic area for easy and quick mapping. You do not always need shape files to make nice maps, as in the map shown above.
More reading materials
- Extract US Census 2010 data with data.table and dplyr: this post explains how the summary files of decennial census 2010 are processed under the hood in
- Proccess 5-digit ZIP Code Tabulation Area (ZCTA5) data with totalcensus package: an application example.
- Using totalcensus package to determine relationship between census geographic entities; an application example.
The package has three functions to read decennial census, ACS 5-year survey, and ACS 1-year survey:
read_acs1year(). They are similar but as these datasets are so different, we prefer to keep three separate functions, one for each.
The function arguments serve as filters to select the data you want:
- year: the year or ending year of the decennial census or ACS estimate.
- states: the states of which you want read geography and data files. In addition to 50 states and "DC", you can choose from "PR" (Puerto Rico), plus a special one "US" for national files.
- table_contents: this parameter specifies which table contents you want to read. Population is always returned even if table_contents is NULL. Users can name the table contents in the format such as
c("male = B01001_002", "female = B01001_026").
- areas: if you know which metropolitan areas, counties, cities and towns you want to get data from, you can specify them here by name or FIPS code, for example,
c("New York metro", "PLACE = UT62360", "Salt Lake City city, UT").
- geo_headers: In case you do not know which areas to extract data, you can read all the geographic headers specified here and select areas after reading.
- summary_level: it determines which summary level data to extract. Common ones like "state", "county", "place", "county subdivision", "tract", "block group", and "block" can be input as plain text. Others have to be given by code.
- geo_comp: specifies data of which geographic component you want. Most common ones are "total", "urban", "urbanized area", "urban cluster", and "rural". Others are provided by code.
read_acs5year() have additional argument:
- with_margin: whether to read margin of error of the estimate.
- dec_fill: whether to fill geo_headers codes with data from decennial census. The codes in ACS summary file are often incomplete. To use decennial census 2010 data to fill the missing values, set the argument to "dec2010".
There are a family of
search_xxx() functions to help find table contents, geoheaders, summary levels, geocomponents, FIPS codes and CBSA codes.
The following examples demonstrate how to use these
Median gross rent in cities with population over 65000
A property management company wants to know the most recent rents in major cities in the US. How to get the data?
We first need to determine which survey to read. For most recent survey data, we want to read 2016 ACS 1-year estimates, which provide data for geographic areas with population over 65000.
We also need to determine which data files to read. We know summary level of cities is "160" or "place". Browsing with
search_summarylevels("acs1"), we see that this summary level is only in state files of ACS 1-year estimates. So we will read all the state files.
Then we need to check if 2016 ACS 1-year estimate has the rent data. We run
search_tablecontents("acs1") to open the dataset with
View() in RStudio. You can provide keywords to search in the function but it is better to do the search in RStudio with filters. There are so many tables that contains string "rent". It takes some time to find the right one if you are not familiar with ACS tables. After some struggle, we think B25064_001 is what we want.
We do not need to specify
geo_headers as we are extracting all geographic areas matches the conditions.
Below is the code that gives what we want. The first time you use
read_xxxx() functions to read data files, you will be asked to download data generated from decennial census 2010 and summary files required for this function call, in this case, 2016 ACS 1-year summary files. Choose 1 to continue.
rent <- read_acs1year( year = 2016, states = states_DC, table_contents = "rent = B25064_001", summary_level = "place" ) # Fisrt 5 rows # GEOID NAME STUSAB population rent GEOCOMP SUMLEV lon lat # 1: 16000US0203000 Anchorage municipality, Alaska AK 298192 1296 total 160 -149.27435 61.17755 # 2: 16000US0107000 Birmingham city, Alabama AL 213434 777 total 160 -86.79905 33.52744 # 3: 16000US0121184 Dothan city, Alabama AL 67714 720 total 160 -85.40682 31.23370 # 4: 16000US0135896 Hoover city, Alabama AL 84943 1021 total 160 -86.80558 33.37695 # 5: 16000US0137000 Huntsville city, Alabama AL 196225 766 total 160 -86.53900 34.78427
It is always nice to visualize them on US map.
library(ggplot2) # ggmap requires library(ggmap) # You need to use your own google clound API key register_google("your_google_api_key") us_map <- get_map("united states", zoom = 4, color = "bw") ggmap(us_map) + geom_point( data = rent[order(-population)], aes(lon, lat, size = population/1e3, color = rent) ) + ylim(25, 50) + scale_size_area(breaks = c(100, 200, 500, 1000, 2000, 5000)) + scale_color_continuous(low = "green", high = "red") + labs( color = "monthly\nrent ($)", size = "total\npopulation\n(thousand)", title = "Monthly rent in cities over 65000 population", caption = "Source: 2016 ACS 1-year estimate" ) + theme( panel.background = element_blank(), axis.title = element_blank(), axis.text = element_blank(), axis.ticks = element_blank(), title = element_text(size = 14) )
Black communities in South Bend city, Indiana at census block level
Only the decennial census has data down to block level. The most recent one is Census 2010.
Knowing names of a city, county, metro area, or town, we can feed them directly to argument
areas. The returned data.table contains the table contents we want as well as population and coordinates. The reading takes a few seconds.
# read data of black population in each block black_popul <- read_decennial( year = 2010, states = "IN", table_contents = "black_popul = P0030003", areas = "South Bend city, IN", summary_level = "block" ) # first 5 rows of black_popul: # area lon lat state population black_popul GEOCOMP SUMLEV # 1: South Bend city, IN -86.21864 41.63613 IN 28 10 all 100 # 2: South Bend city, IN -86.21659 41.63670 IN 0 0 all 100 # 3: South Bend city, IN -86.22172 41.63573 IN 52 16 all 100 # 4: South Bend city, IN -86.22022 41.63182 IN 279 21 all 100 # 5: South Bend city, IN -86.22093 41.63367 IN 42 1 all 100
It is better to separate data manipulation from reading to save reading time as you usually need to try multiple manipulations. Data manipulation can be done with
library(magrittr) # remove blocks where no people lives in and add a column of black percentage. black <- black_popul %>% .[population != 0] %>% # percentage of black population in each block .[, black_pct := round(100 * black_popul / population, 2)]
Again we visualize percentage of black population on map with
south_bend <- get_map("south bend, IN", zoom = 13, color = "bw") ggmap(south_bend) + geom_point( data = black, aes(lon, lat, size = population, color = black_pct) ) + scale_size_area(breaks = c(10, 100, 200, 500)) + scale_color_continuous(low = "green", high = "red") + labs( color = "% black", size = "total\npopulation", title = "Black communities in South Bend city at block level", caption = "Source: Census 2010" ) + theme( panel.background = element_blank(), axis.title = element_blank(), axis.text = element_blank(), axis.ticks = element_blank(), title = element_text(size = 14) )
This package requires downloading census data to your local computer. You will be asked to download data when you call
read_xxxx functions. The downloaded data will be extracted automatically to the folder
A set of data generated from Census 2010 will also be downloaded, which is used to fill missing geographic header records in ACS data.
The census data can be found on Census Bureau's website but you do not need to download them manually. Use the function above.