Exogenous outputs from people and models.
Exogenous outputs from people and models.
Why is it that people seemingly have exogenous outputs but models never do ( well unless it is the trivial example of exogenous-in-exogenous-out ) ?
exogenous output
Hi Mike
Your question is difficult to understand.
What do you mean by exogenous outputs from people?
And what is for you an exogenous output from a model?
Is it the policies, the result from the policies like pay off, for example or the behaviour of any variable, or material flowing out of the model in a well?
What is the trivial example exogenous-in-exogenous-out?
Can you illustrate you question with a simplistic model?
Regards.
JJ
Your question is difficult to understand.
What do you mean by exogenous outputs from people?
And what is for you an exogenous output from a model?
Is it the policies, the result from the policies like pay off, for example or the behaviour of any variable, or material flowing out of the model in a well?
What is the trivial example exogenous-in-exogenous-out?
Can you illustrate you question with a simplistic model?
Regards.
JJ
Good day JJ,
i have been working on a model for a long time and was doing a review by listing each Vensim's View variables in the following categories for each view; 'endogenous inputs', 'endogenous outputs', 'key model variables' , 'exogenous inputs' and ... 'exogenous outputs'.
As soon as i looked at the spreadhseet i laughed at myself for absent-mindedly throwing in the 'exogenous outputs' category; although it did make me think a bit on it. Pondering an 'exogenous output' led me to think of the behavior of some people ... I found all this quite amusing; i guess i have a strange sense of humor.
Regards,
Mike:)
i have been working on a model for a long time and was doing a review by listing each Vensim's View variables in the following categories for each view; 'endogenous inputs', 'endogenous outputs', 'key model variables' , 'exogenous inputs' and ... 'exogenous outputs'.
As soon as i looked at the spreadhseet i laughed at myself for absent-mindedly throwing in the 'exogenous outputs' category; although it did make me think a bit on it. Pondering an 'exogenous output' led me to think of the behavior of some people ... I found all this quite amusing; i guess i have a strange sense of humor.
Regards,
Mike:)
exogenous ouput
Hi Mike
It is eventually possible to make a distinction between endogenous and exogenous data. Although the distinction is not always so clear. Is a parameter unchangeable exogenous or endogenous? It is eventually neither one nor the other. Logically if the parameter is changeable it belongs eventually to the policies and can be considered as endogenous.
But what makes the distinction beween an endogenous and an exogenous output? Results depend on endogenous and eventually exogenous inputs at the same time.
I personnally always build a report of all the constants, auxiliaries and levels, print the report and check their behaviour, at each time step and this for carefully chosen parameter values, range of subscripts and length of simulation to reduce the amount of results. It is very tedious and time consuming but very instructive and to my opinion mandatory to build model's confidence.
See my question about the finding of the proper order of the elements to include in that report.
Regards.
JJ
It is eventually possible to make a distinction between endogenous and exogenous data. Although the distinction is not always so clear. Is a parameter unchangeable exogenous or endogenous? It is eventually neither one nor the other. Logically if the parameter is changeable it belongs eventually to the policies and can be considered as endogenous.
But what makes the distinction beween an endogenous and an exogenous output? Results depend on endogenous and eventually exogenous inputs at the same time.
I personnally always build a report of all the constants, auxiliaries and levels, print the report and check their behaviour, at each time step and this for carefully chosen parameter values, range of subscripts and length of simulation to reduce the amount of results. It is very tedious and time consuming but very instructive and to my opinion mandatory to build model's confidence.
See my question about the finding of the proper order of the elements to include in that report.
Regards.
JJ
Response to JJ’s previous post;
JJ says:
“Is a parameter unchangeable exogenous or endogenous?”
Good point; one could argue that any model ‘item’ not both
influencing another model variable and influenced by another model ‘item’
could be considered ‘exogenous’ in some sense.
JJ says:
“But what makes the distinction beween an endogenous and an exogenous output? Results depend on endogenous and eventually exogenous inputs at the same time.”
Good point; i agree.
In terms of your reports of causality, it sounds like an excellent way to understand
a model well. I am sure Bob at Vensim could devise such a report and automate it;
I just have no idea as to how he views that as a profitable way to spend his time.
I run into similar labor intensive, tedious work doing ‘eigenvalue’ analysis of models
without Jim Hines’ software, which he apparently only makes available to a chosen few.
I hear you in terms of how much work modeling is if one tries to do it well. I think
that is the bane and beauty of doing SD modeling. The difficulty is the ‘bane’ of it; but that also is the beauty of it because that ‘difficulty’ provides a moat that rivals needs to traverse
in order to marginalize any competitive advantage that ‘difficult’ model provides you.
Regards,
Mike
[Edited on 3-28-2009 by mike]
[Edited on 3-28-2009 by mike]
JJ says:
“Is a parameter unchangeable exogenous or endogenous?”
Good point; one could argue that any model ‘item’ not both
influencing another model variable and influenced by another model ‘item’
could be considered ‘exogenous’ in some sense.
JJ says:
“But what makes the distinction beween an endogenous and an exogenous output? Results depend on endogenous and eventually exogenous inputs at the same time.”
Good point; i agree.
In terms of your reports of causality, it sounds like an excellent way to understand
a model well. I am sure Bob at Vensim could devise such a report and automate it;
I just have no idea as to how he views that as a profitable way to spend his time.
I run into similar labor intensive, tedious work doing ‘eigenvalue’ analysis of models
without Jim Hines’ software, which he apparently only makes available to a chosen few.
I hear you in terms of how much work modeling is if one tries to do it well. I think
that is the bane and beauty of doing SD modeling. The difficulty is the ‘bane’ of it; but that also is the beauty of it because that ‘difficulty’ provides a moat that rivals needs to traverse
in order to marginalize any competitive advantage that ‘difficult’ model provides you.
Regards,
Mike
[Edited on 3-28-2009 by mike]
[Edited on 3-28-2009 by mike]
exogenous outputs
Hi Mike.
Tracing the values of variables with a report, is one of the best way I found to find bugs of all sorts and have a better understanding of how a model works.
I have never seen a case when I did not find some sort of bugs when doing this, and often in unexpected places.
I do my models for myself and am very concerned with their quality.
I have three other choices to increase the understanding and checking of models.
First use the Ford method of loop knock out.
A variation was explained and is downloadable from the posters of the last conference. I join the pdf file with this post.
Another way is to use the Digest model that uses another method of studying loop dominance.
It can be donwloaded from the web site of George Richardson teacher at the Albany University.
Anothe way is to use eigenvalues.
And another way is to use Reality checks.
Reality checks is more about comparing the pre-established
understanding of the system that you already have than increasing the understanding of the behaviour of the model, but it still does that but inderectly. It is too automatic.
Reality checks is than more about checking than understanding and digest, eigenvalue and loop knock out is more about understanding the behaviour of the model.
I always prefer simple methods.
Eigen value analysis seems very complicated and I am not sure that it betters the understanding of the model.
Digest method seems too very complicate.
Fords method or its more recent version seems the easiest and I may try it if I have the time.
But the next method I will try and practice is the reality check
method. It certainly will necessitate time to practice it, but it
seems rather natural and straightforward.
I like to understand a model as I understand 1 + 1 = 2
The real problem is that most methods become more difficult to use as the model becomes complex.
For instance, tracing the value of variables becomes very time consuming when the model becomes bigger and there is a strong incentive not to do it, with the high risk ot the model being bugged.
It does not guaranty that there is no bug, because it needs to be done with all the possible combination of the parameters, which is often impossible.
Unit consistency and mass balance are automatic and powerful too, but still not enough.
The interesting thing with the reality check is that it is automatic. But it needs to be practiced to get used to it.
I have never seen any published Vensim model with any reality check being done.
I think that it prooves that the average quality of SD models is poor and that it explains why SD is so badly ignored.
Quality of a product is important when the market is very slowly growing and old. In fast moving market, the commercial side is more important than the quality of a product.
I believe that there is no miracle and you do not have
anything for nothing. You have to work with a model to get something from it, and preferably intelligently and with experience.
File joined, modified Ford method.
Regards.
JJ
PS: I cannot join the file: it is too big. But you can dowload it from the Athen's conference on the SD web site, sessions by thread complexity/agent base/non liner dynamic at the beginning of the list of papers, generalysed loop deactivation method from Phaff.
[Edited on 28-3-2009 by LAUJJL]
[Edited on 28-3-2009 by LAUJJL]
[Edited on 28-3-2009 by LAUJJL]
[Edited on 6-4-2009 by LAUJJL]
Tracing the values of variables with a report, is one of the best way I found to find bugs of all sorts and have a better understanding of how a model works.
I have never seen a case when I did not find some sort of bugs when doing this, and often in unexpected places.
I do my models for myself and am very concerned with their quality.
I have three other choices to increase the understanding and checking of models.
First use the Ford method of loop knock out.
A variation was explained and is downloadable from the posters of the last conference. I join the pdf file with this post.
Another way is to use the Digest model that uses another method of studying loop dominance.
It can be donwloaded from the web site of George Richardson teacher at the Albany University.
Anothe way is to use eigenvalues.
And another way is to use Reality checks.
Reality checks is more about comparing the pre-established
understanding of the system that you already have than increasing the understanding of the behaviour of the model, but it still does that but inderectly. It is too automatic.
Reality checks is than more about checking than understanding and digest, eigenvalue and loop knock out is more about understanding the behaviour of the model.
I always prefer simple methods.
Eigen value analysis seems very complicated and I am not sure that it betters the understanding of the model.
Digest method seems too very complicate.
Fords method or its more recent version seems the easiest and I may try it if I have the time.
But the next method I will try and practice is the reality check
method. It certainly will necessitate time to practice it, but it
seems rather natural and straightforward.
I like to understand a model as I understand 1 + 1 = 2
The real problem is that most methods become more difficult to use as the model becomes complex.
For instance, tracing the value of variables becomes very time consuming when the model becomes bigger and there is a strong incentive not to do it, with the high risk ot the model being bugged.
It does not guaranty that there is no bug, because it needs to be done with all the possible combination of the parameters, which is often impossible.
Unit consistency and mass balance are automatic and powerful too, but still not enough.
The interesting thing with the reality check is that it is automatic. But it needs to be practiced to get used to it.
I have never seen any published Vensim model with any reality check being done.
I think that it prooves that the average quality of SD models is poor and that it explains why SD is so badly ignored.
Quality of a product is important when the market is very slowly growing and old. In fast moving market, the commercial side is more important than the quality of a product.
I believe that there is no miracle and you do not have
anything for nothing. You have to work with a model to get something from it, and preferably intelligently and with experience.
File joined, modified Ford method.
Regards.
JJ
PS: I cannot join the file: it is too big. But you can dowload it from the Athen's conference on the SD web site, sessions by thread complexity/agent base/non liner dynamic at the beginning of the list of papers, generalysed loop deactivation method from Phaff.
[Edited on 28-3-2009 by LAUJJL]
[Edited on 28-3-2009 by LAUJJL]
[Edited on 28-3-2009 by LAUJJL]
[Edited on 6-4-2009 by LAUJJL]
Good day JJ;
I should re-visit the ‘Digest’ software you referenced. Thank you for the heads up.
A good exercise akin to your ‘report’ is to slog through a reduced form version of the model
( dx = Ax + c ; A = matrix , others arrays and x= model stocks ) that one encounters on the way to eigenvalue analysis.
Mike
I should re-visit the ‘Digest’ software you referenced. Thank you for the heads up.
A good exercise akin to your ‘report’ is to slog through a reduced form version of the model
( dx = Ax + c ; A = matrix , others arrays and x= model stocks ) that one encounters on the way to eigenvalue analysis.
Mike