Klaus D. Goepel, BPMSG, Contact, last update: Sep 4, 2017
382 thoughts on “Feedback”
Hello Mr. Klaus,
Introduce, my name is Lince Terang. I am now doing my undergraduate thesis about decision making. As you are very familiar with AHP, I want to ask you, “Can I use AHP with opinions from consumer’s preferences?” Using opinions from consumers later will influence the number of decision makers to become more than 10 decision makers. I plan to have 100 respondents via questionnaire. Thank you in advance..
Dear Lince,
yes, you can use AHP to get opinions from consumer’s preferences. If the number of respondents is 100+, you need to do the consolidation separately; my template only supports 20 max. You might also use my free AHP online software. I am still working on it, but it already allows for group inputs.
Regards, Klaus
Hello sir, due to my theses supervisor, I have to prove that AHP can be used by 100+ respondents, but I got difficulty to find the prove. Do you have any reference books / papers those state that AHP can be used in getting opinions from consumer’s preferences, especially for 100+ respondents?
Thanks a lot for your help Sir.. Have a nice day 🙂
Dear Klaus,
The online software is user friendly. i had a question regarding the inconsistency in AHP. Actually i am getting inconsistency more than 0.1 after using the geometric mean as described by saaty. further can i use those priorities derived out of matrix (with inconsistency) to run a regression analysis.
Dear Santhosh,
Thanks for your feedback. Regarding consistency, please read my post here. It really depends on the actual values of CR and the objectives of your project. Regarding regression analysis: once you have the priorities, you can run any kind of analysis. Also here it depends what is your objective? Just keep in mind that priorities are resulting from a decision problem and are based on individual judgments. Therefore the underlying distribution is not a normal distribution, and many statistical methods are based on the assumption of a normal distribution. That’s the reason I use Shannon entropy and not variance analysis for the consensus indicator.
Regards, Klaus
Dear Mr. Klaus,
I was reviewing the AHP process (Fuzzy AHP in particular–after i saw your video tutorial and read your supporting docs) as a decision support in evaluating criteria for coming up attibutes to consider in developing a new product and I am really in high spirit of having found a decision support tool which may be easily (I hope) understood by not so mathematically inclined managers until i came upon this pdf by Lewis Warren (Uncertainties in the AH process). Its a declassified australian department of defence evaluation of AHP as a decision support tool (DSTO-TN-0597 with url http://www.dsto.defence.gov.au/corporate/reports/DSTO-TN-0597.pdf). He pointed two problems which somehow cast doubt to the robustness of AHP in a way; One is that the “ratio scale” used as proposed by Saaty is not really a ratio scale in a sense ‘that the the linguistic or numerical measures applied are on ordinal scales. So, A/B= 5 can not mean A=5B unless units are assigned…ordinal measures, and this implies that the eigenvalue polynomial computation is inadmissible’. Secondly, ‘normalisation of the weight and alternative preference vectors causes anomalies…, and is one reason for rank reversal.’
Could you please comment on at least these two points (i think there were 5 points raised in the conclusion part ) ?
i am a management professional with physical science background but definitely not a math guru. Much literature has been posted on the rank reversal issue, but the first claim on ratio of scale interpretation is quite disturbing. Please, correct me if i read it wrong.
I hope you had a happy holiday, and very much would wait for your response.
Dear John,
from the same department of defense you find another article from Minh-Tuan Nguyen, comparing different Prioritization Methods incl. AHP. His conclusion: “There is probably no “best method” for the prioritisation process …”, and “The nature of the problem to be tackled will determine the most appropriate method.” About AHP: “AHP may have a major impact on the understanding by the participants of the factors which influence the value of a project. Besides a broad agreement on the ranking of projects, the process provides a profound insight in the art of complex decision making and encourages the participants to pool their knowledge and expertise.”
To this I fully agree; AHP is a decision supporting tool, it does not make the decisions. I know the discussion about rank reversal and the scale, there is a lot of literature published. Therefore I implemented different scales in my template, but finally, for my own projects, I usually come back to the fundamental 1- 9 AHP scale. It seems to be the best compromise between outcome and consistency. I cannot follow the argument of Lewis Warren that the ratings are ordinal measures. But I’m also not a Math guru, just a practitioner. I used AHP in many projects:
It forces you to think through the problem and to break it down in categories and structure it.
It helps to gain a better understanding of, and insight into the decisions problem.
From project stakeholders I get fast agreement on proposed hierarchies and criteria weightings when using AHP.
I get fast consensus in group decision making.
I have a consistent and rational basis to make decision finding more transparent and explainable.
This is exactly, what I expect from a MCDM tool. The absolute numbers or percentages of a weights and alternatives only play a secondary role, they always depend on how you structure the problem, how many criteria and alternatives you consider etc. So for me, this is a quite academic discussion.
Regards and thanks for you feedback, Klaus
Hi Klaus: few questions
1. What would be the number of respondents for a stakeholder group? What would it (the number of respondents) depend on?
2. If you could elaborate a little on the significance of the consensus indicator when the AHP survey is administered amongst different stakeholder groups: the consensus can be less (~50%) considering they are likely to have differing opinions on a topic. and since they actually belong to different groups, assigning the respondents post facto into subgroups of their functions in the organization does not appear to add too much value (http://bpmsg.com/feedback/comment-page-2/#comment-1151).
Thanks in advance, S Roy
Hi SRoy,
ad 1: the number of participants depends on many factors, and has nothing to do with AHP. The project and its stakeholders, political and other circumstances can play a major role. Like in any group decision making, its good to have a group not too large (you will have difficulties to come to a consensus), and not too small (so that all major interests are represented). My experience: around 5 to 10 is not too bad.
ad 2: Luckily people are diverse: even if you group participants according to their functions, interests or expertise, it does not mean consensus within their groups they belong to. Therefore, to analyze consensus always makes sense. The outcome in my specific project was not clear a-priory, because I didn’t select the participants according to their functions. Only the analysis afterwards showed the coincidence with their functions and helped to better understand their motivation in setting the priorities.
Regards, Klaus
Hi Klaus:
Thanks for the input: 5-10 respondents for each stakeholder group.
I am not able to understand the significance of consensus in stakeholder analysis through AHP. If you could suggest some reading material for the same in the context of AHP stakeholder analysis.
Thanks for your help, and regards.
S Roy
Thanks, for the great presentation of Conjoint Analysis.
It is great job to explain a topic of this sort in les than 10 minutes.
Ismail Erdem
Baskent University-Ankara/Turkey
I’m doing my thesis at the Universidad Nacional Agraria la Molina located in Lima, Peru based on a company data with a 3 top managements that want to make the right decision of hiring the best laywer.
Using the AHP excel with 3 differents participants – consolidation, I have a question: Which formula do you use to make the Matrix located in the Summary Sheet? and Which formula is the correct to get the percentage of the normalized principal eigen vector in the same matrix? I’ll really appreciate your help and advice.
Hi Marco,
I use the geometric mean of all participants’ matrix elements to get the consolidated matrix. The normalized principal eigenvector is calculated using the power method with a fixed number of 12 iterations. The description and formulas are given in my PDF document on the templates download page.
Regards, Klaus
Thanks, but consolidate the matrix, the first iteration, the sum of the ranks always divided by 10 but if I only have three criteria should not divide by 3?
Marco, I don’t really understand what you are referring to. The eigenvector calculation? Yes, there I always use 10×10, but it will converge to the correct result after a few iterations.
Hello Mr. Klaus,
Introduce, my name is Lince Terang. I am now doing my undergraduate thesis about decision making. As you are very familiar with AHP, I want to ask you, “Can I use AHP with opinions from consumer’s preferences?” Using opinions from consumers later will influence the number of decision makers to become more than 10 decision makers. I plan to have 100 respondents via questionnaire. Thank you in advance..
Dear Lince,
yes, you can use AHP to get opinions from consumer’s preferences. If the number of respondents is 100+, you need to do the consolidation separately; my template only supports 20 max. You might also use my free AHP online software. I am still working on it, but it already allows for group inputs.
Regards, Klaus
Thank you Mr. Klaus
Hello sir, due to my theses supervisor, I have to prove that AHP can be used by 100+ respondents, but I got difficulty to find the prove. Do you have any reference books / papers those state that AHP can be used in getting opinions from consumer’s preferences, especially for 100+ respondents?
Dear Lince,
one example is this articel, another (60+ participants) here.
Thanks a lot for your help Sir.. Have a nice day 🙂
Dear Klaus,
The online software is user friendly. i had a question regarding the inconsistency in AHP. Actually i am getting inconsistency more than 0.1 after using the geometric mean as described by saaty. further can i use those priorities derived out of matrix (with inconsistency) to run a regression analysis.
Dear Santhosh,
Thanks for your feedback. Regarding consistency, please read my post here. It really depends on the actual values of CR and the objectives of your project. Regarding regression analysis: once you have the priorities, you can run any kind of analysis. Also here it depends what is your objective? Just keep in mind that priorities are resulting from a decision problem and are based on individual judgments. Therefore the underlying distribution is not a normal distribution, and many statistical methods are based on the assumption of a normal distribution. That’s the reason I use Shannon entropy and not variance analysis for the consensus indicator.
Regards, Klaus
Dear Klaus
Thanks for your suggestion. I want to use the AHP prioritized values as independent variables to see its influence on dependent variable.
Even i have structured the AHP problem to suit my objective
Tks for this powerful hardware !!
Thanks for this great info. I’ll be checking back peridodically!
Dear Mr. Klaus,
I was reviewing the AHP process (Fuzzy AHP in particular–after i saw your video tutorial and read your supporting docs) as a decision support in evaluating criteria for coming up attibutes to consider in developing a new product and I am really in high spirit of having found a decision support tool which may be easily (I hope) understood by not so mathematically inclined managers until i came upon this pdf by Lewis Warren (Uncertainties in the AH process). Its a declassified australian department of defence evaluation of AHP as a decision support tool (DSTO-TN-0597 with url http://www.dsto.defence.gov.au/corporate/reports/DSTO-TN-0597.pdf). He pointed two problems which somehow cast doubt to the robustness of AHP in a way; One is that the “ratio scale” used as proposed by Saaty is not really a ratio scale in a sense ‘that the the linguistic or numerical measures applied are on ordinal scales. So, A/B= 5 can not mean A=5B unless units are assigned…ordinal measures, and this implies that the eigenvalue polynomial computation is inadmissible’. Secondly, ‘normalisation of the weight and alternative preference vectors causes anomalies…, and is one reason for rank reversal.’
Could you please comment on at least these two points (i think there were 5 points raised in the conclusion part ) ?
i am a management professional with physical science background but definitely not a math guru. Much literature has been posted on the rank reversal issue, but the first claim on ratio of scale interpretation is quite disturbing. Please, correct me if i read it wrong.
I hope you had a happy holiday, and very much would wait for your response.
Thank You in advance,
JL Willauer
Dear John,
from the same department of defense you find another article from Minh-Tuan Nguyen, comparing different Prioritization Methods incl. AHP. His conclusion: “There is probably no “best method” for the prioritisation process …”, and “The nature of the problem to be tackled will determine the most appropriate method.” About AHP: “AHP may have a major impact on the understanding by the participants of the factors which influence the value of a project. Besides a broad agreement on the ranking of projects, the process provides a profound insight in the art of complex decision making and encourages the participants to pool their knowledge and expertise.”
To this I fully agree; AHP is a decision supporting tool, it does not make the decisions. I know the discussion about rank reversal and the scale, there is a lot of literature published. Therefore I implemented different scales in my template, but finally, for my own projects, I usually come back to the fundamental 1- 9 AHP scale. It seems to be the best compromise between outcome and consistency. I cannot follow the argument of Lewis Warren that the ratings are ordinal measures. But I’m also not a Math guru, just a practitioner. I used AHP in many projects:
This is exactly, what I expect from a MCDM tool. The absolute numbers or percentages of a weights and alternatives only play a secondary role, they always depend on how you structure the problem, how many criteria and alternatives you consider etc. So for me, this is a quite academic discussion.
Regards and thanks for you feedback, Klaus
Hi Klaus: few questions
1. What would be the number of respondents for a stakeholder group? What would it (the number of respondents) depend on?
2. If you could elaborate a little on the significance of the consensus indicator when the AHP survey is administered amongst different stakeholder groups: the consensus can be less (~50%) considering they are likely to have differing opinions on a topic. and since they actually belong to different groups, assigning the respondents post facto into subgroups of their functions in the organization does not appear to add too much value (http://bpmsg.com/feedback/comment-page-2/#comment-1151).
Thanks in advance, S Roy
Hi SRoy,
ad 1: the number of participants depends on many factors, and has nothing to do with AHP. The project and its stakeholders, political and other circumstances can play a major role. Like in any group decision making, its good to have a group not too large (you will have difficulties to come to a consensus), and not too small (so that all major interests are represented). My experience: around 5 to 10 is not too bad.
ad 2: Luckily people are diverse: even if you group participants according to their functions, interests or expertise, it does not mean consensus within their groups they belong to. Therefore, to analyze consensus always makes sense. The outcome in my specific project was not clear a-priory, because I didn’t select the participants according to their functions. Only the analysis afterwards showed the coincidence with their functions and helped to better understand their motivation in setting the priorities.
Regards, Klaus
Hi Klaus:
Thanks for the input: 5-10 respondents for each stakeholder group.
I am not able to understand the significance of consensus in stakeholder analysis through AHP. If you could suggest some reading material for the same in the context of AHP stakeholder analysis.
Thanks for your help, and regards.
S Roy
Hi S Roy,
when you google for AHP consensus or group decision making you will find a lot of articles in the web. Regards, Klaus
Thanks for your help, Klaus
Klaus
This is very well constructed teaching lesson, Deeply Appreciate your sharing the video and the model, which will bey very useful for teaching.
Best rgds
Thanks, for the great presentation of Conjoint Analysis.
It is great job to explain a topic of this sort in les than 10 minutes.
Ismail Erdem
Baskent University-Ankara/Turkey
Dear Ismail, thanks for your positive feedback, I appreciate.
Regards, Klaus
HELLO
I AM IRANINA
TANKYO VERY MUCH FOR TIHUS FILE
Dear Mr. Klaus,
I’m doing my thesis at the Universidad Nacional Agraria la Molina located in Lima, Peru based on a company data with a 3 top managements that want to make the right decision of hiring the best laywer.
Using the AHP excel with 3 differents participants – consolidation, I have a question: Which formula do you use to make the Matrix located in the Summary Sheet? and Which formula is the correct to get the percentage of the normalized principal eigen vector in the same matrix? I’ll really appreciate your help and advice.
Best regards,
Marco Loaiza
Hi Marco,
I use the geometric mean of all participants’ matrix elements to get the consolidated matrix. The normalized principal eigenvector is calculated using the power method with a fixed number of 12 iterations. The description and formulas are given in my PDF document on the templates download page.
Regards, Klaus
Thanks, but consolidate the matrix, the first iteration, the sum of the ranks always divided by 10 but if I only have three criteria should not divide by 3?
Marco, I don’t really understand what you are referring to. The eigenvector calculation? Yes, there I always use 10×10, but it will converge to the correct result after a few iterations.