Sensitivity analysis

[ghost]

Enhancement request: Is there any interest in adding one or more methods for performing sensitivity analysis on the results? The paper A Sensitivity Analysis Approach for Some Deterministic Multi-Criteria Decision Making Methods, for example, presents a few techniques.

[gluc]

Yes, that's been requested before. I am very open for pull requests!

[ravikiranrv]

Hi, The shinny AHP app is a very good tool for solving AHP Problems. It would be also great if we can add sensitivity analysis on the results which can make this package a complete suite for decision making.

[ghost]

Prototyping some ideas here. These manipulate the trees differently from the rest of package code, and they could be nonsense, so it is not contribution-ready.

# Brainstorming some sensitivity charts
library(dplyr)
library(magrittr)
library(tidyr)
library(ggplot2)
library(ahp)

# reuse the car example
ahpFile <- system.file("extdata","car.ahp",package="ahp")
carAhp <- Load(ahpFile)
Calculate(carAhp)

# variant a
# the Hurley method scales the preferences by some coefficient
# priorities obtained from scaled pairwise weights
coefficient_range <- seq(0.1,2.0,0.1)
original <- carAhp$preferences$children$DecisionMaker$pairwiseMatrix$preferences

df <- bind_rows(sapply(coefficient_range, function(j) {
  weights <- original ^ j
  priorities <- PrioritiesFromPairwiseMatrixEigenvalues(weights)
  data.frame(Coefficient=j,t(priorities$priority))
},simplify=FALSE))

# reorganize for plotting
tdf <- df %>% gather(Preference,Weight,-Coefficient)

p <- ggplot(tdf,aes(x=Coefficient,y=Weight,color=Preference)) +
  geom_line() +
  scale_y_continuous(limits=c(0,1)) +
  geom_vline(xintercept=1.0,linetype="dashed") +
  labs(title = "Priorities from Pairwise Matrix Eigenvalues",
       caption = "Hurley-method sensitivity analysis",
       x=expression(paste("Coefficient ",alpha)))
p

# variant b
# manipulating the decision-maker preferences
# scaled coefficient range reused
# not strictly a Hurley variant because it can result in rank reversal
# reload the model 
carAhp <- Load(ahpFile)
Calculate(carAhp)
original <- carAhp$preferences$DecisionMaker$pairwise$preferences$preference

df <- bind_rows(sapply(coefficient_range, function(j) {
  scaled <- original ^ j
  carAhp$preferences$DecisionMaker$pairwise$preferences$preference <- scaled
  Calculate(carAhp)
  adf <- Analyze(carAhp,decisionMaker="Total",variable="weightContribution",sort="orig")
  adf %<>% select(-1,-Weight)
  if ( "Inconsistency" %in% colnames((adf)))
    adf %<>% select(-Inconsistency)
  data.frame(Coefficient=j,adf[1,])
},simplify=FALSE))

# reorganize for plotting
tdf <- df %>% gather(Priority,Weight,-Coefficient)

p <- ggplot(tdf,aes(x=Coefficient,y=Weight,color=Priority)) +
  geom_line() +
  geom_vline(xintercept=1.0,linetype="dashed") +
  labs(title = "Priority by Scaled Pairwise Preferences",
       y="Weight Contribution",
       x=expression(paste("Coefficient ",alpha)))
p

[juntianusj]

Hello,

Is there any development on sensitivity to this package? I would like to see if there is any.

Thanks