Frequently Asked Questions

Weather, Fuel Price, and Emissions Data

Utility Bill Data
  • How should I enter delivered fossil fuels (e.g., fuel oil, propane, kerosene) into BETTER if I only know the delivery dates and not the consumption dates?

    In many cases, buildings use natural gas as well as delivered fossil fuels, such as fuel oil. In those cases, we suggest "allocating" the delivered fuel to mimic the natural gas consumption profile in the building, so the billing start and end dates for the delivered fossil fuel will mimic the billing start and end dates of natural gas. Alternatively, if natural gas is not used by the building, you can approximate the billing start and end dates based on the fuel delivery dates.

  • Can I use BETTER if I don’t know my monthly utility costs?

    BETTER works best when a user enters 12 consecutive months of energy usage and cost (for all fuels used in a building). However, if a user does not have monthly energy cost information, BETTER will estimate the monthly energy cost (based on monthly usage) for the following fuel types according to the fuel unit price information contained in the linked source files:

  • Can BETTER read smart meter data?

    At this time, BETTER can only intake monthly energy usage data (broken down by fuel type). If you have smart meter data that quantifies energy usage at more granular intervals (e.g., hourly, daily, weekly), you will need to aggregate it to monthly energy usage. We are investigating adding capability to BETTER to read smart meter data and will announce it in our News section when available.

  • In which currencies can I input and view energy costs and cost savings in BETTER?

    At this time, BETTER allows a user to input energy costs in the currencies linked here.
    Note: Selecting a currency only enables BETTER to display costs in the selected currency, but not convert currency from one to another.

  • How does BETTER handle negative monthly energy consumption values?

    At this time, BETTER will reject negative energy consumption values and won't generate a change-point model and assess energy efficiency measures (EEM) for that building. BETTER requires 12 consecutive months of positive energy consumption values to assess EEMs.

  • How does BETTER break utility bills into calendar months?

    BETTER calendarize utility bills by allocating the energy consumption by calendar months. Specifically, it allocates the energy consumption in a utility bill to each calendar month based on the number of days in each month. For example, if a utility bill covers the period from January 15 to February 15, BETTER will allocate 16/31 of the energy consumption to January and 15/31 of the energy consumption to February. Then, utility bills of the same type (e.g., electricity) are aggregated by calendar month. The subsequent steps of BETTER analytics are performed on the aggregated calendarized utility bills.

    BETTER model

Building Spaces Types
  • Which building space types does BETTER support?

    BETTER can be used to analyze the performance of any commercial building type for which a user can enter all required data (e.g., gross floor area, type, location, 12 consecutive months of energy use for all fuels) for at least 30 buildings. The following commercial building types are most frequently input into BETTER for analysis: bank branch, courthouse, data center, distribution center, financial office, hospital (general medical and surgical), hotel, K-12 school, multifamily housing, non-refrigerated warehouse, office, refrigerated warehouse, retail store, senior care community, supermarket/grocery store, public library. For definitions of these space types, click here. At this time, BETTER can also be used to analyze the performance of a single building, or a portfolio of less than 30 buildings, for the U.S. office, multifamily residential, public libraries, and K-12 school space types; Mexican offices; and Tunisian hotels.

  • How can I analyze a mixed use building using BETTER?

    First, see if you can determine the building’s primary space type. This is the space type that accounts for more than 50% of the building. For example, if you enter a building that has an office that accounts for 60% of the gross floor area (GFA) (excluding parking) and a retail store that accounts for 40% of the GFA, then the primary space type will be office. This primary space type is used to determine the peer group for benchmarking in BETTER. If no individual space type accounts for more than 50 % of the building space, then it is mixed use. To evaluate energy savings for mixed use buildings in BETTER, we recommend determining the size and monthly energy usage for each space in the building and benchmarking each of these spaces separately.


Analytical Settings
  • What does R2 mean? What is the recommended R2 level?

    R-squared (R2) is a goodness-of-fit measure for regression models. It quantifies the proportion of the variability for a dependent variable that's explained by an independent variable, or variables, in a regression model. In the case of BETTER, R2 indicates to what extent variations in outdoor air temperature explain variations in building energy use intensity. An R2 of 1 means that all changes in energy use intensity is completely explained by changes in outdoor air temperature. In general, we recommend users select an R2 of 0.6 or higher when creating a portfolio or building analytics.

  • Why do my energy savings and energy efficiency recommendations change when I change the R2 level?

    When running BETTER analytics, the R2 threshold sets the minimum goodness-of-fit requirements for change-point models. As a result, when running a portfolio analytics, different R2 thresholds may lead to different results:

    • Lower R2 threshold (e.g., 0.4): More change-point models may be found because of the less rigorous goodness-of-fit requirement. In this case, BETTER will be able to estimate the savings and recommend energy efficiency measures for more buildings, which leads to higher savings estimations at the portfolio level.
    • Higher R2 threshold (e.g., 0.8): Fewer change-point models may be found because of the more rigorous goodness-of-fit requirement. In this case, BETTER will be able to estimate the savings and recommend energy efficiency measures for fewer buildings, which leads to lower savings estimations at the portfolio level.

    In general, we recommend users select an R2 of 0.6 or higher for a portfolio or building.

  • How do the savings targets (e.g., aggressive, nominal, conservative) in BETTER differ?

    When analyzing a portfolio or building, BETTER offers users the ability to select an energy savings target based on the scale of savings desired. Users can select from the following targets each time they run an analytical report for a portfolio or a building:

    • Conservative: The savings target will be one standard deviation worse than the median performance of the benchmarking peer group.
    • Nominal: The savings target will be equal to the median performance of the benchmarking peer group.
    • Aggressive: The savings target will be one half standard deviation better than the median performance of the benchmarking peer group.
  • What is the difference between the “reference” and “generate” benchmark statistics options in BETTER?

    The “reference” benchmark statistics method allows a BETTER user to benchmark a commercial building’s inverse model coefficients (i.e., electric and fossil baseloads, heating and cooling change-points, and heating and cooling slopes) against the median and standard deviation for each of these inverse model coefficients derived from a training dataset. Alternatively, the “generate” benchmark statistics method allows a user with a portfolio of commercial buildings (recommended to be at least 30) of a single space type to benchmark "internally." Thus, instead of comparing a building’s inverse model coefficients to the median and standard deviation of the training dataset, BETTER generates a unique set of statistics (i.e., the median and standard deviation for their own portfolio’s electric and fossil baseloads, heating and cooling change-points, and heating and cooling slopes) based solely on the user’s input portfolio - so the user is essentially comparing buildings within their own portfolio to identify cost-saving energy and emission reduction opportunities. We recommend that users include at least 30 buildings of a single space type when using the “generate” benchmark statistics method of analysis.

  • How are U.S. “reference” benchmark statistics developed?

    We take the following steps to develop U.S. “reference” benchmark statistics for building space types:

    1. Assemble a clean and robust building dataset for a space type that is: (a) composed of building size, location, and 12 consecutive months of energy consumption (all fuels); (b) representative of the U.S. national stock (in terms of climate zone and size distribution) according to the most recent U.S. Commercial Building Energy Consumption Survey (CBECS) or U.S. Residential Energy Consumption Survey (RECS); and (c) provides at least 30 data points for each of the 10 BETTER model coefficients. Note: In the future, for more robust statistics, we aim to have at least 30 data points for each of the 10 BETTER model coefficients for each of eight CBECS size categories in each of the eight International Energy Conservation Code (IECC) climate zones in the United States (and possibly for each of the relevant subtypes A, B, and C in these zones). To contribute anonymous data to this effort, please email support@better.lbl.gov.
    2. Develop two change-point models for each building in the dataset (i.e., one for electricity and one for fossil energy).
    3. Extract change-point model coefficients (i.e., electric and fossil baseloads, heating and cooling change-points, and heating and cooling slopes).
    4. Fit normal distributions to extracted change-point model coefficients.
    5. Evaluate the quality of the distributions of change-point model coefficients utilizing statistical tests (e.g., Kolmogorov–Smirnov (KS) test, which measures how well the normal distribution fits to the change-point model coefficient data points).
    6. Repeat steps 1 to 6 until each distribution is sufficiently robust.
    7. Once the distributions are sufficiently robust, we integrate the U.S. reference statistics (i.e., the median and standard deviation for each inverse model coefficient distribution) into BETTER for field validation.
    8. Please click on the link to see an overview of the U.S. benchmark statistic development process and current statistics.
  • How are international “reference” benchmark statistics developed?

    We take the following steps to develop international “reference” benchmark statistics for building space types:

    1. Assemble a clean and robust building dataset for a space type that is: (a) composed of building size, location, and 12 consecutive months of energy consumption (all fuels); (b) representative of the national stock (in terms of climate zone and size distribution); and (c) provides at least 30 data points for each of the 10 BETTER model coefficients. Note: In the future, for more robust statistics, we aim to have at least 30 data points for each of the 10 BETTER model coefficients for each climate zone in the country. To contribute anonymous data to this effort, please email support@better.lbl.gov.
    2. Develop two change-point models for each building in the dataset (i.e., one for electricity and one for fossil energy).
    3. Extract change-point model coefficients (i.e., electric and fossil baseloads, heating and cooling change-points, and heating and cooling slopes).
    4. Fit normal distributions to extracted change-point model coefficients.
    5. Evaluate the quality of the distributions of change-point model coefficients utilizing statistical tests (e.g., Kolmogorov–Smirnov (KS) test, which measures how well the normal distribution fits to the change-point model coefficient data points).
    6. Repeat steps 1 to 6 until each distribution is sufficiently robust.
    7. Once the distributions are sufficiently robust, we integrate the reference statistics (i.e., the median and standard deviation for each inverse model coefficient distribution) into BETTER for field validation.
    8. Please click on the link to see an overview of the international benchmark statistic development process and current statistics.
  • For which space types does BETTER have “reference” benchmark statistics?

    We have developed “reference” benchmark statistics for U.S. offices, K-12 schools, libraries, and multifamily buildings; Mexican offices; and Tunisian hotels. We are currently working to develop “reference” benchmark statistics for U.S. libraries. Please contact us at (support@better.lbl.gov) if there is an additional space type for which you would like us to develop “reference” benchmark statistics. Click on this ( link ).

  • Are the U.S. “reference” benchmark statistics representative of the U.S. national stock?

    At this time, the U.S. “reference” benchmark statistics are not perfectly representative of the U.S. national stock because the statistics were not developed from the U.S. Energy Information Administration (EIA) Commercial Building Energy Consumption Survey (CBECS) nor the U.S. EIA Residential Energy Consumption Survey (RECS) datasets, which are representative of the U.S. national stock. Instead, these statistics were developed utilizing training datasets developed based on voluntary contributions from U.S. industry that are not fully representative of the U.S. national stock in terms of characteristics such as size and climate zone distribution. We are working to expand these training datasets, and hence improve associated “reference” benchmark statistics, so they are more representative of the U.S. national stock. This includes expanding the training data sets to include: at least 30 data points for each of the 10 BETTER model coefficients for each of the (CBECS size categories) in each of the (eight International Energy Conservation Code (IECC) climate zones) in the United States (and possibly for each of the relevant subtypes A, B, and C in these zones). To contribute anonymous data to this effort, please email (support@better.lbl.gov). Additionally, at this time, “reference” statistics are only available for U.S. offices, K-12 schools, libraries, and multifamily buildings, but we are gradually adding other space types (beginning with U.S. multifamily). To know more about the benchmark statistics development process, click on this (link).

  • How many buildings should be included in a single portfolio when using the “generate” benchmark statistics option?

    Use of BETTER’s analytical engine with over 700 buildings globally indicates that a portfolio size of 30 buildings is generally sufficient in order to “generate” benchmark statistics and benchmark a portfolio “internally” (i.e., compare buildings within a portfolio and not against “U.S. reference” benchmark statistics to identify cost-saving energy and GHG emission reduction opportunities). This means that a portfolio size of 30 allows BETTER to generate two inverse models for each building in the dataset (one electric and one fossil); extract inverse model coefficients; anf fit a normal distribution to the extracted inverse model coefficients that passes the KS test). This being said, we know that some building portfolios are highly uniform in terms of the use of electricity and fossil energy for baseload, heating, and cooling, and a smaller portfolio of 20-25 buildings could generate robust benchmarking statistics for "internal” benchmarking. In other cases, where portfolios have missing data points or variations in outdoor air temperature do not significantly explain variations in energy use intensity, then a portfolio size of 30-50 buildings may be needed to deliver viable statistics for "internal” benchmarking. However, overall work to-date on BETTER’s analytical engine for 700 buildings suggests that 30 buildings “generates” benchmark statistics sufficient to identify buildings in a portfolio with energy saving potential and the set of energy efficiency measures that should be implemented to capture these savings. See the links below for case studies of portfolios that benchmarked “internally” to reduce energy usage and costs.

  • Does BETTER perform analysis utilizing the metric or imperial system of units?

    BETTER offers analysis in both unit systems, and a user can easily switch between the two unit systems in the BETTER web app. Go to the upper right corner of the upper navigation bar of the web app to select the unit system in which you want to enter/view data in BETTER. To view/enter data in the metric or international system (IS) of units (kWh, square meters, °C), select SI. To view/enter data in the imperial system (IP) of units (kBtu, square feet, °F), select IP. You can change the system in which you view/enter data in BETTER at any time by changing the unit system in the upper right corner of the navigation bar.

  • Why can't BETTER sometimes identify a change-point model for my building(s)?

    Change-point model fit is influenced by various factors, including data availability, weather-sensitivity of the building's energy consumption, and operating schedule, etc. Typical reasons for no model-fit include: lack of sufficient energy consumption data (i.e., less than 12 consecutive months); the energy use intensity (EUI) for the building is not weather-sensitive; the energy consumption data entered includes data for the COVID shutdown period and is, therefore, irregular.

    According to our analysis of all user-entered anonymous data by the end of 2023, the change-point model-fit rate varies by space type, location, and the time of usage (e.g., pre-COVID vs. COVID). The heatmaps below compares the model-fit rates of buildings of different space types and climate zones, where each cell shows a "space type & climate group." To avoid randomness and ensure reasonable comparisons, at least 30 buildings with at least 12-months of consecutive utility bills are required in each group to be included. Each cell shows the number of buildings with enough utility bill data and the percentage of buildings that have a valid change-point model (i.e., 1P, 3P or 5P). It can be seen that the model-fit rates are generally lower if all data, including the COVID period data, is used. Using pre-COVID data increased the model-fit rates in most cases. This is because some buildings' operating schedules changed drastically after COVID, whereby the buildings' energy consumption patterns were no longer correlated with weather. For more details, please refer to the link .

    Model-fit Statistics
  • What should I do if BETTER cannot fit a change-point model for my building?

    If a model can't be fit on the first try for a building, consider doing one of the following and re-running the analytics:

    1. Review the Building Information Tab: Utility Consumption Trends chart. Ensure 12 consecutive months of electricity and/or fossil energy consumption data is entered for the building.
    2. Review the Building Information Tab: Utility Consumption Trends chart. Remove the consecutive months of data that have abnormal patterns (e.g., extremely low or high energy use intensity (EUI), sudden changes).
    3. Remove COVID period energy consumption data.
    4. Reduce the number of years of data used to fit a model (generally one to two years is sufficient).
    5. Lower the minimum R² value on the Portfolio Analytics Setting page. (See more details on "What does R2 mean? What is the recommended R2 level?")

Output Reports
  • Does BETTER estimate the potential savings of implementing a single energy efficiency measure?

    At this time, BETTER cannot provide estimated energy, cost, and GHG emissions reductions associated with implementing a single energy efficiency measure (e.g., envelope retrofit). However, BETTER will identify which energy efficiency measures (from a total of 15 measures, ranging from lighting upgrades to envelope upgrades) your building could benefit from, and the total energy, cost, and GHG emissions reductions that could be achieved from implementing all of BETTER’s recommended energy efficiency measures in your building or portfolio.


  • Why are the annual energy consumptions on the output reports different from the entered values?

    BETTER utilizes data-driven techniques to analyze a building's monthly energy consumption in response to weather conditions. Change-point models are fitted between daily energy use intensity (EUI) and outdoor air temperature to determine the weather-sensitive and weather-independent components of electricity and fossil fuel energy consumption, respectively. The energy consumption values reported are predictions generated by the change-point models, which may differ from the actual entered values. The figure on the right displays an example change-point model, where the black dots represent the actual EUI, and the orange dots represent the predicted EUI. Generally, a better change-point model results in smaller discrepancies between the entered and predicted energy consumption values.

    BETTER model

  • How are the annual energy savings estimated?

    Continuing from the previous example, the figure demonstrates how change-point models are utilized to predict the current and target levels of EUIs. These values are then used to estimate energy consumption and calculate savings. The solid lines in the figure depict the current change-point model fitted using actual data, while the dashed lines represent the target change-point model, which shows the improvement in coefficients to reach the target level. In this example, the baseload is reduced, the cooling change-point is shifted to the right, and the cooling slope is made less steep. The current level EUIs (represented by orange dots) and target level EUIs (represented by green dots) are predicted using the actual monthly average outdoor air temperatures. Finally, the predicted EUIs are utilized to estimate annual energy consumption and calculate savings using the formula below (note that the summation is over the last 12 months):

    Σ[(EUIcurrent - EUItarget) × floor_area × number_of_days]
    BETTER model

Miscellaneous