Parameter of the model
Posted: Sat Apr 25, 1998 8:50 pm
Dear Colleagues:
I am a new learner in SD. I found SD as very interesting subject which is good
to represent mental model and mental simulation so that it easier to
communicate with others. It helps presenting and structuring all information
and factors in a quantitative way. However, I have some questions that anyone
may help me:
- How to calibrate the parameter of the model. Is that by trial error or it
has
guideline such as maximum likelihood?
- How do we represent it in the model if the data is not perfectly correlated.
Say, based on empirical data, the cofficient of correlation is 0.45 or -0.65.
- How do we represent the variation and distribution in causal-loop diagram,
instead of using average?
- How do we be so sure that the structure of the model we develop is the
correct
one? Sometimes we do not even know whether there is a relationship between one
factor to another. How do I know that this structure is wrong, something is
missing, or one or two relationship has reverse direction?
- If we can calibrate the parameter of one or two relationship, how do we
be so
sure that in the overall structure, that parameter will work?
Of course, above questions based on my assumption that there is a critical
true
parameter of every structure.
regards,
Kardi
Kardi Teknomo
Traffic Engineering and Transportation Planning Laboratory
Petra Christian University
Jl. Siwalankerto 121-131 Surabaya Indonesia
Phone: +62-31-843-9040 Local 1326
Fac. : +62-31-843-6418
e-mail: kardi@peter.petra.ac.id
I am a new learner in SD. I found SD as very interesting subject which is good
to represent mental model and mental simulation so that it easier to
communicate with others. It helps presenting and structuring all information
and factors in a quantitative way. However, I have some questions that anyone
may help me:
- How to calibrate the parameter of the model. Is that by trial error or it
has
guideline such as maximum likelihood?
- How do we represent it in the model if the data is not perfectly correlated.
Say, based on empirical data, the cofficient of correlation is 0.45 or -0.65.
- How do we represent the variation and distribution in causal-loop diagram,
instead of using average?
- How do we be so sure that the structure of the model we develop is the
correct
one? Sometimes we do not even know whether there is a relationship between one
factor to another. How do I know that this structure is wrong, something is
missing, or one or two relationship has reverse direction?
- If we can calibrate the parameter of one or two relationship, how do we
be so
sure that in the overall structure, that parameter will work?
Of course, above questions based on my assumption that there is a critical
true
parameter of every structure.
regards,
Kardi
Kardi Teknomo
Traffic Engineering and Transportation Planning Laboratory
Petra Christian University
Jl. Siwalankerto 121-131 Surabaya Indonesia
Phone: +62-31-843-9040 Local 1326
Fac. : +62-31-843-6418
e-mail: kardi@peter.petra.ac.id