← Back to Electricity Price Hub

Data Collection and Estimation Methodology

This document details the data collection, normalization, and estimation processes used to compile the estimates presented in the Heatmap-MIT Electricity Price Hub.

I.Data Scope and Sources

The Electricity Price Hub includes data for electric utilities across all 50 states and Washington DC. Specifically, it features estimates of monthly average residential rates and monthly average residential bills from January 2020 through the most recently concluded month. For the majority of utilities serving at least five percent of residential customers in a given state, the Hub also provides an estimated breakdown of monthly rates and bills by charges related to generation, distribution, and transmission, along with a category for other charges.

Data is collected from one of three sources, depending on utility size and information availability: (1) Official reporting by state regulators or agencies (official state sources), (2) Directly from individual utility rate books, (3) Energy Information Agency (EIA) Form EIA-861M.

For 14 states, we source data on rate-regulated utilities from official reporting by official state sources. Regulators or government agencies in these states publish rate and/or bill data for the electric utilities within their jurisdiction–in the form of annual bill or rate survey reports and data dashboards. Those sources include: (1) rate or bill information specific to residential customers, (2) sufficient historic data (going back to at least January 2020), and (3) information disaggregating total rates or bills by major components or specific charges to support our estimates. Appendix 1 lists the 14 PUC states and specific sources used.

For the remaining 36 states and Washington DC, rate data for utilities that serve at least five percent of a state’s residential customers, according to the EIA’s 2024 Annual Electric Power Industry Report (EIA-861), is drawn directly from individual utility rate books. In some limited cases, particularly for non-regulated utilities in these states, data on historic rates is unavailable.

We then supplement data collected from official state reporting and utility rate books with estimates sourced from EIA-861M. We supplement our data collected from official state sources and utility rate books with estimates from EIA-861M. Specifically, we estimate average price and average bill for any utilities included in EIA’s monthly reporting and not covered by data collected from the above sources.

In addition to the above sources of data on utility residential rates, we rely on:

• Form EIA 861-M to estimate average residential usage and month-to-month variation in rates
• Form EIA-861 for state customer count data
• Demand data reported by Independent System Operators (ISOs) and Regional Transmission Organizations (RTOs) as detailed in Appendix 2, and
• Other state-specific sources to produce estimates of monthly average rate and bill, as detailed in Section IV

II.Collecting Rate and Bill Data

For data collected from utility commission reporting, the form and precise content of rate and/or bill data varies from one state to the next, but we collect the data on residential rates or bills as presented by the state. The following sections detail the process applied to normalize the data and produce a standard set of estimates for the full time period covered.

For data collected from utility rate books, we first identified the utility’s standard or default residential rate plan, along with all additional charges applicable to residential customers taking service under that plan. We then constructed a monthly rate history for all applicable charges from January 2020 to present. For rates currently in effect, we use tariff documents published on utility webpages. For historic rates, we compiled historic rate data from tariff repositories published on utility webpages, for those utilities that publish cancelled ratebooks. Otherwise, we requested historic rates data from utilities or relied on regulatory dockets to identify historic rates.

For states with retail competition, we collected data on supply rates under each utility’s standard offer or basic generation service or closest equivalent. For the parts of Texas that are open to competition, the data includes residential rates charged by each of the investor-owned transmission and distribution utilities (TDU), along with an estimated average supply charge, weighted by customer count, for a representative selection of retail electric providers operating in each TDU service territory.

For data collected from EIA-861M, we use reported residential sales, revenue from residential sales, and residential customer count to estimate an average price and bill for each utility and month.

III.Disaggregating Rates by Key Function

Having collected residential rate data from January 2020 to present, we then categorize charges or rate components as related to generation, distribution, or transmission. Charges designed to recover costs not directly tied to one or more of those three categories are categorized as “other.” Additional detail on the types of charges included in the other category is provided below.

We first identified all function-specific charges–charges that, based on definitional language in tariff sheets and accompanying information on how rates are computed, relate entirely or predominantly to one category. Each of these charges is then tagged based on the relevant function. Examples of function-specific charges:

• Generation: Fuel Cost Recovery charge or Nuclear Construction Cost Recovery Schedule
• Distribution: Underground Project Charge
• Transmission: Transmission Rate Adjustment

We then estimate the appropriate breakdown of charges or rate components that cover multiple categories. For base rate charges that are not unbundled by function, this process begins by identifying the general rate cases in which base rates in effect from January 2020 were decided. We then used class cost-of-service studies filed in those cases to identify or, where not directly provided, estimate the functionalized revenue requirement allocated to residential customers. We then sort the functionalization used in the COSS into our four categories and calculate the proportions attributable to each of the four categories. We follow a similar approach to estimate the breakdown of other charges that cover multiple categories, identifying utility regulatory filings detailing the associated revenue requirement and estimating a functional breakdown.

For rate-regulated utilities in the 14 states for which we rely on official state reporting, some of those underlying sources provide functional breakdowns of bills or rates, while others break out individual charges. We adopt commission functionalization for reporting or data disaggregating rates by function. For charge-level data, we apply the same approach outlined above.

There is, of course, enormous variation in the structure of residential rates and number and type of related charges. In all cases, we seek to isolate rate components directly related to generation, transmission, and distribution using the best available information. In almost all cases, this sorting then leaves a set of rate components that do not fit squarely within one of those functional categories. We label these as “other” in the data, and while there are a set of very common categories of charges in this bucket, the composition of this rate component varies across states and utilities.

The most common types of charges in the other category include:

• Charges related to taxes and regulatory fees,
• Charges or credits reflecting changes in tax policy,
• Surcharges supporting low-income and affordability programs or discounts
• Charges covering energy efficiency programs
• Charges covering environmental compliance or remediation costs not associated with a specific generation asset
• Broad-ranging societal benefits or economic development charges
• Charges covering pension costs or insurance premiums

As noted in Section I, we sought to estimate rate breakdowns for all utilities serving at least five percent of residential customers. However, where the information needed to disaggregate non-function-specific charges is not available, we do not attempt to estimate the rate and bill breakdowns. This is most often the case for non-regulated utilities, but there are also a small number of investor-owned utilities where information gaps prevent functional disaggregation.

IV.Normalizing Data and Estimating Average Rates and Bills

Having collected data from January 2020 to present and categorized rate components, we follow a consistent set of processes for normalizing the data to calculate average rates and rate components ($/kWh) and average bills ($/month) for each utility and month.

The applicable normalization processes, outlined below, differ depending on the raw data source, but the basic mechanics of estimating rates and bills applies across the full data set. Specifically, we identify each utility’s monthly average residential usage, using data reported by individual utilities or in EIA-861M. For estimates using EIA-861M, average residential usage is calculated by dividing sales (MWh) by customer count and multiplying by 1000 to get kWh per customer. To estimate a monthly average rate, we divide any fixed rate charges ($/month) by average use and add the resulting amounts to any variable rate charges ($/kWh). To estimate the monthly average bill, we multiply any variable rate charges by average use (or the specified portion of average use for tiered rates) and add the resulting amounts to any fixed rate charges.

A.Data Collected from Utility Commission and State Government Reporting

Electricity price data collected from official state reporting vary in structure and content. To account for these differences, we apply a set of normalization processes adapted to each state reporting practice to compile data in a standardized form that is consistent across states and with data collected directly from utility ratebooks. These methods account for data reported as: (1) average bills and rate component contributions to average bills rather than average prices, (2) rate or bill data reported as annual rather than monthly averages, and (3) monthly rate or bill data with gaps in time period covered.

Some states report utility data in terms of residential bills (e.g., average bill for residential customers using 500 kWh) and rate component or charge contributions to average bills. To convert bill amounts into price details resembling information in a utility rate book, we collect any fixed rate charges as reported and divide all variable rate charges by the reported usage amount.

For states that report rate and rate component data as annual averages rather than on a monthly basis, we estimate the month-to-month variation in prices by applying the monthly variation in the utility’s average price reflected in EIA-861M. In particular, we:

• Use EIA-861M data on a utility’s monthly residential revenue and sales to calculate an average residential rate for each month in a given year
• Calculate an average monthly rate for that year
• Divide each month’s average rate by the annual average rate to get a ratio reflecting the relationship between that month’s average to the annual average
• Multiply each month’s ratio by the annual average rate and variable ($/kWh) rate components reported by the state to estimate monthly averages
• Repeat for each year in the data set

For states that report monthly data at intervals that leave gaps in the data, we apply the same approach to fill those gaps. In those instances, since an annual average price is not provided, we calculate an annual price by averaging the price across all reported months in the year before multiplying by the ratio of the given month’s average price to the annual monthly average price.

If no data has yet been reported for the most recent year, we need to project each charge’s price for the months in the most recent year. To do this, we first calculate the average price for each of these months using EIA-861M. If there is a time lag in this data, we project the average price for these months from EIA-861M data using the method outlined in Section V.B. We then multiply this data by the average historic ratio of residential price reported by a state regulator or agency divided by the EIA-861M price for each of the last twelve months. This accounts for differences in EIA and official reporting practices. Then we calculate each variable charge’s historic contribution to total price, and multiply these shares by the estimated total price to get each charge value for the given month.

We employ any combination of the above methods specific to the given state’s reporting practices. This normalizes the state’s data into a structure consistent with the raw data collected from utility ratebooks. Thus, following these steps, we employ the same process applied to data collected from utility ratebooks to output finalized data for this set of utilities.

B.Data Collected from Utility Rate Books

1.Standard

For electricity price data collected from utility rate books, with the exception of California and Texas, we follow a standard set of operations to calculate the average price ($/kWh) and average bill ($/month) for each utility at a monthly interval, and where data allows, monthly rate and bill broken down by component (generation, transmission, distribution, other).

We first calculate average price, along with component breakdown of average price. Treatment of individual charges depends on charge type and coverage.

Charge types include variable, fixed, daily, peak monthly use, percentage of base rate charges, or percentage of total price. Variable charges are uniformly collected in $/kWh form and left as is in computing average price. Fixed charges expressed as $/month are divided by a utility’s average residential usage for a given month. Fixed charges expressed as $/day are multiplied by the number of days in a given month and divided by a utility’s average residential usage for that month.

Peak monthly use charges ($/kW-month charges) are fixed charges multiplied by the peak kW usage in the month. To estimate this usage, we use hourly demand data on the grid as reported by the utility’s RTO or ISO (see Appendix 2). We identify the peak hour of demand for the utility’s corresponding load zone and scale it, first to account for the utility’s share of load, and second to isolate residential share of load. To account for a given utility’s share of demand, we take the median ratio of RTO/ISO reported monthly demand to the utility’s monthly total sales as reported in EIA-861M. We then calculate the historic share of the utility’s sales attributed to residential use using EIA-861M. Finally, we multiply the median ratio and average share from above to the peak hourly load amount to get a final peak usage amount (kW) for the month. We multiply this value by the reported peak monthly use charge and then follow the same approach applied to a standard fixed cost.

The final two charge types are both percentage based. After all non-percentage based charges have been normalized into their average price contribution ($/kWH), we sum all base rate and all of the non-base rate charges, respectively. Finally, we calculate the charge percentage of the base rate sum to get the percentage of base rate average charge contribution amount and calculate the percentage of the base rate sum plus the non-base rate sum to get the percentage of total price. If there are multiple percentage based prices, they are taken from these sums independent of each other, i.e., we do not add the percentage based amounts into the total summations.

Having normalized all charges into $/kWh amounts, we then need to determine what coverage of the charge applies to the monthly bill. A charge’s coverage may be restricted by usage tiers, time of use conditions, or a change in rates in the middle of a month.

Tiered charges have varying rates for different portions of a customer’s monthly usage. For these, we calculate each tier’s share of total usage for the month to avoid overestimating an individual charge’s contribution to average price. Note that each tier’s contribution to average price may change each month, depending on average use, even if the charge values in the rate book remain unchanged. It is also possible for tiers greater than the first tier to be zeroed out if average residential use for the month is less than the tiers lower bound.

The dataset also includes time-of-use charges, which vary depending on the amount of usage during specified times. To estimate the share attributed to each time, we use hourly demand data reported by the utility’s RTO/ISO (see Appendix 2). We then calculate for the utility’s corresponding load zone, the proportion of demand for the month occurring during hours designated as on-peak versus off-peak. We apply these same proportions to the on-peak and off-peak charge amounts.

The final constraint on a charge’s monthly coverage is if the charge’s rate changes in the middle of the month. In these instances, we take the proportion of the days in the month in which each charge was in place and calculate a weighted average charge amount based on these proportions.

After calculating each charge’s contribution to average price ($/kWh), we are left with our final charge values. We then group each charge (or charge portion, as discussed in Section III) based on its functional classification (generation, transmission, distribution, and other) and sum to get total component price amounts. Finally, we calculate total average price by summing up each of the component prices.

Once the average price is calculated, we can calculate the average bill ($/month) by multiplying the average price by the given month’s average use. For utilities with component breakdowns, the same approach is applied to each component individually to get the component bill amount.

2.California

Rate data for two of California’s largest utilities, Pacific Gas & Electric Co. (PG&E) and Southern California Edison Co (SCE), is collected from utility rate books following the standard approach, but we adapt our rate and bill estimation processes to account for a unique element of their rate design. In particular, PG&E and SCE divide their customers into distinct baseline allowance zones. Each zone has a unique usage threshold where tiered charges transition from the first tier to the second. To account for this in our dataset, we calculate average price and bill for these two utilities at the zonal level.

PG&E and SCE report monthly usage and customer count at the zip code level. By mapping zip codes to each utility’s zones, we can calculate a total usage and customer count for each zone and month. We can then calculate an average use by dividing total residential usage by residential customer count. Given these reports are released on a time lag of quarterly installments, we produce preliminary estimates of average use via methods outlined in Section V.

Finally, we produce average price and bill data using the standard utility ratebook approach with the average use filled in with the method above and with zonal specific thresholds applied to tiered rates for each zone.

3.Texas

A majority of Texas consumers pay for charges from both a transmission/distribution utility (TDU), which bills standard, state-regulated delivery charges, and a retail electric provider (REP), which is chosen by the consumer.

We collect delivery rate data from TDU rate books following our standard approach. However, because there is no single “standard offer” supply service rate for each TDU service territory and because data on REP rates available from other sources is at the state level rather than the TDU level, we collect data on rates offered by a representative subset of the largest REPs servicing a given TDU. For each, we identify the closest equivalent to a standard rate plan. We then calculate an average supply charge (weighted by customer count) for each TDU. Below are the REPs and rate plans included in these averages.

Table 1. Texas Retail Electric Providers and Rate Plans

We collect data on current rates from “Electricity Facts Labels” found on REP webpages. Data on historic rate data is not consistently reported, but the Public Utility Commission of Texas previously compiled standard reports on REP rates, which we used to source historic rate information. Those reports present average REP rates under the above-specified plans for average monthly usage of 500 kWh, 1000 kWh, and 2000 kWh. For gaps in historic rates, we follow the same approach described in Section IV.A for filling gaps in monthly data reported by state regulators or agencies. Next, we calculate an average REP price for each TDU at each tier level by weighting each REP price for the TDU by the REP’s proportional share of customers within the TDU (Customer count data was provided by ERCOT in response to an information request). This calculates a weighted average supply charge at each tier for the given TDU, which we add to the set of all other charges present in the TDU.

We then need to calculate the average residential use in each TDU service area for each of the REPs in our dataset. We first calculate annual average use for each REP using EIA-861 (note, this is the annual form). We then calculate a weighted annual average use across all REPs by using the same shares of TDU customers used above and divide by 12 to get an average monthly usage for the TDU. To estimate monthly variation in usage, we assume that usage within each TDU service area follows similar monthly variations as average use across all of Texas. We thus calculate monthly variation in average Texas residential usage using EIA-861M. For a given year, we calculate average use for each month in the given year for all of Texas. We then take an annual average of this monthly average usage amount and calculate each month’s ratio of its own average to this annual average monthly usage. This gives each month’s relationship to the average monthly usage for all of Texas. We then multiply each monthly ratio by the annual TDU monthly usage to get the given month’s average residential usage for the TDU.

Given the large time lag in annually reported average usage, we need to project average residential usage for more recent years. To do this, we calculate the TDU’s historic share of residential usage and customers across all of Texas. We then apply these shares to monthly usage and customers for all of Texas as reported in EIA-861M to use as proxies for the TDU’s monthly usage and customers which we then use to calculate monthly average residential usage. For the most recent months where monthly data is not available for all of Texas we calculate preliminary data using the method described in Section V.A.

C.Data Collected from Form EIA-861M

We supplement our data collected from official state sources and utility rate books with estimates from EIA-861M. Specifically, we estimate average price and average bill for any utilities included in EIA’s monthly reporting and not covered by data collected from our primary sources. Average price is calculated by dividing residential revenue by residential sales. Average bill is calculated by multiplying average price by average residential use. We follow the process outlined in Section V.B to produce preliminary estimates of these prices and bills through the most recent months for which EIA-861M has yet to be released.

V.Generating Preliminary Estimates for Most Recent Months

A.Preliminary Average Usage

As described above, we source estimates of a utility’s average residential usage from individual utilities or EIA-861M. However, these sources are reported on a lag. EIA-861M, for example, provides data on a 3-month lag. As a result, to produce estimated rates and bills through the most recently-concluded month, we use the following methods to project average usage for months not yet included in officially-reported data.

At a high level, we calculate preliminary average usage estimates for months not yet covered by official reporting by dividing a utility’s estimated residential sales in each month by its residential customer count. See Appendix 3.A for sample back testing results for the methods included in Section V.

1.Projected Residential Sales

We have two alternative approaches for projecting residential sales. Our default approach relies on hourly data for demand on portions of the U.S. grid as reported by the Independent System Operators (ISOs) and Regional Transmission Organizations (RTOs) that manage those portions of the grid. We rely on this approach wherever possible to reflect real time variations in month-to-month electricity usage that would otherwise not be reflected in historic trends.

Where RTO and ISO hourly demand data is accessible, we match utilities to their corresponding load resource zone or load area and pull the corresponding hourly demand data for the relevant months for which we are projecting average use. Aggregating the demand data for every hour in a given month results in monthly demand on the portion of the grid corresponding to the utility’s service area. To properly apportion this demand to residential usage, we multiply the utility’s historic residential share of total sales by the total demand in the load area for the given month. This equates to the residential demand on the grid for a given utility and month, which we use as a proxy for total residential sales for the projected month.

This method is applied to utilities that we can accurately match to a specific load zone (tested for accuracy by comparing historic projections and actual residential sales and comparing results to alternative approach below) within the Midcontinent Independent System Operator (MISO), Pennsylvania-New Jersey-Maryland Interconnection (PJM), and the Southwest Power Pool (SPP). Appendix 2 lists the utilities that we apply this method to.

For all other utilities, for which hourly demand data is not available, we follow an alternative method for projecting residential sales. This approach is based on the tested assumption (see Appendix 3.B) that sales follow similar seasonal patterns.

This method estimates sales in a given month (e.g., March 2026) based on the historic relationship between a utility’s sales in that month and its sales in the most recent month for which we have officially-reported data (e.g., January 2026). For each year from 2020 through the most-recently concluded year, we calculate the ratio of utility’s residential sales in those two months (e.g., March sales/January sales for each year from 2020 to 2025), we take an average of the ratio in each of the preceding years, and multiply the average ratio by data for the most recent month of reported data. In other words, we calculate expected residential sales for March 2026 based on how March sales typically relate to sales in January of the same year.

2.Projected Residential Customer Count

The method we use for estimating residential customer count for months not yet covered in officially reported data follows the same logic as our alternate method for estimating residential sales–specifically, it relies on the tested assumption that monthly customer counts follow similar seasonal patterns.

This method estimates customer count in a given month based on the historic relationship between the number of customers in that month (X) and customer count in the most recent month with reported data (Y). For each year from 2020 through the most recently concluded year, we calculate the ratio of customer count in X to customer count in Y. We take an average and multiple the average ratio by the most recent reported data. The result is our estimated customer count data for the given month.

We then divide estimated residential sales by the estimated number of residential customers to generate an estimate of average residential usage. It is important to note that these methods are only applied to estimate average use for the most recent months where usage data is not yet available through our standard sources, in most instances this applies to the three most recent months of data. Additionally, we treat the resulting data as preliminary, and in each monthly data update, we replace preliminary data with reported usage figures and adjust estimates accordingly.

B.Preliminary Rate and Bill Estimates

As described above, we supplement data collected from utility commission reporting and utility tariffs with data calculated using EIA-861M for a wider set of utilities, including utilities serving less than five percent of a state’s residential customers and utilities for which gaps in historic rates data prevents direct data collection. However, as noted, EIA-861M releases data on a time lag inconsistent with the rest of our dataset. Thus, we apply a projection method that mirrors those used to estimate average usage detailed above to produce preliminary estimates of average price and bill for months where EIA-861M data is not yet available.

The approach relies on the tested assumption that the metrics used to calculate average price and bill (sales, revenue, and customer count) tend to follow similar seasonal patterns. As above, we estimate residential sales, revenue, and customer count for the most recent months by calculating the ratio of each metric in a given month to that metric in the most recent month reported in EIA-861M for each year from 2020 through the most recently concluded year. We then multiply data for the most recent month included in EIA-861M by the average of the annual revenues to estimate sales, revenue, or customer count for the given month not yet covered in the latest reporting.

As above, it is important to note that this approach only applies to rate and bill estimates for the most recent months for which data on sales, revenue, and customer count are not yet reported. Additionally, it is further limited to those utilities for which data is exclusively sourced from Form EIA-861M and not from utility rate books or utility commission reporting. We treat the resulting data as preliminary, and in each monthly data update, we replace preliminary data with reported usage figures and adjust estimates accordingly.

VI.Displaying Rate and Bill Data

A.Geographic Averages

Our data is collected at the utility service level. Data presented at other geographic levels on the Electricity Price Hub are averages of this utility-level data. At the state level, average price and bill data represent an average of all utilities within our dataset weighted by the utilities residential sales. Residential sales data is either collected from EIA-861M or in special cases, such as for California and Texas, calculated using the methods described in Section IV.B.2 and Section IV.B.3. Preliminary sales data, as described in Section V.A.1, is used to fill lags in reported sales data. All geographies below the state level report average price and bill as a simple average among the utilities present in the geographic area.

B.Mapping Utility Service Territory at each Geographic Level

To map utility-level data to different geographies, we label every zip code in a given utility service territory as reported by Open Energy Information and combine zip codes to form the shape of larger geographic levels.

Acknowledgements

The team would like to acknowledge and thank Daniel Posthumus for his invaluable contributions to the project, along with the following Stanford students for their time and effort supporting project data collection: Raphael Barbier, Delia Brown, Charlotte Cao, Jaxon Gonzales, Brendan Kuo, Gerardo Murga, Rikhil Ranjit, Sophia Rubin, Cherry Sung, Hongze Yao, Hannah Zingapan.

Appendix 1. Official State Reporting: States & Sources

Appendix 2. List of Utilities with Corresponding Load Zones in MISO, PJM, and SPP

Appendix 3. Method Testing

A.Projection Method Backtests

The underlying data sources used to produce our retail electricity price and bills estimates are reported on time lags up to three months. To account for this, we employ a set of methods to project data for these months. The projected data is used to produce preliminary estimates and updated each month as official data is reported.

These projection methods are intended to produce preliminary estimates based on reasoned assumptions using a mix of historical trends and available real-time data. Backtesting confirms that these methods are effective at estimating values in line with the actual values in our dataset. Additionally, the estimation accuracy improves as the time lag decreases and with additional historic data to draw from. Based on these results, each monthly data update improves the accuracy of the preliminary data points.

Figure 1 depicts the accuracy of the method described in Section V.A to project average use based on historical ratios for lags of 1-, 2-, or 3-months.

Figure 2 depicts the accuracy of projecting average use with hourly demand data as described in Section V.A, comparing the results to the historical ratio method for the set of utilities with real-time demand data available.

Figures 3 and 4 depict the accuracy of projecting average residential price and bill for lags of 1-, 2-, or 3-months for the utilities where we estimate data entirely from Form EIA-861M as described in Section V.B.

B.Seasonal Trends

The historic ratio projection method described in Section V relies on the assumption that seasonal trends in residential sales, revenue, and customer count is largely consistent from one year to the next. Figures 5, 6, and 7 test that assumption.

C.Data Source Comparison

We tested rate data collected from utility ratebooks or state government reporting against rate estimates from existing data sources to ensure a level of consistency across sources. Figure 8 depicts data for these utilities from our dataset compared to average price estimates from EIA-861M. This figure shows the two datasets to be heavily correlated.

← Back to Electricity Price Hub