How (not) to model social processes: Thick and thin models in the social sciences
From Santa Fe Institute Event Wiki
Contents |
People involved
(please put your name)
Status
We had a first meeting on Friday (week 1) and have met again on Tuesday. This discussion should lead to a project soon.
Resources
(please add more)
- Epstein, Generative Science: [1]
- one critique by Gruene-Yanoff: [2]
- Robert Sugden's view on the use of abstract models in economics: [3] (email Kai Spiekermann for the pdf)
- Robert Clower's (1994) critique of the modern current of abstract economic theorizing [4]
How to specify models
To specify models in the social sciences we need to set many parameters and make many assumptions:
- Agents: How much information?, How rational (bounded rationality, heuristics)?, stochastic elements (e.g. "trembling", random mistakes...)
- World: Abstract or high fidelity? Stochastic or deterministic? Static or dynamic (does the world change in time)?
- Interaction Agents-Agents: Fully defined as a (possibly multistage) game? If not, how to specify interaction. Link with bounded rationality?
- Interaction Agents-World: Static or dynamic (does the world change in time)?
Maybe it's useful to link these questions with some modelling theory in philosophy of science. A good intro can be found in chapter 1 of Michael Weisberg's thesis [5]. He also has some nice papers out [6].
The standard conception proposed by Giere consists of three elements:
- A model description
- A model
- A target system in the world.
A model description is usually a set of mathematical equations or a computer program. Model descriptions are typically uninstantiated, i.e. no assumptions regarding parameters are made. The model description typically singles out a family of models.
The model is one instantiated member of the family of models given by the model description. It is typically instantiated, that is the parameters are now set. The model describes a set of trajectories through the state space.
The model should stand in a similarity relation to the target system in the world. Defining or even measuring similarity is notoriously difficult.
With this conceptual framework, we can now think about models in the social sciences. I think we have at least three big, conceptually different issues:
- We can think about appropriate model descriptions, i.e. how we choose specific model descriptions. This includes debate on the basic assumptions about persons, rationality, the world, etc.
- We can ask why we look only at specific parameter assumptions within our model and we can question whether we are looking at the relevant instantiated models.
- We can ask how similar our models should be to the target system, and what should constitute similarity.
This is still a bit vague, but I hope it structures the debate! Kai Kai Spiekermann
Earlier Discussion
Models [7] have fascinated philosophers of science for a long time. However, computational modeling is quite a new technique for the social sciences, and there relatively little has been written on the use of computational models in the social sciences from a philosophy of science perspective.
Here are some questions:
- Is there a systematic difference between modeling in the (natural) sciences and modeling in the social sciences?
- How detailed should our models be? (Two paradigms seem to be evolving: On the one hand, some researchers try to "grow societies" from the bottom up. These are sometimes called "thick" models. On the other hand, there are "thin" models, which try to model only one specific social mechanism with highly idealized models.)
- How are models related to "reality"?
- How do we make sure that our models are relevant and avoid producing artefacts?
We could discuss several examples of models in the social sciences and maybe read some of the (few) existing papers in this field and try to define our own position. If you are interested, please leave a comment or send me an email to k dot p dot spiekermann at lse dot ac dot uk.
Kai --Spiekermann 17:38, 5 June 2007 (MDT)
I'm interested in this topic. I think it is important to consider the history of modeling in the social sciences through the so-called 'Quantitative Revolution.' I would also like to further explore the entrenched (or imposed?) quantitative/qualitative divide across the social sciences and possible stigmas or clasifications that result from these distinctions.
An additional potential topic: validating and replicating results in the (hard) sciences versus the social sciences.
Liz (emullane at ucla dot edu)
Fascinating discussion. Can you post a list of sources in addition to the links? The Wiley link was broken when I tried it. I'd be happy to track down a copy from the bibliography.
Cheers- Ben Mazzotta
Hi Kai, I'd be very interested to participate and think about the issue (especially the two last themes you mentioned -- How are models related to "reality"? and -- How do we make sure that our models are relevant and avoid producing artefacts?. Amelie 00:48, 6 June 2007 (MDT)
Hey, I posted a suggested meeting time on the related thread, Representing People, for lunch on friday after Scott Page's lectures. Perhaps from there we could break off to one of the conference rooms in Santa Fe Hall.
GREAT! Thanks for all the comments so far (more always welcome). Let's meet for lunch on Friday. I think some people want to go on excursions on Friday afternoon, but we can decide over lunch how we want to proceed! Kai
