Hello,
I am doing a review on an article dealing with SD-model validation.
Now I do not get the point, what the difference between direct and indirect structure tests is.
Is it possible to explain in two sentences, what a structure test for model validation is?
Thank you,
Lukas S
structure tests for SD models
Not sure I am answering your specific question but here is 2 cents on the subject:
Vensim's "Reality Checks" may help you gain confidence in a model and a sense of its relevance and practicality in helping solve a problem. I would venture to guess that most SDers strive for a model to be useful as opposed to being 'valid'.
Regards,
Mike
Vensim's "Reality Checks" may help you gain confidence in a model and a sense of its relevance and practicality in helping solve a problem. I would venture to guess that most SDers strive for a model to be useful as opposed to being 'valid'.
Regards,
Mike
structure tests
The notion of structure is already vague. Do the parameters belong to the structure for instance?
The notion of validity is something too subjective.
Can you give the link for the article you are writing about.
I share Mike's idea about usefulness. A model is automatically valid if it is useful. And I do not mind if the model is 'valid' or not as long as it is useful.
Regards.
JJ
The notion of validity is something too subjective.
Can you give the link for the article you are writing about.
I share Mike's idea about usefulness. A model is automatically valid if it is useful. And I do not mind if the model is 'valid' or not as long as it is useful.
Regards.
JJ
This is how I worked through that myself.
A model just says "If these structures and decision rules (in the model) are correct, THEN this is what will happen".
The only thing that the structures can be tested against is reality, as in "is this river really flowing?". Am I missing something? The only thing that a graph can show is that a combination of factors can potentially create that graph...but I don't think it can say that the combination of factors is an accurate depiction of reality.
To use a simple obvious model as an example: If a farmer sprays his garden for weeds and his honey bees die...The model just says "IF weedkiller kills bees and as long as some bee decline isn't being caused by something else that's NOT in the model...then this could potentially be causing the honeybee decline". In that case, it seems like you could then assume that every aspect of your structure is questionable...Does weedkiller kill bees? Can it reach the bees? Is anything else causing the decline? The outcome in that case would be a bunch of questions that were formally buried under oversimplification.
A model just says "If these structures and decision rules (in the model) are correct, THEN this is what will happen".
The only thing that the structures can be tested against is reality, as in "is this river really flowing?". Am I missing something? The only thing that a graph can show is that a combination of factors can potentially create that graph...but I don't think it can say that the combination of factors is an accurate depiction of reality.
To use a simple obvious model as an example: If a farmer sprays his garden for weeds and his honey bees die...The model just says "IF weedkiller kills bees and as long as some bee decline isn't being caused by something else that's NOT in the model...then this could potentially be causing the honeybee decline". In that case, it seems like you could then assume that every aspect of your structure is questionable...Does weedkiller kill bees? Can it reach the bees? Is anything else causing the decline? The outcome in that case would be a bunch of questions that were formally buried under oversimplification.
Hi Barry
You are totally right. 'All models are wrong, some are useful' as somebody said. I prefer to say 'no models are right, some (few) are useful'. Another saying that I like is Coyle's: 'The key to sucessful modelling is to keep one's understanding of the model and what it says about the problem ahead of its size'.
And I would add 'well ahead of its size'. I do not believe in big models, because I cannot understand them and before using SD, I try to use much simpler techniques and common sense which is generally largely sufficient. And if I use SD, I try to get the solution out of very simple models and I spend a lot of time thinking with them. If their structure is wrong I will quickly see it, otherwise if it is a big model, it is easy to work with a structure that is wrong because it comes out of our hopes instead of a reality wrongly perceived and to realize it, a long time afterwards, having spent a lot of useless efforts. It is too much more difficult to see the link between the behaviour of a big model and the reality than with a small one.
Oversimplification is not bad if it helps you to think and you are conscious of the oversimplification. It is often useful because it is simple to understand and can deliver quick understanding even if does not give an 'overall result'. In real life overall result and explanation do not exist anyway in social matters.
Regards.
JJ
You are totally right. 'All models are wrong, some are useful' as somebody said. I prefer to say 'no models are right, some (few) are useful'. Another saying that I like is Coyle's: 'The key to sucessful modelling is to keep one's understanding of the model and what it says about the problem ahead of its size'.
And I would add 'well ahead of its size'. I do not believe in big models, because I cannot understand them and before using SD, I try to use much simpler techniques and common sense which is generally largely sufficient. And if I use SD, I try to get the solution out of very simple models and I spend a lot of time thinking with them. If their structure is wrong I will quickly see it, otherwise if it is a big model, it is easy to work with a structure that is wrong because it comes out of our hopes instead of a reality wrongly perceived and to realize it, a long time afterwards, having spent a lot of useless efforts. It is too much more difficult to see the link between the behaviour of a big model and the reality than with a small one.
Oversimplification is not bad if it helps you to think and you are conscious of the oversimplification. It is often useful because it is simple to understand and can deliver quick understanding even if does not give an 'overall result'. In real life overall result and explanation do not exist anyway in social matters.
Regards.
JJ