Consumer Expenditure Survey (CES)

A household budget survey designed to guide major economic indicators like the Consumer Price Index.

• One table of survey responses per quarter with one row per sampled household (consumer unit). Additional tables containing one record per expenditure.

• A complex sample survey designed to generalize to the civilian non-institutional U.S. population.

• Released annually since 1996.

• Administered by the Bureau of Labor Statistics.

---

Recommended Reading

Four Example Strengths & Limitations:

✔️ Detailed expenditure categories

✔️ Respondents diary spending for two consecutive 1-week periods

❌ Measures purchases but not consumption

❌ Consumer unit definition differs from households or families in other surveys

Three Example Findings:

1. In 2022, one third of total nationwide expenditures were attributed to housing-related expenses.

2. Between 2015 and early 2022, male household heads consumed a greater proportion of resources (33%) compared to female household heads (28%), who, in turn, consume more than children (23%).

3. In 2020, if income increased by $100, spending on all food and alcohol increased by $14 on average.

Two Methodology Documents:

Consumer Expenditure Surveys Public Use Microdata Getting Started Guide

Wikipedia Entry

One Haiku:

# price indices and
# you spent how much on beans, jack?
# pocketbook issues

---

Download, Import, Preparation

Download both the prior and current year of interview microdata:

library(httr)

tf_prior_year <- tempfile()

this_url_prior_year <- "https://www.bls.gov/cex/pumd/data/stata/intrvw22.zip"

dl_prior_year <- GET( this_url_prior_year , user_agent( "email@address.com" ) )

writeBin( content( dl_prior_year ) , tf_prior_year )

unzipped_files_prior_year <- unzip( tf_prior_year , exdir = tempdir() )

tf_current_year <- tempfile()

this_url_current_year <- "https://www.bls.gov/cex/pumd/data/stata/intrvw23.zip"

dl_current_year <- GET( this_url_current_year , user_agent( "email@address.com" ) )

writeBin( content( dl_current_year ) , tf_current_year )

unzipped_files_current_year <- unzip( tf_current_year , exdir = tempdir() )

unzipped_files <- c( unzipped_files_current_year , unzipped_files_prior_year )

Import and stack all 2023 quarterly files plus 2024’s first quarter:

library(haven)

fmli_files <- grep( "fmli2[3-4]" , unzipped_files , value = TRUE )

fmli_tbls <- lapply( fmli_files , read_dta )

fmli_dfs <- lapply( fmli_tbls , data.frame )

fmli_dfs <- 
    lapply( 
        fmli_dfs , 
        function( w ){ names( w ) <- tolower( names( w ) ) ; w }
    )

fmli_cols <- lapply( fmli_dfs , names )

intersecting_cols <- Reduce( intersect , fmli_cols )

fmli_dfs <- lapply( fmli_dfs , function( w ) w[ intersecting_cols ] )

ces_df <- do.call( rbind , fmli_dfs )

Scale the weight columns based on the number of months in 2023:

ces_df[ , c( 'qintrvyr' , 'qintrvmo' ) ] <-
    sapply( ces_df[ , c( 'qintrvyr' , 'qintrvmo' ) ] , as.numeric )

weight_columns <- grep( 'wt' , names( ces_df ) , value = TRUE )

ces_df <-
    transform(
        ces_df ,
        mo_scope =
            ifelse( qintrvyr %in% 2023 & qintrvmo %in% 1:3 , qintrvmo - 1 ,
            ifelse( qintrvyr %in% 2024 , 4 - qintrvmo , 3 ) )
    )

for ( this_column in weight_columns ){
    ces_df[ is.na( ces_df[ , this_column ] ) , this_column ] <- 0
    
    ces_df[ , paste0( 'popwt_' , this_column ) ] <-
        ( ces_df[ , this_column ] * ces_df[ , 'mo_scope' ] / 12 )   
    
}

Combine previous quarter and current quarter variables into a single variable:

expenditure_variables <- 
    gsub( "pq$" , "" , grep( "pq$" , names( ces_df ) , value = TRUE ) )

# confirm that for every variable ending in pq,
# there's the same variable ending in cq
stopifnot( all( paste0( expenditure_variables , 'cq' ) %in% names( ces_df ) ) )

# confirm none of the variables without the pq or cq suffix exist
if( any( expenditure_variables %in% names( ces_df ) ) ) stop( "variable conflict" )

for( this_column in expenditure_variables ){

    ces_df[ , this_column ] <-
        rowSums( ces_df[ , paste0( this_column , c( 'pq' , 'cq' ) ) ] , na.rm = TRUE )
    
    # annualize the quarterly spending
    ces_df[ , this_column ] <- 4 * ces_df[ , this_column ]
    
    ces_df[ is.na( ces_df[ , this_column ] ) , this_column ] <- 0

}

Append any interview survey UCC found at https://www.bls.gov/cex/ce_source_integrate.xlsx:

ucc_exp <- c( "450110" , "450210" )

mtbi_files <- grep( "mtbi2[3-4]" , unzipped_files , value = TRUE )

mtbi_tbls <- lapply( mtbi_files , read_dta )

mtbi_dfs <- lapply( mtbi_tbls , data.frame )

mtbi_dfs <- 
    lapply( 
        mtbi_dfs , 
        function( w ){ names( w ) <- tolower( names( w ) ) ; w }
    )

mtbi_dfs <- lapply( mtbi_dfs , function( w ) w[ c( 'newid' , 'cost' , 'ucc' , 'ref_yr' ) ] )

mtbi_df <- do.call( rbind , mtbi_dfs )

mtbi_df <- subset( mtbi_df , ( ref_yr %in% 2023 ) & ( ucc %in% ucc_exp ) )

mtbi_agg <- aggregate( cost ~ newid , data = mtbi_df , sum )

names( mtbi_agg ) <- c( 'newid' , 'new_car_truck_exp' )

before_nrow <- nrow( ces_df )

ces_df <-
    merge(
        ces_df ,
        mtbi_agg ,
        all.x = TRUE
    )

stopifnot( nrow( ces_df ) == before_nrow )

ces_df[ is.na( ces_df[ , 'new_car_truck_exp' ] ) , 'new_car_truck_exp' ] <- 0

Save Locally  

Save the object at any point:

# ces_fn <- file.path( path.expand( "~" ) , "CES" , "this_file.rds" )
# saveRDS( ces_df , file = ces_fn , compress = FALSE )

Load the same object:

# ces_df <- readRDS( ces_fn )

Survey Design Definition

Construct a multiply-imputed, complex sample survey design:

Separate the ces_df data.frame into five implicates, each differing from the others only in the multiply-imputed variables:

library(survey)
library(mitools)

# create a vector containing all of the multiply-imputed variables
# (leaving the numbers off the end)
mi_vars <- gsub( "5$" , "" , grep( "[a-z]5$" , names( ces_df ) , value = TRUE ) )

# loop through each of the five variables..
for ( i in 1:5 ){

    # copy the 'ces_df' table over to a new temporary data frame 'x'
    x <- ces_df

    # loop through each of the multiply-imputed variables..
    for ( j in mi_vars ){
    
        # copy the contents of the current column (for example 'welfare1')
        # over to a new column ending in 'mi' (for example 'welfaremi')
        x[ , paste0( j , 'mi' ) ] <- x[ , paste0( j , i ) ]
        
        # delete the all five of the imputed variable columns
        x <- x[ , !( names( x ) %in% paste0( j , 1:5 ) ) ]

    }
    
    assign( paste0( 'imp' , i ) , x )

}

ces_design <- 
    svrepdesign( 
        weights = ~ finlwt21 , 
        repweights = "^wtrep[0-9][0-9]$" , 
        data = imputationList( list( imp1 , imp2 , imp3 , imp4 , imp5 ) ) , 
        type = "BRR" ,
        combined.weights = TRUE ,
        mse = TRUE
    )

Variable Recoding

Add new columns to the data set:

ces_design <- 
    update( 
        ces_design , 
        
        one = 1 ,
        
        any_food_stamp = as.numeric( jfs_amtmi > 0 ) ,
        
        bls_urbn = factor( bls_urbn , levels = 1:2 , labels = c( 'urban' , 'rural' ) ) ,
        
        sex_ref = factor( sex_ref , levels = 1:2 , labels = c( 'male' , 'female' ) )
        
    )

---

Analysis Examples with the survey library  

Unweighted Counts

Count the unweighted number of records in the survey sample, overall and by groups:

MIcombine( with( ces_design , svyby( ~ one , ~ one , unwtd.count ) ) )

MIcombine( with( ces_design , svyby( ~ one , ~ bls_urbn , unwtd.count ) ) )

Weighted Counts

Count the weighted size of the generalizable population, overall and by groups:

MIcombine( with( ces_design , svytotal( ~ one ) ) )

MIcombine( with( ces_design ,
    svyby( ~ one , ~ bls_urbn , svytotal )
) )

Descriptive Statistics

Calculate the mean (average) of a linear variable, overall and by groups:

MIcombine( with( ces_design , svymean( ~ totexp ) ) )

MIcombine( with( ces_design ,
    svyby( ~ totexp , ~ bls_urbn , svymean )
) )

Calculate the distribution of a categorical variable, overall and by groups:

MIcombine( with( ces_design , svymean( ~ sex_ref ) ) )

MIcombine( with( ces_design ,
    svyby( ~ sex_ref , ~ bls_urbn , svymean )
) )

Calculate the sum of a linear variable, overall and by groups:

MIcombine( with( ces_design , svytotal( ~ totexp ) ) )

MIcombine( with( ces_design ,
    svyby( ~ totexp , ~ bls_urbn , svytotal )
) )

Calculate the weighted sum of a categorical variable, overall and by groups:

MIcombine( with( ces_design , svytotal( ~ sex_ref ) ) )

MIcombine( with( ces_design ,
    svyby( ~ sex_ref , ~ bls_urbn , svytotal )
) )

Calculate the median (50th percentile) of a linear variable, overall and by groups:

MIcombine( with( ces_design ,
    svyquantile(
        ~ totexp ,
        0.5 , se = TRUE 
) ) )

MIcombine( with( ces_design ,
    svyby(
        ~ totexp , ~ bls_urbn , svyquantile ,
        0.5 , se = TRUE ,
        ci = TRUE 
) ) )

Estimate a ratio:

MIcombine( with( ces_design ,
    svyratio( numerator = ~ totexp , denominator = ~ fincbtxmi )
) )

Subsetting

Restrict the survey design to california residents:

sub_ces_design <- subset( ces_design , state == '06' )

Calculate the mean (average) of this subset:

MIcombine( with( sub_ces_design , svymean( ~ totexp ) ) )

Measures of Uncertainty

Extract the coefficient, standard error, confidence interval, and coefficient of variation from any descriptive statistics function result, overall and by groups:

this_result <-
    MIcombine( with( ces_design ,
        svymean( ~ totexp )
    ) )

coef( this_result )
SE( this_result )
confint( this_result )
cv( this_result )

grouped_result <-
    MIcombine( with( ces_design ,
        svyby( ~ totexp , ~ bls_urbn , svymean )
    ) )

coef( grouped_result )
SE( grouped_result )
confint( grouped_result )
cv( grouped_result )

Calculate the degrees of freedom of any survey design object:

degf( ces_design$designs[[1]] )

Calculate the complex sample survey-adjusted variance of any statistic:

MIcombine( with( ces_design , svyvar( ~ totexp ) ) )

Include the complex sample design effect in the result for a specific statistic:

# SRS without replacement
MIcombine( with( ces_design ,
    svymean( ~ totexp , deff = TRUE )
) )

# SRS with replacement
MIcombine( with( ces_design ,
    svymean( ~ totexp , deff = "replace" )
) )

Compute confidence intervals for proportions using methods that may be more accurate near 0 and 1. See ?svyciprop for alternatives:

# MIsvyciprop( ~ any_food_stamp , ces_design ,
#   method = "likelihood" )

Regression Models and Tests of Association

Perform a design-based t-test:

# MIsvyttest( totexp ~ any_food_stamp , ces_design )

Perform a chi-squared test of association for survey data:

# MIsvychisq( ~ any_food_stamp + sex_ref , ces_design )

Perform a survey-weighted generalized linear model:

glm_result <- 
    MIcombine( with( ces_design ,
        svyglm( totexp ~ any_food_stamp + sex_ref )
    ) )
    
summary( glm_result )

---

Replication Example

This example matches the number of consumer units and the Cars and trucks, new rows of Table R-1:

result <-
    MIcombine( with( ces_design , svytotal( ~ as.numeric( popwt_finlwt21 / finlwt21 ) ) ) )

stopifnot( round( coef( result ) , -3 ) == 134556000 )

results <- 
    sapply( 
        weight_columns , 
        function( this_column ){
            sum( ces_df[ , 'new_car_truck_exp' ] * ces_df[ , this_column ] ) / 
            sum( ces_df[ , paste0( 'popwt_' , this_column ) ] )
        }
    )

stopifnot( round( results[1] , 2 ) == 2896.03 )

standard_error <- sqrt( ( 1 / 44 ) * sum( ( results[-1] - results[1] )^2 ) )

stopifnot( round( standard_error , 2 ) == 225.64 )

# note the minor differences
MIcombine( with( ces_design , svymean( ~ cartkn ) ) )