Introduction to ipumsr - IPUMS Data in R

Minnesota Population Center

2022-06-04

The ipumsr package allows you to read data from your IPUMS extract into R along with the associated metadata like variable labels, value labels and more. IPUMS is a great source of international census and survey data.

IPUMS provides census and survey data from around the world integrated across time and space. IPUMS integration and documentation makes it easy to study change, conduct comparative research, merge information across data types, and analyze individuals within family and community context. Data and services are available free of charge.

Learn more here: https://www.ipums.org/mission-purpose

Additional Vignettes

This vignette gives the basic outline of the ipumsr package. Additional vignettes provide guidance on working with IPUMS value labels, IPUMS geographic data, and data from three specific IPUMS projects – CPS, NHGIS, and Terra.

To view any of these additional vignettes, run one the following commands after installing the ipumsr package:

vignette("value-labels", package = "ipumsr")
vignette("ipums-geography", package = "ipumsr")
vignette("ipums-cps", package = "ipumsr")
vignette("ipums-nhgis", package = "ipumsr")
vignette("ipums-terra", package = "ipumsr")

Getting Data from the IPUMS website

Registered users can download IPUMS data from our website at https://www.ipums.org. The website provides an interactive extract system that allows you to select only the samples and variables that are relevant to your research question.

For microdata projects (all supported projects except NHGIS and IPUMS Terra), once you have created your extract, you should choose to download your data in either a fixed-width text (.dat file extension) or comma delimited (.csv file extension) format.

Once your extract is complete, download the data file and the DDI. Downloading the DDI is a little bit different depending on your browser. On most browsers you should right-click the file and select “Save As…”. If this saves a file with a .xml file extension, then you should be ready. However, Safari users must select “Download Linked File” instead of “Download Linked File As”. On Safari, selecting the wrong version of these two will download a file with a .html file extension instead of a .xml extension.

Annotated screenshot for downloading microdata (may look different on your computer)

For NHGIS, download the table data in a comma delimited format, and, if you want the associated mapping data, the GIS data. NHGIS provides the option to download comma-delimited files with an extra header row; it does not matter which option you select.

For IPUMS Terra, download the .zip extract “bundles”. If you want the associated mapping data, select the “include boundary files” option. You do not need to unzip the data.

Import Functions

Once your extract is downloaded, the ipumsr package functions read_*() help you load the data into R.

Metadata Functions

Once the data is in R, you can view information about the extract using the metadata functions.

Grab Bag

Survey Weights

The data from most IPUMS projects contain some form of weighting variable that should be used to calculate estimates that are representative of the whole population. Many projects also provide specifications to help estimate variance given the complex design of the survey, such as replicate weights or design variables like STRATUM and PSU. The survey package provides functions that allow you to estimate variance taking this into account, and the srvyr package implements dplyr-like syntax for survey analysis, using the survey package’s functions.

For more information about what these variables mean and how to use them, see the website for the project you are interested in.

Non-Extract Data

Some projects have data that is not contained within the extract system, such that no DDI is provided for this data. In this case, either use the comma delimited file (.csv file extension) if available, or use the haven package to read one of the files intended for another statistical software (like Stata, SAS or SPSS).

Value Labels

The way that IPUMS treats value labels does not align with factors (the main way that R is able to store values associated with labels). R’s factor variables can only store values as an integer sequence (1, 2, 3, …), but IPUMS conventions are to store missing and not-in-universe codes as large numbers, to distinguish them from the normal values.

Therefore, the ipumsr package uses the labelled class from the haven package to store labelled values. See the “value-labels” vignette for more information (vignette('value-labels')).

If you want to use IPUMS value labels attached to a variable, it is generally best to convert from the labelled class to factor early on in your data analysis workflow. This is because many data manipulation functions will drop the labels stored in a variable with class labelled. The function as_factor() is the main function to create factors from labels, but often you will need to do more manipulation before that.

library(ipumsr)
library(dplyr, warn.conflicts = FALSE)

# Note that you can pass in the loaded DDI into the `read_ipums_micro()`
cps_ddi <- read_ipums_ddi(ipums_example("cps_00006.xml"))
cps_data <- read_ipums_micro(cps_ddi, verbose = FALSE)

# Show which variables have labels
cps_data %>%
  select_if(is.labelled)
#> # A tibble: 7,668 × 3
#>          STATEFIP     MONTH                               INCTOT
#>         <int+lbl> <int+lbl>                            <dbl+lbl>
#>  1 55 [Wisconsin] 3 [March]     4883                            
#>  2 55 [Wisconsin] 3 [March]     5800                            
#>  3 55 [Wisconsin] 3 [March] 99999998 [Missing.]                 
#>  4 27 [Minnesota] 3 [March]    14015                            
#>  5 27 [Minnesota] 3 [March]    16552                            
#>  6 27 [Minnesota] 3 [March]     6375                            
#>  7 19 [Iowa]      3 [March] 99999999 [N.I.U. (Not in Universe).]
#>  8 19 [Iowa]      3 [March]        0                            
#>  9 19 [Iowa]      3 [March]      600                            
#> 10 19 [Iowa]      3 [March] 99999999 [N.I.U. (Not in Universe).]
#> # … with 7,658 more rows

# Notice how the tibble print function shows the dbl+lbl class on top

# Investigate labels
ipums_val_labels(cps_data$STATEFIP)
#> # A tibble: 75 × 2
#>      val lbl                 
#>    <int> <chr>               
#>  1     1 Alabama             
#>  2     2 Alaska              
#>  3     4 Arizona             
#>  4     5 Arkansas            
#>  5     6 California          
#>  6     8 Colorado            
#>  7     9 Connecticut         
#>  8    10 Delaware            
#>  9    11 District of Columbia
#> 10    12 Florida             
#> # … with 65 more rows

# Convert the labels to factors (and drop the unused levels)
cps_data <- cps_data %>%
  mutate(STATE_factor = as_factor(lbl_clean(STATEFIP)))

table(cps_data$STATE_factor, useNA = "always")
#> 
#>         Iowa    Minnesota North Dakota South Dakota    Wisconsin         <NA> 
#>         1892         2362          188          227         2999            0
# Manipulating the labelled value before as_factor 
# often leads to losing the information...
# Say we want to set Iowa (STATEFIP == 19) to missing
cps_data <- cps_data %>%
  mutate(STATE_factor2 = as_factor(ifelse(STATEFIP == 19, NA, STATEFIP)))
# ipumsr provides helpers for these kinds of tasks, like lbl_na_if().
# See the value-labels vignette for more information
cps_data <- cps_data %>%
  mutate(STATE_factor3 = as_factor(lbl_na_if(STATEFIP, ~.val == 19)))

# The as_factor function also has a "levels" argument that can 
# put both the labels and values into the factor
cps_data <- cps_data %>%
  mutate(STATE_factor4 = droplevels(as_factor(STATEFIP, levels = "both")))

table(cps_data$STATE_factor4, useNA = "always")
#> 
#>         [19] Iowa    [27] Minnesota [38] North Dakota [46] South Dakota 
#>              1892              2362               188               227 
#>    [55] Wisconsin              <NA> 
#>              2999                 0

Other IPUMS attributes

As with value labels, the other attributes that ipumsr stores about the data are often lost during an analysis. One way to deal with this is to load the DDI or codebook in addition to the actual data using the functions read_ipums_ddi() and read_ipums_codebook(). This way, when you wish to refer to variable labels or other metadata, you can use the DDI object, which does not get modified during your analysis.

library(ipumsr)
library(dplyr, warn.conflicts = FALSE)

# Note that you can pass in the loaded DDI into the `read_ipums_micro()`
cps_ddi <- read_ipums_ddi(ipums_example("cps_00006.xml"))
cps_data <- read_ipums_micro(cps_ddi, verbose = FALSE)

# Currently variable description is available for year
ipums_var_desc(cps_data$YEAR)
#> [1] "YEAR reports the year in which the survey was conducted.  YEARP is repeated on person records."

# But after using ifelse it is gone
cps_data <- cps_data %>%
  mutate(YEAR = ifelse(YEAR == 1962, 62, NA))
ipums_var_desc(cps_data$YEAR)
#> [1] NA

# So you can use the DDI
ipums_var_desc(cps_ddi, "YEAR")
#> [1] "YEAR reports the year in which the survey was conducted.  YEARP is repeated on person records."

# The DDI also has file level information that is not available from just
# the data.
ipums_file_info(cps_ddi, "extract_notes") %>% cat()
#> User-provided description:  Minimal test extract
#> Samples: 1962, 1963
#> Variables: STATEFIP, INCTOT (automatically Year, SERIAL, HWTSUPP, MONTH, WTSUPP)
#> Select Cases: State - Minnesota, Iowa, Wisconsin, South Dakota, North Dakota

“dplyr select-style” Syntax

Several functions within the ipumsr package allow for “dplyr select-style” syntax. This means that they accept either a character vector of values (e.g. c("YEAR", "AGE")), bare vectors of values (e.g. c(YEAR, AGE)) and the helper functions allowed in dplyr::select() (e.g. one_of(c("YEAR", "AGE"))).

library(ipumsr)
library(dplyr, warn.conflicts = FALSE)

# The vars argument for `read_ipums_micro` uses this syntax
# So these are all equivalent
cf <- ipums_example("cps_00006.xml")
read_ipums_micro(cf, vars = c("YEAR", "INCTOT"), verbose = FALSE) %>%
  names()
#> [1] "YEAR"   "INCTOT"

read_ipums_micro(cf, vars = c(YEAR, INCTOT), verbose = FALSE) %>%
  names()
#> [1] "YEAR"   "INCTOT"

read_ipums_micro(cf, vars = c(one_of("YEAR"), starts_with("INC")), verbose = FALSE) %>%
  names()
#> [1] "YEAR"   "INCTOT"

# `data_layer` and `shape_layer` arguments to `read_nhgis()` and terra functions
# also use it.
# (Sometimes extracts have multiple files, though all examples only have one)
nf <- ipums_example("nhgis0008_csv.zip")
ipums_list_files(nf)
#> # A tibble: 1 × 2
#>   type  file                                       
#>   <chr> <chr>                                      
#> 1 data  nhgis0008_csv/nhgis0008_ds135_1990_pmsa.csv

ipums_list_files(nf, data_layer = "nhgis0008_csv/nhgis0008_ds135_1990_pmsa.csv")
#> # A tibble: 1 × 2
#>   type  file                                       
#>   <chr> <chr>                                      
#> 1 data  nhgis0008_csv/nhgis0008_ds135_1990_pmsa.csv

ipums_list_files(nf, data_layer = contains("ds135"))
#> # A tibble: 1 × 2
#>   type  file                                       
#>   <chr> <chr>                                      
#> 1 data  nhgis0008_csv/nhgis0008_ds135_1990_pmsa.csv

Hierarchical data structures

For certain IPUMS projects, the data is hierarchical, multiple people are included in a single household, or multiple activities are performed by a single person. The ipumsr package provides two data structures for storing such data (for users who did not select the “rectangularize” option on the website). The data can be loaded as a "list" or "long".

List data loads each record type into a separate data.frame. The names of the recordtype data.frames are the value of the RECTYPE variable (e.g. “H” and “P”). Use the function read_ipums_micro_list() to load the data this way.

Long data has one row per unit, regardless of what type of record the unit is. Therefore, datasets loaded this way often contain variables with a large number of missings, for the variables that only apply to certain record types. Use the function read_ipums_micro() to load the data this way.

library(ipumsr)
library(dplyr, warn.conflicts = FALSE)

# List data
cps <- read_ipums_micro_list(
  ipums_example("cps_00010.xml"),
  verbose = FALSE
)

cps$PERSON
#> # A tibble: 7,668 × 6
#>    RECTYPE            YEAR SERIAL PERNUM WTSUPP                           INCTOT
#>    <chr+lbl>         <dbl>  <dbl>  <dbl>  <dbl>                        <dbl+lbl>
#>  1 P [Person Record]  1962     80      1  1476.     4883                        
#>  2 P [Person Record]  1962     80      2  1471.     5800                        
#>  3 P [Person Record]  1962     80      3  1579. 99999998 [Missing.]             
#>  4 P [Person Record]  1962     82      1  1598.    14015                        
#>  5 P [Person Record]  1962     83      1  1707.    16552                        
#>  6 P [Person Record]  1962     84      1  1790.     6375                        
#>  7 P [Person Record]  1962    107      1  4355. 99999999 [N.I.U. (Not in Univer…
#>  8 P [Person Record]  1962    107      2  1386.        0                        
#>  9 P [Person Record]  1962    107      3  1629.      600                        
#> 10 P [Person Record]  1962    107      4  1432. 99999999 [N.I.U. (Not in Univer…
#> # … with 7,658 more rows

cps$HOUSEHOLD
#> # A tibble: 3,385 × 6
#>    RECTYPE               YEAR SERIAL HWTSUPP       STATEFIP     MONTH
#>    <chr+lbl>            <dbl>  <dbl>   <dbl>      <int+lbl> <int+lbl>
#>  1 H [Household Record]  1962     80   1476. 55 [Wisconsin] 3 [March]
#>  2 H [Household Record]  1962     82   1598. 27 [Minnesota] 3 [March]
#>  3 H [Household Record]  1962     83   1707. 27 [Minnesota] 3 [March]
#>  4 H [Household Record]  1962     84   1790. 27 [Minnesota] 3 [March]
#>  5 H [Household Record]  1962    107   4355. 19 [Iowa]      3 [March]
#>  6 H [Household Record]  1962    108   1479. 19 [Iowa]      3 [March]
#>  7 H [Household Record]  1962    122   3603. 27 [Minnesota] 3 [March]
#>  8 H [Household Record]  1962    124   4104. 55 [Wisconsin] 3 [March]
#>  9 H [Household Record]  1962    125   2182. 55 [Wisconsin] 3 [March]
#> 10 H [Household Record]  1962    126   1826. 55 [Wisconsin] 3 [March]
#> # … with 3,375 more rows

# Long data
cps <- read_ipums_micro(
  ipums_example("cps_00010.xml"),
  verbose = FALSE
)

cps
#> # A tibble: 11,053 × 9
#>    RECTYPE     YEAR SERIAL HWTSUPP STATEFIP    MONTH PERNUM WTSUPP        INCTOT
#>    <chr+lbl>  <dbl>  <dbl>   <dbl> <int+lb> <int+lb>  <dbl>  <dbl>     <dbl+lbl>
#>  1 H [Househ…  1962     80   1476. 55 [Wis…  3 [Mar…     NA    NA  NA           
#>  2 P [Person…  1962     80     NA  NA       NA            1  1476.  4.88e3      
#>  3 P [Person…  1962     80     NA  NA       NA            2  1471.  5.8 e3      
#>  4 P [Person…  1962     80     NA  NA       NA            3  1579.  1.00e8 [Mis…
#>  5 H [Househ…  1962     82   1598. 27 [Min…  3 [Mar…     NA    NA  NA           
#>  6 P [Person…  1962     82     NA  NA       NA            1  1598.  1.40e4      
#>  7 H [Househ…  1962     83   1707. 27 [Min…  3 [Mar…     NA    NA  NA           
#>  8 P [Person…  1962     83     NA  NA       NA            1  1707.  1.66e4      
#>  9 H [Househ…  1962     84   1790. 27 [Min…  3 [Mar…     NA    NA  NA           
#> 10 P [Person…  1962     84     NA  NA       NA            1  1790.  6.38e3      
#> # … with 11,043 more rows

Geospatial Packages: sf vs sp

The ipumsr package allows for loading geospatial data in two formats (sf for Simple Features and sp for Spatial). The sf package is relatively new, and so does not have as widespread support as the sp package. However, (in my opinion) it does allow for easier analysis, and so may be a better place to start if you have not used GIS data in R before.

For more details about how to load geographic data using ipumsr, see the vignette “ipums-geography” (vignette("ipums-geography", package = "ipumsr"))