read_ipums_micro_chunked: Read data from an IPUMS microdata extract by chunk

View source: R/micro_read_chunked.R

read_ipums_micro_chunkedR Documentation

Read data from an IPUMS microdata extract by chunk

Description

Read a microdata dataset downloaded from the IPUMS extract system in chunks.

Use these functions to read a file that is too large to store in memory at a single time. The file is processed in chunks of a given size, with a provided callback function applied to each chunk.

Two files are required to load IPUMS microdata extracts:

  • A DDI codebook file (.xml) used to parse the extract's data file

  • A data file (either .dat.gz or .csv.gz)

See Downloading IPUMS files below for more information about downloading these files.

read_ipums_micro_chunked() and read_ipums_micro_list_chunked() differ in their handling of extracts that contain multiple record types. See Data structures below.

Note that Stata, SAS, and SPSS file formats are not supported by ipumsr readers. Convert your extract to fixed-width or CSV format, or see haven for help loading these files.

Usage

read_ipums_micro_chunked(
  ddi,
  callback,
  chunk_size = 10000,
  vars = NULL,
  data_file = NULL,
  verbose = TRUE,
  var_attrs = c("val_labels", "var_label", "var_desc"),
  lower_vars = FALSE
)

read_ipums_micro_list_chunked(
  ddi,
  callback,
  chunk_size = 10000,
  vars = NULL,
  data_file = NULL,
  verbose = TRUE,
  var_attrs = c("val_labels", "var_label", "var_desc"),
  lower_vars = FALSE
)

Arguments

ddi

Either a path to a DDI .xml file downloaded from IPUMS, or an ipums_ddi object parsed by read_ipums_ddi(). See Downloading IPUMS files below.

callback

An ipums_callback object, or a function that will be converted to an IpumsSideEffectCallback object. Callback functions should include both data (x) and position (pos) arguments. See examples.

chunk_size

Integer number of observations to read per chunk. Higher values use more RAM, but typically result in faster processing. Defaults to 10,000.

vars

Names of variables to include in the output. Accepts a vector of names or a tidyselect selection. If NULL, includes all variables in the file.

For hierarchical data, the RECTYPE variable is always included even if unspecified.

data_file

Path to the data (.gz) file associated with the provided ddi file. By default, looks for the data file in the same directory as the DDI file. If the data file has been moved, specify its location here.

verbose

Logical indicating whether to display IPUMS conditions and progress information.

var_attrs

Variable attributes from the DDI to add to the columns of the output data. Defaults to all available attributes. See set_ipums_var_attributes() for more details.

lower_vars

If reading a DDI from a file, a logical indicating whether to convert variable names to lowercase. Defaults to FALSE for consistency with IPUMS conventions.

This argument will be ignored if argument ddi is an ipums_ddi object. Use read_ipums_ddi() to convert variable names to lowercase when reading a DDI file.

Note that if reading in chunks from a .csv or .csv.gz file, the callback function will be called before variable names are converted to lowercase, and thus should reference uppercase variable names.

Value

Depends on the provided callback object. See ipums_callback.

Data structures

Files from IPUMS projects that contain data for multiple types of records (e.g. household records and person records) may be either rectangular or hierarchical.

Rectangular data are transformed such that each row of data represents only one type of record. For instance, each row will represent a person record, and all household-level information for that person will be included in the same row.

Hierarchical data have records of different types interspersed in a single file. For instance, a household record will be included in its own row followed by the person records associated with that household.

Hierarchical data can be read in two different formats:

  • read_ipums_micro_chunked() reads each chunk of data into a tibble where each row represents a single record, regardless of record type. Variables that do not apply to a particular record type will be filled with NA in rows of that record type. For instance, a person-specific variable will be missing in all rows associated with household records. The provided callback function should therefore operate on a tibble object.

  • read_ipums_micro_list_chunked() reads each chunk of data into a list of tibble objects, where each list element contains only one record type. Each list element is named with its corresponding record type. The provided callback function should therefore operate on a list object. In this case, the chunk size references the total number of rows across record types, rather than in each record type.

Downloading IPUMS files

You must download both the DDI codebook and the data file from the IPUMS extract system to load the data into R. ⁠read_ipums_micro_*()⁠ functions assume that the data file and codebook share a common base file name and are present in the same directory. If this is not the case, provide a separate path to the data file with the data_file argument.

If using the IPUMS extract interface:

  • Download the data file by clicking Download .dat under Download Data.

  • Download the DDI codebook by right clicking on the DDI link in the Codebook column of the extract interface and selecting Save as... (on Safari, you may have to select Download Linked File as...). Be sure that the codebook is downloaded in .xml format.

If using the IPUMS API:

  • For supported collections, use download_extract() to download a completed extract via the IPUMS API. This automatically downloads both the DDI codebook and the data file from the extract and returns the path to the codebook file.

See Also

read_ipums_micro_yield() for more flexible handling of large IPUMS microdata files.

read_ipums_micro() to read data from an IPUMS microdata extract.

read_ipums_ddi() to read metadata associated with an IPUMS microdata extract.

read_ipums_sf() to read spatial data from an IPUMS extract.

ipums_list_files() to list files in an IPUMS extract.

Examples

suppressMessages(library(dplyr))

# Example codebook file
cps_rect_ddi_file <- ipums_example("cps_00157.xml")

# Function to extract Minnesota cases from CPS example
# (This can also be accomplished by including case selections
# in an extract definition)
#
# Function must take `x` and `pos` to refer to data and row position,
# respectively.
filter_mn <- function(x, pos) {
  x[x$STATEFIP == 27, ]
}

# Initialize callback
filter_mn_callback <- IpumsDataFrameCallback$new(filter_mn)

# Process data in chunks, filtering to MN cases in each chunk
read_ipums_micro_chunked(
  cps_rect_ddi_file,
  callback = filter_mn_callback,
  chunk_size = 1000,
  verbose = FALSE
)

# Tabulate INCTOT average by state without storing full dataset in memory
read_ipums_micro_chunked(
  cps_rect_ddi_file,
  callback = IpumsDataFrameCallback$new(
    function(x, pos) {
      x %>%
        mutate(
          INCTOT = lbl_na_if(
            INCTOT,
            ~ grepl("Missing|N.I.U.", .lbl)
          )
        ) %>%
        filter(!is.na(INCTOT)) %>%
        group_by(STATEFIP = as_factor(STATEFIP)) %>%
        summarize(INCTOT_SUM = sum(INCTOT), n = n(), .groups = "drop")
    }
  ),
  chunk_size = 1000,
  verbose = FALSE
) %>%
  group_by(STATEFIP) %>%
  summarize(avg_inc = sum(INCTOT_SUM) / sum(n))

# `x` will be a list when using `read_ipums_micro_list_chunked()`
read_ipums_micro_list_chunked(
  ipums_example("cps_00159.xml"),
  callback = IpumsSideEffectCallback$new(function(x, pos) {
    print(
      paste0(
        nrow(x$PERSON), " persons and ",
        nrow(x$HOUSEHOLD), " households in this chunk."
      )
    )
  }),
  chunk_size = 1000,
  verbose = FALSE
)

# Using the biglm package, you can even run a regression without storing
# the full dataset in memory
if (requireNamespace("biglm")) {
  lm_results <- read_ipums_micro_chunked(
   ipums_example("cps_00160.xml"),
   IpumsBiglmCallback$new(
     INCTOT ~ AGE + HEALTH, # Model formula
     function(x, pos) {
       x %>%
         mutate(
           INCTOT = lbl_na_if(
             INCTOT,
             ~ grepl("Missing|N.I.U.", .lbl)
           ),
           HEALTH = as_factor(HEALTH)
         )
     }
   ),
   chunk_size = 1000,
   verbose = FALSE
 )

 summary(lm_results)
}

ipumsr documentation built on Oct. 20, 2023, 5:10 p.m.