(Monthly) Rolling averages

Author

Zane Mokhiber and Emma Cohn

Published

March 18, 2026

This script will teach you how to calculate simple monthly rolling averages using CPS data and employment-to-population ratios (EPOPs). It accounts for missing months of data, such as the October 2025 gap due to the government shutdown and resulting lapse in Bureau of Labor Statistics appropriations.

Load required libraries

The following chunk of code loads the R libraries necessary for this exercise. You may need to install them to run this code. If you haven’t yet set up your computer to run EPI CPS microdata extracts, complete that process before running this code.

#Load necessary libraries
library(tidyverse) #this package contains most common R functions
library(epiextractr) #allows you to get CPS (e.g., Basic, ORG) extracts
library(slider) #contains the sliding mean function

Import and clean data

Note: Don’t forget to update years to match your setup before running the script.

Running this script chunk will call the BLS Current Population Survey Basic data required to calculate rolling average EPOPs.

# Step 1: Import CPS Basic data
cps_data <- load_basic(2023:2026, year, month, basicwgt, emp, age, cow1) |>  
    mutate(weight = basicwgt)

Calculate EPOPs

This code chunk creates a universe variable that restrictions the data to observations with employment data (i.e., excludes NA responses). The the code divides the total weighted number of employed people (count) by the total universe count. The output lists EPOPs, total counts, and sample sizes by month and year.

# Step 2: calculate monthly weighted counts, universe, and sample size 
monthly_data <- cps_data |> 
    mutate(universe = if_else(!is.na(emp), 1, 0)) |> 
        #swap out emp here for another variable to calculate a different statistic
        summarize(count = sum(emp * weight, na.rm = TRUE), 
                  universe_total = sum(universe * weight, na.rm = TRUE), 
                  sample_size = sum(universe, na.rm = TRUE), .by = c(year, month)) |> 
    mutate(percent = count / universe_total)

Calculate rolling average

This section smooths the data over a calendar year. The slide_index function takes a certain number of calendar months as an input, rather than a fixed number of rows (in this case, months). This means that it consistently calculates the mean over a set time period and adjusts the denominator based on whether months during that window are missing. Alternative methods often only accept a constant number of rows, and so will look back farther (i.e., at extra months) if any are missing.

In other words, a 12-month rolling average for November 2025 goes back to October 2024 and ‘counts’ the blank data in October 2025 as a month (while still averaging over 11 months’ worth of data). This method will adapt to any gaps in the data and maintain a consistent average.

In order to change the rolling length, change the months(x) value. This code calculates 12-month rolling averages; a smaller number of months will be more susceptible to short-term changes such as business cycles or seasonal employment. On the flip side, a high number of months will smooth out shorter-term or temporary changes, like the effects of recent policy, but will include more robust sample sizes. Consider this when deciding how many months (or other measure) to include.

smoothed_monthly_data <- monthly_data |> 
    mutate(date = as.Date(paste(year, month, "01", sep = "-"))) |>  #format date label               
    arrange(date) |> 
    mutate(percent = slide_index_dbl(percent, date, mean, .before = months(11), .complete = TRUE), 
          count = slide_index_dbl(count, date, mean, .before = months(11), .complete = TRUE), 
          sample_size = slide_index_dbl(sample_size, date, sum, .before = months(11), .complete = TRUE))

And that’s it! You can alter this code to look at specific demographic cuts, such as filtering to specific racial/ethnic groups or to prime-age (24-54 year old) workers. You can also use this code to look at other variables–the rolling average section is written ambigously enough to allow for many different variables.

I’d also recommend benchmarking your data to the State of Working America Data Library to ensure reliability. As always, be sure to keep an eye on those sample sizes!

Back to top