validateR

Introduction

Real-world data is rarely clean. A column might have missing values that are easy to overlook. A few extreme (outlier) values might sneak into an otherwise normal numeric column. Rows sometimes get duplicated by accident during data collection or when merging files. A column that should be numeric can end up stored as text, just because a few entries contain things like "N/A" or "unknown". And a column of dates can have a few entries that don’t match the format the rest of the column uses.

validateR is a small set of tools that helps you catch these problems before they affect your analysis. It checks a data frame for five common issues:

• Missing values — how many values are missing in each column, and what percentage that represents
• Outliers — unusually extreme numeric values, detected using the IQR method
• Duplicate rows — rows that are exact copies of each other, grouped so you can see which rows match
• Numeric-like inconsistencies — text columns that look like they were meant to hold numbers, but have a few values that don’t convert cleanly
• Date-like inconsistencies (optional) — text columns that look like they were meant to hold dates in a format you specify, but have a few values that don’t parse cleanly

This vignette walks through a complete example: starting with a messy data frame, running a check, and reading the results.

library(validateR)

A messy example

Let’s create a small data frame with a few realistic problems: a missing value, an outlier, a duplicated row, and a text column that should probably be numeric.

messy_data <- data.frame(
  age = c(25, 12, NA, 17, 19), 
  weight = c("45", "34", "25", "N/A", "32"),
  price = c(100, 120, 140, 1500, 125)
)
messy_data <- rbind(messy_data, messy_data[3, ])
messy_data
#>    age weight price
#> 1   25     45   100
#> 2   12     34   120
#> 3   NA     25   140
#> 4   17    N/A  1500
#> 5   19     32   125
#> 31  NA     25   140

Running validate_df()

The main entry point is validate_df(). It runs four checks automatically, and returns a single report object. A fifth check, for date-like columns, is available too, but only runs if you ask for it — more on that later in this vignette.

report <- validate_df(messy_data)
report
#> < validation_report for 'messy_data' >
#> 6 rows, 3 columns
#> Checked: 2026-08-31 19:26:23 
#> 
#> Missing values:          1  column(s) affected
#> Outliers:                1  column(s) affected
#> Duplicates:              1  group(s) affected
#> Numeric-like issues:     1  column(s) affected
#> Date-like issues:        NOT checked 
#> 
#> For details, inspect report$results$missing, $outliers, $duplicates, $numeric_like, $date_like

Looking at the details

The printed summary tells you how many columns were affected by each check, but not the specifics. The full results are stored inside the report object, under report$results, and are worth exploring directly when you need to act on what was found.

report$results$missing
#>   column_name n_missing pct_missing
#> 1         age         2       33.33
#> 2      weight         0        0.00
#> 3       price         0        0.00

report$results$outliers
#>   column_name n_outliers pct_outliers
#> 1         age          0         0.00
#> 2       price          1        16.67

report$results$duplicates
#> [[1]]
#> [1] 3 6

report$results$numeric_like
#> $weight
#> $weight$expected_type
#> [1] "numeric"
#> 
#> $weight$proportion_convertible
#> [1] 83.33
#> 
#> $weight$bad_rows
#> [1] 4

Visualizing the results

For a quick visual overview, plot() shows two bar charts: missing value counts and outlier counts, one column per bar.

plot(report)

Running checks individually

Sometimes you don’t need the full report — maybe you already know your data is fine except for one specific concern. Each check function used by validate_df() is also exported and can be used on its own.

check_outliers(messy_data)
#>   column_name n_outliers pct_outliers
#> 1         age          0         0.00
#> 2       price          1        16.67

This runs just the outlier check, without computing the others. The same applies to check_missing(), check_duplicates(), check_numeric_like(), and check_date_like().

How the thresholds work

Some of the checks make a judgment call about what counts as a problem. It’s worth understanding how each one decides, so you know when to trust the default behavior and when you might want to inspect the results more closely yourself.