Since this is a technical forum and considering the paper that Nikola Kadoic has written and published in RG, I understand that it is subject to discussion.
I understand that Nikola, in this elaborated paper, expresses certain technical concepts with which I do not concur, and that can lead to the practitioner to accept attributes that the method does not posses, and I explained which is the reason of my disagreement.
It is not my intention to discredit this paper, only to point out some aspects, with which I believe are incorrect.
From that paper
NK- ANP is a method that takes into account the most data about decision-making problem compared to other multi-criteria decision-making methods.
NM- I don’t think so. I can post more than 25 occurrences in real-world projects that can’t be modelled in ANP.
NK - By using the ANP, it is possible to model dependencies and feedbacks in the network structure of a
Problem.
NM- This is incorrect. The author’s source of this assertion is a very well-known paper from Saaty, who did not either explain which is feed-back. He only says that there is feedback when two events are mutually linked, that is A to B and B to A. Months ago I posted here in RG the question of what is feed-back according to Saaty, and nobody answered. Why?
Because it does not exist, it is an incorrect appraisal to credit advantages to a method does not have, and with no mathematical support, other that the use of the eigenvector procedure, and that the author of this paper does not explain either.
Feed-back is the reaction of the event B to the action of another event A. The fact that A is linked to B, and B to A, does not mean that there is a reaction from B that is fed-back to A. That is, there is no guarantee that A induces a reaction on B, and that B addresses A about that reaction.
A trivial example: In a road, speed of your car in some way depends on the weather, but weather does not depend on your speed. You can reduce your speed because it started to snow heavily, but that is not a feed back because it does not alter the weather. It is indifferent if you reduce or not your speed
A practitioner said that it exists because the matrix is multiplied by itself many times, and the value in B is added to the value of A, but that is not feed-back; it is just incorporating in A the value of B. In network theory it has a clear explanation and result, related with the distance between two points, but not in ANP.
NK- In general, the term dependency is the opposite to influence
NM- Curiously, the author does not explain this concept. In my understanding there could be dependency with and without influence. A child depends on his parents but perhaps is not influenced by them. Instead, he can have a friend, from who he does not depend, but that greatly influences him.
NK- Also, we do not know the strength (intensity) of this dependency (influence)
NM- The author assumes that strength is the same as intensity and dependency is equivalent to influence
I can understand that he equals strength with intensity but not dependency with influence, both are different things. According to the Dictionary, dependence is: the state of relying on or being controlled by someone or something else.
And the definition of influence is: the capacity to have an effect on the character, development, or behavior of someone or something, or the effect itself.
Now, could someone explain me that if the DM says that considering a certain goal, criterion GDP is 3 times more important than criterion Water contamination, where is the dependency and influence on one on the other?
Does the contamination depend on the GDP? Of course, Not
Is the contamination influenced by the GDP? Of course, Not
There are other aspects on this paper, but I believe that the short comments and citations about that illustrate my point:
There are some assertions in that paper that need to be reviewed and revised, because they are incorrect
Dear Dr Munier
Thank you for your discussion. I am busy and often not available on this network so I didn't reply sofar.
I'll try to explain some of the questionable remarks from my paper.
- ANP is a method that takes into account the most data about decision-making problem compared to other multi-criteria decision-making methods.
... this sentence must be analyzed in the context of the whole paragraph. The research (and the mentioned project) is focused on 'clean' multi-criteria methods (in the narrow sense) in the higher education area, not including LP methods, or uncertainty&risks methods... So, we talk about multi-criteria methods only and when ANP is compared to other methods (Electre, Topsis, Promethee, Evenswaps ...), the ANP takes into account the most information about decision making problem - reference 15 contains the table which presents the parallel analysis of the methods considering the MCDM features. Of course, there are some other methods that can deal with even more information about the alternatives, but they might not be clean MCDM methods in the narrow sense. This was my reasoning process here.
- By using the ANP, it is possible to model dependencies and feedbacks in the network structure of a problem.
... I think that the claim is properly cited, even though it is not described more in detail or discussed if it is correct/wrong. From the paper cited, and from many other prof. Saaty papers, it can be concluded that in ANP, we can model dependencies between the criteria. Also, the main idea of the ANP is to structure the feedback system.
From how I see the feedback - The feedback system is actually a network which is consists of elements: criteria and alternatives - which are grouped into clusters (sets of elements).
The dependency is a situation when certain element somehow depends on the other element (the element "speed" depends on the element "weather").
The feedback is analyzed as a cycle or loop dependency on the cluster level (not on element level).
- (i,j) ... contains information about the dependency of element j on element i
- (j, i) ... contains information about the dependency of element i on element j
- Particularly, (speed, weather)=0, and (weather, speed)=1
- The ANP does not assume that there is a loop or cycle between every two elements or even clusters - but when there is a dependency - we can model it; but when there isn't - we will not say and model that there it is.
- Sometimes, there will be dependencies in both directions.
- In general, the term dependency is the opposite to influence.
.... Indeed, the concepts of dependency and influence are not explained because I considered them as known concepts for the ANP community and the focus of the paper is to discuss some characteristics of ANP (sections 2, 3). Even Dr. Saaty didn't explain them explicitly in his papers. But, he did it implicitly:
Additionally, there are many authors which combined the DEMATEL method with ANP. In DEMATEL, we model "influences" between the elements. In several types of DEMATEL-ANP integrations "influences" are directly transformed into "dependencies" in ANP. One of the best papers (in my opinion) on that topic is [5].
To go back to the supermatrix:
- (i, j) ... contains information about the dependency of element j on element i; if (i, j)=1: element j depends on element i = element i influences the element j
- (j, i) ... contains information about the dependency of element i on element j; if (j, i)=1: element i depends on element j = element j influences the element i
To conclude on this topic:
- The network is consists of elements that are grouped into clusters.
- Some element can influence any other element from the same or other clusters. This means that those "other elements" depend on "some element". This directly means: If the child depends on his parents, the parents influence the child.
- I am aware that the terms "influences" and "dependencies" might have the meanings that will not be directly opposite, but in terms of the ANP, I find them directly opposite.
- Also, we do not know the strength (intensity) of this dependency (influence)
... This claim does not mean that dependency = influence. In particular example, criterion2 influences criterion7. In other words, criterion7 depends on criterion2. So, there is a relation between criterion2 and criterion7 which can be described through the previous two statements which are equivalent in terms of the ANP (1: criterion2 influences criterion7. 2: criterion7 depends on criterion2). This relation (dependency/influence) has some kind of strength. The DEMATEL introduced scale 0-4 for the strength of influences. In ANP, in that particular case (second sub-bullet on pg 242), the supermatrix does not have the information about the strength of the relation.
I hope I clarified some of the questionable remarks. If you have additional questions, let me know.
Best regards,
Nikola
__________
References:
[1] Saaty, T. L., & Vargas, L. G. (2006). Decision Making with the Analytic Network Process: Economic, Political, Social and Technological Applications with Benefits, Opportunities, Costs and Risks. Springer; Softcover reprint of hardcover 1st ed. 2006 edition (December 28, 2009). https://doi.org/10.1007/978-1-4614-7279-7
[2] http://www.cashflow88.com/decisiones/saaty91.pdf
[3] http://www.cashflow88.com/decisiones/saaty92.pdf
[4] Saaty, T. L., & Cillo, B. (2008). A Dictionary of Complex Decision Using the Analytic Network Process, The Encyclicon, Volume 2 (2nd ed.). Pittsburgh: RWS Publications.
[5] Gölcük, İ., & Baykasoğlu, A. (2016). An analysis of DEMATEL approaches for criteria interaction handling within ANP. Expert Systems with Applications, 46, 346–366. https://doi.org/10.1016/j.eswa.2015.10.041
ur discussion - I think the majority of our discussion is related to two aspects: (1) dependency/influence and (2) what is feedback.
(1.) What is a dependency, and what is influence? I said that in ANP they are terms with opposite meanings. Why did I say that prof Saaty implicitly said that? When explaining the ANP, in terms of graphical representation - arrows represent the "dependencies" between the elements but at the same time:
Nolberto
I agree with indicating dependency by arrows, is the usual methods used in networks
- the author introduces the "Influence matrix" which fits the graphical representation of the problem (created using the dependencies) 100%. Table of influences is zero/one table that at position (i, j) contains 1 if there is an influence from element i to element j, and it contains 0 it there is no influence.
I agree with the binary matrix and that (1) means dependency and (0) not. But again, where is the influence? And how is it expressed?
Ex. in the attached demo-example, there is an arrow from co to pa (courseware depends on the number of papers) - in terms of the influence matrix this becomes equal to "number of papers influence the courseware". ... (a similar conclusion mechanism is used in creating the zero-one matrix of this problem) ... This guides us (or at least some of us) to the conclusion that the influences are arrows with opposite directions than dependencies.
As I understand it, papers depend of courseware because they are the way the course is imparted, the more intensive, the more reports and documentation.
Now, how the number of papers may influence the courseware kit? If we have a large amount of papers, that is articles, drawings, reports, etc. how can they influence de courseware? It is like a snake eating its tail!
The 0-1 matrix is different. It clearly expresses that there is dependency only in one sense. Now, if you put 1-1, then there will be dependency in he two senses, but even in this last case, where is the influence?
However, if you tell me that students are asking for some additional subject, that, I can understand, because the courseware kit is modified in order to produce the additional material. If that is the case, yes there is feedback from the students, because it modifies the original number of papers. Is my reasoning correct?
I agree with you about what you say that dependency arrows are opposite to influence arrows, but in your former letter you said that dependency and influence were opposite, and they are not, what is opposite is their graphic representation. Observe that in case of feedback, both, dependency and influence have the same sense
- to fill the data in the supermatrix, the questions that are asked use the word "influence" - not "dependency" - even though the questions are the result of the graphical model created using the "dependencies".
Yes, and for me, that is incorrect.
Ex. From the attached figure we can see that criterion "ci" depends on the criteria "pa" and "pr" (arrows from "ci" to "pa" and "pr").
According to your drawing, c1 depends from pr, and indirectly from pa, through pr
At the step of filling the unweighted supermatrix, (among many comparisons,) the decision will have to compare the criteria "pa" and "pr" with respect to criterion "ci". By ANP theory the question that will be asked is: Given the criterion "ci", which of two criteria, "pa" or "pr", has a greater influence on criterion "ci"?
Agreed, but in this context, the word ‘influence’, is a simple linguistic way and it does not mean that either pa or pr can MODIFY c1. By the way this is one of the problems with pair-wise comparisons, since many times the comparison value depends in the way the question is formulated. Many authors, not me, have addressed this issue.
I think that there are some other ANP author's states which can lead us to the conclusion that "dependency" is in relation to "influence" as "opposite meaning" in the environment of ANP.
As said earlier, in all DEMATEL-ANP integration, and also BSC-ANP researches - in which the influences between the elements are modeled - the influences are directly transformed into ANP models and dependencies changing the direction of arrows in the models.
Remember that DEMATEL indicates a cause-effect relationship, and then it can detect influences, but in my opinion, that you can change the directions of arrows, is a little naïve and without any grounds,
Those are some of the reasons for the conclusion on dependency and influence as terms with opposite meanings.
Also, in ANP, there are different types of elements (goals, criteria, alternatives). Because in ANP, we want to express the connections between different types of elements, maybe the "dependency" or "influence" have more general meaning than we think. It is easier to talk about the connections between the criteria, but it becomes harder when we connect alternatives and criteria. How to logically interpret the content of connections between alternatives and criteria and vice-versa... Just thinking at loud :)
Well, may be. However, links alternatives and criteria can follow the same binary procedure. I don’t see any problem on it.
My last thought here - is it possible that something is lost in the translation. When I translate the dependency and influence on my language, it really has the opposite meaning,
Again, I don’t think so, as I commented above. You can indicate with the same arrow dependency and influence. As a matter of fact, dependency is implicit if you consider influence, since the latter can’t take place without the former
or at least, I can't find two elements in which the dependency won't be opposed to "influence". Indeed, If I go back to your problem - children depend on parents, it seems to be logically that the parents influence the children.
Exactly
I find dependency as "B will be changed if A changes" and influence as "change of A changes the B". If children depend on parents because of some reason (financial, emotional etc.), because of the same reason the parents influence the children. I see similar logic in other examples you mentioned.
Well, for me, dependency simply indicates precedence. If B changes when A changes is not influence but feedback, because the change is produced in the output, and the inputs reacts to it. If B changes because of A it is influence
I find your reasoning using the word "precede" interesting. I don't find it equal to dependency. Can you direct me to Saaty's material where that word is used to study more?
No, my friend, I can’t.
I use precedence as equal to dependency. That is B can’t start if A is not finished. This is the fundamental issue in the PERT system used in construction. However, precedence may have a wider meaning, such as a matter takes precedence over another.
(2.) Feedback. You maybe don't share the opinion that the same term can have two or more approximately meanings or totally different meanings. But there are cases for that. Ex. the word "agent" has one meaning in terms of police investigators, and at least slightly different in terms of multiagents systems.
You are using a linguistic argument, similar as I used above. Remember that in all languages a word normally may have different meanings
The meaning of the root of the word might be or often is kept, but the interpretations can differ from field to field. Maybe the feedback is one of them. You know a definition of feedback in engineering. But, here, maybe and just maybe, it does not mean totally the same.
Well, in this I agree with you, however, in the same sense that it has a clear definition in engineering, it should have a clear definition in MCDM, and it does not. I don’t see anything wrong with this procedure, as long as it is explained, and Saaty does not explain it.
At a very general level, but in terms of ANP, I would say that feedback is any situation (any network) in which we can find at least one loop or cycle at the cluster level.
The attached network is an example of the feedback network. We have two loops (Yellow arrows on the RIGHT figure): SS, TT; and 12 cycles on the cluster level: STAS, SATS, SAS, TAT, TSAT, TAST, TST, STS, ASA, ATA, ASTA, ATSA.
The loop SS appeared because there is at least one dependency between two criteria in S. There are 5 of them presented: number of citations depends on the number of papers, the number of papers depends on the number of projects...
The feedback does not require the loop on the element level (certain element depends on itself) - event hough I saw paper where the loop on the element level.
In ANP, we always have mutual dependency between the cluster of alternatives and clusters of criteria.
I am not sure if you expecting more from the definition of feedback, but that's it: at a very low level of explaining - a network that has at least one loop, or one cycle - this is a total explanation for feedback. I find it understandable. This was a novelty comparing to the AHP in which there was only a linear dependency (it was impossible that criteria depend on other criteria, or alternatives depend on the criteria). The mutual dependency between any two clusters is not required to have a feedback.
You are very kind and I really appreciate your efforts, but there is still no explanation of the meaning of feed back, you are always talking about relationships and dependencies, but in no place, it is explained how B may influence A, following the dependency of B from A.
I am sorry, I am not stubborn, I am an engineer and then I am not easily convinced with words, I need proofs, if not tangibles, at least supported by logic or by a theorem
I am simply trying to see clearly how feed back means, and unfortunately, I could not get it
I have enjoyed and learnt from this interesting interchange of ideas, for which I thank you.
If you are working in MCDM perhaps you would be interesting in exploring the SIMUS method, based in Linear Programming, and which does not require weights or pair-wise comparisons. If you are interested it would me my pleasure to send you the SIMUS software, it its full capacity.
Of course, we can communicate via Skype or WhatsApp. Only consider that I am about 7 hours behind you
WhatsApp: 1 613 770 7123
Skype: juanlarch
Best regards
NM- Dear Dr. Kanoik
NK- In terms of our discussion - I think the majority of our discussion is related to two aspects: (1) dependency/influence and (2) what is feedback.
(1.) What is a dependency, and what is influence? I said that in ANP they are terms with opposite meanings. Why did I say that prof Saaty implicitly said that? When explaining the ANP, in terms of graphical representation - arrows represent the "dependencies" between the elements but at the same time:
NM- I agree with indicating dependency by arrows, is the usual methods used in networks
NK- - the author introduces the "Influence matrix" which fits the graphical representation of the problem (created using the dependencies) 100%. Table of influences is zero/one table that at position (i, j) contains 1 if there is an influence from element i to element j, and it contains 0 it there is no influence.
NM- I agree with the binary matrix and that (1) means dependency and (0) not. But again, where is the influence? And how is it expressed?
NK- Ex. in the attached demo-example, there is an arrow from co to pa (courseware depends on the number of papers) - in terms of the influence matrix this becomes equal to "number of papers influence the courseware". ... (a similar conclusion mechanism is used in creating the zero-one matrix of this problem) ... This guides us (or at least some of us) to the conclusion that the influences are arrows with opposite directions than dependencies.
NM- As I understand it, papers depend of courseware because they are the way the course is imparted, the more intensive, the more reports and documentation.
Now, how the number of papers may influence the courseware kit? If we have a large amount of papers, that is articles, drawings, reports, etc. how can they influence de courseware? It is like a snake eating its tail!
The 0-1 matrix is different. It clearly expresses that there is dependency only in one sense. Now, if you put 1-1, then there will be dependency in he two senses, but even in this last case, where is the influence?
However, if you tell me that students are asking for some additional subject, that, I can understand, because the courseware kit is modified in order to produce the additional material. If that is the case, yes there is feedback from the students, because it modifies the original number of papers. Is my reasoning correct?
I agree with you about what you say that dependency arrows are opposite to influence arrows, but in your former letter you said that dependency and influence were opposite, and they are not, what is opposite is their graphic representation. Observe that in case of feedback, both, dependency and influence have the same sense
NK- - to fill the data in the supermatrix, the questions that are asked use the word "influence" - not "dependency" - even though the questions are the result of the graphical model created using the "dependencies".
Yes, and for me, that is incorrect.
Ex. From the attached figure we can see that criterion "ci" depends on the criteria "pa" and "pr" (arrows from "ci" to "pa" and "pr").
NM- According to your drawing, c1 depends from pr, and indirectly from pa, through pr
NK- At the step of filling the unweighted supermatrix, (among many comparisons,) the decision will have to compare the criteria "pa" and "pr" with respect to criterion "ci". By ANP theory the question that will be asked is: Given the criterion "ci", which of two criteria, "pa" or "pr", has a greater influence on criterion "ci"?
NM- Agreed, but in this context, the word ‘influence’, is a simple linguistic way and it does not mean that either pa or pr can MODIFY c1. By the way this is one of the problems with pair-wise comparisons, since many times the comparison value depends in the way the question is formulated. Many authors, not me, have addressed this issue.
NK- I think that there are some other ANP author's states which can lead us to the conclusion that "dependency" is in relation to "influence" as "opposite meaning" in the environment of ANP.
As said earlier, in all DEMATEL-ANP integration, and also BSC-ANP researches - in which the influences between the elements are modeled - the influences are directly transformed into ANP models and dependencies changing the direction of arrows in the models.
NM- Remember that DEMATEL indicates a cause-effect relationship, and then it can detect influences, but in my opinion, that you can change the directions of arrows, is a little naïve and without any grounds,
NK- Those are some of the reasons for the conclusion on dependency and influence as terms with opposite meanings.
Also, in ANP, there are different types of elements (goals, criteria, alternatives). Because in ANP, we want to express the connections between different types of elements, maybe the "dependency" or "influence" have more general meaning than we think. It is easier to talk about the connections between the criteria, but it becomes harder when we connect alternatives and criteria. How to logically interpret the content of connections between alternatives and criteria and vice-versa... Just thinking at loud :)
NM- Well, may be. However, links alternatives and criteria can follow the same binary procedure. I don’t see any problem on it.
NK- My last thought here - is it possible that something is lost in the translation. When I translate the dependency and influence on my language, it really has the opposite meaning,
NM-Again, I don’t think so, as I commented above. You can indicate with the same arrow dependency and influence. As a matter of fact, dependency is implicit if you consider influence, since the latter can’t take place without the former
NK-or at least, I can't find two elements in which the dependency won't be opposed to "influence". Indeed, If I go back to your problem - children depend on parents, it seems to be logically that the parents influence the children.
NM- Exactly
NK-I find dependency as "B will be changed if A changes" and influence as "change of A changes the B". If children depend on parents because of some reason (financial, emotional etc.), because of the same reason the parents influence the children. I see similar logic in other examples you mentioned.
NM- Well, for me, dependency simply indicates precedence. If B changes when A changes is not influence but feedback, because the change is produced in the output, and the inputs reacts to it. If B changes because of A it is influence
NK-I find your reasoning using the word "precede" interesting. I don't find it equal to dependency. Can you direct me to Saaty's material where that word is used to study more?
NM- No, my friend, I can’t.
I use precedence as equal to dependency. That is B can’t start if A is not finished. This is the fundamental issue in the PERT system used in construction. However, precedence may have a wider meaning, such as a matter takes precedence over another.
NK- (2.) Feedback. You maybe don't share the opinion that the same term can have two or more approximately meanings or totally different meanings. But there are cases for that. Ex. the word "agent" has one meaning in terms of police investigators, and at least slightly different in terms of multiagents systems.
NM- You are using a linguistic argument, similar as I used above. Remember that in all languages a word normally may have different meanings
NK- The meaning of the root of the word might be or often is kept, but the interpretations can differ from field to field. Maybe the feedback is one of them. You know a definition of feedback in engineering. But, here, maybe and just maybe, it does not mean totally the same.
NM-Well, in this I agree with you, however, in the same sense that it has a clear definition in engineering, it should have a clear definition in MCDM, and it does not. I don’t see anything wrong with this procedure, as long as it is explained, and Saaty does not explain it.
NK-At a very general level, but in terms of ANP, I would say that feedback is any situation (any network) in which we can find at least one loop or cycle at the cluster level.
The attached network is an example of the feedback network. We have two loops (Yellow arrows on the RIGHT figure): SS, TT; and 12 cycles on the cluster level: STAS, SATS, SAS, TAT, TSAT, TAST, TST, STS, ASA, ATA, ASTA, ATSA.
The loop SS appeared because there is at least one dependency between two criteria in S. There are 5 of them presented: number of citations depends on the number of papers, the number of papers depends on the number of projects...
The feedback does not require the loop on the element level (certain element depends on itself) - event hough I saw paper where the loop on the element level.
In ANP, we always have mutual dependency between the cluster of alternatives and clusters of criteria.
I am not sure if you expecting more from the definition of feedback, but that's it: at a very low level of explaining - a network that has at least one loop, or one cycle - this is a total explanation for feedback. I find it understandable. This was a novelty comparing to the AHP in which there was only a linear dependency (it was impossible that criteria depend on other criteria, or alternatives depend on the criteria). The mutual dependency between any two clusters is not required to have a feedback.
NM- You are very kind and I really appreciate your efforts, but there is still no explanation of the meaning of feed back, you are always talking about relationships and dependencies, but in no place, it is explained how B may influence A, following the dependency of B from A.
I am sorry, I am not stubborn, I am an engineer and then I am not easily convinced with words, I need proofs, if not tangibles, at least supported by logic or by a theorem
I am simply trying to see clearly how feed back means, and unfortunately, I could not get it
I have enjoyed and learnt from this interesting interchange of ideas, for which I thank you.
If you are working in MCDM perhaps you would be interesting in exploring the SIMUS method, based in Linear Programming, and which does not require weights or pair-wise comparisons. If you are interested it would me my pleasure to send you the SIMUS software, it its full capacity.
Of course, we can communicate via Skype or WhatsApp. Only consider that I am about 7 hours behind you
WhatsApp: 1 613 770 7123
Skype: juanlarch
Best regards
Nolberto
Sorry for the delay. Thank you for the discussion.
(1) Dependency/Influence.
- I think that your own reasoning on "dependency as precedence" might be a source of your misunderstanding of the method. I would say that precedence has a similar meaning to influence. If A precedes B, then A influences B (and B depends on A).
- I really think that we should seek for the more general meaning of terms dependency/influence. Try to formulate the statement using the word "precedence" between elements (1) criterion "number of papers" and (2) alternative "scientist 1", in both directions.
- Again, I think that the original idea by Saaty was to describe a relation between two elements with two equal statements: If A->B (if the arrow -> is a dependency), then "A depends on B" = "B influences A". This allowed Saaty to use the "influence" matrix to describe the "dependency" graph. If at a certain position (i, j) in the influence matrix we have a 1, this means that "i influences j" = "j depends on i". So, one number (value, cell) from the influence matrix has two meanings. This is only possible if "dependency" and "influence" have something in common. ... I am not sure how to explain the idea of the influence/dependency matrix more simpler. Maybe, there was no need for more same-meaning statements for one thing, but here we are now. This situation reminds me of the following: 1 < 2- This can be explained as "1 is smaller than 2" and "2 is larger than 1" ... small/large are two words that describe one relation between two numbers.
- There is an arrow "co -> pa" (arrow presents the dependency in the attached figure) ... that means that the "courseware depends on the papers" = "papers influence the courseware". So, the content you present to your students at teaching hours can (among other things) include results you published in papers = published papers have an influence on what you can present to the students. In addition, the arrow "co -> pa" does NOT mean that courseware influence papers, and that papers depend on courseware. Please notice that there is NO arrow "pa -> co".
- if you agree that "that dependency arrows are opposite to influence arrows", why then dependency and influence weren't opposite terms if the same problem can be represented with two models, one dependency and other influence and they differ only in directions of arrows?
- you said "If B changes when A changes is not influence" and "If B changes because of A it is influence". Why in the first case (or both) we do not talk about the influence of A on B? Maybe some examples of illustration...
(2) Feedback.
- You said there is still no definition of feedback. Why can't you accept the "low-level" definition "a network that has at least one loop, or one cycle at the cluster level"?
- Respecting the example from the attached figure, here is the reasoning:
On the end, in general, there are differences in different sciences. In social science, there is much more fuzziness than in engineering - it is harder to explicitly and formally describe some artifacts. I am not sure if you are familiar with the Cynefin Framework. This framework describes different contexts: simple and complicated contexts assume an ordered universe, where cause-and-effect relationships are perceptible, and the right answers can be determined based on the facts. Complex and chaotic contexts are unordered—there is no immediately apparent relationship between cause and effect, and the way forward is determined based on emerging patterns. See https://www.academia.edu/522169/A_leaders_framework_for_decision_making
So, in complex and chaotic contexts, math and engineering methods seem to often be almost helpless. Maybe the ANP is a try to deal with complex/chaotic using the mechanisms that are appropriate for simple and complicated.
Related to SIMUS, I downloaded one of your paper and now I am in a process of step-by-step analysis of the algorithm.
Thank you for discussion,
Best regards,
Nikola
Dear Nikola
(1) Dependency/Influence.
NK- I think that your own reasoning on "dependency as precedence" might be a source of your misunderstanding of the method. I would say that precedence has a similar meaning to influence. If A precedes B, then A influences B (and B depends on A).
NM - OK, assume that I have a misunderstanding of the method. May I remind you that this same argument was used by Saaty collaborators in the 90s to refute Dyer criticism?
This were their comment:
…..’this criticism arises out of a lack of understanding of the theory underlying the AHP (Harker and Vargas , 1990)’
Regarding your last paragraph, suppose you have to fly. You have among others two clearly defined activities
1- Go to the airport
2. Catch a flight
Obviously the first precedes the second, because if you do not reach the airport you can’t catch the flight
Now, can you explain me why the fact that you are in the airport influence the flight? The latter is independent that you arrive or not in the airport
Dependency is a different matter: The plane can’t fly if it does not have the airport control tower approval
I would say that your interpretation is a little ‘loose’, and then it is very convenient to explain something, that nobody, let alone Saaty, explained properly. You use the word ‘similar’, which in mathematics, our field, translates as a homothetic transformation, that is, that the value in B is multiplied by a positive constant because the influence of A, and that is not the case in most projects..
This is something that invariably I found in AHP ‘explanations’. All is very vaguely expressed, and the reader has to guess what the writer wanted to say.
You are a learned scholar, however, you don’t justify what you say, only use words, while I have put examples and reasoning.
Can I be mistaken? Of course, I can be, but I need to be convinced, with reasoning, using mathematics or common sense that I am mistaken. Just asserting it is not enough.
NK - I really think that we should seek for the more general meaning of terms dependency/influence. Try to formulate the statement using the word "precedence" between elements (1) criterion "number of papers" and (2) alternative "scientist 1", in both directions.
NM. It appears, that you use an argument that has been refuted by other colleagues, in the sense that the answer to a pair-wise comparison may depend on the way the question is formulated!
You are now relating criterion ‘papers ‘with alternative ‘scientist 1’. I don’t need to remind you that AHP does not consider alternatives to determine criteria weights, but alternatives are weighted according to criteria.
NK - Again, I think that the original idea by Saaty was to describe a relation between two elements with two equal statements: If A->B (if the arrow -> is a dependency), then "A depends on B" = "B influences A".
NM-See the vagueness? What arguments used Saaty to make that description? You know the answer: None
To influence on something, you need to produce changes. Which changes are produced in A because B? None, unless, as I said in my last letter, if changes in A modify A, and that is feedback
NK- This allowed Saaty to use the "influence" matrix to describe the "dependency" graph. If at a certain position (i, j) in the influence matrix we have a 1, this means that "i influences j" = "j depends on i".
NM- It appears that this was a Saaty assumption but he did not explain the grounds for this assumption. This is what I criticize. You see that you repeat what you say before, you use a binary matrix format, and that is correct, but you don’t explain WHY and HOW you use the binary matrix, you are on ly saying something that everyone knows, that is, to indicate with ‘1’ a relationship and with’0’ a lack of it. What does it prove?
NK- So, one number (value, cell) from the influence matrix has two meanings. This is only possible if "dependency" and "influence" have something in common. ... I am not sure how to explain the idea of the influence/dependency matrix more simpler.
NM- Very convenient comment that I am afraid clarifies nothing, although I reckon it is original. However, you should explain how this duality works. A number such as 8 may have different meanings, but it does not mean that they are related. It can indicate a quantity of boxes, or that a candidate is in the 8th place in a ranking, or it may be the number of a city bus, or 8 Euros. What do they have in common? Only a number.
NK-Maybe, there was no need for more same-meaning statements for one thing, but here we are now. This situation reminds me of the following: 1 < 2- This can be explained as "1 is smaller than 2" and "2 is larger than 1" ... small/large are two words that describe one relation between two numbers.
NM. Sorry my friend, but small/large do not express the relation between two numbers, there are rather absolute values. If you say ‘smaller than’ or ‘lager than’, yes, you are right
Your example is a simple comparison and a conclusion. If 11. What you did is simply change the mathematical Symbols < and >. It looks as an axiom, like saying: if A=2, then its reciprocal is ½. An what does it prove? Only that t heir product ids ‘1’.
NK - There is an arrow "co -> pa" (arrow presents the dependency in the attached figure) ... that means that the "courseware depends on the papers" = "papers influence the courseware".
NM- I agree that a course may depend on papers, and say, the more material the better. However, the number of papers may not be related with their quality, and you can have for the same course a reduced number of papers with high quality. In both cases the papers can certainly influence the prestige and utility of the course, but it is not related with number of papers but with their quality, which is an attribute of the material and also not related with the course, but with an attribute of it.
As a matter of fact, the material maybe poor, and the course director does nothing about it.
NK- So, the content you present to your students at teaching hours can (among other things) include results you published in papers = published papers have an influence on what you can present to the students.
NM- You are corroborating what I said, of course the quality of the paper may influence the students.
NK-In addition, the arrow "co -> pa" does NOT mean that courseware influence papers, and that papers depend on courseware. Please notice that there is NO arrow "pa -> co".
NM- That is correct, and again is a simple use of binary, that says there is only one direction, but what does it prove?
NR-- if you agree that "that dependency arrows are opposite to influence arrows", why then dependency and influence weren't opposite terms if the same problem can be represented with two models, one dependency and other influence and they differ only in directions of arrows?
NM. Sorry, I don’t agree. The direction of arrows is a simple geometrical convention. A child is dependent of his/her parents and as the same time receives their influence (two arrows with different meaning and in the same direction),so, there is simultaneously dependency and influence, and there is no doubt that the child can influence the parents in different ways, changing the lifestyle, economy, etc., which will be an arrow in the opposite direction, but only on influence.
NK- you said "If B changes when A changes is not influence" and "If B changes because of A it is influence". Why in the first case (or both) we do not talk about the influence of A on B? Maybe some examples of illustration...
NM- Yes, because if B changes when A changes it may be due to feedback, that modifies B until the system comes to the former equilibrium. It is the typical case of a room thermostat.
The second case happens for instance, when A (a child) influences his/her parents B, given them joy, with his/her school positive marks or sport achievements.
(2) Feedback.
NK- You said there is still no definition of feedback. Why can't you accept the "low-level" definition "a network that has at least one loop, or one cycle at the cluster level"?
NM – I will accept it if you explain it. Of course, there is a loop in feedback, but if you don’t explain why that loop is produced, I can’t accept it. It appears as an intuitionist approach, instead of a reasonable one. In addition, what a ‘low level’ means?
NK- Respecting the example from the attached figure, here is the reasoning:
1. "co -> pa" ...number of papers influences the courseware (if you have more research papers, you have more teaching materials to present; or if you don't have them, it also has the influence on teaching materials that will be delivered to the students).
NM- Not necessarily, the students may think that the number of papers is enough.
NK 2. "pr -> co" ... the courseware influences the number of projects. If you have challenging teaching activities with research projects with your (masters and Ph.D.) students, it can result in new projects.
NM- Yes, Y agree
NK – 3. Finally, in the first statement, we have an influence from one cluster (system) Science to another cluster (system) Teaching, and in the second statement, we have an influence of Teaching (back) to Science. I think that this is an example of the feedback in terms of the ANP.
NM- Does it mean that you are going to influence science according to teaching? Very strange conclusion indeed. What you can change is t he way science is taught.
NK- On the end, in general, there are differences in different sciences. In social science, there is much more fuzziness than in engineering - it is harder to explicitly and formally describe some artifacts. I am not sure if you are familiar with the Cynefin Framework. This framework describes different contexts: simple and complicated contexts assume an ordered universe, where cause-and-effect relationships are perceptible, and the right answers can be determined based on the facts. Complex and chaotic contexts are unordered—there is no immediately apparent relationship between cause and effect, and the way forward is determined based on emerging patterns. See https://www.academia.edu/522169/A_leaders_framework_for_decision_making
So, in complex and chaotic contexts, math and engineering methods seem to often be almost helpless.
NM- Have you heard of the mathematical theory of chaos, or from Artificial Intelligence and Machine Learning, or from Experts Systems?
NK-Maybe the ANP is a try to deal with complex/chaotic using the mechanisms that are appropriate for simple and complicated.
NM- Absolutely! That is the reason why I say that complex scenarios can’t be solved by AHP with hierarchical structure and not even by ANP by its network structure. They are clearly not enough.
NK-Related to SIMUS, I downloaded one of your paper and now I am in a process of step-by-step analysis of the algorithm.
NM- I really appreciate your effort. I only recommend you to analyze why SIMUS can solve complex scenarios, including dependencies, influences, several scenarios, multiple objectives, etc.
However, I would very much appreciate your opinion of the method and especially your criticism
Best regards
Nolberto
Dear prof. Munier
(1) Dependency/Influence.
- Related to the flying example, I do not want to comment on the cases that are not fully defined - in this case, it is not obvious what is the decision-making goal, what are criteria and what are the alternatives (even though, for me is it obvious that the fact that you are in the airport or not influence cathing the flight - if you're not there, it is ensured that you won't ever catch it. However, it is hard to generalize because I am not sure if "going to the airport"/"catching the flight" are the alternatives or criteria in the decision-making model).
- my suggestion to you from my previous reply - in which I asked you to formulate simple statements using the word "precedence" between elements (1) criterion "number of papers" and (2) alternative "scientist 1", in both directions - had a purpose. The answer is: "the number of papers precedes the scientist1", and "the scientist 1 precedes the number of papers". And, both of the statements are harder to understand or imagine than the relation between two criteria, ex. "the courseware precedes the number of projects". My point here was to suggest that "influence" and "dependence" (or precedence) should be "used" or "interpreted" at some more general level of meaning which you probably don't accept.
- also, I think that you didn't figure out my point in 1
Nikola
Sorry for the delay in answering your letter. Its seemed to me that I had already answered it, but I did not.
(1) Dependency/Influence.
NK- Related to the flying example, I do not want to comment on the cases that are not fully defined - in this case, it is not obvious what is the decision-making goal, what are criteria and what are the alternatives (even though, for me is it obvious that the fact that you are in the airport or not influence cathing the flight - if you're not there, it is ensured that you won't ever catch it. However, it is hard to generalize because I am not sure if "going to the airport"/"catching the flight" are the alternatives or criteria in the decision-making model).
NM -OK, you have a point on this, my mistake, they are not alternatives. The alternatives could be to take a plane or take the train, and the reasoning does not change, since the flight and the train are independent of what decide
NK - my suggestion to you from my previous reply - in which I asked you to formulate simple statements using the word "precedence" between elements (1) criterion "number of papers" and (2) alternative "scientist 1", in both directions - had a purpose. The answer is: "the number of papers precedes the scientist1", and "the scientist 1 precedes the number of papers".
NM. Could you please explain why the numbers of papers precede scientist 1? I don’t see why.
NK-And, both of the statements are harder to understand or imagine than the relation between two criteria, ex. "the courseware precedes the number of projects". My point here was to suggest that "influence" and "dependence" (or precedence) should be "used" or "interpreted" at some more general level of meaning which you probably don't accept.
NM- No, I don’t accept it because I so doing you are arranging the words according to your convenience.
NK- also, I think that you didn't figure out my point in 1
Thank you for reply.
We can only agree that we disagree!
Hope that someone will resolve those issues!
Best regards,
Nikola
Dear Nikola
We don't disagree my friend, it is simply that we have different point of views.
AHP and ANP supporters don't have arguments to hold what they believe or what they were told, while I work only with proofs. Why don 't they use them?
Because they can't.
I doubt that someone will solve these issues which are unsolved, or by putting it mildly, unexplained. You, at least, tried, and that, for me has an enormous value, for which I am thankful.
Please remember that I requested your good will in examining the SIMUS method.
Don't be shy on your criticism!
Thank you for your reply.
Still, I believe that there are answers to this issue.
Dear Nikola
Maybe, but we need to hear them, and to be honest, after 5 years that nobody responded, being you the only exception, I am in doubt that we will receive those answers.
The reasons? I have said many times that there are not valid arguments to support AHP or ANP. All are assumptions, such as equating trade-offs with criteria weights, or fantasies such as assuming that complex problems can be solved by preferences, or that the DM may interpret the real world without taking it in account, or the absurd that a DM has to delete or modify which in good faith he thought, just because an invented formula says so.
I can continue but to no avail. Do you remember the saying?
There's no better blind man than the one who doesn't want to see, and not better deaf man that the one who doesn't want to hear"
Thank you for reply prof. Munier,
Well, maybe you should send your question directly to someone in the "ANP group". They maybe don't use Researchgate, or they didn't see your question (I can't find it too). I neither didn't saw your original question in this thread from November 30 2019 until a few weeks ago.
Or, maybe they saw the question and just assumed that the discussion makes no sense considering that you are an engineer and maybe not very open-minded.
For me too, it was hard to discuss it with you.
I find it strange that you in the ANP interpretation use the concept of "precedence" which is not represented in ANP literature at all.
You also agreed that two words can have different meanings in different fields, but at the same time, you don't want to apply it to dependency/influence case.
Also, you said that, in terms of influence matrix, you don't see the problem in connections between alternatives and criteria (they follow the same binary procedure), and when I asked you to formulate the statement for connection between some alternative (scientist1) and some criterion (numbers of papers), you refused even though in ANP there are links between criteria and alternatives in both directions (scientist1 influences number of papers and number of papers influences scientist1). The understanding of the "influence" here in the alternative-criterion case is different than in the criterion-criterion case....
Well, I tried to avoid any further discussion but didn't succeed.
Still, I think that we should not reject the method because we think that there are no valid arguments. At least, as an intriguing research case, this can be always something to reflect on.
Best regards
Thank you, Prof Kadoic
N.K-Well, maybe you should send your question directly to someone in the "ANP group". They maybe don't use Researchgate, or they didn't see your question (I can't find it too). I neither didn't saw your original question in this thread from November 30 2019 until a few weeks ago.
NM- That is correct, that was my last request. Prior to that, and since 2 or 3 years ago I asked the same question via emails directly to three learned colleagues on ANP. None of them was capable to give me a coherent answer, and that is the reason I published it as a general question. In these two months I did not receive any answer.
NK-Or, maybe they saw the question and just assumed that the discussion makes no sense considering that you are an engineer and maybe not very open-minded.
NM- Well, yes, that could be. The problem is that this field it is not a matter of being open-minded. We engineers are trained to reason, not to accept speculations, and certainly not to accept unproved assumptions. We need axioms, theorems or tangible proofs about something, and like it or not, our world works in that way.
Your Nikola Tesla was one of the best exponents of an engineer. He worked with facts in his fight with Edison on the use of the AC in lieu of the CC. Edison was a brilliant engineer, but without a doubt not open-minded in this aspect because he has to support his electric company (General Electric), and he was adamant to recognize that his CC could only reach less than two km from the power house. He did not work on facts while Tesla did.
In my humble opinion there is a similitude between Edison and Saaty, a brilliant mathematician, but that worked with assumptions without any mathematical support, such as assuming that trade-offs may be considered as criteria weights, or that dependency also means influence. I am only asking for a demonstration of this. Does it mean that I am not open-minded?
NK. For me too, it was hard to discuss it with you.
I find it strange that you in the ANP interpretation use the concept of "precedence" which is not represented in ANP literature at all.
NM. In network theory, which is the base for ANP, dependence and precedence are equivalent. If A precedes B, or A>B, B can’t start until A is completed, consequently B depends on A.
NK. You also agreed that two words can have different meanings in different fields, but at the same time, you don't want to apply it to dependency/influence case.
NM. Because influence and dependency are different things. I put several times the example that children are dependent of their patens because food and shelter, but it does not mean that parents can always influence children. It is difficult for me to look for a clearer example. And I am awaiting yours.
Also, you said that, in terms of influence matrix, you don't see the problem in connections between alternatives and criteria (they follow the same binary procedure),
NM- Not quite. I always say that criteria depend on alternatives, since they are which define the necessary criteria, and also, as it is my custom, I put examples.
NK. and when I asked you to formulate the statement for connection between some alternative (scientist1) and some criterion (numbers of papers), you refused even though in ANP there are links between criteria and alternatives in both directions (scientist1 influences number of papers and number of papers influences scientist1). The understanding of the "influence" here in the alternative-criterion case is different than in the criterion-criterion case...
NM. It is really for me very difficult to understand your example. In my opinion the scientist determines the content and number of papers, as a writer determines the content and extension of his book; he creates those instruments, but he can’t influence them, unless, of course, if he decides to add, delete or correct something.
Influence means the capacity to change a behaviour. How can you influence a piece of paper? If you tell me that the writer can influence the readers, yes, I agree, and the readers, in turn, sometimes, because their comments, can influence the writer, that is, the writer receives a reaction of the readers. This is feedback.
Now, your example poses an interesting question: If the scientist creates a paper and if a paper influences the scientist back, it appears that the scientist is influenced by himself…….
You are right in the sense that alternatives define criteria, that is, criteria depend on the nature of the project. But criteria do not influence alternatives. What influences alternatives are the attributes of those criteria, that is its series of values from alternatives, according to the fundamental theorem of Information Theory developed by Shannon. It is so true, that if a criterion has the same numeric value for several alternatives, its influence is zero. Can you refute me this?
Regarding criteria, there could be dependency in the sense that one of them depends on the other, only if there is correlation between t hem, but of course, the correlation concept is alien to ANP.
NK. Well, I tried to avoid any further discussion but didn't succeed.
NM. The problem my friend, is that I am hard to convince if don’t see proofs. Just show me one, and I will publicly recognize my misconceptions
Still, I think that we should not reject the method because we think that there are no valid arguments. At least, as an intriguing research case, this can be always something to reflect on.
NM. Of course, I don’t reject the method. I think that it is valuable, for certain problems, but it does not mean that I have to believe everything that Saaty said, based in his assumptions. As a matter of fact, I had discussions with him, about three years ago.
We could be in disagreement of matters such as religions or about which football team is best, that is legitimate; the first is a matter of faith, and on that there is no room for discussions. The second, is a matter of point of view, and if somebody thinks that way, well, let it be. But what I can’t accept is somebody asserting to that team A is 3 times better than team B, or because Saaty said that there is feedback, which he never explained what really it is.
Best regards
Nolberto
I read your paper “Characteristics of the Analytic Network Process, a Multi-Criteria Decision-Making Method” and I certainly enjoyed it!
It is the best paper I have read on ANP, clearly written, easily understandable, but most important, it is very honest, since in several parts you don’t hesitate in mentioning its drawbacks. Please accept my compliments on such a fine work.
However, I want to comment on some aspects of the paper.
1- You use the concept of feedback but you never defined it.
2- In your example of procurement you say that it is a good example of the fact that alternatives are not known, and this is correct up to a certain point in time. If the government calls for bids, say for desks for schools, and details the features that they have to match, or criteria, this is fine, and the unknown bidders (that is options or alternatives), bid according to them.
The government establishes the criteria, not the bidders, and certainly there is no reason for them to make pair-wise comparisons.
When the government receives the offers, it knows which the alternatives or bidders are, and at that moment it can use a MCDM method to select the most convenient offer. At that point it can determine the criteria weights by using a pair-wise comparison or any other procedure. Therefore, what you say in your paper, is true until the offers are received, but not before.
The government however, can’t follow the AHP or ANP method, because the trade-offs are not constant, as AHP and ANP sustain, which is another convenient assumption which can be easily rebutted.
You pair-wise compare two alternatives at a time regarding each criterion, determine your preference and the multiply it by the trade-off value of the criterion. Then you assume that the other alternatives valuations are multiplied by a constant. However, that trade-off may be not constant because for its calculation you ignore the values of the alternatives.
If the pair-wise comparisons between criteria, say for instance that quality is 2 times more important that price, that can hold when comparing quality and price between alternative A to B. However, when comparing A with C it may not hold. The reason is simple, the trade-off application for it, depends on the values of the alternatives, that you don’t have. This simple and evident fact is ignored by AHP and ANP.
It even applies for Promethee and Topsis, since in both you have reliable values for each alternative and each criterion, but then you apply the same trade-off to all of hem, replicating what you do in AHP.
3. You don’t define influence either. You only say “e.g. a certain leaf could influence other leaves”, which in my opinion it explains nothing. As a matter of fact, Saaty equates influence with importance (Saaty) The Analytic Network Process
Dear prof. Munier
Thank you for your reply.
Wow, you talked with prof. Saaty - I can imagine it was a very interesting conversation. Too bad that it wasn't in the form of some (Oxford style) debate and recorded. That would be an interesting scientific forum.
Related to my paper, yes, I didn't give any definitions because I implied them as Saaty "defined" them. I just wanted to put some accents on some additional issues I found in ANP.
I never had the problem with understanding the feedback from the Saaty's materials in which he graphically explained what he considers as feedback in terms of ANP (existence of loops/cycles between clusters in the network).
Also, with dependence/influence - I find them as opposite meanings terms because of the direct conversion of the dependency graph into the influence network. Also, I find their definitions in some higher-level (more general) meaning because we have different types of elements in the network (criteria and alternatives) and strict definition of ex. the influence between one alternative and one criterion is blurry to understand.
Related to the parents-children case - again, without the whole network it is hard to analyze the case for me. What are parents/children here - criteria or alternatives, and what is the decision-making goal? However, if children depend on their parents (ex. money), then they through that dependency can influence them (if they give or not give the appropriate sum of money). This does not mean that parents always influence the children just like that the children always depend on parents. In the influence matrix, parents can have many 1s in the row (they influence many other elements, among which, one element is "the children"). Also, the children can have many 1s in the column (they depend on many other elements, one of them is "the parents").
Best regards,
Nikola
Dear Prof Kadoic
NM- I appreciate your fast reply
NK- Thank you for your reply.
Wow, you talked with prof. Saaty - I can imagine it was a very interesting conversation. Too bad that it wasn't in the form of some (Oxford style) debate and recorded. That would be an interesting scientific forum.
NM- I did not really talk with Dr. Saaty. He and I had a brief interchange of opinions using emails, on about March 2017. A little before that time I wrote and I sent to about 40 researchers my idea that MCDM needed a revamping, citing about 30 features that should be considered in real-life projects, and they were not. Only one person responded, and he was Dr. Saaty, and agreeing with my proposal in a 100 %, according his own words, even when most of my suggestions meant lacks in the AHP method.
You can appreciate by this fact the nobility showed by this great scholar and decent human being, in aligning with me even against his own ideas. From there we enjoyed a short period of discussions. It was for me a privilege in having this opportunity; unfortunately, his passing away in August 2017 interrupted this healthy interchange.
NK- Related to my paper, yes, I didn't give any definitions because I implied them as Saaty "defined" them. I just wanted to put some accents on some additional issues I found in ANP.
NM- Well, that is the problem, because he did not.
NK - I never had the problem with understanding the feedback from the Saaty's materials in which he graphically explained what he considers as feedback in terms of ANP (existence of loops/cycles between clusters in the network).
NM. If I am allowed to correct you, Saaty never explained it. He presented a drawing with two nodes A and B, with arrows from A to B, and from B to A ,and he called that ‘feedback’.
On what grounds? He did not say.
NK- Also, with dependence/influence - I find them as opposite meanings terms because of the direct conversion of the dependency graph into the influence network.
NM- And what an ‘influence network’ is in decision-making? You don’t explain it. I can understand influence network in social issues but not in mathematics
NK- Also, I find their definitions in some higher-level (more general) meaning because we have different types of elements in the network (criteria and alternatives) and strict definition of ex. the influence between one alternative and one criterion is blurry to understand.
NM- Sorry Nikola, but you don’t explain here de dependence/influence subject. You simply speak in a language that can have several interpretations.
I don’t think that is blurry to understand the relationship between an alternative and a criterion. The first defines the second, while the latter evaluates the first. Where is it blurry?
NK-Related to the parents-children case - again, without the whole network it is hard to analyze the case for me. What are parents/children here - criteria or alternatives, and what is the decision-making goal?
NM- I don’t remember saying that in this case are alternatives and criteria, there is a simple dependency which most of the time involves influencing the children, but this s not a law, because sometimes in does not happen. If you tell me that the new baby in a house will most probably change the way of life of his parents, yes, I believe it, he will influence his parents. There are lots of examples where you can influence a person, even if he does not depend on you or even know you
NK-However, if children depend on their parents (ex. money), then they through that dependency can influence them (if they give or not give the appropriate sum of money).
NM. Yes, I agree, and it can produce different effects on the children, but again, that is not a law
NK - This does not mean that parents always influence the children just like that the children always depend on parents. In the influence matrix, parents can have many 1s in the row (they influence many other elements, among which, one element is "the children"). Also, the children can have many 1s in the column (they depend on many other elements, one of them is "the parents").
NM- Agreed, but again it is not a law, and in addition, in MCDM we are not talking about human beings. We are talking about facts, things, inanimated objects. The main difference is human beings can influence other human beings, but can’t influence things, because they don’t have the capacity to decide.
If an alternative is location A for building a paper mill, it certainly will influence people because new jobs opportunities, but it can’t influence the plot of land where it will be installed. It will change the land use but there is no influence here. The land has no opinion. Is the same case with your example of a proffers and his papers; the latter do not have any opinion. The plot will continue to be as it always was, only that now there is a plant on it. There is however, a case where there is really an influence. It is believed that the creation of a big lake for a hydro power plant may induce tremors. In this case, you changed the land use, and ‘it reacted’ by inducing a tremor. This is feedback.
NM- Agreed, but again it is not a law, and in addition, in MCDM we are not talking about human beings. We are talking about facts, things, unanimated objects. The main difference is human beings can influence other human beings, but can’t influence things, because they don’t have the capacity to decide.
Nolberto
Best regards
Before you said that there in terms of influence matrix, you don't see the problem in connections between alternatives and criteria. And now you said "I don’t think that is blurry to understand the relationship between an alternative and a criterion. The first defines the second, while the latter evaluates the first. Where is it blurry?"
- Agree, it is not. But it is harder to explain those relationships using the word "influence". You changed the word in your explanation to "defines" and "evaluates" even though the name of the matrix is influence matrix. It is harder (or blurry) to use the word "influence" between criteria and alternatives than between the criteria only (ex. the number of papers influences the projects). This was my point - to suggest that the influence/dependency in ANP cannot be considered by strict definitions. They have more general meanings to cover all possible relations of affecting, independently of the elements which can be human beings, criteria, things, etc. I think that our whole influence/dependency discussion is only a matter of language and how we interpret things. While I think that the meaning of those terms is much wider in terms of ANP, you insist on engineering definitions, but on the other hand in explanation change the interpretations from "influence" to "define/evaluate" (or "precedence" in some previous replies).
At one place you said that "If B changes because of A it is influence". If A is a "number of papers" and B is "Scientist1" - how will we formulate the statements that describe that A influences B and B influences A (in ANP, there are connections in both directions). Well, hardly - if we follow strict definitions and the facts "Scientist1" is a human being, and "a number of papers" isn't. But then, this is in contradiction with the first sentence in this reply (understanding the relationships between the criteria and alternatives in the influence matrix by you).
You said - "in MCDM we are not talking about human beings". Why then we talk about the parents and children? I am confused now.
In ANP, when talking about the influences, I agree with you that many things are not law - ex. if children depend on their parents (ex. money), then they through that dependency can influence them (if they give or not give the appropriate sum of money). This is the reason why I avoid to analyze general cases. Also, in ANP, through priority size, we can describe the "level" of the influence. Something similar like in correlation. Variable X and Y can be in "higher" and "lower" correlation.
In the paper mill case, I am not an expert in the domain, and without a complete network of criteria and alternatives, I can't discuss the ANP analysis of the problem. I can only disagree that the paper mill can’t influence the plot of land where it will be installed. It will - because the paper mill will define its purpose and it will not be possible to build ex. the hotel in the same place.
This is your first reply where you introduced "ability to have the opinion" as the part of the definition of the term "influence". In all my replies, I am encouraging you to relax (unloose) the definition of the term "influence", but you are doing totally opposite: narrowing down the definition :)
Best regards
Dear Prof. Kadoic
NK- Before you said that there in terms of influence matrix, you don't see the problem in connections between alternatives and criteria. And now you said "I don’t think that is blurry to understand the relationship between an alternative and a criterion. The first defines the second, while the latter evaluates the first. Where is it blurry?"
- Agree, it is not. But it is harder to explain those relationships using the word "influence". You changed the word in your explanation to "defines" and "evaluates" even though the name of the matrix is influence matrix.
It is harder (or blurry) to use the word "influence" between criteria and alternatives than between the criteria only (ex. the number of papers influences the projects). This was my point - to suggest that the influence/dependency in ANP cannot be considered by strict definitions.
NM- Well, if ANP cannot be considered by strict definitions, that is very convenient! Don't You think?
You can adopt the most fitting definitions to support you point of view. It is like saying that Pythagoras theorem does not apply in some right triangle case. Therefore, sorry, but that argument is not valid.
It is harder to explain because you can’t explain something that is not true, and hard does not mean blurry. For instance, tensor analysis is hard to understand, but it is not blurry
They have more general meanings to cover all possible relations of affecting, independently of the elements which can be human beings, criteria, things, etc. I think that our whole influence/dependency discussion is only a matter of language and how we interpret things.
NM- Agreed, but we are in mathematics not in literature, psychology or philosophy. We have to find the best solution, and that is not a matter of interpretations. Leave that for religions.
NK- While I think that the meaning of those terms is much wider in terms of ANP, you insist on engineering definitions,
NM- Perhaps I did not express myself vey well. Since ANP works with a network, we should use some clear definitions. Lets’ see what the Dictionary says:
Precedence: The condition of being dealt with, before other things
Therefore, A precedes B if A developed before B, that is, it is a time related factor.
Dependency: Something that is dependent of something else, and in a network.
That is, B depends on A, which must also be precedent, since you can’t depend on the future with is unknown or unpredictable.
Influence: The capacity to have an effect, or the capacity to produce effects on the actions opinions, behaviours of others.
This is what I have been saying: A having the capacity to modify or influence B
Feedback: Information about reactions. Or the transmission of evaluative or corrective action about an action, event, process to the source. Or the return to the input of art of the output.
Again, this is what I said. If B reacts to A for whatever reasons, the it can feedback that change to A, which can modify or not its former position. Not talking in engineering terms. Is a child that has been influenced by his parents to follow certain norms, reacts with cries, mutism or range, it is his reaction, which can or can’t modify his parent’s decision.
Since ANP deals with a network, you need to have precedence, that is A precedes B. In that precedence you can have
NK- but on the other hand in explanation change the interpretations from "influence" to "define/evaluate" (or "precedence" in some previous replies).
NM- Sorry, I don’t understand what you are trying to say
NK - At one place you said that "If B changes because of A it is influence". If A is a "number of papers" and B is "Scientist1" - how will we formulate the statements that describe that A influences B and B influences A (in ANP, there are connections in both directions).
NM. I support what I said, I agree that a novel, a writing, the Bible, an article in a newspaper, can influence and change you character and your beliefs, but is does not mean that the reciprocal is true. Can you change the Bible? Well, you can, if you write a new one.
As Saaty said, you can express mutual influences by opposite arrows, but the fact that you can geometrically express it, does not mean that it is true. If you are a professor that write a paper for your students, I don’t see how the same paper THAT YOU WROTE influences you. You can correct it, because you found that something is wrong, or because a student discovered an error, but that is not influence. It influences you because there is a mistake. In the first case YOU discovered errors, in the second another person, not related with the paper discovered them.
NK-Well, hardly - if we follow strict definitions and the facts "Scientist1" is a human being, and "a number of papers" isn't. But then, this is in contradiction with the first sentence in this reply (understanding the relationships between the criteria and alternatives in the influence matrix by you).
NM- What I said is that criteria depend on the alternatives because if we don’t now them, we are not able to select which criteria to use.
Alternatives do not depend on criteria; they are evaluated by them.
Saaty in one of his papers proposes an interesting example: A bridge has to be built, and there are two options, one of them in concrete, which is strong but ugly, the other one using steel, not so strong but is beautiful. Considering criterion ‘Aesthetics’ the DM, give a 4 to the beautiful bridge and a 2 to the ugly one. Now, how this subjective selection can influence the bridges? It can’t. You can talk with the architects and they may agree to aesthetically improve the ugly bridge, this is feed back, because you change the input based in your judgement of the output, but if you do nothing the bridges will not change their characteristics because your judgement. And this is not only engineering, in any human activity a feedback means that you alter the input because a change or appraisal of the output.
NK-You said - "in MCDM we are not talking about human beings". Why then we talk about the parents and children? I am confused now.
NM- Because it is only an example.
NK - In ANP, when talking about the influences, I agree with you that many things are not law - ex. if children depend on their parents (ex. money), then they through that dependency can influence them (if they give or not give the appropriate sum of money). This is the reason why I avoid to analyze general cases. Also, in ANP, through priority size, we can describe the "level" of the influence. Something similar like in correlation. Variable X and Y can be in "higher" and "lower" correlation.
NM- Well, if you believe that give me and example and reasoning. I don’t buy words. And what is ‘priority size’?
Sorry, I don’t see the relation with correlation. In here it is the measure by which 2 variables change together. In our case, the influence of A on B is a cause-effect relationship, which is absent in correlation.
NK- In the paper mill case, I am not an expert in the domain, and without a complete network of criteria and alternatives, I can't discuss the ANP analysis of the problem. I can only disagree that the paper mill can’t influence the plot of land where it will be installed. It will - because the paper mill will define its purpose and it will not be possible to build ex. the hotel in the same place.
NM- It is not the papermill that defines the land use, the stakeholders do that.
NK- This is your first reply where you introduced "ability to have the opinion" as the part of the definition of the term "influence". In all my replies, I am encouraging you to relax (unloose) the definition of the term "influence", but you are doing totally opposite: narrowing down the definition :)
NM-It is unrealistic to even consider that the papermill will itself defines its purpose propose the land to be used for a hotel….Come on Nikola, you know that it does not have any sense,
What you address is called analysis for different land uses, which could be the subject for a MCDM analysis.
It appears them that Saaty develop a new definition or mea ning for feed back. This is legitimate, but has to be proved, and he dd not.i
Best regards
Nolberto
- Badges
- Science topic
- Similar topics
- Psychology