Regression Events

Regression events are an IPMVP recommended method to account for unplanned or non-recurring fluctuations in energy usage that falls outside the scope of a performance contract. These events can be anything from unplanned occupancy changes, renovations/remodels, or natural disasters.

EnergyTracer accounts for these events in performance analysis by taking a factor or constant value adjusting the monthly usage by this factor. Scale factors affect most values by simply multiplying the factor by the event value. Constant factors only affect total usage for bills and may affect other values, like cost or GHG, depending on how they are calculated (more about that can be found below).  These factors should be used sparingly and with confidence that the factor accurately reflects the change in energy usage. These factors should not be used to correct routine energy fluctuations (like winter or summer holidays), instead a production regressor should be tried first.

Regression Events belong to an 'event collection' and this whole collection is applied to a regression model. It is recommended to have one collection per regression but this is not required. If you are confident the collection is applicable to multiple regressions, it may be used this way. 

Creating collections is done on the analysis tab under the 'Regression Events' section. First give the collection a name (name does not affect anything it is just an identifier). Then add the scale and constant factors that are applicable. Events can be both for the baseline and performance period and either on expected or performance data, more information about these categories below. Scale factors are applied first then constant factor (example: y=mx+b). Once the collection is created it can be added to a new regression model in the model setting are by the date range select. 

After the collection is added to a regression in can be edited anytime, expect for data in that regression's baseline period. If data is edited within the baseline period for any regressions using this collection, a warning will be displayed and the regression will be removed from any projects it is a part of. 

Regression Events on Actual Data

'Actual Data' refers to any data that is taking right off a bill. This data is used for two main purposes: 1. Creating a regression model with baseline actual data 2: Calculating savings with actual data in the performance period. 

1. Regression events used on actual baseline data only affect usage. This is because this is the only value that is relevant for calculating the model.
2. Regression events can affect usage, cost, and demand for actual performance data but it matters which type of savings method is being used. If you wish to affect data in the performance period for the purpose of adjusting savings it is recommended to do this by using regression events on expected data. This is because when adjusting performance values of actual data cost and GHG may not change in the way that is desired based on how they are calculated. 
   1. Normalized savings
      1. Usage is directly affected by both scale and constant factors.
      2. Cost is indirectly affected by scale and constant factors as it is just usage multiplied by a rate.
      3. GHG is directly affected by scale factors and not at all by constant. Constant factor don't work because GHG is not purely calculated from total cost. 
      4. Demand is only affected by scale factors. 
   2. Performance Savings
      1. Usage is directly affected by both scale and constant factors.
      2. Cost is directly affected by scale and not by constant factors.
      3. GHG is directly affected by scale factors and not at all by constant. Constant factor don't work because GHG is not purely calculated from total cost. 
      4. Demand is only affected by scale factors. 

Regression Events on Expected Data

'Expected Data' refers to data that is calculated from a regression model and is used to establish an adjusted baseline. This data is used to for calculating savings in the performance period and has no affect in the baseline period or if no regression is used for savings calculations. 

1. Usage is directly affected by both scale and constant factors 
2. Cost is indirectly affected by scale and by constant factors as it calculated from expected usage which is affected by both these factor. 
3. GHG is indirectly affected by scale and by constant factors as it calculated from expected usage which is affected by both these factor. 
4. Demand is only  affected by scale factors.