Applying the Churchman/Ackoff Value Estimation Procedure to Spatial Modeling
-
Upload
mechelle-hinton -
Category
Documents
-
view
37 -
download
3
description
Transcript of Applying the Churchman/Ackoff Value Estimation Procedure to Spatial Modeling
Applying the Churchman/AckoffValue Estimation
Procedure to Spatial ModelingSusan L. Ose
MGIS Capstone PresentationPenn State University - World Campus
17 June 2008
GIS Models - A Refresher
• GIS models provide– Decision Support– Prediction– Cost assessment, etc.
• Weighting values– Emphasize/De-emphasize impact of
inputs– Model adjusted via tweaking weights
• How does an analyst determine weights?
Figure source: Joseph K. Berry, University of Denver
Estimating Weights - Gathering Expert Opinions via Delphi Process
?? CompileResults
Experts
Questionnaire
Discussion
• Anonymous Questionnaires
• Statistical aggregation
• Controlled feedback
• Avoids groupthink
• Consensus determines number of rounds
Questionnaire Format
• The ideal questionnaire:– Guides the user through the weight estimation
process– Focuses on expert knowledge - not statistics– Emphasizes relative not numeric values
• Software module– Stores and compiles results– Allows easy distribution to expert group– Can be used in online meeting
Churchman/Ackoff Procedure
• Focuses on relative value of layers – Compares layer against combination of other layers– Uses recursive procedures– Breaks down larger groups of layers for easier
assessment
• Recommended for group decision making• Project adapted method to browser-based
software module
Churchman/Ackoff Procedure Step 1: Rank Layers
Layers are dragged and dropped into desired order
Churchman/Ackoff Procedure Step 2: Assign Initial Weights
Layers divided subdivided into equal groups of no more than four.
Control layer randomly chosen, assigned value of 50, added to each group
Expert inputs initial estimate of weight
Group 1
Group 2
Churchman/Ackoff Procedure Step 3: Judge Importance
Expert chooses one of the three conditions and clicks on it.
Algorithm adjusts values accordingly.
Churchman/Ackoff Procedure :Results Screen
Procedure Test Case
• Potential for rain-fed agriculture in Liberia taking into account cost to market
• Expert panel consisted of MDA Federal, Inc. employees experienced in modeling
• Used two Delphi rounds – Compiled statistics – Ran model using average, high, and low values
Model Diagram
First Round Results• Reference layers received lowest scores• Proximity to water and land layers received highest
scores• Model output using average score and 8 highest
weighted layers consistent with results obtained using different methodology
• Discussion focused on experience with procedure• Panel agreed on top 7 layers to include in final model
Second Round Results• Results reflected post-first round discussion
– Access to water considered key, water layers grouped near top of ranking
– Land cover shows potential land to be transformed to agriculture, thus higher score than previous
• Higher variance in scores than in first round– Probably due to score distribution among less layers – First round lower variance may not have happened if
expert could discard layers
• Model more definitive around water features
Round Comparison
Model Results - Average
2nd round1st round
Does the method work?
• Model result viable • Considering one vs. many values more difficult than
considering one vs. one (pairwise method)– Deliberately designed that way to challenge one's opinions
– May frustrate participants - "weighting fatigue"
• Random groupings confused some participants– Difficult to recall original ranking, weight
– However randomness focuses judgment on subgroup
• Can assist individual in examining own conclusions
What happens when someone joins the group
later?• One expert did not attend discussions, used tool with
minimal guidance• First round results changed slightly• Average Z-score highest of the group
Second Round with Additional Person
• Pronounced difference in results• Greater difference in layer order in mid to lower layers• Average Z-score still highest of the group• Demonstrates importance of group discussion
Z-ScoresNew
expert Expert 1 Expert 2 Expert 3 Expert 4 Expert 5
First Round 1.21 0.45 0.66 0.67 0.65 0.88
Second Round 1.20 0.46 0.83 0.88 0.55 0.77
Further Development/Study
• Improve the software tool– Rework the math algorithm that calculates the weights– Offer opportunity to discard layers– Provide comparison of start/finish results– Provide progress bar
• Application of the method– Categorical values within layers need to be rated first – Orientation meetings important
• Conduct an experiment using both this method and a pairwise approach and compare results
Acknowledgements
• Dr. Gregory Koeln, President, MDA Federal Inc. for sponsoring this project
• Michael Schreiber, Webmaster, MDA Federal Inc. for transforming the procedure from paper to software
• Dr. Todd Bacastow and Dr. Douglas Miller, Penn State World Campus, for their advice and guidance
• Dr. Douglas Way, MDA Federal Inc. for his invaluable guidance
• David Cunningham, Dr. Anna Oldak, Dr. François Smith, and Dr. Andrew Ralowicz, MDA Federal Inc. for their expertise
• Gregory A. Ose, for his unfailing support