Addressing missing consumer expenditure data due to the 2025 lapse in appropriations

From October 1, 2025, to November 12, 2025, a lapse in federal appropriations halted data collection activities for the Consumer Expenditure Surveys (CE). Although the federal government received funding again on November 13, 2025, data collection for the surveys did not resume until December 5, 2025. As a result, no data were collected in October and November 2025.

The U.S. Bureau of Labor Statistics (BLS) uses data from the CE to publish estimates of consumer spending and to update the spending weights of the Consumer Price Index (CPI). To mitigate the impacts of missing data on the accuracy of these estimates, BLS considered three approaches: adjusting survey weights, imputing unit-level data, and modeling domain-level estimates. When considering approaches, BLS evaluated operational complexities and potential impacts on publication timeliness. In a May 2026 article, BLS summarized the analysis of these approaches based on data simulations.¹

BLS engaged the National Association for Business Economics (NABE) and the Committee on National Statistics (CNSTAT) for additional expert input on the approaches under consideration. Both expert panels advised BLS to prioritize the timely publication of estimates and noted that survey weight adjustment sufficiently mitigated the negative impacts to data quality. In addition, both panels stressed the importance of BLS working on future improvements to reduce risks to timely publication associated with other approaches.

Based on BLS analysis and input from the expert panels, BLS plans to implement an adjustment to the survey weights to address the missing data from October and November 2025. This article describes additional analysis to support this decision and provides more detail on the implementation of the adjustments.

All estimates using the 2025 CE data are expected to remain on schedule. First, BLS plans to release the final estimates of the Chained CPI-U (C-CPI-U) for July, August, and September 2025 in August 2026. Next, BLS will publish the 2025 annual CE estimates and public use microdata in October 2026. Then, BLS will publish the final estimates of the C-CPI-U for October, November, and December 2025 in November 2026. Finally, BLS will issue the 2027 CPI weight update in February 2027. BLS also plans to summarize this information for data users on the BLS website prior to each data release.

Background research

This section of the article describes the approaches considered by BLS to handle the disruption to CE data collection in October and November 2025, as well as further discussions with expert panels to handle the missing data.

Approaches considered by BLS

The CE consists of two separate surveys, the Interview Survey and the Diary Survey. The Interview Survey captures spending on major expenses using a 3-month recall design. The Diary Survey captures spending on smaller purchases using two consecutive weekly diaries.

The disruption to CE data collection in October and November 2025 led to missing expenditure data. Due to the recall design of the Interview Survey, even though data were not collected in October and November, at least some expenditures in all reference months in 2025 were collected. BLS expects approximately 33 percent fewer expenditures reported for July and October, and 66 percent fewer expenditures reported for August and September. For the Diary Survey, no data were collected in October and November. Since the collection month is the same as the reference months in the Diary Survey, there will not be expenditure data in all reference months in 2025. In a May 2026 article, BLS published a detailed explanation of the impact of the lapse in appropriations.²

BLS integrates data from both CE surveys to create annual estimates of spending and spending weights for the CPI. BLS updates the CPI for All Urban Consumers (CPI-U) and CPI for Wage Earners and Clerical Workers (CPI-W) annually with January indexes to reflect spending 2 years earlier. BLS produces another measure of inflation, the C-CPI-U, using monthly spending weights. Because monthly spending weights are only available with a lag, the C-CPI-U is produced in preliminary form and revised 10 to 12 months later.

Missing CE data affects BLS estimates of consumer spending and consumer price indexes. BLS considered three approaches to address the missing CE data. The first approach was to adjust the CE survey weights to approximate the missing months of data collection. This approach can be implemented with the least intervention into BLS production systems and minimizes the risk of disrupting publication timelines. The specific proposal was to adjust survey weights for:

• August and September expenditures from the Interview Survey by a factor of 3

• July and October expenditures from the Interview Survey by a factor of 3/2

• September and December expenditures from the Diary Survey by a factor of 2

The second approach was unit-level modeling, whereby missing monthly spending for the consumer unit (CU) is modeled.³ Investigating possible predictors, model specification, and model development would almost certainly risk disrupting publication schedules. BLS considered a simple year-over-year model, where missing unit-level data were modeled as the prior year with an inflation factor applied. Although the simulation results for the C-CPI-U were promising, BLS determined the complexities to implement this approach added risk to the overall publication schedule.

The third approach was to improve estimates at the domain (aggregate) level by incorporating time series options: linear interpolation, Auto Regressive Integrated Moving Average (ARIMA), and Echo State Network (ESN) multivariate forecast methods. ARIMA and ESN have the appeal of addressing seasonality concerns with monthly spending weights for the C-CPI-U. Again, BLS deemed the complexities to implement these methods too risky to address the missing 2025 data. However, in the event of another case of extensive missing data, BLS plans to further investigate this method for possible future implementation.

Expert panels

BLS engaged the National Association for Business Economics (NABE) and the Committee on National Statistics (CNSTAT) to conduct expert panels regarding approaches for handling missing October and November 2025 CE data.

The NABE panel consisted of 10 expert panelists from a variety of public and private institutions. The CNSTAT panel consisted of eight expert panelists, again from a variety of organizations. Panelists brought knowledge of a wide range of topics including imputation, forecasting, time series econometrics, economic measurement, monetary policy, price measurement, and more. A full list of panelists is available on the BLS website.⁴

The panels provided input on the approaches under consideration by BLS to address the missing data. Both panels recommended that BLS prioritize the timely publication of data. Both panels noted the importance of addressing seasonal variations in expenditures, which is especially important for the C-CPI-U. If the approaches presented minimal risks to the timeliness and integrity of the data, the panels suggested:

• Investigating the survey weight adjustments. If the survey weight adjustments can be optimized based on historical data, there could be marginal gains in the accuracy of estimates.

• Exploring the use of Interview Survey data to supplement missing Diary Survey data. There is overlap in some item categories between the Interview and Diary Surveys. Since expenditures were collected for October and November 2025 in the Interview Survey, the collected data could be leveraged to address the missing data in the Diary Survey.

• Blending options for estimation of C-CPI-U monthly spending weights. Given the results of simulations performed by BLS, combining several options appears to have the best result compared to published index values in the past.

Finally, both panels recommended that BLS continue research in domain estimation and enhancements of production systems to accommodate future instances of extensive missing data.

Additional analysis

This section discusses additional analysis that BLS has conducted to address some of the recommendations received from the NABE and CNSTAT expert panels.

Determining the survey weight factor adjustment

BLS identified proposed factors for the Interview and Diary Surveys. For the Interview Survey, BLS explored using factors that are closely tied to the survey design. Given the survey design with a 3-month recall, it is expected that one third of expenditures for a reference month will be reported in each of the 3 collection months. Relative to the survey design, 33 percent fewer expenditures are expected in July and October 2025, whereas 66 percent fewer expenditures are expected in August and September due to the missed data collection in October and November.

Based on this, BLS proposed factors of 3/2 for July and October 2025 expenditures and 3 for August and September expenditures. That is, since 2/3 of the data were collected in July and October, multiplying them by 3/2 provides an estimate of what the full sample would have yielded if collected; and multiplying the 1/3 of data collected for August and September provides a similar estimate.

In practice, the share of missing expenditures will vary depending on the CUs surveyed.⁵ Additionally, recall decay could result in a share of missing expenditures that is systematically different from expectations. The theory of recall decay suggests that more distant reference period months tend to have fewer expenditures reported when compared with more recent months. Additional research is needed to explore the appropriateness of making assumptions about the missing data patterns. This could include analyzing the impact of recall decay, particularly at lower levels of detail, or researching the assumption that the expenditure data are missing at random within the weighting adjustment cells.

In 2024, approximately one third of reference month expenditures were reported in each of the 3 collection months. If October and November data collection were missing in 2024, that would have resulted in fewer expenditures in July (33.4 percent), August (67.5 percent), September (66.5 percent), and October (34.5 percent). This suggests the data missing in 2025 should not meaningfully deviate from expectations at the aggregate level and the factors of 3/2 and 3 remain appropriate for the survey weight adjustment.

For the Diary Survey, BLS experimented with two proposed factors. Reweighting September and December diaries by a factor of 2 can reasonably compensate for the missing October and November expenditures.⁶ BLS also considered a factor of 1.5 applied to July, August, September, and December. This approach would evenly distribute the weight across the available 4 months of data over the 6-month period (July–December). Tables 1 and 2 display summary statistics on the percent difference in mean and standard errors between production data and the two simulations: reweighting 2 months of data or reweighting 4 months of data. The results are based on the statistics calculated for each Universal Classification Code (UCC) aggregated across all CUs. UCCs are the lowest level of expenditure category detail estimated in the CE. The 546 UCCs used to calculate CE annual estimates are included in this analysis.⁷

Applying the factor to 2 months produced a more accurate point estimate (smaller difference in mean) but with less precision (larger difference in standard error) than applying the factor to 4 months. BLS decided to prioritize accuracy and use the 2-month weight adjustment for Diary Survey data.

CE survey source evaluation

The Interview and Diary Surveys collect redundant information for a subset of categories. Because the Interview Survey conducted in December, January, and February collected reference expenditure data for October and November, BLS considered whether the Interview Survey could be leveraged to address missing data in the Diary Survey.

Annually, BLS decides which survey (Interview or Diary) is the better source for use in publication for cases where both surveys capture spending on the same UCCs. BLS determines the best source by evaluating overlapping UCCs for the prior 3 years of data against several criteria. If there are enough individual records in both surveys and no clear choice based on the criteria, a decision is made based on the current year’s source, an item’s seasonality, and the level of the mean (favoring the survey that generates the highest mean value).⁸

BLS simulated the missing data in 2025 by using 2024 production data and removing data collected in October and November. For the overlapping UCCs that are found in both surveys, BLS compared the means, standard errors, and ratios of the standard error to the mean (or relative standard errors) across surveys. BLS considers estimates with a relative standard error of 25 or more to be unreliable.⁹ All the simulated estimates deemed unreliable had very high relative standard errors in both surveys. Since data quality was not significantly improved by changing survey source, BLS did not recommend additional survey source changes to address missing data in 2025.

Combining approaches to address seasonal variation

Within the domain estimation approach, BLS explored three options: linear interpolation, Auto Regressive Integrated Moving Average (ARIMA), and Echo State Network (ESN) multivariate forecast methods.¹⁰

BLS also prepared estimates of the impact on the CPI-U. BLS ran data simulations from December 2023 to December 2025 to mimic the options to address missing CE data in October and November. For the simulations, October and November CE data from 2022 and 2023 were removed, and then five adjustment options were explored. The simulated 2022 CE data were used to estimate the impact to the 2024 CPI-U. Similarly, the simulated 2023 CE data were used to estimate the impact to the 2025 CPI-U.

When rounded to the level of precision for publication (one decimal place), all options are indistinguishable from the production annualized percent change over this period, 2.8 percent. (See table 3.) This means the difference between options is imperceptible at the level of publication precision.

The choice of options also has impacts on the C-CPI-U. In addition to the index results researched by BLS and discussed in a May 2026 article, BLS considered which method best estimated spending weights at a monthly level.¹¹ BLS simulated missing data in the years from 2022 to 2024 and applied five options to address these missing data. One measure of data quality is the difference in expenditures between the simulated data and the production data. Table 4 summarizes the mean difference in monthly expenditures (absolute value) over the period. All options result in monthly expenditure levels that are very close to production data (with differences of 0.10 percent or less).

The survey weight adjustment method performed well. For the C-CPI-U, the CNSTAT panel remarked that the use of other methods could better address the seasonal variation of missing data. The NABE panel noted that combining various options could reduce the overall deviation of simulated data from production data. In practice, the survey weight adjustment and unit-level imputation approaches are applied at different points in the production process. Although a post-hoc average of the results minimizes the error to historical data, combining several options in the production process would be complex and detrimental to timeliness.

BLS approach to address missing data

After careful consideration, BLS is implementing the survey weight adjustment method to address missing CE data. For the CE annual estimates and weight update for the CPI-U and CPI-W, expenditures collected in the Interview Survey in August, September, and December 2025 and January 2026 will be adjusted by a factor of 3 or 3/2 to ensure they represent a full month of spending. Expenditures collected in the Diary Survey during September and December 2025 will be adjusted by a factor of 2 to represent the missing expenditures in October and November. For the C-CPI-U, this is implemented as a forecast and backcast of the data to explicitly impute the missing October and November 2025 estimates.

Finalization of Chained CPI-U, third quarter of 2025

BLS plans to publish final estimates of the C-CPI-U for the third quarter (i.e., July, August, and September) of 2025 on August 12, 2026. To calculate monthly spending weights, BLS makes several adjustments to CE microdata prior to the data smoothing steps described in the BLS Handbook of Methods.¹² Since July, August, and September expenditures collected in the Interview Survey are impacted by the missing data, an adjustment is needed for the third quarter’s release. During microdata processing, BLS will apply the survey weight adjustment factor to the Interview Survey reference month expenditures in August and September. (See equation 1.) No adjustment is needed to the Diary Survey expenditures for the third quarter.

Equation 1. Formula to calculate adjusted monthly expenditures with a survey weight adjustment factor, Interview Survey

${\widehat{E}}_{i,j}=E_{i,j}{f}_j$

where

${\widehat{E}}_{i,j}$ is the adjusted expenditure per CU, i, in reference month, j;

i is the interview consumer unit;

j is the reference month;

E is the unadjusted expenditure;

$f_j$ is the survey weight factor. $f$ = 3/2 when j is July and $f$ = 3 when j is August or September.

The adjusted expenditure, ${\widehat{E}}_{i,j}$, is input into the typical spending weight calculations. BLS maps consumer spending by CUs at the UCC level to the CPI item categories, adjusting for scope and definitional differences. Monthly expenditures for a CPI item category in each geographic area are calculated from the adjusted expenditures.¹³

Consumer spending, 2025 annual estimates

To calculate annual estimates of consumer spending in 2025, BLS will adjust the survey weights. Equation 2 displays the formula to calculate average annual expenditures using data from the Interview and Diary Surveys.¹⁴ The formulas differ to reflect the different survey designs. The survey weight adjustment factor also differs by survey: $f_{i,j}$ for the Interview Survey and $f_i$ for the Diary Survey.

Equation 2. Formula to calculate average annual expenditures with a survey weight adjustment factor

Interview Survey:

$\bar{y}=12\times \frac{\sum _{i\in S}{w}_i(\sum _{j=1}^3{f}_{i,j}{y}_{i,j})}{\sum _{i\in S}{w}_i(\sum _{j=1}^3{f}_{i,j})}$ 

Diary Survey:

$\bar{y}=52\times \frac{\sum _{i\in S}\left(w_i{f}_i\right)y_i}{\sum _{i\in S}\left(w_i{f}_i\right)}$

where

$\bar{y}$ is the average yearly expenditure per CU on the item;

$y_{i,j}$ is the monthly expenditure made by the ith CU in the sample in their jth reference month (j = 1, 2, or 3) on the item (Interview Survey only);

$y_j$ is the weekly expenditure made by the ith CU in the sample on the item (Diary Survey only);

$w_i$ is the weight of the ith CU in the sample;

$f_{i,j}$ is the weight adjustment factor for the expenditures of the ith CU in the sample in their jth reference month (Interview Survey only): 3 when the reference month is August or September, 3/2 when the reference month is July or October, 1 when the reference month is any other month in 2025, and 0 when the reference month is in 2024 or 2026;

$f_i$ is the weight adjustment factor for the expenditures of the ith CU in the sample (Diary Survey only): 0 for CUs in the October and November samples, 2 for CUs in the September and December samples, and 1 otherwise;

S is the set of CUs in the sample that participated in the survey.

For expenditures sourced from the Interview Survey, the adjustment is applied to expenditures made in July, August, September, and October. As shown in equation 2, average annual expenditures ($\bar{y}$) are calculated as the sum of weighted monthly expenditures ($y_{i,j}$) divided by a population estimate. Because of the recall period, 15 months of collected data are used to calculate an annual estimate. An adjustment is needed to correct the population estimates ($\sum _{i\in S}{w}_i$) to an annual total.

The adjustment is called “months-in-scope,” and it counts the number of reference months that a CU’s expenditures are in the in-scope period. It usually equals 0,1, 2, or 3, but in 2025, it equals $m_i=\sum _{j=1}^3{f}_{i,j}.$The survey weight, $w_i$, is calculated for a CU and does not vary by recall period. For expenditures sourced from the Diary Survey, the adjustment is applied to the survey weight. The survey weight is doubled for CUs collected in September and December via the adjustment factor. Because the survey is collected for one week, estimates are annualized by applying a factor of 52.

Consumer spending, public use microdata

Annually, BLS publishes the Public Use Microdata (PUMD), which are a subset of consumer spending microdata files with perturbations applied for disclosure avoidance. While missing 2 months of collected data impacts the PUMD, BLS plans to publish them without adjustment to account for missing data in October and November 2025.

It is important for microdata users to know that CUs from the fourth quarter of 2025 will only represent one third of the number of CUs in the country. This is because December includes only one third of CUs in the quarter, and October and November are missing. This is true for data from both the Interview and Diary Surveys. Microdata users will need to decide how to address the missing data for themselves on a case-by-case basis.

With the scheduled release of 2025 annual consumer spending data on October 29, 2026, BLS plans to update the PUMD Getting Started Guide with further information on how to replicate BLS annual estimates with the missing data factors applied.¹⁵ Depending on the specific application, microdata users may want to make adjustments that mimic the CE published estimates. This would require adjustments to the Interview Survey costs and months in scope, as well as to the Diary Survey weights.

Finalization of Chained CPI-U, fourth quarter of 2025

BLS plans to publish final estimates of the C-CPI-U for October, November, and December 2025 on November 10, 2026. To calculate monthly spending weights, BLS makes several adjustments to CE microdata prior to the data-smoothing steps described in the BLS Handbook of Methods.¹⁶ Like the third quarter, an adjustment is needed to calculate October data collected in the Interview Survey. (See equation 3.) During microdata processing, BLS will apply the survey weight adjustment factor to October 2025 expenditures.

Equation 3. Formula to calculate adjusted monthly expenditures with a survey weight adjustment factor, Interview Survey

${\widehat{E}}_{i,j}=E_{i,j}{f}_j$

where

${\widehat{E}}_{i,j}$ is the adjusted expenditure per CU, i, in reference month, j, for an item;

i is the interview consumer unit;

j is the reference month;

E is the unadjusted expenditure;

$f_j$ is the survey weight factor. $f$ = 3/2 when j is October.

An additional adjustment is needed for Diary Survey data missing in October and November 2025. Because the formula used to calculate the final C-CPI-U index requires monthly spending weights, BLS must adjust Diary Survey data differently for the C-CPI-U than for the CE annual estimates. Rather than apply an adjustment to September and December survey weights, BLS will create October spending weights as a copy of September and November spending weights as a copy of December. (See equation 4.)

Equation 4. Formula to calculate adjusted monthly expenditures with a survey weight adjustment factor, Diary Survey

${\widehat{E}}_{i,j}=E_{i,j*}$

where

${\widehat{E}}_{i,j}$ is the adjusted expenditure per CU, i, in reference month, j, for an item;

i is the diary consumer unit;

j* is the source reference month. When the reference month j is October, then j* = j-1; when the reference month is November, then j* = j+1.

The adjusted expenditure, ${\widehat{E}}_{i,j}$, is input into the typical spending weight calculations. As with third quarter data, BLS maps consumer spending by CUs at the UCC level to the CPI item categories and calculates monthly expenditures for a CPI item category in each geographic area.

2027 CPI weight update

The survey weight adjustments applied to calculate monthly expenditures for the C-CPI-U will also be applied to calculate annual expenditures for the CPI-U and CPI-W in 2027. As with the monthly spending weights for the C-CPI-U, the adjusted expenditure, ${\widehat{E}}_{i,j}$, is input into the typical spending weight calculations.¹⁷ The aggregation weight is calculated from annual expenditures for an item category in each geographic area for each population (CPI-U and CPI-W). The annual expenditures are derived from the adjusted expenditures calculated for monthly spending weights.

Next steps

BLS will finalize implementation plans for the survey weight adjustments as described in this article. Since the adjustments are interventions to the CE and CPI production systems, BLS will thoroughly test these adjustments prior to implementation. BLS plans to release the final C-CPI-U estimates, CE annual data, and 2027 CPI weight update as scheduled. (See table 5.)

To inform data users of the changes described in this article, BLS plans to publish additional information on the BLS website prior to release. This could include additional frequently asked questions to describe the methods considered and selected.¹⁸

BLS is actively conducting research into domain estimation to improve consumer price indexes. The first project is motivated by an interest in constructing estimates of the CPI for all Core-Based Statistical Areas (CBSAs) and states, despite a geographically sparse set of sampled CBSAs. This research effort develops spatial-temporal hierarchical Bayesian models while accounting for local spatial-temporal correlations to compensate for geographical sparseness of the collected survey data and to allow for the construction of reliable estimates in small areas. Through a comparison of the model-based estimates of fuel prices from the CPI with estimates from a large administrative data set, this research has demonstrated improvements in robustness of CBSA- and state-level estimates by accounting for spatial correlations, sociodemographic fixed effects, and temporal trends.

The second effort, although not initially developed specifically as a method to account for the missing CE and CPI data due to the 2025 lapse in appropriations, proposes a method to potentially improve preliminary estimation of the C-CPI-U using layered ESNs to forecast item-area expenditure shares, which are normally unavailable for 10 to 12 months. The current implementation of the ESN methodology incorporates expenditure sample variances, handles both exogenous and endogenous inputs, and optionally includes seasonal Fourier components. Prototype ESN models are being built and compared with predictions from the current constant elasticity model.

In terms of systems development, BLS continues to modernize IT infrastructure and software systems. While such modernization efforts are happening, BLS will be limited in its ability to make many changes to current systems. Eventually, BLS aims to introduce greater flexibility to its IT infrastructure and software systems to accommodate methods to address large-scale missing data, such as the missing data in 2025.