Use of alternative data and methods in the CPI for wireless telephone services : U.S. Bureau of Labor Statistics

Bureau of Labor Statistics > Consumer Price Index > Methods > Additional Resources

Use of alternative data and methods in the CPI for wireless telephone services

Starting with the release of July 2025 indexes in August 2025, the BLS will begin using secondary source data and non-traditional index methods to measure price change for the wireless telephone services category of the CPI. One secondary source of data provides a near universe of wireless telephone service plans available to consumers and their offer prices and characteristics, and another source provides consumer expenditure data used to weight the service plans. The non-traditional index methodology uses prices estimated from hedonic regression models to control for quality change. These estimated prices are then aggregated using a Törnqvist index formula that accounts for changes in the relative weighting of the service plans. The use of these data reduces sampling error in the index, and the non-traditional index methodology improves the measurement of quality-adjusted price change.

This document includes further detail on the alternative data and methods, and how they are used in the calculation of the wireless telephone services index. The Monthly Labor Review article “Alternative data sources for high-tech products in the CPI” describes the research behind this methodology change.

Traditional methodology

Traditionally, all prices used in the CPI for wireless telephone services have been collected using data from the Commodities and Services (C&S) Pricing Survey. Businesses (called “outlets” in CPI terminology) that provide wireless telephone services are sampled based on results of the Consumer Expenditure Survey (CE).

CPI data collectors work with survey respondents from the sampled outlets to scientifically select specific services to price on a monthly or bimonthly basis. Respondents provide offer prices for each sampled service. The characteristics of the originally selected service, such as the exact type of service plan and any other service plan characteristics are held constant over time until the unique service plan becomes unavailable for purchase.

For more information on how C&S survey data are collected and calculated, see the BLS Handbook of Methods and the telecommunications services factsheet for CPI.

Service plan data

BLS purchases wireless telephone service plan price and characteristics data from a market research firm that specializes in the telecommunications industry. The vendor uses both web-scraping and nonautomated methods of data collection to monitor and record the advertised prices and characteristics of service plans offered by wireless telecommunications providers to both new customers and existing customers who are upgrading their plans. The data includes service plans offered by mobile network operators (MNOs) and mobile virtual network operators (MVNOs). Service plans and their prices are available nationwide with national pricing. BLS receives the service plan data in the middle of the month in which they will be used in index calculations.

After receiving the data, BLS processes the data to refine and format it for use. BLS removes ineligible service plans, such as those requiring bundling with other telecommunications services (like home internet services) and those that do not include voice services. BLS derives average monthly plan prices for month-to-month plans (postpaid contracts) and prepaid plans after considering applicable promotions. Additional processing compares the characteristics of service plans from the current month to service plans from the previous month to identify changes to the structure of the data or changes to how data elements have been coded by the vendor.

Information in certain data elements is standardized and formatted. Notable examples are the amount of data included with the service plan (and used directly by the phone associated with the plan) and hotspot data allowances. BLS creates two categories – phone data and hotspot data – and each service plan is assigned a value in these categories based on its characteristics. The values are:

Unlimited premium (unlimited amounts of high-speed data)
Unlimited throttled (unlimited amounts of data delivered at a slower speed)
Data cap (high-speed data up to a specified threshold)
No data (no data offered with the plan)

Each service plan is also given a unique identifier based on its provider, plan name, type of contract (postpaid or prepaid), number of serviced lines associated with the plan (only postpaid plans with five or less and prepaid plans with two or less are eligible), and other plan characteristics. The unique plan identifiers are used to determine which plans are entering the data (only appearing in the current month’s dataset), are exiting the data (only appearing in the previous month’s dataset), or are continuing (appearing in both the current and previous months’ datasets).

Weighting data

BLS uses household survey data purchased from a market research firm to derive weights for wireless telephone service plans. The vendor conducts annual studies of wireless services by surveying households regarding their wireless telephone ownership experiences. Data purchases and weighting data updates occur every two years. As described in detail in the “Index Methodology” section below, the weight data are used to properly weigh service plans in regression models and in the aggregation of price relatives.

After receiving the survey data, BLS processes the data to prepare it for use. First, survey responses where the respondent did not provide a monthly wireless telephone service bill amount are removed from the data. Next, BLS eliminates survey responses where the amount of data included with a service plan is not indicated or where the number of lines serviced by the plan is missing. These steps are necessary because survey responses without a billed amount or key plan characteristics are not useful for our purposes and can skew results. Lastly, survey responses where prepaid contracts were identified as servicing more than two telephone lines are removed from the data to align with service plan eligibility rules. The final preparation step is to classify specific service providers as either MNOs or MVNOs.

The survey data are further refined through an outlier identification and removal process. The first step in the process is to subdivide the survey data into groups based on the type of service plan contract (either postpaid or prepaid) and whether the plan services one telephone line or more than one telephone line. Those groups are further subdivided by the amount of data included with a service plan. Those four categories are: no data, 1 to less than 10 gigabytes (GBs), 10 or more GBs, and unlimited GBs. The group distributions are stratified into quartiles based on the reported billed amount, and outliers are identified. An outlier is defined as any survey response located outside of 1.5 times the interquartile range on either side of the 75^th percentile (upper quartile) and the 25^th percentile (lower quartile).

The survey response data do not include specific plan identifiers that could be used to link the survey data to the service plan data described in the section above. Therefore, to derive weighting data, BLS developed a process for summarizing the billed amounts provided in the survey response data into useful groupings. These groupings are based on the type of plan contract (postpaid or prepaid), the number of telephone lines serviced per plan (1-5 for postpaid and 1-2 for prepaid), the amount of high-speed data included with the plan (no data, less than 10GBs, 10GBs or more, unlimited), and the type of service provider (MNO or MVNO). The results of this calculation are group level shares of consumer-reported expenditure by the type of service provider.

The last step to prepare the survey data for linkage to the service plan data is to create provider-specific expenditure shares for the MNO providers. To do this, an adjustment is made to the group-level expenditure totals based on the market shares of each MNO provider. MVNO providers are considered as a block representing a share of total expenditures. The results of this adjustment are group level shares of consumer-reported expenditure by MNO providers and a block representing all MVNO providers.

To illustrate these results, consider a scenario where a total of $1,000,000 in expenditure is reported for all survey responses. In this example, MNO Provider A has 60 percent market share, MNO Provider B has 30 percent, and MVNO providers collectively account for 10 percent. Before adjusting the data for market share, a group level share of 20 percent was calculated for postpaid plans servicing one line with unlimited data through MNO providers. Table 1 below illustrates the calculation of deriving the final shares of total expenditure spent on postpaid plans servicing one line with unlimited data for each of the three hypothetical providers. The group expenditure shares are recalculated every two years.

Table 1: Example of group level expenditure share derivation
Provider	Market Share	Group Level Share (Postpaid, 1-Line)	Equation	Group Expenditure Share
MNO – A	60%	20%	($1,000,000 X 0.6 X 0.2)	$120,000 (or 12% of total)
MNO – B	30%	20%	($1,000,000 X 0.3 X 0.2)	$60,000 (or 6% of total)
MVNOs	10%	20%	($1,000,000 X 0.1 X 0.2)	$20,000 (or 2% of total)

BLS combines the weighting data with the price and characteristics data by determining which expenditure share group each service plan maps to (based on provider, contract type, number of serviced lines, and amount of included data) and then dividing the expenditure share of each group equally among all service plans mapped to that group. To illustrate this, if the expenditure share group described in the previous paragraph as having 12 percent of the weight has four service plans that map to it, each of those four service plans will receive a final expenditure share of 3 percent. The plan-level expenditure shares are recalculated each month to account for changes in service plan availability.

Index methodology

Prior to the release of July 2025 indexes in August 2025, the BLS methodology for calculating the wireless telephone service price index included the application of hedonic quality adjustments only when a service plan was replaced by a newer plan. In contrast, the methodology employed starting in July 2025 is not reliant on item replacement. All service plans in the monthly data are used to build a predictive regression on price, which serves as the foundation of index calculation. Thus, quality improvements introduced by new service plans are immediately and fully integrated into estimates of constant-quality price change.

The use of hedonic imputation methods was first explored for this item out of the prospect of accounting for the effect of unobserved product characteristics in the price estimations of exiting products. Unobserved characteristics are features that improve product quality but cannot be measured (and thus cannot be included in a hedonic regression on price).

Erickson and Pakes (2011)^[1] propose that it is possible to predict the effect of unobserved product characteristics by using observable characteristics, thereby reducing the omitted-variable bias inherent in some hedonic regressions. This approach is appropriate for many high-tech products; however it relies on the prices of continuing goods varying over time, and many telecommunications services have price changes primarily at the time of product turnover. Therefore, the methodology suggested by Erickson and Pakes is not recommended for wireless telephone services. Instead, a less complex hedonic imputation index methodology, as described by Pakes (2003)^[2], was adopted for wireless telephone services. The general form of the log-level hedonic regression model can be specified as shown in equation 1.

Equation 1: Log-level hedonic regression model

where Z_kis a vector of observable characteristics for product k. The function h_t() is estimated with a weighted least squares regression. This hedonic equation varies over time and is estimated each month, allowing the function to detect changes in consumer valuations of the characteristics of wireless telephone services. BLS analysts evaluate the hedonic regression models used in this process using statistical tests and knowledge of the economics of the market.

Blending the plan price data with the estimated expenditure share weights data allows for the use of weighted regression functions. The effect of using expenditure share-weighted hedonic regressions is that the service plans with the highest expenditure share (the most popular items) are emphasized. The weighted regression model more accurately maps the relationship between prices and characteristics for the mix of service plans purchased by consumers. With this function, hedonic imputation methods (namely, using observable characteristics to impute the missing prices for entering and exiting goods) can be used to correct for quality change introduced by product turnover. See Appendix C of the Monthly Labor Review article “Alternative data sources for high-tech products in the CPI” for more information about the hedonic regression model used in the research that motivated this change to the CPI for wireless telephone services.

Using the weighted least squares regression, prices are predicted for all service plans in both the current month (t) and previous month (t−1). Additionally, the methodology is extended to predict prices for plans that were not observed - the current month’s prices for exiting plans and the previous month’s prices for entering plans. After predicting prices for time t and t−1, price relatives are calculated and aggregated into an index using a Törnqvist index formula. Plan level weight shares are assigned to the service plans for both the current month (t) and previous month (t−1). In months when a service plan was not observed, a weight of zero is assigned. The plan’s average weight share is used to weigh the plan’s price relative in the aggregation. Since there is no regional variation in the prices of wireless telephone service plans, a single national index is calculated using equation 2.

Equation 2: Törnqvist index formula

The resulting hedonic index is derived completely from prices predicted by hedonic regression models.

Taxes, surcharges, and fees

To account for taxes, surcharges, and fees, the national index is replicated to the 32 basic CPI areas and a tax rate specific to wireless telephone services is applied to the untaxed index relatives, or the ratio of the current month’s index to the previous month’s index. For this process, BLS uses tax information gathered by a publicly available, third-party source. This source publishes an annual summary of taxes and fees for wireless services, including average wireless-specific tax rates for both federal and state/local taxes and fees. A process of population weighting the state rates is used to calculate an average rate for CPI areas that cross state lines. A similar process of population weighting is used to derive average rates for the nine CPI areas that are aggregates of smaller metropolitan areas. Tax data are updated annually in January.

Finally, BLS aggregates the taxed area relatives using area weights provided by the Consumer Expenditure (CE) survey to create the U.S.-level index.

Sources

[1] Erickson, Tim, and Ariel Pakes. 2011. "An Experimental Component Index for the CPI: From Annual Computer Data to Monthly Data on Other Goods." American Economic Review 101 (5): 1707–38.

[2] Pakes, Ariel. 2003. "A Reconsideration of Hedonic Price Indexes with an Application to PC's." American Economic Review 93 (5): 1578–1596.

Last Modified Date: July 15, 2025