Creating data overview reports

By now, you might expect dataMaid to be a virtual robot ready for doing all your dirty work. You might even worry about the early coming of the robot overtake, which might then again make you slightly iffy about this whole dataMaid idea. But fear not. We have not (yet) succeeded in creating artificial intelligence for data cleaning. Therefore, human decision making is still very much needed. What dataMaid does do is to make the basis of these decisions well-documented, easy to read and easy to share – even with colleagues that are not fluent in R.

The main capability of dataMaid is autogenerating data overview documents that summarize each variable in a dataset and flags typical problems, depending on the data type of the variable. Let us consider an example dataset that is part of the dataMaid package

suppressPackageStartupMessages(library("dataMaid")) # Load package
data(toyData)                                       # Get data
toyData                                             # Show data
## # A tibble: 15 x 6
##    pill  events region change id    spotifysong
##    <fct>  <dbl> <fct>   <dbl> <fct> <fct>      
##  1 red        1 a      -0.626 1     Irrelevant 
##  2 red        1 a       0.184 2     Irrelevant 
##  3 red        1 a      -0.836 3     Irrelevant 
##  4 red        2 a       1.60  4     Irrelevant 
##  5 red        2 a       0.330 5     Irrelevant 
##  6 red        6 b      -0.820 6     Irrelevant 
##  7 red        6 b       0.487 7     Irrelevant 
##  8 red        6 b       0.738 8     Irrelevant 
##  9 red      999 c       0.576 9     Irrelevant 
## 10 red       NA c      -0.305 10    Irrelevant 
## 11 blue       4 c       1.51  11    Irrelevant 
## 12 blue      82 .       0.390 12    Irrelevant 
## 13 blue      NA " "    -0.621 13    Irrelevant 
## 14 <NA>     NaN other  -2.21  14    Irrelevant 
## 15 <NA>       5 OTHER   1.12  15    Irrelevant

The toyData data frame contains 6 variables and if we want assist the investigators in indentifying possible errors in the dataset then we call the makeDataReport() function, which is the main function of the dataMaid package:

makeDataReport(toyData)  # Produce data report

This produces and opens a pdf report in which an overview of the dataset is given, and each variable is presented in turn and submitted to various checks for typical problems. The overview is a simple summary and the results from the data cleaning process and looks like this

The main part of the document provides the reader information about each variable and it looks like this

The report also includes an overview of what checks are performed for each variable and the default options are presented here:

The output report is intended as a guide for identifying potential errors and as a blueprint for discussion between the data analyst and the field experts, and since the report is readable by data analysts and field experts alike, it contributes to an inter-field dialogue about the data at hand. Oftentimes, e.g., distinguishing between faulty codings of a numeric value and unusual, but correct, values requires problem-specific expertise that might not be held by the data analyst. Hopefully, having easy access to data descriptions through dataMaid will help this sharing of knowledge.

Using dataMaid interactively

We have found the data overview report to be a very convenient tool for documenting the data cleaning process and for discussing problems in the data with collaborators. But sometimes it is also nice to be able to work interactively in the R-console. The interactive functionality of dataMaid is centered around three steps of data assesment:

• Summarize: What is in the variable?
• Visualize: What does the distribution look like?
• Check: What potential problems do we find in the variable?

Each of these three steps have a dedicated function, named summarize(), visualize() and check(), respectively, that can be called on datasets as well as singular variables:

summarize(toyData[, c("pill", "events")]) #summarize a subset of a dataset
## $pill
## $pill$variableType
## Variable type: factor
## $pill$countMissing
## Number of missing obs.: 2 (13.33 %)
## $pill$uniqueValues
## Number of unique values: 2
## $pill$centralValue
## Mode: "red"
## 
## $events
## $events$variableType
## Variable type: numeric
## $events$countMissing
## Number of missing obs.: 3 (20 %)
## $events$uniqueValues
## Number of unique values: 8
## $events$centralValue
## Median: 4.5
## $events$quartiles
## 1st and 3rd quartiles: 1.75; 6
## $events$minMax
## Min. and max.: 1; 999

summarize() shows the output that is put into each variable overview, and contains information about features that depend on the variable type. It is possible to modify the summarize() function and/or to provide you own custom information to summarize() (for more details on writing your own custom summarize() function see the documentation at the GitHub page).

check() runs through the variable-type-dependent checks

check(toyData$events) #check a single variable
## $identifyMissing
## The following suspected missing value codes enter as regular values: 999, NaN.
## $identifyOutliers
## Note that the following possible outlier values were detected: 82, 999.

For a numeric variable, two default checks are run: are there any elements that could be potential miscodings of NA (in this case 999 and NaN are flagged as possible errors), and possible outliers are identified. 82 and 999 are sufficiently far away from the other observations to be flagged as potential outliers, so the check draws the attention to those values.

Visualizations are done with the visualize() function and they provide a graphical overview of the distribution of the data. Factors are shown as bar plots while numeric variables are shown as histograms.

visualize(toyData) #visualize a dataset

Figur 1: The result of calling visualize() on the full toyData dataset

One can also apply a single, specific summary- or check-function to a variable. All available summary-, visual- or check-functions can be seen by calling allXXXFunctions() where XXX is either summary, visual or check. For instance, we find all available checks by typing:

allCheckFunctions()
----------------------------------------------------------------------------------------
name                      description                      classes                      
------------------------- -------------------------------- -----------------------------
identifyCaseIssues        Identify case issues             character, factor            

identifyLoners            Identify levels with < 6 obs.    character, factor            

identifyMissing           Identify miscoded missing        character, factor, integer,  
                          values                           labelled, logical, numeric   

identifyNums              Identify misclassified numeric   character, factor, labelled  
                          or integer variables                                          

identifyOutliers          Identify outliers                Date, integer, numeric       

identifyOutliersTBStyle   Identify outliers (Turkish       Date, integer, numeric       
                          Boxplot style)                                                

identifyWhitespace        Identify prefixed and suffixed   character, factor, labelled  
                          whitespace                                                    

isCPR                     Identify Danish CPR numbers      character, Date, factor,     
                                                           integer, labelled, logical,  
                                                           numeric                      

isEmpty                   Check if the variable contains   character, Date, factor,     
                          only a single value              integer, labelled, logical,  
                                                           numeric                      

isKey                     Check if the variable is a key   character, Date, factor,     
                                                           integer, labelled, logical,  
                                                           numeric                      

isSupported               Check if the variable class is   character, Date, factor,     
                          supported by dataMaid.           integer, labelled, logical,  
                                                           numeric                      
----------------------------------------------------------------------------------------

Then, we can identify values in a variable that look like miscoded missing values by calling

identifyMissing(toyData$events)
## The following suspected missing value codes enter as regular values: 999, NaN.

Note that the problematic values found by this function can be extracted, as all check functions return checkResult objects, that are really just named lists:

missVar2 <- identifyMissing(toyData$events)
str(missVar2)
## List of 3
##  $ problem      : logi TRUE
##  $ message      : chr "The following suspected missing value codes enter as regular values: \\\"999\\\", \\\"NaN\\\"."
##  $ problemValues: num [1:2] 999 NaN
##  - attr(*, "class")= chr "checkResult"

This implies that the miscoded missing value problem can be solved by writing e.g.

toyData$events[toyData$events %in% missVar2$problemValues] <- NA

thereby replacing the 999 and NaN values in the var2 variable with NA values.