Tutorial on cleaning the LTDB

# load necessary packages ----
library( dplyr )
library( here )
library( knitr )
library( pander )

# import specific functions
# note: all of these are R objects that will be used throughout this .rmd file
import::here("clean_d",
             "tidy_up_data",
             "build_year",
             "RELEVANT_FILES",
             "obtain_crosswalk",
             "create_final_metadata_file",
             # notice the use of here::here() that points to the .R file
             # where all these R objects are created
             .from = here::here("labs/utilities.R"),
             .character_only = TRUE)

Inspect Data

First, let’s inspect the raw data. Note: please do not import files using static file paths. Notice the use of here::here() down below.

# load all data as character vecs
d.2010.samp <- read.csv( here::here("data/raw/ltdb_std_2010_sample.csv"),
                         colClasses="character" )
str( d.2010.samp[1:10] )

## 'data.frame':    73056 obs. of  10 variables:
##  $ tractid : chr  "1001020100" "1001020200" "1001020300" "1001020400" ...
##  $ statea  : chr  "01" "01" "01" "01" ...
##  $ countya : chr  "001" "001" "001" "001" ...
##  $ tracta  : chr  "020100" "020200" "020300" "020400" ...
##  $ pnhwht12: chr  "85.31999969" "37.02000046" "79.77999878" "92.59999847" ...
##  $ pnhblk12: chr  "11.52999973" "56.27000046" "17.14999962" "1.450000048" ...
##  $ phisp12 : chr  "0" "2.519999981" "1.769999981" "2.630000114" ...
##  $ pntv12  : chr  "0.170000002" "0" "0" "0.829999983" ...
##  $ pasian12: chr  "0" "2.839999914" "1.080000043" "0" ...
##  $ phaw12  : chr  "0" "0" "0" "0" ...

head( d.2010.samp$p65hsp12 ) # missing values coded as -999

## [1] "-999"        "0"           "0"           "9.649999619" "0"          
## [6] "6.440000057"

sum( d.2010.samp$p65hsp12 == "-999" )

## [1] 3874

summary( as.numeric(d.2010.samp$p65hsp12) )

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## -999.00    0.00    1.76  -46.81    7.35  100.00

# first four columns are unique IDs - leave them as character vectors
d.2010.samp <- clean_d( d.2010.samp, start_column=5 )

str( d.2010.samp[1:10] )

## 'data.frame':    73056 obs. of  10 variables:
##  $ tractid : chr  "1001020100" "1001020200" "1001020300" "1001020400" ...
##  $ statea  : chr  "01" "01" "01" "01" ...
##  $ countya : chr  "001" "001" "001" "001" ...
##  $ tracta  : chr  "020100" "020200" "020300" "020400" ...
##  $ pnhwht12: num  85.3 37 79.8 92.6 75.3 ...
##  $ pnhblk12: num  11.53 56.27 17.15 1.45 18.1 ...
##  $ phisp12 : num  0 2.52 1.77 2.63 2.53 ...
##  $ pntv12  : num  0.17 0 0 0.83 0.18 ...
##  $ pasian12: num  0 2.84 1.08 0 2.41 ...
##  $ phaw12  : num  0 0 0 0 0 0 0 0 0 0 ...

summary( d.2010.samp$p65hsp12 ) %>% pander()

Tidy Up Dataframes

We want to standardize datasets across all of the years so that they are all clean, have the same structure, same variable name conventions, etc.

Test code:

The following is set to eval=FALSE because it’s not required for you to generate the final outputs.

Note: tidy_up_data() is able to read in the data because it is not importing files using static file paths.

file.name <- "ltdb_std_2010_sample.csv"
d.2010.s <- tidy_up_data( file.name )
head( d.2010.s[1:20] ) %>% pander()

file.name <- "LTDB_Std_2010_fullcount.csv"
d.2010.f <- tidy_up_data( file.name )
head( d.2010.f[1:20] ) %>% pander()

d2 <- bind_rows( d.2010.s, d.2010.f )


file.name <- "ltdb_std_2000_sample.csv"
d.2010.s <- tidy_up_data( file.name )
head( d.2010.s[1:20] ) %>% pander()

file.name <- "LTDB_Std_2000_fullcount.csv"
d.2010.f <- tidy_up_data( file.name )
head( d.2010.f[1:20] ) %>% pander()

d2 <- bind_rows( d.2010.s, d.2010.f )

Clean and tidy all data from the same year, then combine sample and full dataframes into a single table. Notice the use of here::here() down below can also be used when telling R where to save a file.

# for each relevant file, run the build_year() function
# note: this populates the data/rodeo/ directory with clean files
for (relevant_file in RELEVANT_FILES) {
  print(paste0("Starting on ", relevant_file[["year"]]))
  build_year(fn1 = relevant_file[["fullcount"]],
             fn2 = relevant_file[["sample"]],
             year = relevant_file[["year"]])
  if (relevant_file[["year"]] < 2010) {
    print("Finished! Moving onto the next decade.")
  } else {
    print("Finished! No more data to parse.")
  }
  
}

## [1] "Starting on 1970"
## [1] "Finished! Moving onto the next decade."
## [1] "Starting on 1980"
## [1] "Finished! Moving onto the next decade."
## [1] "Starting on 1990"
## [1] "Finished! Moving onto the next decade."
## [1] "Starting on 2000"
## [1] "Finished! Moving onto the next decade."
## [1] "Starting on 2010"
## [1] "Finished! No more data to parse."

Check a file:

Note: Notice the use of here::here() below when importing data.

# import the clean file 
d <- readRDS( here::here( "data/rodeo/LTDB-2000.rds" ) )
head( d ) %>% pander()

Metro Area Metadata

Metro areas are designated by the US Census as Core-Based Statistical Areas (CBSA).

“A core-based statistical area (CBSA) is a U.S. geographic area defined by the Office of Management and Budget (OMB) that consists of one or more counties (or equivalents) anchored by an urban center of at least 10,000 people plus adjacent counties that are socioeconomically tied to the urban center by commuting. Areas defined on the basis of these standards applied to Census 2000 data were announced by OMB in June 2003. These standards are used to replace the definitions of metropolitan areas that were defined in 1990. The OMB released new standards based on the 2010 Census.” cite

Note that these are defined as sets of counties, so the definition files are organized with one county per row, and attributes associated with the county.

Census data files do not always have info about metro areas. If we need this information for our analysis we can get a crosswalk file from the National Bureau of Economic Research:

https://data.nber.org/data/cbsa-msa-fips-ssa-county-crosswalk.html

Note: Notice the absence of here::here(). It is not necessary here because the file lives outside of our directory.

# load the crosswalk
# note: this stores a copy in the data/raw/ directory
cw <- obtain_crosswalk()

# view all metro areas in the country
sort( unique( cw$cbsaname ) ) %>% head() %>% pander()

__, Abilene, TX, Aguadilla-Isabela-San Sebastián, PR, Akron, OH, Albany-Schenectady-Troy, NY and Albany, GA

There are 3,292 counties in 2010. Of these, 35% are urban, 65% are rural.

# note in the data dictionary for CBSA Name (copied below): “blanks are rural”
table( cw$urban ) %>% pander()

It’s not technically not strictly raw data because we created a new variable and dropped some columns, but it’s input data we are grabbing from an external site as meta-data, and it will not be a final research dataset used for analysis, so we can put it into the raw folder.

# DATA DICTIONARY FOR CROSSWALK

  1.  cbsatocountycrosswalk2005 set up by Jean Roth , jroth@nber.org , 20 Dec 2016
  2.  Source: fr05_cbsa_msa_xwalk_pub.txt
  3.  NBER URL: http://www.nber.org/data/cbsa-msa-fips-ssa-county-crosswalk.html
  4.  Source Page: http://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/AcuteInpatientPPS/Acute-Inpatient-Files-for-Download-Items/CMS022637.html
  5.  Source File URL: http://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/AcuteInpatientPPS/Downloads/fr05_cbsa_msa_xwalk_pub.zip
  6.  by Jean Roth , jroth@nber.org , 28 Nov 2016

ssacounty:
  1.  Los Angeles FIPS 06037 can have two SSA county codes: 05210 and 05200

  obs:         3,293                          
 vars:            21                          20 Dec 2016 11:41
 size:       757,390                          (_dta has notes)
-----------------------------------------------------------------------------------------------------------
              storage   display    value
variable name   type    format     label      variable label
-----------------------------------------------------------------------------------------------------------
countyname      str26   %26s                  County Name
state           str2    %9s                   State
ssacounty       str5    %9s                 * SSA County Code
fipscounty      str5    %9s                   FIPS County Code
msa             str6    %9s                   Old MSA
l               str1    %9s                   Lugar
msaname         str48   %48s                  Old MSA Name
cbsa            str5    %9s                   CBSA - if blank then rural area (set equal to first 2 digits of ssa code)
cbsaname        str50   %50s                  CBSA Name
cbsaold         long    %12.0g                 (Blanks are Rural)
cbsanameold     str42   %42s                   (Blanks are Rural)
ssast           str2    %9s                   SSA State code
fipst           str2    %9s                   FIPS State code
y2005           float   %9.0g                 Present in 2005 source file
y2011           float   %9.0g                 Present in 2011 source file
y2012           float   %9.0g                 Present in 2012 source file
y2013           float   %9.0g                 Present in 2013 source file
y2014           float   %9.0g                 Present in 2014 source file
y2015           float   %9.0g                 Present in 2015 source file
y2016           float   %9.0g                 Present in 2016 source file
y2017           float   %9.0g                 Present in 2017 source file
                                            * indicated variables have notes
------------------------------------------------------------------------------------------------------------
Sorted by: fipscounty  ssacounty

Create Meta-Data Table

Each of the file contains redundant meta-data. We can remove it to make merges easier, and consolidate all of the meta-data (attributes of counties and census tracts) into a single file for ease of use.

We need one per year from 1980 to 2000 to grab all of the unique meta-data in the files.

# create the final meta data file
# note: this stores a copy within the data/rodeo/ directory
create_final_metadata_file(file_names = RELEVANT_FILES,
                           crosswalk = cw)

# view the results
md_complete = readr::read_rds(here::here("data/rodeo/LTDB-META-DATA.rds"))
pander::pander(head(md_complete))

Alternative Approach

Build one large stacked dataset:

Hard to use because you don’t know which panel years exist for each variable.

d.list <- NULL
loop.count <- 1

for( i in these )
{
  file.name <- i
  d.i <- tidy_up_data( file.name )
  d.list[[ loop.count ]] <- d.i
  loop.count <- loop.count + 1
}

d <- bind_rows( d.list )

Then you can reshape the dataset as needed:

dat <- dplyr::filter( dat, year %in% c(2000,2010) )

library(data.table)   # CRAN version 1.10.4
setDT(world)   # coerce to data.table
data_wide <- dcast(world, Country ~ Year, 
                   value.var = c("Growth", "Unemployment", "Population"))

reshape(world, direction = "wide", timevar = "Year", idvar = "Country")

d2 <- d[1:20]

reshape( d2, direction="wide", timevar="year", idvar="tractid" )

+---------+------+--------+--------------+------------+
| Country | Year | Growth | Unemployment | Population |
+---------+------+--------+--------------+------------+
| A       | 2015 |      2 |          8.3 |         40 |
| B       | 2015 |      3 |          9.2 |         32 |
| C       | 2015 |    2.5 |          9.1 |         30 |
| D       | 2015 |    1.5 |          6.1 |         27 |
| A       | 2016 |      4 |          8.1 |         42 |
| B       | 2016 |    3.5 |            9 |       32.5 |
| C       | 2016 |    3.7 |            9 |         31 |
| D       | 2016 |    3.1 |          5.3 |         29 |
| A       | 2017 |    4.5 |          8.1 |       42.5 |
| B       | 2017 |    4.4 |          8.4 |         33 |
| C       | 2017 |    4.3 |          8.5 |         30 |
| D       | 2017 |    4.2 |          5.2 |         30 |
+---------+------+--------+--------------+------------+
  
  +---------+-------------+-------------------+-----------------+-------------+-------------------+-----------------+
| Country | Growth_2015 | Unemployment_2015 | Population_2015 | Growth_2016 | Unemployment_2016 | Population_2016 |
+---------+-------------+-------------------+-----------------+-------------+-------------------+-----------------+
| A       |           2 |               8.3 |              40 |           4 |               8.1 |              42 |
| B       |           3 |               9.2 |              32 |         3.5 |                 9 |            32.5 |
| C       |         2.5 |               9.1 |              30 |         3.7 |                 9 |              31 |
| D       |         1.5 |               6.1 |              27 |         3.1 |               5.3 |              29 |
+---------+-------------+-------------------+-----------------+-------------+-------------------+-----------------+