Introduction to crosswalkr

Benjamin Skinner

2019-12-30

Researchers often must compile master data sets from a number of smaller data sets that are not consistent in terms of variable names or value encodings. This can be especially true for large administrative data sets that span multiple years and/or departments. Other times, teams of researchers must work together to maintain a master data set and it is important for replicability and future collaboration that the team rely on consistent naming and encoding conventions.

For example, let’s say there are three flat files of student information that need to be merged into a single large data set for analysis.

File 1

sid lname state t_score
1 Jackson VA 74
2 Harrison KY 86
3 Nixon IL 78

File 2

stu_id last_name st test_score
4 Washington 35 92
5 Roosevelt 11 67
6 Taylor 47 68

File 3

s_id name sta score
7 Tyler North Dakota 91
8 Grant South Dakota 82
9 Adams Illinois 89

It is clear that these files contain the same basic information, but neither the names nor encodings for state | st | sta are consistent.

One solution is to just fix these one at a time before joining them. For example:

library(crosswalkr)
library(dplyr)
library(labelled)
library(haven)
df1 <- file_1 %>%
    rename(id = sid,
           last_name = lname,
           stabbr = stat,
           score = t_score)

df2 <- file_2 %>%
    rename(id = stu_id,
           stabbr = st,
           score = test_score) %>%
    mutate(stabbr = as.character(stabbr))

df3 <- file_3 %>%
    rename(id = s_id,
           stabbr = sta,
           last_name = name)

df <- rbind(df1, df2, df3)
df
##   id  last_name       stabbr score
## 1  1    Jackson           VA    74
## 2  2   Harrison           KY    86
## 3  3      Nixon           IL    78
## 4  4 Washington           35    92
## 5  5  Roosevelt           11    82
## 6  6     Taylor           47    89
## 7  7      Tyler North Dakota    91
## 8  8      Grant South Dakota    82
## 9  9      Adams     Illinois    89

The problem, of course, is there is a lot of room for error since the renaming process has to be repeated for each data frame.

Using a crosswalk file

Instead, it makes more sense to create a crosswalk data set that aligns old (or raw) column names with new (or clean) column names and, if desired, labels. The crosswalk to join these files could be:

clean label file_1_raw file_2_raw file_3_raw
id Student ID sid stu_id s_id
last_name Student last name lname last_name name
stabbr State abbreviation stat st sta
score Test score t_score test_score score

The crosswalk file (cw_file) could be:

  1. Data frame object already in memory
  2. A string with path and name (e.g., './path/to/crosswalk.csv') of a flat file of one of the following types:
    1. Comma separated (*.csv)
    2. Tab separated (*.tsv)
    3. Other delimited (*.txt) with delimiter option set to delimiter string (e.g., delimiter = '|')
    4. Excel (*.xls or *.xlsx) with sheet option set to sheet number or string name (defaulting to the first sheet)
    5. R data (*.rdata, *.rda, *.rds)
    6. Stata data (*.dta)

If given a string to the cw_file argument, renamefrom() and encodefrom() determine the type of file by its ending.

Renaming

To rename using the renamefrom() command:

df1 <- renamefrom(file_1, cw_file = crosswalk, raw = file_1_raw, clean = clean, label = label)
df2 <- renamefrom(file_2, cw_file = crosswalk, raw = file_2_raw, clean = clean, label = label)
df3 <- renamefrom(file_3, cw_file = crosswalk, raw = file_3_raw, clean = clean, label = label)

df <- rbind(df1, df2, df3)
df
##   id  last_name       stabbr score
## 1  1    Jackson           VA    74
## 2  2   Harrison           KY    86
## 3  3      Nixon           IL    78
## 4  4 Washington           35    92
## 5  5  Roosevelt           11    82
## 6  6     Taylor           47    89
## 7  7      Tyler North Dakota    91
## 8  8      Grant South Dakota    82
## 9  9      Adams     Illinois    89

And check out the labels:

var_label(df)
## $id
## [1] "Student ID"
## 
## $last_name
## [1] "Student last name"
## 
## $stabbr
## [1] "State abbreviation"
## 
## $score
## [1] "Test score"

As new raw data files are added to the project, they could simply be given a new column in the crosswalk file that mapped their raw column names to the clean versions.

Encoding

These same example files have inconsistent encodings for state: one uses two-letter abbreviations, another the FIPS code, and another the full name. Again, instead of fixing each one at a time, a separate crosswalk for encoding these values could be used. The crosswalkr package includes a state-level crosswalk, stcrosswalk:

data(stcrosswalk)
stcrosswalk
## # A tibble: 51 x 7
##    stfips stabbr stname               cenreg cenregnm  cendiv cendivnm          
##     <int> <chr>  <chr>                 <int> <chr>      <int> <chr>             
##  1      1 AL     Alabama                   3 South          6 East South Central
##  2      2 AK     Alaska                    4 West           9 Pacific           
##  3      4 AZ     Arizona                   4 West           8 Mountain          
##  4      5 AR     Arkansas                  3 South          7 West South Central
##  5      6 CA     California                4 West           9 Pacific           
##  6      8 CO     Colorado                  4 West           8 Mountain          
##  7      9 CT     Connecticut               1 Northeast      1 New England       
##  8     10 DE     Delaware                  3 South          5 South Atlantic    
##  9     11 DC     District of Columbia      3 South          5 South Atlantic    
## 10     12 FL     Florida                   3 South          5 South Atlantic    
## # … with 41 more rows

The encodefrom() function works much like renamefrom(). The only difference is that a vector of encoded values is returned that can be added to an existing dataframe.

encodefrom() returns either base R factors or labels depending on whether the input data frame is a tibble.

factor

df1$state <- encodefrom(file_1, var = stat, stcrosswalk, raw = stabbr, clean = stfips, label = stname)
df1
##   id last_name stabbr score    state
## 1  1   Jackson     VA    74 Virginia
## 2  2  Harrison     KY    86 Kentucky
## 3  3     Nixon     IL    78 Illinois
sapply(df1, class)
##          id   last_name      stabbr       score       state 
##   "integer" "character" "character"   "numeric"    "factor"

labelled vector

file_1_ <- file_1 %>% tbl_df()
df1$state <- encodefrom(file_1_, var = stat, stcrosswalk, raw = stabbr,
                        clean = stfips, label = stname)
as_factor(df1)
##   id last_name stabbr score    state
## 1  1   Jackson     VA    74 Virginia
## 2  2  Harrison     KY    86 Kentucky
## 3  3     Nixon     IL    78 Illinois
zap_labels(df1)
##   id last_name stabbr score state
## 1  1   Jackson     VA    74    51
## 2  2  Harrison     KY    86    21
## 3  3     Nixon     IL    78    17

Combined example: dplyr chain

The renamefrom() and encodefrom() functions can be combined in a dplyr chain.

df <- rbind(file_1 %>%
            tbl_df() %>%
            renamefrom(., crosswalk, file_1_raw, clean, label) %>%
            mutate(stabbr = encodefrom(., stabbr, stcrosswalk, stabbr, stfips, stname)),

            ## append file 2
            file_2 %>%
            tbl_df() %>%
            renamefrom(., crosswalk, file_2_raw, clean, label) %>%
            mutate(stabbr = encodefrom(., stabbr, stcrosswalk, stfips, stfips, stname)),

            ## append file 3
            file_3 %>%
            tbl_df() %>%
            renamefrom(., crosswalk, file_3_raw, clean, label) %>%
            mutate(stabbr = encodefrom(., stabbr, stcrosswalk, stname, stfips, stname)))

df
## # A tibble: 9 x 4
##      id last_name                     stabbr score
##   <int> <chr>                      <int+lbl> <dbl>
## 1     1 Jackson    51 [Virginia]                74
## 2     2 Harrison   21 [Kentucky]                86
## 3     3 Nixon      17 [Illinois]                78
## 4     4 Washington 35 [New Mexico]              92
## 5     5 Roosevelt  11 [District of Columbia]    82
## 6     6 Taylor     47 [Tennessee]               89
## 7     7 Tyler      38 [North Dakota]            91
## 8     8 Grant      46 [South Dakota]            82
## 9     9 Adams      17 [Illinois]                89
as_factor(df)            
## # A tibble: 9 x 4
##      id last_name  stabbr               score
##   <int> <chr>      <fct>                <dbl>
## 1     1 Jackson    Virginia                74
## 2     2 Harrison   Kentucky                86
## 3     3 Nixon      Illinois                78
## 4     4 Washington New Mexico              92
## 5     5 Roosevelt  District of Columbia    82
## 6     6 Taylor     Tennessee               89
## 7     7 Tyler      North Dakota            91
## 8     8 Grant      South Dakota            82
## 9     9 Adams      Illinois                89