class: center, middle, inverse, title-slide # Analyzing 2020 Census Data with R and tidycensus ## SSDAN Workshop Series ### Kyle Walker ### March 10, 2022 --- ## About me .pull-left[ * Associate Professor of Geography at TCU * Spatial data science researcher and consultant * R package developer: __tidycensus__, __tigris__, __mapboxapi__ * Book: [_Analyzing US Census Data: Methods, Maps and Models in R_](https://walker-data.com/census-r/) - Available for free online right now; - To be published in print with CRC Press in fall 2022 ] .pull-right[ <img src=img/photo.jpeg> ] --- ## SSDAN workshop series * Today: an introduction to 2020 US Census data * Next Friday (March 18): Mapping 2020 Census data * Friday, March 25: a first look at the 2016-2020 American Community Survey data with R and __tidycensus__ --- ## Today's agenda * Hour 1: Getting started with 2020 Census data in __tidycensus__ * Hour 2: Wrangling Census data with __tidyverse__ tools * Hour 3: Visualizing 2020 US Census data --- class: middle, center, inverse ## Part 1: Getting started with 2020 Census data in tidycensus --- ## Typical Census data workflows <img src=img/spreadsheet.png style="width: 900px"> --- ## The Census API * [The US Census __A__pplication __P__rogramming __Interface__ (API)](https://www.census.gov/data/developers/data-sets.html) allows developers to access Census data resources programmatically * R packages to interact with the APIs: censusapi, acs * Other languages: cenpy (Python), citySDK (JavaScript) --- ## tidycensus * R interface to the Decennial Census, American Community Survey, Population Estimates Program, and Public Use Microdata Series APIs * Key features: - Wrangles Census data internally to return tidyverse-ready format (or traditional wide format if requested); - Automatically downloads and merges Census geometries to data for mapping (next week's workshop!); - Includes tools for handling margins of error in the ACS and working with survey weights in the ACS PUMS; - States and counties can be requested by name (no more looking up FIPS codes!) --- ## Development of tidycensus * Mid-2010s: I started accumulating R scripts that did the same thing over and over (download Census data from the API, transform to tidy format, join to spatial data) * (Very) early implementation: [acs14lite](https://rpubs.com/walkerke/acs14lite) * 2017: first release of tidycensus following the implementation of a "tidy spatial data model" in the sf package * 2020: Matt Herman joins as co-author; support for ACS microdata (PUMS) in tidycensus --- ## Getting started with tidycensus * To get started, install the packages you'll need for today's workshop * If you are using the RStudio Cloud environment, these packages are already installed for you ```r install.packages(c("tidycensus", "tidyverse", "geofacet", "ggridges")) ``` --- ## Your Census API key * To use tidycensus, you will need a Census API key. Visit https://api.census.gov/data/key_signup.html to request a key, then activate the key from the link in your email. * Once activated, use the `census_api_key()` function to set your key as an environment variable ```r library(tidycensus) census_api_key("YOUR KEY GOES HERE", install = TRUE) ``` --- class: middle, center, inverse ## Basic usage of tidycensus --- ## 2020 Census data: what we have now * The currently available 2020 Census data come from the PL94-171 Redistricting Summary File, which is used for congressional apportionment & redistricting * Variables available include total counts (population & households), occupied / vacant housing units, total and voting-age population breakdowns by race & ethnicity, and group quarters status * Later in 2022 (expected): the Demographic and Housing Characteristics summary files, which will include other variables typically included in the decennial Census data (age & sex breakdowns, detailed race & ethnicity) --- ## 2020 Census data in tidycensus * The `get_decennial()` function is used to acquire data from the decennial US Census * The two required arguments are `geography` and `variables` for the functions to work; for 2020 Census data, use `year = 2020`. ```r pop20 <- get_decennial( geography = "state", variables = "P1_001N", year = 2020 ) ``` --- * Decennial Census data are returned with four columns: `GEOID`, `NAME`, `variable`, and `value` ```r pop20 ``` ``` ## # A tibble: 52 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 01 Alabama P1_001N 5024279 ## 2 02 Alaska P1_001N 733391 ## 3 04 Arizona P1_001N 7151502 ## 4 05 Arkansas P1_001N 3011524 ## 5 06 California P1_001N 39538223 ## 6 08 Colorado P1_001N 5773714 ## 7 09 Connecticut P1_001N 3605944 ## 8 10 Delaware P1_001N 989948 ## 9 11 District of Columbia P1_001N 689545 ## 10 16 Idaho P1_001N 1839106 ## # … with 42 more rows ``` --- ## Understanding the printed messages * When we run `get_decennial()` for the 2020 Census for the first time, we see the following messages: ``` Getting data from the 2020 decennial Census Using the PL 94-171 Redistricting Data summary file Note: 2020 decennial Census data use differential privacy, a technique that introduces errors into data to preserve respondent confidentiality. ℹ Small counts should be interpreted with caution. ℹ See https://www.census.gov/library/fact-sheets/2021/protecting-the-confidentiality-of-the-2020-census-redistricting-data.html for additional guidance. This message is displayed once per session. ``` --- ## Understanding the printed messages * The Census Bureau is using _differential privacy_ in an attempt to preserve respondent confidentiality in the 2020 Census data, which is required under US Code Title 13 * Intentional errors are introduced into data, impacting the accuracy of small area counts (e.g. some blocks with children, but no adults) * Advocates argue that differential privacy is necessary to satisfy Title 13 requirements given modern database reconstruction technologies; critics contend that the method makes data less useful with no tangible privacy benefit --- ## Requesting tables of variables * The `table` parameter can be used to obtain all related variables in a "table" at once ```r table_p2 <- get_decennial( geography = "state", * table = "P2", year = 2020 ) ``` --- ```r table_p2 ``` ``` ## # A tibble: 3,796 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 01 Alabama P2_001N 5024279 ## 2 02 Alaska P2_001N 733391 ## 3 04 Arizona P2_001N 7151502 ## 4 05 Arkansas P2_001N 3011524 ## 5 06 California P2_001N 39538223 ## 6 08 Colorado P2_001N 5773714 ## 7 09 Connecticut P2_001N 3605944 ## 8 10 Delaware P2_001N 989948 ## 9 11 District of Columbia P2_001N 689545 ## 10 16 Idaho P2_001N 1839106 ## # … with 3,786 more rows ``` --- class: middle, center, inverse ## Understanding geography and variables in tidycensus --- ## US Census Geography <img src=img/census_diagram.png style="width: 500px"> .footnote[Source: [US Census Bureau](https://www2.census.gov/geo/pdfs/reference/geodiagram.pdf)] --- ## Geography in tidycensus * Information on available geographies, and how to specify them, can be found [in the tidycensus documentation](https://walker-data.com/tidycensus/articles/basic-usage.html#geography-in-tidycensus-1) <img src=img/tidycensus_geographies.png style="width: 650px"> --- ## Querying by state .pull-left[ * For geographies available below the state level, the `state` parameter allows you to query data for a specific state * __tidycensus__ translates state names and postal abbreviations internally, so you don't need to remember the FIPS codes! * Example: data on the Hispanic population in Michigan by county ] .pull-right[ ```r mi_hispanic <- get_decennial( geography = "county", variables = "P2_002N", * state = "MI", year = 2020 ) ``` ] --- ```r mi_hispanic ``` ``` ## # A tibble: 83 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 26001 Alcona County, Michigan P2_002N 122 ## 2 26003 Alger County, Michigan P2_002N 115 ## 3 26005 Allegan County, Michigan P2_002N 9389 ## 4 26007 Alpena County, Michigan P2_002N 417 ## 5 26009 Antrim County, Michigan P2_002N 459 ## 6 26011 Arenac County, Michigan P2_002N 270 ## 7 26013 Baraga County, Michigan P2_002N 102 ## 8 26015 Barry County, Michigan P2_002N 2142 ## 9 26017 Bay County, Michigan P2_002N 5930 ## 10 26019 Benzie County, Michigan P2_002N 391 ## # … with 73 more rows ``` --- ## Querying by state and county * County names are also translated internally by __tidycensus__ for sub-county queries, e.g. for Census tracts, block groups, and blocks ```r washtenaw_hispanic <- get_decennial( geography = "tract", variables = "P2_002N", state = "MI", * county = "Washtenaw", year = 2020 ) ``` --- ```r washtenaw_hispanic ``` ``` ## # A tibble: 107 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 26161400100 Census Tract 4001, Washtenaw County, Michigan P2_002N 108 ## 2 26161400300 Census Tract 4003, Washtenaw County, Michigan P2_002N 271 ## 3 26161400400 Census Tract 4004, Washtenaw County, Michigan P2_002N 144 ## 4 26161400500 Census Tract 4005, Washtenaw County, Michigan P2_002N 457 ## 5 26161400600 Census Tract 4006, Washtenaw County, Michigan P2_002N 358 ## 6 26161400700 Census Tract 4007, Washtenaw County, Michigan P2_002N 121 ## 7 26161400800 Census Tract 4008, Washtenaw County, Michigan P2_002N 264 ## 8 26161402100 Census Tract 4021, Washtenaw County, Michigan P2_002N 200 ## 9 26161402201 Census Tract 4022.01, Washtenaw County, Michigan P2_002N 323 ## 10 26161402300 Census Tract 4023, Washtenaw County, Michigan P2_002N 99 ## # … with 97 more rows ``` --- ## Searching for variables * To search for variables, use the `load_variables()` function along with a year and dataset * The `View()` function in RStudio allows for interactive browsing and filtering ```r vars <- load_variables(2020, "pl") View(vars) ``` --- ## Tables available in the 2020 Census PL file * H1: Occupancy status (by household) * P1: Race * P2: Race by Hispanic origin * P3: Race for the population 18+ * P4: Race by Hispanic origin for the population 18+ * P5: Group quarters status --- class: middle, center, inverse ## Data structure in tidycensus --- ## "Tidy" or long-form data .pull-left[ * The default data structure returned by __tidycensus__ is "tidy" or long-form data, with variables by geography stacked by row ] .pull-right[ ```r group_quarters <- get_decennial( geography = "state", table = "P5", year = 2020 ) ``` ] --- ```r group_quarters ``` ``` ## # A tibble: 520 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 01 Alabama P5_001N 127934 ## 2 02 Alaska P5_001N 30291 ## 3 04 Arizona P5_001N 160269 ## 4 05 Arkansas P5_001N 82518 ## 5 06 California P5_001N 917932 ## 6 08 Colorado P5_001N 126848 ## 7 09 Connecticut P5_001N 108002 ## 8 10 Delaware P5_001N 22745 ## 9 11 District of Columbia P5_001N 40682 ## 10 16 Idaho P5_001N 49729 ## # … with 510 more rows ``` --- ## "Wide" data .pull-left[ * The argument `output = "wide"` spreads Census variables across the columns, returning one row per geographic unit and one column per variable ] .pull-right[ ```r group_quarters_wide <- get_decennial( geography = "state", table = "P5", year = 2020, * output = "wide" ) ``` ] --- ```r group_quarters_wide ``` ``` ## # A tibble: 52 × 12 ## GEOID NAME P5_001N P5_002N P5_003N P5_004N P5_005N P5_006N P5_007N P5_008N ## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> ## 1 01 Alabama 127934 70648 39749 1479 27869 1551 57286 45489 ## 2 02 Alaska 30291 7177 4842 457 1781 97 23114 1472 ## 3 04 Arizona 160269 89904 64154 2331 21938 1481 70365 38945 ## 4 05 Arkans… 82518 48001 27079 1248 19266 408 34517 26887 ## 5 06 Califo… 917932 344896 201570 8966 124804 9556 573036 230361 ## 6 08 Colora… 126848 55851 32307 1525 21379 640 70997 38819 ## 7 09 Connec… 108002 38022 13581 910 22264 1267 69980 51718 ## 8 10 Delawa… 22745 9755 4801 114 4585 255 12990 11045 ## 9 11 Distri… 40682 5606 2278 315 2727 286 35076 23802 ## 10 16 Idaho 49729 21271 10931 570 8955 815 28458 22521 ## # … with 42 more rows, and 2 more variables: P5_009N <dbl>, P5_010N <dbl> ``` --- ## Using named vectors of variables .pull-left[ * Census variables can be hard to remember; using a named vector to request variables will replace the Census IDs with a custom input * In long form, these custom inputs will populate the `variable` column; in wide form, they will replace the column names ] .pull-right[ ```r vacancies_wide <- get_decennial( geography = "county", state = "MI", * variables = c(vacant_households = "H1_003N", * total_households = "H1_001N"), output = "wide", year = 2020 ) ``` ] --- ```r vacancies_wide ``` ``` ## # A tibble: 83 × 4 ## GEOID NAME vacant_households total_households ## <chr> <chr> <dbl> <dbl> ## 1 26001 Alcona County, Michigan 5356 10263 ## 2 26003 Alger County, Michigan 2560 6169 ## 3 26005 Allegan County, Michigan 6244 51789 ## 4 26007 Alpena County, Michigan 2744 15645 ## 5 26009 Antrim County, Michigan 7391 17538 ## 6 26011 Arenac County, Michigan 2873 9504 ## 7 26013 Baraga County, Michigan 1724 5052 ## 8 26015 Barry County, Michigan 3275 27351 ## 9 26017 Bay County, Michigan 3557 48562 ## 10 26019 Benzie County, Michigan 4346 12099 ## # … with 73 more rows ``` --- ## Part 1 exercises 1. Review the available geographies in tidycensus from the tidycensus documentation. Acquire data on total households (variable `H1_001N`) for a geography we have not yet used. 2. Use the `load_variables()` function to find a variable that interests you that we haven't used yet. Use `get_decennial()` to fetch data from the 2020 Census for counties in a state of your choosing. --- class: middle, center, inverse ## Part 2: Wrangling Census data with tidyverse tools --- ## The tidyverse ```r library(tidyverse) tidyverse_logo() ``` ``` ## ⬢ __ _ __ . ⬡ ⬢ . ## / /_(_)__/ /_ ___ _____ _______ ___ ## / __/ / _ / // / |/ / -_) __(_-</ -_) ## \__/_/\_,_/\_, /|___/\__/_/ /___/\__/ ## ⬢ . /___/ ⬡ . ⬢ ``` * The [tidyverse](https://tidyverse.tidyverse.org/index.html): an integrated set of packages developed primarily by Hadley Wickham and the RStudio team --- ## tidycensus and the tidyverse * Census data are commonly used in _wide_ format, with categories spread across the columns * tidyverse tools work better with [data that are in "tidy", or _long_ format](https://vita.had.co.nz/papers/tidy-data.pdf); this format is returned by tidycensus by default * Goal: return data "ready to go" for use with tidyverse tools --- class: middle, center, inverse ## Exploring 2020 Census data with tidyverse tools --- ## Finding the largest values * dplyr's `arrange()` function sorts data based on values in one or more columns, and `filter()` helps you query data based on column values * Example: what are the largest and smallest counties in Texas by population? ```r library(tidycensus) library(tidyverse) tx_population <- get_decennial( geography = "county", variables = "P1_001N", year = 2020, state = "TX" ) ``` --- ```r arrange(tx_population, value) ``` ``` ## # A tibble: 254 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 48301 Loving County, Texas P1_001N 64 ## 2 48269 King County, Texas P1_001N 265 ## 3 48261 Kenedy County, Texas P1_001N 350 ## 4 48311 McMullen County, Texas P1_001N 600 ## 5 48033 Borden County, Texas P1_001N 631 ## 6 48263 Kent County, Texas P1_001N 753 ## 7 48443 Terrell County, Texas P1_001N 760 ## 8 48393 Roberts County, Texas P1_001N 827 ## 9 48345 Motley County, Texas P1_001N 1063 ## 10 48155 Foard County, Texas P1_001N 1095 ## # … with 244 more rows ``` --- ```r arrange(tx_population, desc(value)) ``` ``` ## # A tibble: 254 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 48201 Harris County, Texas P1_001N 4731145 ## 2 48113 Dallas County, Texas P1_001N 2613539 ## 3 48439 Tarrant County, Texas P1_001N 2110640 ## 4 48029 Bexar County, Texas P1_001N 2009324 ## 5 48453 Travis County, Texas P1_001N 1290188 ## 6 48085 Collin County, Texas P1_001N 1064465 ## 7 48121 Denton County, Texas P1_001N 906422 ## 8 48215 Hidalgo County, Texas P1_001N 870781 ## 9 48141 El Paso County, Texas P1_001N 865657 ## 10 48157 Fort Bend County, Texas P1_001N 822779 ## # … with 244 more rows ``` --- ## What are the counties with a population below 1,000? * The `filter()` function subsets data according to a specified condition, much like a SQL query ```r below1000 <- filter(tx_population, value < 1000) below1000 ``` ``` ## # A tibble: 8 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 48393 Roberts County, Texas P1_001N 827 ## 2 48033 Borden County, Texas P1_001N 631 ## 3 48261 Kenedy County, Texas P1_001N 350 ## 4 48263 Kent County, Texas P1_001N 753 ## 5 48269 King County, Texas P1_001N 265 ## 6 48301 Loving County, Texas P1_001N 64 ## 7 48311 McMullen County, Texas P1_001N 600 ## 8 48443 Terrell County, Texas P1_001N 760 ``` --- ## Using summary variables * Many decennial Census and ACS variables are organized in tables in which the first variable represents a _summary variable_, or denominator for the others * The parameter `summary_var` can be used to generate a new column in long-form data for a requested denominator, which works well for normalizing estimates --- ## Using summary variables ```r race_vars <- c( Hispanic = "P2_002N", White = "P2_005N", Black = "P2_006N", Native = "P2_007N", Asian = "P2_008N", HIPI = "P2_009N" ) az_race <- get_decennial( geography = "county", state = "AZ", variables = race_vars, * summary_var = "P2_001N", year = 2020 ) ``` --- ```r az_race ``` ``` ## # A tibble: 90 × 5 ## GEOID NAME variable value summary_value ## <chr> <chr> <chr> <dbl> <dbl> ## 1 04001 Apache County, Arizona Hispanic 3861 66021 ## 2 04003 Cochise County, Arizona Hispanic 42615 125447 ## 3 04005 Coconino County, Arizona Hispanic 21719 145101 ## 4 04007 Gila County, Arizona Hispanic 9283 53272 ## 5 04009 Graham County, Arizona Hispanic 11428 38533 ## 6 04011 Greenlee County, Arizona Hispanic 4376 9563 ## 7 04012 La Paz County, Arizona Hispanic 4197 16557 ## 8 04013 Maricopa County, Arizona Hispanic 1351415 4420568 ## 9 04015 Mohave County, Arizona Hispanic 34126 213267 ## 10 04017 Navajo County, Arizona Hispanic 10887 106717 ## # … with 80 more rows ``` --- ## Normalizing columns with `mutate()` * dplyr's `mutate()` function is used to calculate new columns in your data; the `select()` column can keep or drop columns by name * In a tidyverse workflow, these steps are commonly linked using the pipe operator (`%>%`) from the magrittr package ```r az_race_percent <- az_race %>% * mutate(percent = 100 * (value / summary_value)) %>% * select(NAME, variable, percent) ``` --- ```r az_race_percent ``` ``` ## # A tibble: 90 × 3 ## NAME variable percent ## <chr> <chr> <dbl> ## 1 Apache County, Arizona Hispanic 5.85 ## 2 Cochise County, Arizona Hispanic 34.0 ## 3 Coconino County, Arizona Hispanic 15.0 ## 4 Gila County, Arizona Hispanic 17.4 ## 5 Graham County, Arizona Hispanic 29.7 ## 6 Greenlee County, Arizona Hispanic 45.8 ## 7 La Paz County, Arizona Hispanic 25.3 ## 8 Maricopa County, Arizona Hispanic 30.6 ## 9 Mohave County, Arizona Hispanic 16.0 ## 10 Navajo County, Arizona Hispanic 10.2 ## # … with 80 more rows ``` --- class: middle, center, inverse ## Group-wise Census data analysis --- ## Group-wise Census data analysis * The `group_by()` and `summarize()` functions in dplyr are used to implement the split-apply-combine method of data analysis * The default "tidy" format returned by tidycensus is designed to work well with group-wise Census data analysis workflows --- ## What is the largest group by county? ```r largest_group <- az_race_percent %>% * group_by(NAME) %>% * filter(percent == max(percent)) ``` --- ```r largest_group ``` ``` ## # A tibble: 15 × 3 ## # Groups: NAME [15] ## NAME variable percent ## <chr> <chr> <dbl> ## 1 Santa Cruz County, Arizona Hispanic 83.1 ## 2 Yuma County, Arizona Hispanic 63.8 ## 3 Cochise County, Arizona White 54.4 ## 4 Coconino County, Arizona White 53.0 ## 5 Gila County, Arizona White 61.5 ## 6 Graham County, Arizona White 52.9 ## 7 Greenlee County, Arizona White 46.5 ## 8 La Paz County, Arizona White 54.7 ## 9 Maricopa County, Arizona White 53.3 ## 10 Mohave County, Arizona White 75.1 ## 11 Pima County, Arizona White 51.5 ## 12 Pinal County, Arizona White 56.4 ## 13 Yavapai County, Arizona White 77.6 ## 14 Apache County, Arizona Native 70.4 ## 15 Navajo County, Arizona Native 43.6 ``` --- ## What are the median percentages by group? ```r az_race_percent %>% * group_by(variable) %>% * summarize(median_pct = median(percent)) ``` ``` ## # A tibble: 6 × 2 ## variable median_pct ## <chr> <dbl> ## 1 Asian 1.14 ## 2 Black 0.967 ## 3 HIPI 0.114 ## 4 Hispanic 28.6 ## 5 Native 2.88 ## 6 White 53.0 ``` --- class: middle, center, inverse ## Analyzing change since 2010 --- ## How have areas changed since the 2010 Census? * A common use-case for the 2020 decennial Census data is to assess change over time * For example: which areas have experienced the most population growth, and which have experienced the steepest declines? * __tidycensus__ allows users to access the 2000 and 2010 decennial Census data for comparison, though variable IDs will differ --- ## Getting data from the 2010 Census ```r county_pop_10 <- get_decennial( geography = "county", * variables = "P001001", * year = 2010 ) county_pop_10 ``` ``` ## # A tibble: 3,221 × 4 ## GEOID NAME variable value ## <chr> <chr> <chr> <dbl> ## 1 05131 Sebastian County, Arkansas P001001 125744 ## 2 05133 Sevier County, Arkansas P001001 17058 ## 3 05135 Sharp County, Arkansas P001001 17264 ## 4 05137 Stone County, Arkansas P001001 12394 ## 5 05139 Union County, Arkansas P001001 41639 ## 6 05141 Van Buren County, Arkansas P001001 17295 ## 7 05143 Washington County, Arkansas P001001 203065 ## 8 05145 White County, Arkansas P001001 77076 ## 9 05149 Yell County, Arkansas P001001 22185 ## 10 06011 Colusa County, California P001001 21419 ## # … with 3,211 more rows ``` --- ## Cleanup before joining * The `select()` function can both subset datasets by column and rename columns, "cleaning up" a dataset before joining to another dataset ```r county_pop_10_clean <- county_pop_10 %>% * select(GEOID, value10 = value) county_pop_10_clean ``` ``` ## # A tibble: 3,221 × 2 ## GEOID value10 ## <chr> <dbl> ## 1 05131 125744 ## 2 05133 17058 ## 3 05135 17264 ## 4 05137 12394 ## 5 05139 41639 ## 6 05141 17295 ## 7 05143 203065 ## 8 05145 77076 ## 9 05149 22185 ## 10 06011 21419 ## # … with 3,211 more rows ``` --- ## Joining data * In __dplyr__, joins are implemented with the `*_join()` family of functions ```r county_pop_20 <- get_decennial( geography = "county", variables = "P1_001N", year = 2020 ) %>% select(GEOID, NAME, value20 = value) county_joined <- county_pop_20 %>% * left_join(county_pop_10_clean, by = "GEOID") ``` --- ```r county_joined ``` ``` ## # A tibble: 3,221 × 4 ## GEOID NAME value20 value10 ## <chr> <chr> <dbl> <dbl> ## 1 19013 Black Hawk County, Iowa 131144 131090 ## 2 19003 Adams County, Iowa 3704 4029 ## 3 19007 Appanoose County, Iowa 12317 12887 ## 4 19009 Audubon County, Iowa 5674 6119 ## 5 19015 Boone County, Iowa 26715 26306 ## 6 19019 Buchanan County, Iowa 20565 20958 ## 7 19023 Butler County, Iowa 14334 14867 ## 8 19025 Calhoun County, Iowa 9927 9670 ## 9 19029 Cass County, Iowa 13127 13956 ## 10 19031 Cedar County, Iowa 18505 18499 ## # … with 3,211 more rows ``` --- ## Calculating change * __dplyr__'s `mutate()` function can be used to calculate new columns, allowing for assessment of change over time ```r county_change <- county_joined %>% * mutate( * total_change = value20 - value10, * percent_change = 100 * (total_change / value10) * ) ``` --- ```r county_change ``` ``` ## # A tibble: 3,221 × 6 ## GEOID NAME value20 value10 total_change percent_change ## <chr> <chr> <dbl> <dbl> <dbl> <dbl> ## 1 19013 Black Hawk County, Iowa 131144 131090 54 0.0412 ## 2 19003 Adams County, Iowa 3704 4029 -325 -8.07 ## 3 19007 Appanoose County, Iowa 12317 12887 -570 -4.42 ## 4 19009 Audubon County, Iowa 5674 6119 -445 -7.27 ## 5 19015 Boone County, Iowa 26715 26306 409 1.55 ## 6 19019 Buchanan County, Iowa 20565 20958 -393 -1.88 ## 7 19023 Butler County, Iowa 14334 14867 -533 -3.59 ## 8 19025 Calhoun County, Iowa 9927 9670 257 2.66 ## 9 19029 Cass County, Iowa 13127 13956 -829 -5.94 ## 10 19031 Cedar County, Iowa 18505 18499 6 0.0324 ## # … with 3,211 more rows ``` --- ## Caveat: changing geographies! * County names and boundaries can change from year to year, introducing potential problems in time-series analysis * This is particularly acute for small geographies like Census tracts & block groups, which we'll cover on March 25! ```r filter(county_change, is.na(value10)) ``` ``` ## # A tibble: 4 × 6 ## GEOID NAME value20 value10 total_change percent_change ## <chr> <chr> <dbl> <dbl> <dbl> <dbl> ## 1 02066 Copper River Census Area, A… 2617 NA NA NA ## 2 02158 Kusilvak Census Area, Alaska 8368 NA NA NA ## 3 46102 Oglala Lakota County, South… 13672 NA NA NA ## 4 02063 Chugach Census Area, Alaska 7102 NA NA NA ``` --- ## Part 2 exercises With the `county_change` object, use __tidyverse__ tools to answer these questions: * Which counties gained and lost the most people during the 2010s? * How many counties in the US grew by 40 percent or more during the 2010s? * How many counties in the US lost 20 percent or more of their populations during the 2010s? --- class: middle, center, inverse ## Part 3: Visualizing US Census data --- ## Visualizing US Census data * __tidycensus__ is designed with ggplot2-based visualization in mind, the core framework for data visualization in the tidyverse * ggplot2 along with its extensions can be used for everything from simple graphics to complex interactive plots --- ## Data setup: Hispanic population by county in Georgia ```r library(tidycensus) library(tidyverse) ga_hispanic <- get_decennial( geography = "county", variables = c(total = "P2_001N", hispanic = "P2_002N"), state = "GA", year = 2020, output = "wide" ) %>% mutate(percent = 100 * (hispanic / total)) ``` --- ```r ga_hispanic ``` ``` ## # A tibble: 159 × 5 ## GEOID NAME total hispanic percent ## <chr> <chr> <dbl> <dbl> <dbl> ## 1 13009 Baldwin County, Georgia 43799 1139 2.60 ## 2 13015 Bartow County, Georgia 108901 10751 9.87 ## 3 13019 Berrien County, Georgia 18160 1045 5.75 ## 4 13025 Brantley County, Georgia 18021 326 1.81 ## 5 13031 Bulloch County, Georgia 81099 4180 5.15 ## 6 13037 Calhoun County, Georgia 5573 149 2.67 ## 7 13043 Candler County, Georgia 10981 1378 12.5 ## 8 13049 Charlton County, Georgia 12518 2036 16.3 ## 9 13055 Chattooga County, Georgia 24965 1297 5.20 ## 10 13061 Clay County, Georgia 2848 41 1.44 ## # … with 149 more rows ``` --- ## Exploring data with visualization .pull-left[ * Graphics in __ggplot2__ are initialized with the `ggplot()` function, in which a user typically supplies a dataset and aesthetic mapping with `aes()` * Graphical elements are then "layered" onto the ggplot object, consisting of a "geom", or geometric object (`geom_*()`) and custom styling elements linked with the `+` operator * Histograms can be created with `geom_histogram()`; the `bins` argument controls the number of bins on the plot ] .pull-right[ ```r ggplot(ga_hispanic, aes(x = percent)) + geom_histogram(bins = 10) ``` <!-- --> ] --- ## Univariate visualization .pull-left[ * Other univariate visualization methods in __ggplot2__ include `geom_boxplot()` for box and whisker plots, `geom_density()` for kernel density plots, and `geom_violin()` for violin plots ] .pull-right[ ```r ggplot(ga_hispanic, aes(x = percent)) + * geom_boxplot() ``` <!-- --> ] --- # Multivariate visualization .pull-left[ * A second variable can be mapped to the second axis for visualization of multivariate relationships * A _scatterplot_ is commonly used to visualize the joint distributions of two quantitative variables, implemented with `geom_point()`. ] .pull-right[ ```r options(scipen = 999) # Disable scientific notation ggplot(ga_hispanic, aes(x = total, y = percent)) + geom_point() ``` <!-- --> ] --- ## Layering multiple geoms .pull-left[ * Multiple geoms can be represented on the same plot by adding additional calls to `geom_*()` to the graphic's code * Shown here: a regression line superimposed over the scatterplot to show the linear relationship ] .pull-right[ ```r ggplot(ga_hispanic, aes(x = total, y = percent)) + geom_point() + * geom_smooth(method = "lm") ``` <!-- --> ] --- ## Modifying axis scales .pull-left[ * In many cases, the baseline populations of Census units will vary dramatically (counties in Georgia have populations ranging from 1500 to 1 million) * Changing a scale from linear to logarithmic can help with exploratory visualization when data is heavily skewed in this way ] .pull-right[ ```r ggplot(ga_hispanic, aes(x = total, y = percent)) + geom_point() + * scale_x_log10() + geom_smooth() ``` <!-- --> ] --- class: middle, center, inverse # Customizing styling of Census plots with ggplot2 --- * Prompt: comparing vacant household percentages by county in a state ```r nj_vacancies <- get_decennial( geography = "county", variables = c(total_households = "H1_001N", vacant_households = "H1_003N"), state = "NJ", year = 2020, output = "wide" ) %>% mutate(percent_vacant = 100 * (vacant_households / total_households)) ``` --- ## Comparative plots * A categorical variable (rather than a numeric one) can be mapped to the second axis to compare Census data by category (e.g. by county) ```r ggplot(nj_vacancies, aes(x = percent_vacant, y = NAME)) + geom_col() ``` --- <img src="img/first_bar.png" style="width: 800px"> --- ## Improving your plot * In the code below, we format the axis tick labels with functions and apply custom labels to chart elements like the axis and plot titles ```r library(scales) ggplot(nj_vacancies, aes(x = percent_vacant, y = reorder(NAME, percent_vacant))) + geom_col() + * scale_x_continuous(labels = label_percent(scale = 1)) + * scale_y_discrete(labels = function(y) str_remove(y, " County, New Jersey")) + * labs(x = "Percent vacant households", * y = "", * title = "Household vacancies by county in New Jersey", * subtitle = "2020 decennial US Census") ``` --- <img src="img/second_bar.png" style="width: 800px"> --- ## Styling your plot * Once you have settled on a general format, you can style the plot to your liking with fonts, colors and more! ```r ggplot(nj_vacancies, aes(x = percent_vacant, y = reorder(NAME, percent_vacant))) + * geom_col(fill = "navy", color = "navy", alpha = 0.5) + * theme_minimal(base_family = "Verdana") + scale_x_continuous(labels = label_percent(scale = 1)) + scale_y_discrete(labels = function(y) str_remove(y, " County, New Jersey")) + labs(x = "Percent vacant households", y = "", title = "Household vacancies by county in New Jersey", subtitle = "2020 decennial US Census") ``` --- <img src="img/third_bar.png" style="width: 800px"> --- class: middle, center, inverse ## Visualizing group-wise comparisons --- * Prompt: how do the distributions of percentage Black population by Census tract vary among the five boroughs of New York City? ```r nyc_percent_black <- get_decennial( geography = "tract", variables = "P2_006N", summary_var = "P2_001N", state = "NY", county = c("New York", "Kings", "Queens", "Bronx", "Richmond"), year = 2020 ) %>% mutate(percent = 100 * (value / summary_value)) ``` --- ```r nyc_percent_black ``` ``` ## # A tibble: 2,327 × 6 ## GEOID NAME variable value summary_value percent ## <chr> <chr> <chr> <dbl> <dbl> <dbl> ## 1 36047057300 Census Tract 573, Kings Cou… P2_006N 49 2590 1.89 ## 2 36047057000 Census Tract 570, Kings Cou… P2_006N 257 3534 7.27 ## 3 36047057100 Census Tract 571, Kings Cou… P2_006N 42 4267 0.984 ## 4 36047057200 Census Tract 572, Kings Cou… P2_006N 2585 5221 49.5 ## 5 36047057400 Census Tract 574, Kings Cou… P2_006N 57 2560 2.23 ## 6 36047057500 Census Tract 575, Kings Cou… P2_006N 76 4902 1.55 ## 7 36047057600 Census Tract 576, Kings Cou… P2_006N 56 2912 1.92 ## 8 36047057800 Census Tract 578, Kings Cou… P2_006N 74 3332 2.22 ## 9 36047057901 Census Tract 579.01, Kings … P2_006N 70 1416 4.94 ## 10 36047057902 Census Tract 579.02, Kings … P2_006N 0 0 NaN ## # … with 2,317 more rows ``` --- ## Separating columns * The `separate()` function splits values in a single column into multiple columns * This function can be used to parse the `NAME` column returned by __tidycensus__ to obtain tract, county, and state information ```r nyc_percent_black2 <- nyc_percent_black %>% separate(NAME, into = c("tract", "county", "state"), sep = ", ") ``` --- ```r nyc_percent_black2 ``` ``` ## # A tibble: 2,327 × 8 ## GEOID tract county state variable value summary_value percent ## <chr> <chr> <chr> <chr> <chr> <dbl> <dbl> <dbl> ## 1 36047057300 Census Tract 5… Kings… New … P2_006N 49 2590 1.89 ## 2 36047057000 Census Tract 5… Kings… New … P2_006N 257 3534 7.27 ## 3 36047057100 Census Tract 5… Kings… New … P2_006N 42 4267 0.984 ## 4 36047057200 Census Tract 5… Kings… New … P2_006N 2585 5221 49.5 ## 5 36047057400 Census Tract 5… Kings… New … P2_006N 57 2560 2.23 ## 6 36047057500 Census Tract 5… Kings… New … P2_006N 76 4902 1.55 ## 7 36047057600 Census Tract 5… Kings… New … P2_006N 56 2912 1.92 ## 8 36047057800 Census Tract 5… Kings… New … P2_006N 74 3332 2.22 ## 9 36047057901 Census Tract 5… Kings… New … P2_006N 70 1416 4.94 ## 10 36047057902 Census Tract 5… Kings… New … P2_006N 0 0 NaN ## # … with 2,317 more rows ``` --- ## Visualizing data by group .pull-left[ * Mapping a categorical variable to the `fill` or `color` aesthetics (depending on the geom used) will draw one geom per category on the plot ] .pull-right[ ```r ggplot(nyc_percent_black2, aes(x = percent, fill = county)) + geom_density(alpha = 0.3) ``` <!-- --> ] --- ## Faceted visualization * The `facet_wrap()` function splits plots into separate panels by category, creating "small multiples" visualizations that are excellent for making comparisons ```r ggplot(nyc_percent_black2, aes(x = percent)) + geom_density(fill = "darkgreen", color = "darkgreen", alpha = 0.5) + * facet_wrap(~county) + * scale_x_continuous(labels = label_percent(scale = 1)) + * theme_minimal(base_size = 14) + * theme(axis.text.y = element_blank()) + * labs(x = "Percent Black", * y = "", * title = "Black population shares by Census tract, 2020") ``` --- <img src=img/nyc_facets.png style="width: 800px"> --- ## Ridgeline plots * The __ggridges__ package implements _ridgeline plots_, which visualize overlapping density plots among categories ```r library(ggridges) ggplot(nyc_percent_black2, aes(x = percent, y = county)) + * geom_density_ridges() + * theme_ridges() + labs(x = "Percent Black, 2020 (by Census tract)", y = "") + scale_x_continuous(labels = label_percent(scale = 1)) ``` --- <img src=img/nyc_ridgeline.png style="width: 800px"> --- class: middle, center, inverse ## Advanced example: geo-faceted plots --- ## ggplot2 extensions * [Highly customized Census data visualizations are possible with extensions to ggplot2](https://exts.ggplot2.tidyverse.org/gallery/) <img src="img/extensions.png" style="width: 650px"> --- ## Step 1: acquire data for all Census tracts in the US ```r us_percent_white <- map_dfr(c(state.abb, "DC"), function(state) { get_decennial( geography = "tract", variables = "P2_005N", summary_var = "P2_001N", state = state, year = 2020 ) %>% mutate(percent = 100 * (value / summary_value)) %>% separate(NAME, into = c("tract", "county", "state"), sep = ", ") }) ``` --- ## Step 2: build a geo-faceted plot ```r library(geofacet) ggplot(us_percent_white, aes(x = percent)) + geom_histogram(fill = "navy", alpha = 0.8, bins = 30) + theme_minimal() + scale_fill_manual(values = c("darkred", "navy")) + facet_geo(~state, grid = "us_state_grid2", label = "code", scales = "free_y") + theme(axis.text = element_blank(), strip.text.x = element_text(size = 8)) + labs(x = "", y = "", title = "Non-Hispanic white population shares among Census tracts", fill = "", caption = "Data source: 2020 decennial US Census & tidycensus R package\nX-axes range from 0% white (on the left) to 100% white (on the right). Y-axes are unique to each state.") ``` --- <img src=img/geofacet.png style="width: 800px"> --- ## Part 3 exercises * Choose one of the example charts and try customizing its appearance. Some tips on styling are found at https://ggplot2.tidyverse.org/articles/ggplot2-specs.html. * Try customizing the New Jersey vacancies example for a different variable (challenge: express it as an appropriate percentage!) and a different state. --- class: middle, center, inverse ## Thank you!