Very Interesting discussion!
I feel we have to clarify 3 things before advancing in this discussion
1- Social science is not special
2- Mathematics is the language we ask questions to nature and simulations are sophisticated math, not a science
3- Science is not a religion. Science is not interested in truth but in better explanations and deeper understanding of nature
For example: Simulations with Socyiodynamica show unexpected dynamic features in the working of a free market of 2 goods; or in the establishment of cooperative societies. An empirical search for these features finds them in real economies. The simulations helped us explore reality, not testing the truth of theories. Another example is models of subatomic particle physics predicting the existence of certain particles which are later found by experimental physicist. Simulations are fundamental in these models
Some Social scientists thing social science is special because it studies humans which are special. I see this as a pre-Galilean anthropocentric arrogance that is unscientific. Science is about humility and incremental increases of knowledge, not about truth or definitive proof. See the book What is Science?, nicely criticized by non-experimental “Social Scientists” in JASSS for. more details
Happy New Year
Klaus Jaffe Carbonell
http://atta.labb.usb.ve/Klaus/klaus.htm
El 1/3/2014 7:18 AM, Andy Turner escribió:
Hi,
I agree with the points that Eliot and Francesco make.
It is possible to prove the falsity of a theory by contradiction using simulation, but if the domain is unbounded or if some parts of the domain cannot be completely tested, then there are no guarantees that such a contradiction will be found. There can be a lot of seemingly wasted effort in trying to find contradictions, but the accumulated evidence of not finding them does give us some idea about the uncertainties. In mathematics, if you find the contradiction it is proof (albeit that some would argue a relatively week reductio ad absurdum proof) that a theory is false (founded on some assumptions that are not completely true).
In reality there are often weird exceptions to the general ways we might expect things to behave. Interesting excitations abound!
For stochastic probabilistic models there are some things (perhaps they can be called proofs) that we can use to confirm that the probabilities we are using are unbiased. But in reality things are complex and we lose the ability to prove or confirm the correctness of the probabilities we use in our models and have to resort to some form of model fitting exercise to get our simulated results match observations on average. I argue that when we develop stochastic models, we need to allow for just the right amount of randomness so that the extreme results are at least theoretically possible. It turns out that this is extremely hard to do and you can use up a lot of energy trying to come up with a better pseudo-random number generator!
As an example one thing I've considered is the simple case of the probability of death of an individual of a given type in a given time period. Suppose we have a count of the deaths of this type of individual in the complete system within a time period. Suppose we also know the population of that type of individual at the start of the period. Then assuming an even distribution of deaths over the period, it should be possible to estimate the probability that an individual of the given type will die in any sub-period. There are many complications with this, not least being that the type of individual may change during the time period. Consider for example changes of age, and changes of health status. I could digress, but let me try to get to the point. If we estimate the probabilities, we can then run a set of simulations from the same start point using different random seeds for the stochastic simulation. If we run this lots of times and the average simulated death rat
e is see
n to be
biased we can scratch our heads and wonder if we have done all the integral calculations right. We can make adjustments based upon our rethinking and try again. Soon we may learn our mistakes and everything might seem logical and nice again... As well as rethinking, we can also make adjustments based upon a large number of runs that are used to estimate what the probabilities should be. Whilst this is not ideal, I think it is perhaps the only practicable way to proceed as the statistical modelling gets very complicated very quickly. With colleagues, I did manage to get a model working that had not just death, but pregnancy, miscarriage, birth and migration in it and the results without migration did on average produce the expected number of deaths, miscarriages, births (twins and triplets). Migration added lots of complexity, but that is another story. I probably should write this work up, but we ran out of time and funding...
It is often not the average answer that we look for, but the variation and some idea of the uncertainty at the extremes.
May peace be with you.
Andy
-----Original Message-----
From: News and discussion about computer simulation in the social sciences [mailto:[log in to unmask]] On Behalf Of Fr. Kalvas
Sent: 03 January 2014 10:36
To: [log in to unmask]
Subject: Re: [SIMSOC] FW: [SIMSOC] ABM To Test Theory ...
Dear all,
I am popperian guy in the case of proofs of social science theories. I am convinced we can not prove them and because models are also theories, I am convinced we can not prove ABMs, as well.
I see strict difference between math and sciences. Math, as a humanly created system (or language according C. P. Snow), has its axioms, so we can link these axioms with our statement through the means of logic and by doing so we can prove it. But in the sciences we are looking for such axioms in fact, for axioms of our world. That is why we can't prove theories - we have no axioms for proving theories and we can't prove axioms (whatever we can prove it is not axiom, in fact).
So, my opinion is that we can't prove models and we also can't use them for proving theories. But we can prove whether theory states clear and deliberative relations between micro and macro levels/features/nodes/actors/agents. I know, it is less than proof, but it still worths.
I also think it worths to translate our thoughts, hypotheses, and theories about real world into ABMs. I think it is better to have consistent ones than inconsistent, and I also think it is better to have model of reality and resign to bring its final proof than resign to construct any model. Because as popperian guy I am convinced that we have rejected and yet non rejected models and that there are no proven models.
I wish you all the best,
Francesco
"Rich, Eliot" <[log in to unmask]>napsal/a:
Dear Sylvie (and list members)
Thank you for the reference to your paper.
I have a continuing interest in how simulation scientists validate (gain strength and confidence in) and verify (demonstrate correctness) models, so please forgive me for inserting a possibly contentious statement about the premise of proof. Proof of a theory is different than demonstrating a "clear and deliberate relationship" as described in Edmund's post. In mathematics, we can construct proofs by bounding the rules and logics acceptable for our problem. Unlike mathematics, social simulators do not have the tools or logics needed to ensure provability in our chosen reality.
Models, whether ABM or others, simulate properties that exist only in reality. When a simulation provides evidence that supports a theory, it can help us learn about the theory's strengths and its applicability to problems, but it does not prove that our simulation is correct. Nor can it, unless we believe we are exactly replicating the processes that happen in society.
In my own work in dynamic simulation, we attempt to disprove theory through simulation, rather than prove it. I argue that the application of simulation to Reductio ad absurdum is an important and valid application of our craft. But still, it's not proof.
A wonderful statement of the concern was provided by Oreskes, N. et al., "Verification, validation, and confirmation of numerical models in the earth sciences". Science February 4, 1994, pp 641-646.
What do you (and others) think? Does Proof vs. "Clear and Deliberate Relationship" matter?
Best,
Eliot
Eliot Rich
Associate Professor
Department of Information Technology Management School of Business University at Albany State University of NY, USA
[log in to unmask]
-----Original Message-----
From: News and discussion about computer simulation in the social sciences [mailto:[log in to unmask]] On Behalf Of Huet Sylvie
Sent: Thursday, January 02, 2014 12:59
To: [log in to unmask]
Subject: Re: [SIMSOC] ABM To Test Theory ...
I would say the Leviathan model starting from the Hobbes hypothesis on the individual dynamics, and not only since I have participated to this work (or at least I think so). That is a very innovative work proving the emergence of various leaderships (as the Leviathan one). It is based on a coupled individual dynamics of vanity and opinion propagation through gossip. It shows each one self-esteem is built through individual interactions which in turns lead to different organisations of people in terms of relations. These organisations can be usefully compared to various power structure forms. This is a recent work. It is continuing and promises a lot in terms of understanding. However, the first published paper cited below has already proved the interest of the individual-based model.
The Leviathan Model: Absolute Dominance, Generalised Distrust, Small Worlds and Other Patterns Emerging from Combining Vanity with Opinion Propagation. Guillaume Deffuant, Timoteo Carletti and Sylvie Huet (2013). Journal of Artificial Societies and Social Simulation 16 (1) 5. <http://jasss.soc.surrey.ac.uk/16/1/5.html>
Sylvie Huet
Laboratoire d'Ingénierie pour les Systèmes Complexes (LISC) IRSTEA
CS20085
9 avenue Blaise Pascal
Campus des Cézeaux
63178 AUBIERE CEDEX - France
Tél. (33) (0)4.73.44.06.15
http://motive.cemagref.fr/people/sylvie.huet
-----Message d'origine-----
De : News and discussion about computer simulation in the social sciences [mailto:[log in to unmask]] De la part de Edmund Chattoe-Brown Envoyé : jeudi 2 janvier 2014 16:52 À : [log in to unmask] Objet : [SIMSOC] ABM To Test Theory ...
Dear All,
What would you say are the best/most thought provoking/most persuasive attempts to test or build on existing theory using ABM? By this I mean not just _any_ ABM (which could be argued in a general sense to build
theory) but one that has some clear and deliberate relationship with a reasonably well known piece of published theory (like labelling theory, Marxism, the Hobbesian state of nature, Friedman's claim that profit making firms will drive out non profit making firms, the theory of Habermasian communicative action and so on.)
I'll summarise back to the list.
Happy 2014,
Edmund
--
Edmund Chattoe-Brown
[log in to unmask]
--
http://www.fastmail.fm - A no graphics, no pop-ups email service
