Labels and Missing Values

Overview

Survey data from SPSS, Stata, and SAS stores two kinds of labels:

Variable labels describe what a variable measures (e.g., gender → “Respondent’s gender”)
Value labels map numeric codes to text (e.g., 1 → “Male”, 2 → “Female”)

In R, these labels are stored as attributes on haven_labelled columns. mariposa provides 10 functions for inspecting, modifying, and converting labelled data — plus tools for declaring missing values and searching variables.

Function	Purpose
`var_label()`	Get or set variable labels
`val_labels()`	Get or set value labels
`find_var()`	Search variables by name or label pattern
`to_label()`	Convert labelled → factor
`to_character()`	Convert labelled → character
`to_numeric()`	Convert factor/labelled → numeric
`to_labelled()`	Convert factor/character → labelled
`set_na()`	Declare values as missing (tagged NAs)
`unlabel()`	Strip all label metadata
`copy_labels()`	Restore labels after dplyr operations
`drop_labels()`	Remove unused value labels

Inspecting Labels

Variable Labels

Variable labels describe what each column contains. Use var_label() to retrieve them:

# Get labels for specific variables
var_label(survey_data, gender, education, life_satisfaction)
#>                                            gender 
#>                                          "Gender" 
#>                                         education 
#>                  "Highest educational attainment" 
#>                                 life_satisfaction 
#> "Life satisfaction (1=dissatisfied, 5=satisfied)"

# Get labels for all variables
var_label(survey_data)
#>                                                id 
#>                                                NA 
#>                                               age 
#>                                    "Age in years" 
#>                                            gender 
#>                                          "Gender" 
#>                                            region 
#>                              "Region (East/West)" 
#>                                         education 
#>                  "Highest educational attainment" 
#>                                            income 
#>                  "Monthly household income (EUR)" 
#>                                        employment 
#>                               "Employment status" 
#>                             political_orientation 
#>         "Political orientation (1=left, 5=right)" 
#>                             environmental_concern 
#>           "Environmental concern (1=low, 5=high)" 
#>                                 life_satisfaction 
#> "Life satisfaction (1=dissatisfied, 5=satisfied)" 
#>                                  trust_government 
#>        "Trust in government (1=none, 5=complete)" 
#>                                       trust_media 
#>             "Trust in media (1=none, 5=complete)" 
#>                                     trust_science 
#>           "Trust in science (1=none, 5=complete)" 
#>                                   sampling_weight 
#>                                "Weighting factor" 
#>                                           stratum 
#>                         "Stratification variable" 
#>                                    interview_mode 
#>                                  "Interview mode"

Value Labels

Value labels map numeric codes to meaningful text. Use val_labels() to retrieve them:

# Get value labels for a single variable
val_labels(survey_data, gender)
#> NULL

# Get value labels for multiple variables
val_labels(survey_data, education, employment)
#> $education
#> NULL
#> 
#> $employment
#> NULL

Finding Variables

SPSS datasets often have cryptic variable names like q104a_1 or v23. Use find_var() to search by name or label:

# Search in both names and labels (default)
find_var(survey_data, "trust")
#>   col             name                                    label
#> 1  11 trust_government Trust in government (1=none, 5=complete)
#> 2  12      trust_media      Trust in media (1=none, 5=complete)
#> 3  13    trust_science    Trust in science (1=none, 5=complete)

# Search only in variable labels
find_var(survey_data, "satisfaction", search = "label")
#>   col              name                                           label
#> 1  10 life_satisfaction Life satisfaction (1=dissatisfied, 5=satisfied)

# Search only in variable names
find_var(survey_data, "age|income", search = "name")
#>   col   name                          label
#> 1   2    age                   Age in years
#> 2   6 income Monthly household income (EUR)

Setting Labels

Setting Variable Labels

# Set labels for specific variables
labeled_data <- var_label(survey_data,
  age = "Age of respondent in years",
  income = "Monthly net income in euros"
)

# Verify
var_label(labeled_data, age, income)
#>                           age                        income 
#>  "Age of respondent in years" "Monthly net income in euros"

Setting Value Labels

# Set value labels
labeled_data <- val_labels(survey_data,
  gender = c("Male" = 1, "Female" = 2)
)

# Verify
val_labels(labeled_data, gender)
#>   Male Female 
#>      1      2

# Add labels without replacing existing ones
labeled_data <- val_labels(survey_data,
  gender = c("Diverse" = 3),
  .add = TRUE
)

val_labels(labeled_data, gender)
#> Diverse 
#>       3

Converting Between Formats

Survey data often needs conversion between labelled, factor, character, and numeric formats depending on the analysis.

Labelled → Factor

Use to_label() when you need factors for plotting or statistical models:

# Convert gender from labelled to factor
factor_data <- to_label(survey_data, gender, education)

# Check the result
class(factor_data$gender)
#> [1] "factor"
levels(factor_data$gender)
#> [1] "Male"   "Female"
head(factor_data$gender)
#> [1] Female Male   Male   Female Male   Female
#> Levels: Male Female

# Create ordered factors (useful for ordinal variables)
ordered_data <- to_label(survey_data, education, ordered = TRUE)
levels(ordered_data$education)
#> [1] "Basic Secondary"        "Intermediate Secondary" "Academic Secondary"    
#> [4] "University"

Labelled → Character

Use to_character() for string-based operations:

char_data <- to_character(survey_data, gender, region)
head(char_data$gender)
#> [1] "Female" "Male"   "Male"   "Female" "Male"   "Female"

Factor → Numeric

Use to_numeric() to convert factors or labelled vectors back to numbers:

# First convert to factor, then back to numeric
factor_data <- to_label(survey_data, education)

# Parse numeric values from factor levels
numeric_data <- to_numeric(factor_data, education)
head(numeric_data$education)
#> [1] 2 3 3 1 1 2

Numeric/Factor → Labelled

Use to_labelled() to add labels to plain numeric or factor columns:

# Convert a factor back to haven_labelled
plain_data <- data.frame(
  gender = factor(c("Male", "Female", "Male")),
  score = c(3, 4, 2)
)

labelled_data <- to_labelled(plain_data, gender)
class(labelled_data$gender)
#> [1] "haven_labelled" "vctrs_vctr"     "double"

Declaring Missing Values

Setting Values as Missing

Use set_na() to declare specific numeric codes as missing values:

# After importing SPSS data where -9 = refused, -8 = don't know
data <- read_spss("survey.sav", tag_na = FALSE)

# Declare -9 and -8 as missing across all numeric columns
data <- set_na(data, -9, -8, tag = TRUE)

# Declare different missing codes for specific variables
data <- set_na(data, q1 = c(-9, -8), q2 = c(99))

When tag = TRUE (the default), each missing code becomes a distinct tagged NA, so you can later distinguish “refused” from “don’t know” responses.

Removing All Labels

Use unlabel() when you need plain numeric data without any label metadata:

# Strip all labels from entire dataset
plain_data <- unlabel(survey_data)

# Strip labels from specific variables only
plain_data <- unlabel(survey_data, gender, education)

This converts haven_labelled columns to plain numeric or character, tagged NAs to regular NA, and removes all label attributes.

Preserving Labels Through Pipelines

The Problem

dplyr operations like filter(), mutate(), and select() can strip label attributes from columns:

filtered <- survey_data %>%
  filter(age >= 30) %>%
  select(gender, education, income)

# Labels may be lost after certain operations

The Solution: copy_labels()

Use copy_labels() to restore labels from the original data:

# Restore labels from the source dataset
filtered <- copy_labels(filtered, survey_data)

# Verify labels are back
var_label(filtered, gender, education)
#>                           gender                        education 
#>                         "Gender" "Highest educational attainment"

Cleaning Unused Labels

After filtering, some value label categories may no longer appear in the data. Use drop_labels() to clean them up:

# Subset to one region only
subset_data <- survey_data %>% filter(region == 1)

# Remove value labels for regions that are no longer in the data
clean_data <- drop_labels(subset_data, region)
val_labels(clean_data, region)
#> NULL

Complete Example

A typical workflow for preparing SPSS data for analysis:

# 1. Import SPSS file
data <- read_spss("survey_2024.sav")

# 2. Explore what's in the data
codebook(data)
find_var(data, "satisf")
find_var(data, "trust")

# 3. Check labels
var_label(data, q1, q2, q3)
val_labels(data, q1)

# 4. Rename variables for clarity
data <- data %>%
  rename(life_satisfaction = q1, income = q2, education = q3)

# Update variable labels
data <- var_label(data,
  life_satisfaction = "Overall life satisfaction (1-5)",
  income = "Monthly net income in euros",
  education = "Highest education level"
)

# 5. Convert categorical variables for analysis
data <- to_label(data, education, gender)

# 6. Analyze
data %>%
  describe(life_satisfaction, income, weights = sampling_weight)

data %>%
  t_test(life_satisfaction, group = gender, weights = sampling_weight)

Practical Tips

Keep labels as long as possible. Labels carry important context. Only convert to factor or character when a specific function requires it.
Use find_var() instead of names(). It searches both names and labels, which is essential for SPSS datasets with non-descriptive variable names.
Use copy_labels() after complex pipelines. If your pipeline involves joins, reshaping, or other operations that might strip attributes, restore labels from the original data.
Prefer to_label() over as.factor(). to_label() uses the value labels as factor levels, giving you meaningful names instead of numeric codes.
Use set_na() early in your workflow. Declaring missing values immediately after import ensures they are handled correctly in all downstream analyses.

Summary

var_label() and val_labels() get and set variable and value labels
find_var() searches variables by name or label pattern
to_label(), to_character(), to_numeric(), to_labelled() convert between formats
set_na() declares values as missing; unlabel() strips all metadata
copy_labels() restores labels after dplyr operations; drop_labels() cleans up unused labels