Thanks for "forecasting" instead of "prediction." Excellent choice for
discussing most social/economic models. I took "predictive" from
"predictive hypothesis" but as a noun (prediction) it carries perhaps too
much finality or connotations of precision.
As you point out with "ontological prediction," this predictability does not
need to be future oriented. It can be timeless (as with planetary orbits)
or even in the past (as with predicting new forms of fossils from those
already observed). The _future_ component lies in our observation, not in
the phenomena being observed. This is what makes hypotheses predictive,
even when the phenomena being observed is itself past, present, or future.
The SFI model of the Balinese water temples is a great example, btw, of how
models can explicate and simplify (no "central actor" was needed as we might
naively assume) -- but only insofar as the model can be shown to correspond
to elements in the real world. We don't know from the Balinese rice
management model if this is how the current system came to be (certainly not
with precision), but the model shows that it _could_ have happened this way
as a course of succeeding local minima (i.e., no improbable input of human
energy much less deus ex machina required). This is significant, in
particular if the model's explanation and correspondence leads to additional
questions and hypotheses.
As to whether model-building is academically justified, I personally would
go back to their forecasting/predictive nature. A model that makes possible
new hypotheses, and which adds to its own foundation of correspondence to
the physical world seems to me to be entirely justified. But then, as one
coming at this from more of a commercial than strictly academic view, my
bias is probably suspect.
Mike Sellers
> -----Original Message-----
> From: News and discussion about computer simulation in the
> social sciences [mailto:[log in to unmask]] On Behalf Of Dan Olner
> Sent: Thursday, June 11, 2009 2:33 AM
> To: [log in to unmask]
> Subject: Re: [SIMSOC] what is the point?
>
> Greetings,
>
> A few thoughts on the point of it all.
>
> 1. First-off, here's a link to a recent, very useful,
> conference paper, "what use are computational models of
> cognitive processes?" - from computational neuroscience, but
> entirely applicable to modelling generally:
>
> http://dl.getdropbox.com/u/306562/stafford123.pdf
>
> - all models are tautologies, different from "1 + 2 = 3"
> only in their complexity; it's up to the modeller to justify
> its connection to the real world, and which elements are of concern.
>
> 2. We need some better words to describe model purpose. I
> would distinguish two -
>
> a. Forecasting (not prediction) - As Mike Sellers notes,
> future prediction is usually "inherently probabalistic" - we
> need to know whether our models can do any better than
> chance, and how that success tails off as time passes. Often
> when we talk about "prediction" this is what we mean -
> prediction of a more-or-less uncertain future. I can't think
> of a better word than forecasting.
>
> b. Ontological prediction (OK, that's two words!) - a term
> from Gregor Betz, Prediction Or Prophecy (2006). He gives the
> example of the prediction of Neptune's existence from
> Newton's laws - Uranus' orbit implied that another body must
> exist. Betz's point is that an ontological prediction is
> "timeless" - the phenomenon was always there. Einstein's
> predictions about light bending near the sun is another:
> something that always happened, we just didn't think to look
> for it. (And doubtless Eddington wouldn't have considered
> *how* to look, without the theory.)
>
> This - I think - fits Epstein's distinction between
> earthquakes and tectonic plates. Discovery of the latter is
> an ontological find. And - vitally - the ontological
> prediction does indeed allow for some solid future prediction
> beyond forecasting (which I'm defining as "probabalistic
> prediction, becoming les distinguishable from chance as time
> passes"). It places a boundary around possible future
> outcomes that we didn't know about before: Earthquakes will
> mostly happen on or near lines of tectonic plates.
>
> How valuable ontological prediction can be depends entirely
> on context. J Stephen Lansing and James Kremer's model of
> Balinese rice management is my favourite example. Attempts to
> apply Green Revolution-style growing techniques to Bali were
> breaking their system. Lansing had tried to tell the
> authorities about the role that the Balinese water temples
> played in finding an optimal trade-off between fallow and
> growing that maximised yield / minimised pests, but no-one
> listened. They *did* listen, however, when Kremer helped
> Lansing model the system.
>
> So they made an ontological claim: here's a simple model that
> demonstrates a mechanism that manages rice production, in a
> way that appears to capture the salient features of the
> real-world system. (Cf. the paper above - the point isn't the
> simplicity of the model, but the ability of the modeller to
> theoretically justify its connection to real-world features -
> what Craik calls a "similar relation structure.")
>
> It's a story with a happy ending too: the model was
> convincing enough to get official support for traditional
> methods of rice management. If one were cynical, there's an
> argument to say all this shows is that policymakers have an
> unhealthy propensity to prefer "quantitative-looking
> answers", regardless of their actual philosophical merit.
> This may be self-serving, but I prefer to think that
> "ontological models" like this *can* reveal hidden phenomena
> in exactly the way Betz describes, and that - as the Bali
> example shows - this revealing of dynamics previously
> invisible to the eyes of policymakers is an entirely sound
> justification for modelling.
>
> This is, perhaps, what Epstein claims is happening in
> "generative" social science. Here, I agree with Mike Sellers
> - when Epstein says "growing it" can count as "explaining
> it", he's completely wrong. Generating a model phenomenon
> offers only a *hypothesis* about something that might be
> happening. As with the above paper, the tautology of the
> model still needs to be theoretically tied to the real world.
> This implies, I guess, that generative social science is a
> hypothesis engine, not an explanatory tool.
>
> There's also another very important reason to model, I'd
> argue, due to the double-edged nature of models: they can do
> exactly the opposite of the Bali model - keep things
> invisible by insisting they've already captured everything
> important about a system. Friedman's classic positivist
> argument underlines this: it doesn't matter at all, he tells
> us, whether the assumptions of one's model are realistic or
> not - as long as the model predicts successfully. That
> entirely fails in economics, where powerful policy bodies
> have consistently argued e.g. that a country liberalised too
> fast, or too slow - but at any rate, it certainly wasn't the
> fault of the model they were using if things didn't work out.
> Actually, Friedman is right - but as regards ontological
> prediction. Uncovering previously unseen dynamics in social
> systems - rather than forecasting - is a very useful outcome
> of a model. (J.C. Scott's ecological morality tale of German
> forest death, in Seeing Like a State, is a great analogy of
> the dangers of this kind of modelling and its relation to
> their implementaton by powerful bodies, this is why
> continuing to model systems we think we understand is so important.)
>
> 3. Dr Stafford also mentions models can be tools for
> cultivating a researcher's intuitions, but argues that can't
> count as an academic justification for modelling. Many of
> Epstein's 16 reasons come into this category. This is a
> tricky one: its an absolutely vital part of the modelling
> research process - right at the minute, I'm having to let
> myself off the 'academically justified' hook so I can get on
> with model-building to test out my thinking. It won't be
> good enough to be the foundation of my final PhD, but that's
> not the point - it's an absolutely necessary step to a useful
> result, I can use it to illustrate to others what I'm trying
> to get at, and it's a stage that cannot be skipped. I think
> we need to recognise the value of modelling in the research
> process, without getting snotty about it, while still
> striving to achieve something of academic worth at the end of
> it. I don't think there's any way to the lofty mountains of
> "academically worthy" except through the dark damp forest of
> modelling "what the hell I'm doing."
>
> To finish - Craik also notes: "[a] model need not resemble
> the real object pictorially; Kelvin's tide-predictor, which
> consists of a number of pulleys on levers, does not resemble
> a tide in appearance, but it works in the same way in certain
> essential respects..." [The Nature of Explanation, 1967,
> p.51] Here at Leeds, we've got the original hydraulic
> economic model, which - again - looks nothing like an economy:
>
> http://en.wikipedia.org/wiki/Moniac
>
> I quote that because I think it's vital to keep an eye on
> building 'relation stuctures' and not getting lost in pursuit
> of 'correspondences' with real-world entities; agent-based
> modelling structurally lends itself to the latter, of course
> - but (cf. Friedman above) that doesn't necessarily make it realistic.
>
> Cheers,
>
> Dan
|