Medical Expenditure Panel Survey

Methodology

All estimates are produced using the R survey package following AHRQ-recommended methods for complex survey data. Each year is analyzed independently as a cross-sectional snapshot of the U.S. civilian noninstitutionalized population.

Survey Weights

MEPS oversamples certain populations—including low-income families, racial and ethnic minorities, and the elderly—to ensure they are adequately represented. Each observation carries a person-level final weight (e.g., PERWT22F for 2022) that adjusts for this oversampling as well as survey nonresponse and post-stratification to Census population totals. All means shown in the heatmap are weighted, so they represent the full U.S. population, not just the ~30,000 survey respondents.

Variance Estimation

Because MEPS uses a stratified, multi-stage cluster design, standard formulas for simple random samples would understate the true uncertainty. This analysis specifies the stratum (VARSTR) and primary sampling unit (VARPSU) variables and uses Taylor-series linearization to produce correct standard errors. In practical terms, this means the confidence intervals properly reflect the fact that people within the same geographic cluster tend to be more similar to each other than to the population at large.

Domain (Subgroup) Analysis

When computing estimates for subgroups (e.g., adults age 45–64 in the South), the full survey design is preserved rather than subsetting the data first. This is critical: dropping observations outside the subgroup would remove information the variance estimator needs, producing standard errors that are too small. The R survey package’s subset() method on a svydesign object handles this correctly by zeroing out weights for out-of-domain observations while retaining them in the design structure.

Expenditures Are in Nominal Dollars

Dollar amounts shown in the heatmap (Mean Total Expenditure and Mean Out-of-Pocket) are reported in nominal dollars for each survey year—that is, they reflect the prices that prevailed in that year without adjustment for inflation. Between 2018 and 2022, the CPI for Medical Care rose roughly 10–12%, meaning that a portion of any increase in expenditures across years reflects higher prices rather than greater utilization of services. Because of this, direct year-over-year comparison of the raw dollar values in the tooltip can overstate real spending growth. To make expenditures comparable across time, they would need to be deflated to a common base year using an appropriate price index such as the CPI for Medical Care or the Personal Consumption Expenditures (PCE) Health price index. The heatmap’s color scale is not affected by this issue: normalization is performed independently within each year, so the color pattern shows relative differences between region-age groups for a given year, not absolute dollar comparisons across years.

No Pooling Across Years

Each survey year is analyzed as a standalone cross-section. The data are not pooled across years, so no weight adjustment (dividing by the number of pooled years) is necessary. This approach is appropriate for examining how population-level estimates change over time but means that rarer conditions or smaller subgroups may exhibit more year-to-year volatility than they would in a pooled multi-year analysis.

Data Sources

Agency for Healthcare Research and Quality. (2024). Medical Expenditure Panel Survey. [Data set]. U.S. Department of Health and Human Services. https://meps.ahrq.gov/mepsweb/

For accessing MEPS data in R, see the MEPS R Package maintained by AHRQ.

R Core Team. (2024). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/

Lumley, T. (2004). Analysis of Complex Survey Samples. Journal of Statistical Software, 9(8), 1–19. R package version 4.4. https://cran.r-project.org/package=survey