ENERGY STAR Score for Retail Stores in Canada

---

Overview

The ENERGY STAR score for Retail Stores applies to both retail stores and wholesale club/supercenters. The objective of the ENERGY STAR score is meant to fairly assess how a property’s energy use measures up against similar properties considering the climate and business activities. A statistical analysis of the peer population is performed to identify the aspects of property activity that are significant drivers of energy use and to normalize for those same factors. The result of this analysis is an equation that predicts the energy use of a property, based on its business activities. This prediction is compared to the property’s actual energy use to yield a 1 to 100 percentile ranking in relation to the national population of properties.

• Property types. The ENERGY STAR score for retail stores applies to two property types: retail stores and wholesale club/supercenters. The score applies to individual stores and is not available for entire strip malls or enclosed malls. To be eligible for an ENERGY STAR score, a retail store must be a single store that is at least 464.5 m² (5,000 ft²) and has an exterior entrance to the public.
• Reference data. The analysis for retail stores in Canada is based on data from the Survey on Commercial and Institutional Energy Use (SCIEU), which was commissioned by Natural Resources Canada (NRCan) and carried out by Statistics Canada, and represents the energy use for the year 2014.
• Adjustments for weather and business activity. The analysis includes adjustments for:
  • Worker density (the number of workers per 100 m²)
  • Percentage of the building that is heated
  • Climate (using heating degree days, retrieved based on postal code)
  • Weekly operating hours
  • Total length of commercial refrigeration/freezer units
  • Whether or not the building is large
• Release date. This is the original release of the ENERGY STAR score for retail stores in Canada.

This document details the development of the 1 – 100 ENERGY STAR score for retail store properties. For more information on the methodology used to develop ENERGY STAR scores, go to the Technical Reference for the ENERGY STAR Score (PDF, 709 KB).

The following sections explain how the ENERGY STAR score for retail stores is developed:

• Overview
• Reference Data & Filters
• Variables Analyzed
• Regression Equation Results
• Energy star Score Lookup Table
• Example Calculation

Reference Data & Filters

The reference data used to form the peer property population relies on the Survey on Commercial and Institutional Energy Use (SCIEU), which was commissioned by Natural Resources Canada and conducted by Statistics Canada in late 2015 and early 2016. The energy data for the survey was from the calendar year 2014. The raw collected data file for this survey is not publicly available, but a report providing summary results is available on Natural Resources Canada’s website at: Survey of Commercial and Institutional Energy Use (SCIEU) - Buildings 2014 – Data Tables.

Four types of filters are applied to analyze the building energy and operating characteristics in the survey. They are set to define the peer population for comparison and to overcome any technical limitations. Those filters are: Building Type Filters, Program Filters, Data Limitation Filters, and Analytical Filters.

A complete description of each category is given in the Technical Reference for the ENERGY STAR Score (PDF, 709 KB). Figure 1 summarizes each filter used to develop the ENERGY STAR score for retail stores model and the rationale that supports the filter. After all filters are applied, the remaining data set has 117 observations. Due to the confidentiality of the survey data, NRCan is not able to identify the number of cases after each filter.

Of the filters applied to the reference data, some result in constraints on calculating a score in Portfolio Manager, and others do not. Building Type and Program Filters are used to limit the reference data to include only properties that are eligible to receive a score in Portfolio Manager and are therefore related to eligibility requirements. In contrast, Data Limitation Filters account for limitations in the data available during the analysis, but do not apply in Portfolio Manager. Analytical Filters are used to eliminate outlier data points or different subsets of data and may or may not affect eligibility. In some cases, a subset of the data has a different behaviour from the rest of the properties (e.g. retail stores smaller than 464.5 m² do not behave the same way as larger buildings), in which case an Analytical Filter is used to determine eligibility in Portfolio Manager. In other cases, Analytical Filters exclude a small number of outliers with extreme values that skew the analysis, but do not affect eligibility requirements. A full description of the criteria you must meet to obtain a score in Portfolio Manager is available at Benchmarking - Frequently Asked Questions.

Related to the filters and eligibility criteria described above, another consideration is how Portfolio Manager treats properties situated on a campus. The main unit for benchmarking in Portfolio Manager is the property, which may be used to describe either a single building or campus of buildings. The applicability of the ENERGY STAR score depends on the type of property. For retail stores, the score is based on individual establishments and is not available for entire strip malls or enclosed malls.

Eligible store configurations include: free standing stores; stores located in open air or strip centers (a collection of attached stores with common areas that are not enclosed); and mall anchors (stores located in enclosed malls with exterior entrances).

Retail configurations not eligible to receive an ENERGY STAR score include: enclosed malls; individual stores located within enclosed malls without exterior entrances; lifestyle centers; strip malls; and individual stores that are part of a larger non-mall building (i.e. office or hotel). To receive an ENERGY STAR score, a retail store must be a single store that is at least 464.5 m² (5,000 ft²) with an exterior entrance to the public.

Variables Analyzed

To normalize for differences in business activity, NRCan performed a statistical analysis to understand what aspects of building activity are significant with respect to energy use. The filtered reference data set, described in the previous section, was analyzed using a weighted ordinary least squares regression, which evaluated energy use relative to business activity (e.g. number of workers, operating hours per week, floor area, and climate). This linear regression gives an equation used to compute energy use (also called the dependent variable) based on a series of characteristics that describe the business activities (also called independent variables). This section details the variables used in the statistical analysis for retail stores in Canada.

Dependent Variables

The dependent variable is what NRCan tries to predict with the regression equation. For the retail store analysis, the dependent variable is energy use, expressed in source energy use intensity (source EUI). This is equal to the total source energy use of the property divided by the gross floor area. The regression analyzes the key drivers of source EUI—those factors that explain the variation in source energy use per square metre in retail stores. The units for source EUI in the Canadian model are annual gigajoules per square metre (GJ/m²).

Independent Variables

The reference survey contains numerous property operation questions that NRCan identified as likely to be important for retail stores. Based on a review of the variables found in the reference data, and following the criteria for inclusion in Portfolio Manager,^{Footnote 1} NRCan initially analyzed the following variables in the regression analysis:

• Gross floor area (m²)
• Cooling degree days (CDD)
• Heating degree days (HDD)
• Percentage of floor space that is cooled
• Percentage of floor space that is heated
• Weekly hours of operation
• Number of workers during the main shift
• Length of all open/closed refrigeration/freezer units
• Area of walk-in refrigeration
• Number of vending machines
• Months in operation in 2014
• Number of commercial appliances
• Presence of a commercial kitchen
• Number of computers
• Number of cash registers
• Number of televisions/electronic displays/LCDs
• Year of construction

NRCan, with the advice of the Environmental Protection Agency (EPA), performed an extensive review on all of these operational characteristics individually and in combination with each other (e.g. heating degree days times percent heated). As part of the analysis, some variables were reformatted to reflect the physical relationships of building components. For example, the number of workers on the main shift can be evaluated in a density format: workers per 100m². The worker density (as opposed to the gross number of workers) is more closely related to the energy use intensity. In addition, using analytical results and residual plots, variables were assessed using different transformations (such as the natural logarithm, abbreviated as Ln). Overall, the analysis consists of multiple regression formulations, structured to find the combination of statistically significant operating characteristics that explained the greatest amount of variance in the dependent variable: source EUI.

The final regression equation includes the following variables:

• Number of workers per 100m² during main shift (worker density)
• Number of Heating Degree Days times the percentage of the building that is heated (PercHeat x HDD)
• Weekly operating hours (For large buildings)
• Length of Commercial Refrigeration/Freezer units per 100 m² (For large buildings)
• Larger than 750 m² Yes/No

These variables are used together to compute the predicted source EUI for retail stores. The predicted source EUI is the mean EUI for a hypothetical population of buildings that share the same values for each of these characteristics. It is the mean energy for buildings that operate like your building.

Worker Density Analysis

Worker density is an essential variable in the retail store model, as it was found to best represent levels of business activity. While cash register density and computer density can represent sales activities, worker density was found to best predict the EUI and explain variance in these variables.

Climate (HDD and CDD) Analysis

The analysis looked at the heating degree days, cooling degree days, percent of the building that is heated, and percent of the building that is cooled.

There was a strong positive correlation between the EUI and both the number of heating degree days (HDD), as well as the percentage of the building that is heated (PercHeat). The combination of these two terms (PercHeat x HDD) created a more robust variable that accounted for variations in percentage of the building heated and climate.

The climate analysis also included a review of two cooling terms: the percentage of the building that is cooled and cooling degree days (CDD). However, neither term nor any combination of terms containing percent cooled or CDD had a statistically significant relationship with EUI.

Large Yes/No

Through model analysis, it was observed that smaller retail stores did not behave in a similar predictable fashion as larger retail stores. The EUI of stores larger than 750m² was also influenced by the number of weekly operating hours, as well as the length of refrigeration and freezers. Therefore, buildings larger than or equal to 750 m² receive an adjustment for hours, length of refrigeration/freezers, and size.

Testing

NRCan further analyzed the regression equation using actual data entered in Portfolio Manager. In addition to the SCIEU data, this analysis provided another set of buildings to examine the ENERGY STAR scores and distributions to assess the impacts and adjustments. It also confirmed that there are minimal biases when it comes to fundamental operational characteristics, such as worker density or percent heated, and that there was no regional bias or bias for the type of energy used for heating.

It is important to reiterate that the final regression equation is based on the nationally representative reference data from SCIEU 2014, not on data previously entered into Portfolio Manager.

Regression equation results

The final regression is a weighted ordinary least squares regression across the filtered data set of 117 observations. The dependent variable is source EUI. Each independent variable is centred relative to the weighted mean value, presented in Figure 2. The final equation is presented in Figure 3. All variables in the regression equation are considered significant at a 90% confidence level or better, as shown by their respective significance levels.

The regression equation has a coefficient of determination (R²) value of 0.314, indicating that this equation explains 31.4% of the variance in source EUI for retail stores. Because the final equation is structured with energy per unit area as the dependent variable, the explanatory power of the area is not included in the R² value, and thus this value appears artificially low. Recomputing the R²value in units of source energy^{Footnote 2} demonstrates that the equation actually explains 97.5% of the variation in total source energy of retail stores. It is an excellent result for a statistically based energy model.

For detailed information on the ordinary least squares regression approach, go to the Technical Reference for the ENERGY STAR Score (PDF, 709 KB).

Figure 3 – Final Regression Results

Energy Star Score Lookup Table

The final regression equation (presented in Figure 3) gives a prediction of source EUI based on a building’s operating characteristics. Some buildings in the SCIEU data sample use more energy than predicted by the regression equation, while others use less. The actual source EUI of each reference data observation is divided by its predicted source EUI to calculate an energy efficiency ratio:

An efficiency ratio lower than one indicates that a building uses less energy than predicted, and consequently is more efficient. A higher efficiency ratio indicates the opposite.

The efficiency ratios are sorted from smallest to largest, and the cumulative percent of the population at each ratio is computed using the individual observation weights from the reference data set. Figure 4 presents a plot of this cumulative distribution. A smooth curve (shown in orange) is fitted to the data using a two-parameter gamma distribution. The fit is performed in order to minimize the sum of squared differences between each building’s actual percent rank in the population and each building’s percent rank with the gamma solution. The final fit for the gamma curve gives a shape parameter (alpha) of 11.29 and a scale parameter (beta) of 0.0864. The sum of the squared error for this fit is 0.485.

The final gamma shape and scale parameters are used to calculate the efficiency ratio at each percentile (1 to 100) along the curve. For example, the ratio on the gamma curve at 1% corresponds to a score of 99; only 1% of the population registers such a small or even smaller ratio. The ratio on the gamma curve at the value of 25% corresponds to the ratio for a score of 75; only 25% of the population registers such a small or even smaller ratio. Figure 5 shows the complete score lookup table.

Figure 5 – ENERGY STAR Score Lookup Table for Retail Store

Example Calculation

According to the Technical Reference for the ENERGY STAR Score (PDF, 709 KB), there are five steps to compute a score for retail stores. Below is a specific example:

1 User enters building data into Portfolio Manager

• 12 months of energy use information for all energy types (annual values, entered in monthly meter entries)
• Physical building information (size, location, etc.) and use details describing building activity (hours, etc.)