I'm supposed to do other things today but I can't resist. Here's a comment on
John Campbell's very thoughtful reply on ratch.

>Date: Tue, 12 Sep 1995
>From: "John C. Campbell"

>In near exhaustion after reading McKean and Krauss, let me ask a naive
question. Maybe in principle ratch models allow individuals all sorts of
(exogenous) preferences, but as a practical matter, at least for these fairly
primitive models we Japanologists are worried about, don't they generally assume
a single preference across a class of individuals? E.g. politicians want to get
reelected and bureaucrats (in Silberman's useful formulation) want predictable
and comfortable careers. We then figure out the logic of that model, what it
would predict for dependent variables of interest (what institutions look like,
content of public policy, whatever), and compare with reality. It is just the
same as simple economic supply and demand models, or maybe a hair more complex
since those can say that both buyers and sellers want to maximize money.
MMy sense is that once we look at more than one preference for a class of
individuals--much less let each individual have his own--that the models get too
complicated to be of much use. Of course they are more realistic, but we can't
use them for anything.

McKean replies:
Yep. That's absolutely right. Our research problem may well goad us into
assuming uniform preferences across a class of individuals. This can still be
useful though. If the predictions one makes on this basis hold, you can safely
conclude that you got enough right, both the preferences you assumed and the
strategies you expected people to follow. You win and your research is over. But
more interesting things happen if your predictions don't work out. Then you know
that EITHER you were wrong about the preferences, or about their uniformity
across the class, or about the strategies that were really available to the
individuals you examined. Without abandoning the initial ratch assumption (that
individuals rationally pursue their preferences), you can decide to go back and
re- examine the preferences you assumed (is there a better way of getting at
them perhaps?) or the context in which your individuals operate. Bewildering
findings force a closer examination. Thus (is Chalmers [Johnson -- moderator]
out there) research built on tautology (treating behavior at time 1 as revealed
preference and then examining behavior at time 2 and time 3) can force us to
look harder, dig deeper, try to figure out what we got WRONG about either the
preferences or the institutional/structural/cultural context so that we can
improve our predictions on the next round. Thus I've always been very happy
combining ratch theory with close area study work -- they fit smoothly together
for me.

Often ratch is just a heuristic device to make us work harder. Maybe we clumsily
assume that political candidates seek re-election but of course experience a law
of diminishing returns in terms of the value to them of votes beyond what they
need to secure re-election, and they work less hard to get those "extra" votes.
We can then explore why some politicians seem to want LOTS of extra votes (maybe
there are special bennies available to elected representatives who clearly come
from safe seats, like Takeshita and Hashimoto???? minor bennies like control of
the party????) and why politicians with NO chance at winning run anyway (Steve
Reed always asks about the JCP, and one could also ask about the 72 candidates
for the 4 local seats in Tokyo during the July 95 election). Of course some may
run repeatedly because they somehow expect to do better each time and eventually
to win. But still, we are perplexed. Even a religious ratcher has to admit being
perplexed.

Finally (sorry this is already so long) sometimes one can break up the "class"
of individuals into a couple of different classes without going all the way down
to interviewing each individual's psychiatrist to get the preferences. A couple
of really interesting papers on this were presented at APSA in Chicago, and
although I haven't read them the presentations were fascinating. Bob Bates and
Barry Weingast did a formal simulation of the conflict in Serbia in which they
tried to figure out if a "pivot" group of rabid Serb nationalists within an
otherwise milquetoast population could operate to enable Milosovic the rabid
into winning political power. The idea was to break the population into
subgroups to see what would happen. A second paper like this was by David Laitin
and James Fearon, which similarly modelled imaginary ethnic conflicts by
breaking the ethnic groups into leaders, polite followers, and terrorist
followers who keep breaking the peace between the groups. Pretty interesting
stuff. (And of course, methodologically safe, since both papers were formal
simulations using invented people, subgroups, and preferences.) But they came
out with interesting observations and possibilities about the emergence of
ethnic conflict and the potential for controlling ethnic conflict.

The best and safest use of ratch theory in empirical work is when we don't have
to fear that we've got the preferences too far wrong.

Back to other stuff for now.
Margaret A. McKean
Department of Political Science
Duke University
Box 90204
Durham, North Carolina 27708-0204
direct telephone with voice mail: 919-660-4340 departmental FAX: 919-660-4330