In EMIjess/evalwaterfallr: Rearranges Multiplicative Factors into an Order-independent Waterfall

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)
library(dplyr,warn.conflicts = FALSE)
library(ggplot2)
library(scales)
source('~/Documents/R_working/evalwaterfallr/R/waterfallPlot.R')
source('~/Documents/R_working/evalwaterfallr/R/waterfallPrep.R')
source('~/Documents/R_working/evalwaterfallr/R/addwaterfallPrep.R')
source('~/Documents/R_working/evalwaterfallr/R/wParamPermute.R')

Motivation

This package provides a simple visualization tool for the display of energy efficiency program evaluation results as argued by @kasmanetaliepec2015. Differences between claimed (ex ante) and evaluated (ex post) savings are most informative and actionable if the reasons behind discrepancies are quantified and reported. A best practice for reporting variations in ex ante and ex post savings is the use of dimensionless multiplicative impact parameters, such as an Hours of Use (HOU) factor that gives the ratio of ex post HOU to ex ante HOU. When impact parameters are reported in this manner, a powerful visualization of quantified discrepancies is made possible. However, the magnitude of a savings discrepancy associated with a given impact parameter is dependent upon the other impact parameters. The interdependence of impact parameter savings adjustments also creates an issue of order-dependence when constructing a graphical representation of ex ante/ex post adjustment. The permutation procedure in @kasmanetaliepec2015 ensures that graphics are order-independent and normalized.

From that paper, consider a lighting program with the following key values provided:

Gross.XA <- 100 # reported Gross Savings (arbitrary units)
NTG.XA <- 0.8 # 90% NTG reported
Net.XA <- Gross.XA*NTG.XA
Net.XA #80
NTG.XP <- 0.6 # 80% NTG found in evaluation
Gross.XP <- 50 # Gross Savings found in evaluation (arbitrary units)
Net.XP <- Gross.XP*NTG.XP
Net.XP #30

The following impact parameters are reported:

# define impact parameters
HOU <- 0.7 # Hours of lighting use (HOU_expost/HOU_exante) smaller as evaluated
deltaWatts <- 1.14 # difference in lighting watts higher as evaluated
ISR <- 0.63 # Installation Rate (IRR_expost/IRR_exante) smaller as evaluated

Here we develop waterfall graphics to quantify the ex ante/ex post savings adjustments as well as net to gross (NTG) adjustments.

Simple Visualization of Non-Permuted Values

First, we create a waterfall plot that presents an order-dependent application of impact parameters.

# ensure that we have a dataframe ready to put into waterfallPrep()
myparamdf <- data.frame( # lighting example
                          params = c("HOU","deltaWatts","ISR"),
                          value = c(0.7, 1.14, 0.63),
                          stringsAsFactors = FALSE
                         )
lighting_given <- waterfallPrep(myparamdf, 
                                gross.report=100, NTG.report=0.8, NTG.eval=0.6,
                          altparamnames = NULL,
                                output="none") # none means no permutation
lighting_given

waterfallPlot(lighting_given)

Upon changing the order of the impact parameters, the associated step sizes also change.

# change the dataframe  to put into waterfallPrep()
myparamdf <- data.frame( # lighting example WITH CHANGED ORDER
                          params = c("deltaWatts","ISR","HOU"),
                          value = c(1.14, 0.63, 0.7),
                          stringsAsFactors = FALSE
                         )
lighting_given <- waterfallPrep(myparamdf, 
                                gross.report=100, NTG.report=0.8, NTG.eval=0.6,
                          altparamnames = NULL,
                                output="none") # none means no permutation
lighting_given

waterfallPlot(lighting_given)

Permuted Values

For an accurate representation of step size that solves the order-dependence issue, a complicated mathematical procedure is required in which the average of all possible permutations is taken for each impact parameter. This package greatly simplifies creating permuted values (compared to doing the matrix calculations in a spreadsheet application, such as MS Excel). The function waterfallPrep() calculates the tables for no permutation (as shown above), gross permutation, and net permutation all as detailed in @kasmanetaliepec2015

lighting_all <- waterfallPrep(myparamdf, 
                          gross.report=100, NTG.report=0.8, NTG.eval=0.6,
                          altparamnames = NULL,
                                output="all")
lighting_gross <- lighting_all[[2]]
lighting_net <- lighting_all[[3]]

waterfallPlot(lighting_gross) # gross permutation plot
waterfallPlot(lighting_net) # net permutation plot

These graphics are now order-independent and show accurate savings step sizes in both the gross and net domains.

References

EMIjess/evalwaterfallr documentation built on May 6, 2019, 3:09 p.m.