During the last month I’ve learned a great
deal about the fundamental aspects of modeling. I thought that work for the
first paper was almost finished. So I started to seek information and write
about the theoretical aspects of modeling. During the process I found such a
good references that it actually helped me to figure out some totally new
functional forms that I haven’t thought before. It didn’t take long until I got
20 functional forms for the model. I looked for reading which would help me to
determine which functional forms to apply in the final model. I learned that
there are at least three fundamental aspects that one needs to consider. First,
one needs to think carefully what is the phenomenon that one is trying to
model. That is, what assumptions one have to make about the properties of the
model. Second, one takes a careful look at the data and examines it as deeply
as possible. Third, one have to consider what for the model is to be built.
That is, in what applications the model is to be used. The model have to work
in those applications.
My supervisor Kari Hyytiäinen told me that
model builders could basically be divided into process based and data based
people. It seems that economist tend to be often process based whereas
ecologist tend to be more data based. This takes us to the resent problems I’ve
been struggling with. When we started this modeling process we made some
assumptions about the properties of the model. This was before we even looked
at any data. Then we contacted ecologists and got involved with the great data.
We got into some arguments with the ecologists because they wouldn’t accept the
assumptions we made because, although the assumptions were reasonable, the data
didn’t support those.
So I jumped into the data based train for a
moments; I removed the theoretical element related to the yield increasing
effect of the soil phosphorus (STP) from the model. The ecologists were happy
and I managed to build a model that delivered the aspects that were observable
in the data. Then I started to focus on the application of the model:
optimization and simulations. That’s where I run into trouble again. First I
noticed that there was something suspicious going on with the optimization
process: why the optimal soil phosphorus path didn’t go to zero if it was only
bad in the model? I also noticed that absolute amount of the yield decreased as
the STP increased when the model was simulated. This happened because STP was
the bad guy in the model, but this was highly unintuitive and it wouldn’t make
sense in the real life. I realized that the problems related to the missing
model element describing the response to STP and the transition function
describing the soil phosphorus dynamics.
At this point I had a meeting with a great
environmental economist, Antti Iho, who knows a great deal about phosphorus
modeling, among other things. He also noticed what was wrong. We realized that
the transition function, which was suggested by the ecologists, was a failure.
It only worked on a very limited data range. This is the problem with highly
data based modeling; the produced model makes sense on a limited domain and
thus it is not useful for any applications. I don’t think it’s a good approach.
Anyway, Iho also said to me that we just couldn’t remove the increasing element
of STP; the simulation and optimization result will be biased. This was true of
course. It only confirmed my suspicions; the initial assumptions of us were
correct. The problems won’t disappear if we just look the other way for a
Hence, here I go again, in the conflict of
the two modeling camps. But now I know, that the tree fundamental aspects of
modelling must hold simultaneously. Otherwise the model just isn’t good.
Currently I try to overcome this challenge, related to the mystery guy, STP, by
applying additional data from the great report by Saarela (1995). Hope the
model will be ready soon, before I completely loose my head with it.
Best regards: Matti Sihvonen