Transcript - Lorentz Center

Where Explanation Ends:
Understanding as the Place the
Spade Turns in the Social
Sciences
Stephen Turner
The End of Explanation
•
Explanation as “substituting one mystery for another” Mill
•
A well-known scientist (some say it was the philosopher Bertrand Russell)
once gave a public lecture on astronomy. He described how the Earth orbits
around the sun and how the sun, in turn, orbits around the centre of a vast
collection of stars called our galaxy.
•
At the end of the lecture, a little old lady at the back of the room got up and
said: “What you have told us is rubbish. The world is really a flat plate
supported on the back of a giant tortoise.”
•
The scientist gave a superior smile before replying, “What is the tortoise
standing on?”
•
“You’re very clever, young man, very clever,” said the old lady. “But it’s
turtles all the way down!”
Which Gives us the Problem
Turtles all the way down
or
Ending Some place
What I am looking for
• The explanatory element that will be of concern to me
here is the something that, added to the particular facts,
makes for an explanation: the special inference ticket
that allows us to go from particulars to an account of the
thing.
• Ryle: A law is used as, so to speak, an inference-ticket
(a season ticket) which licenses its possessors to move
from asserting factual statements to asserting other
factual statements. It also licenses them to provide
explanations of given facts and to bring about desired
states of affairs by manipulating what is found existing or
happening.
Diversity, or One end point?
• are the social sciences themselves a unity, or
do they amount to a collection of Dupré-like
semi-independent domains, each of which has
its own explanatory end-points?
• Are the kinds of things social scientists ordinarily
work with, such as rational choice models,
“mechanisms,” and causal models, actually
explanatory or do they depend on something
else which is the real end-point of these
explanations? E.g. evolutionary psychology and
sexual selection?
The Forms Answer
• When is an explanation complete, complete in the sense
that it is satisfactory “on its own terms?”
• That will depend, of course, on what these “terms” are.
But here we can arrive at some answers:
• For a covering law explanation, it is complete when the
premises for the argument that makes up the deduction
are filled in by true generalizations, including the ultimate
inference ticket, a genuine law.
• For a narrative explanation of a particular kind, it is when
the elements of the narrative form can be filled in with
the right kind of facts. The form itself is the ticket that
makes the particulars into an explanation.
Contextualism of a Kind
•
If we stop here, with the recognition that there are
multiple forms of explanation in social science, each of
which has a conventional stopping point, we get a kind of
contextual account.
• We also get a kind of answer to the problem of the role
and nature of understanding: understanding is relative to
explanatory forms, forms which are themselves
paradigms of the understood, for which no further
understanding is required. To fit an account into one of
these pre-understood forms, such as a rational
explanation, is the same as providing an understanding.
And this comports well with actual social science, with its
extraordinary diversity of approaches and explanatory
practices.
But, no such happy ending
• the situation in social science is not like
the felicitous menagerie of domain
relevant explanations in John Dupré’s
descriptions of natural science. The
animals in social science don’t stay in their
cages, but compete with and not
infrequently attempt to devour one
another, or at least to discredit and
supplant one another.
Two Big Problems
• Intelligibility: In the case of physics, it was more
or less accepted that explanations would not
make sense, and that whatever sense the
physicist could made of predictions was enough.
• Social science has some analogues to this, but
they are controversial and rare. So one problem
with social science is the role of intelligibility: do
social science explanations have to end in
something intelligible?
Second Problem: Models
• No genuine laws
• But plenty of good, intelligible, and semipredictive models, which are nevertheless
literally false.
• So there is a question about how to understand
these models: are they the kind which depend
on some sort of intelligibility which we share
with the subjects whose doings are explained by
the model, or is it external intelligibility, the
sort involved in making sense of predictively
successful models, by analogies, for example.
Correlations
• what is a “successful” explanation in social
science?
• If prediction plus intelligibility is enough,
does it matter that the predictive
successes are different– mainly involving
correlations– than in physics or other
natural sciences?
Causal Models as a Case
•
These are models that are made up of correlations.
• From the usual practitioners point of view, modeling is a kind of
hypothesis testing, in which the researcher starts out with some
expectations about what is likely to correlate with what, a list of
standard nuisance variables or confounders that are known to
obscure relationships and can be controlled for statistically, and
relationships that are understood, in the sense that they are known
to be causal or potentially causal, and a data set of ordered tuples.
• The researcher then guesses at a causal interpretation of some
phenomenon of interest– executive pay levels, political instability,
pregnancy, a disease, success in school– and tries to construct a
model, in the form of a set of causal arrows, which indicates what
the lines of causal influence are. The data, or considerations of fit,
determine which elements of the model, which causal relationships,
are actually predictively relevant, and thus explanatorily relevant.
Where is the inference ticket?
• In the ‘assumptions’ about what causes
what, which are not tested by
considerations of fit.
• So the usual strategy is to minimize these
or rely on those that are too innocuous to
reject.
• Then one can apply technical means to
deal with confounding, ordering, etc.
Techie answer to being complete
• the statistics themselves tell us when the explanation is done:
classically, as in Yule’s study of the effects of outdoor-relief,
variables are added to the regression equation until they don’t
change anything about the predictions.
• If all the variables that might be expected to make a difference,
namely those that cannot be placed into the “not affected by”
category a priori, have been eliminated on this “makes no
difference” ground, the model is as complete as it can get.
• More recent methods designed to detect spurious relations work on
the same principle, but they are directed at excess correlation in a
path on a graph, which indicates the presence of some causal
influence that is not included in the model.
Slouching Toward Plausibility
• Being “done” in these cases, in short, comes down to
two things: claiming that there is nothing that could be
added that would make a difference to the predictive
results of the model, and claiming that the causal
assumptions have been reduced to innocuous and
uncontroversial background knowledge.
• Often, of course, innocuous background knowledge is
not enough, and the more flexible standard of
“plausibility” is applied. This standard slides gradually
into something else: intelligibility or causal plausibility.
And in practice, the list of variables, and the way they
are conceived and measured, reflects and reproduces
background knowledge about what can affect what.
•
Robustness or Apply when they
apply
• a kind of completion that is not provided: there is
no sense in which one expects to be able to say
under what conditions the model is applicable.
Doing so would amount to universalization.
• But for these models, one can only say that they
apply when they apply. The basic relationships
themselves may simply cease to operate under
novel conditions, or merely different
circumstances, and there is nothing general that
may be said about this.
So what does one get?
• the kind of causality that amounts to little
more than a predictive relationship that is
not excluded as non-causal a priori.
• Everything else one gets in the way of
understanding or causal content one gets
by adding background knowledge.
The Alternative: Mechanisms
• Mechanisms differ from causal models in that they have more
content. Some of what figures as “background knowledge” or means
of interpretation of relationships for causal models appears within
the model of the mechanism itself.
• A typical example of a mechanism is Merton’s account of machine
politics: the “Bosses” secure the support of disprivileged groups,
such as recent immigrants, by providing services or enabling access
to public services. Support, which appears irrational, now makes
sense as rational by identifying the processes by which each side
benefits. This mechanism does not automatically develop out of the
conditions of exclusion and so forth, but needs to be employed by a
Boss in order to work, like any complex scheme of exchange. But it
is also a model that can be employed by others in similar settings.
Also apply when they apply
• Mechanisms in this sense have the same “applies when
it applies” character of causal models: they are there to
plug in to account for input-output relationships with
similar inputs and outputs, but the inputs and outputs do
not determine the mechanism or produce it.
• The epistemic function of mechanisms is this: they are
free-floating intelligibility-producing devices that fill in
between inputs and outputs in a way that is more
satisfying– more understanding-producing– than
“predictors which are not a priori no affect relations.”
When are mechanism explanations
complete?
• They seem to be complete when the elements of
the mechanism are screwed together and work
to connect the inputs to the outputs.
• The elements which form the connections,
however, are the puzzle. If mechanisms are
better explainers than correlations, it would be
odd if the elements were merely correlations
themselves. Then one would have to ask why
casual models are any different than
mechanisms.
So they have to be something more
• One answer might be that mechanisms
resemble genuinely explanatory models
that are already intelligible to us– this
would be Intelligibility Type II, the
intelligibility of models as models. But this
would make the intelligibility that is
conferred into analogy.
• Another– the elements are intelligible acts.
The Analogy Issue
• The concerns of the promoters of mechanisms may not
seem obvious, but it should be clear that they intend to
contrast mechanism explanations from the kind of
psuedo-mechanisms that flourished under Talcott
Parsons, systems theory, Jürgen Habermas, Niklas
Luhmann, and Pierre Bourdieu, in which “society” did
things like “equilibrate” or have “steering mechanisms.”
These were overtly analogical “explanations” in which
the parts that would enable the analogy to work were,
routinely, either missing or entirely hypothetical.
But these analogies do confer
understanding
• What makes these accounts attractive is that
they confer a kind of understanding, analogous
to the kind of understanding one gets of a
physical science model, but without anything
very predictive going on.
• An end, but also a dead-end: no good way of
selecting between the analogies, or getting rid of
them on the grounds that the underlying model
is false or non-predictive, since there is no
predictive empirical underlying model.
Why Mechanisms are better
• mechanisms in sociology, in contrast, has focused on an attempt to
be more concrete than correlation, which means filling in or
replacing correlations with something substantive. Human action is
at the core of most mechanism explanations, so writers like Peter
Hedstrom have moved on to argue that the kinds of mechanisms
needed are made up of human actions, accounted for in terms of
desires, beliefs, and opportunities.
• This is essentially the model of action explanation found in Donald
Davidson and Max Weber, so the issues with it are well-understood.
2 Andrew Sayer. 1992. Method in Social Science: A Realist
Approach, Thousand Oaks, CA: Sage, pp. 104-105.
• 3 Andrew Collier. 1994. Critical Realism: An Introduction to Roy
Bhaskar’s Philosophy, New York: Verso, p. 20.
– 4 Collier. Critical Realism: An Introduction to Roy Bhaskar’s Philosophy,
p. 20.
Critical Realism
• a third way of talking about mechanisms that appears to
differ both from the analogical mechanisms of
teleological social theory and the kind of in-filling rational
action models.
• basic idea of critical realism is that the observable reality
is produced by unobservable mechanisms that are real,
have generative or causal powers as well as causal
liabilities or “vulnerabilities to being impacted by the
world in various ways.” Mechanisms operate when
something “triggers” them. These powers and so forth
are accessible through “conditions for the possibility”
reasoning, rather than in the way theoretical objects are
supposed to be accessed.
Where this works
• Power is real– exercised or not.
• account of power as something that people have but
need not actually use if others know they have this
power and act in ways which respect this fact.
• One could say that the existence of power which is never
exercised and thus never observed is a condition for the
possibility of the kind of power relations that actually
exist in politics, namely the kind in which people make
decisions in terms of the potential triggering of
mechanisms that really exist but are never used, and are
therefore inaccessible to “positivism.”
Where it collapses into action
• Is this use of power another analogical usage, or is it something that
can be cashed in by translating it into the language of action?
Certainly the literature in this area favors the use of physical,
analogical language.
• But this isn’t the kind of language that stands on its own: we either
accept it because there is some compelling reason to accept it, or
because we can translate it into something we already accept, and
accept as a genuine completion of the explanation.
• The power of a boss rests on his ability to produce results, results
that typically depend on people co-operating with them because
they believe that the Boss could and would act in particular ways
that would harm or benefit them.
• If this is what we mean by power, the term is analogical, and
translates fully into action language. Terms like “triggering” simply
conceal the fact that for someone to trigger the power of the political
machine people need to act on their beliefs and in terms of their
opportunities and desires.
So– action or bust.
• When we look at mechanisms made up of
actions, or make sense of correlations by
identifying typical action patterns (decisions, for
example), where do these end?
• As explanations, they are ideal-types. So the
claim is that they a) make sense (are more
intelligible than the morass of individual actions)
yet represent the actions that make up the
pattern.
Older Issues with Ideal-types
• He calls ideal-types “intuitive idealizations” to
distinguish them from idealizations in physics
(which he suggests, using the ideal gas laws as
a model, are “ideal” only in the sense that they
involve extreme values not to be found in real
cases),
• Hempel– didn’t make sense other than as yet to
be completed nomic explanations, by supplying
the full conditions for the application. But
acknowledges their role in lending intelligibility.
Hempel’s Bottom Line
•
For Hempel, they are treated as sources of hypotheses which might
be made into empirical theories: “their function is to aid in the
discovery of regular connections between various constituents of
some social structure or process.”
• Treating ideal-types as true subject to ceteris paribus clauses, he
notes, will not suffice, because these qualifications make the
formulation irrefutable and empirically irrelevant. Consider the claim
“Q will be realized whenever P is realized” all other things being
equal: “since the protective clause does not specify what factors
other than P have to be equal, (i.e. constant) or irrelevant if the
prediction is q is to be warranted, the hypothesis is not capable of
predictive application to concrete phenomena.
• i.e. they apply when they apply, but this is not good enough for
explanation.
• The only thing that would make them explanatory would
be for them to function as theories, that is, for them to be
claimed to be true.
• But economics, on this account, into the category of
irrefutable and thus empirically irrelevant. Deducing
results from “postulates” which represent ideal forms of
behavior precludes a “theoretical basis for an appraisal
of the idealization involved”. But he thinks that economic
theory can be saved if economic theory could be
deduced as a special case from a more general theory of
social action, of the kind which, he optimistically notes,
were being proposed at the time (by Parsons!).
• The theoretical systems are then understood in
accordance with the standard conception of
empirical theory, and are thus testable on the
basis of predictions about observable
phenomena, because, as special cases, their
‘area of application’ is defined. Open ceteris
paribus clauses mean that any predictive failure
can be claimed to be the result of “external”
conditions.
• This is an answer to the completion question:
completion occurs when there is a valid
empirical theory.
A More Basic Distinction
• The type is an idealized example of an
intelligible action– a narrative. These examples
of intelligible actions are already described in a
way that they are “explained.” They don’t need
to be backed by anything else.
• The problem is whether the idealizations apply
to anything.
• Theories, in contrast, need to meet at least three
standards: to be intuitively intelligible, to be in a
form that is explanatory, and to be true.
Intentions
• If we grant this– that individual intentional
explanations are complete and intelligible
without reference to general laws– just by
filling a narrative form, we get back to an
earlier answer.
• But this answer has its own troubles: are
narratives merely interpretations, of which
there may be many consistent with the
facts?
Hermeneutic Circles
• One answer is yes: that they are never
complete– that attributions of intentions are acts
of interpretation.
• We can revise the hypothesis and return to the
evidence and test it again, in a perpetual
hermeneutic circle. Interpretation is
interpretation all the way down. But the core fact,
the intention, is a mystery which is inaccessible,
because the minds of others are inaccessible.
demystifications
• The locus classicus for the view that intentional
explanations are hypotheses like any other, is
Hempel’s discussion of rational action, in which
rationality is analyzed as a disposition.
• This has the effect of making rational intentions
into an unobservable, and specific attributions of
intentions into derivative by-products of
generalizations about dispositions and their
outcome in action under difference
circumstances.
Anscombe
• Variation of the narrative version:
• the only thing one needs or can get in the way of
an account of intentional action is a correct
description. Explanation is impossible, because
the correct descriptions are non-explanatory and
typologizing rather than causal. These
descriptions are a dead-end. No additional
causes or explanations are needed or possible.
So the end point of these explanations becomes
not a law but a concept, the concept used in the
description.
Why do Narratives Explain
• Why are some descriptions explanatory
and others not?
• Anscombe’s own account seems to be
this: the redescriptions are backed by
practical syllogisms.
• But this in itself is not enough, nor is it
correct, as one can see from her own
examples of Aristotle reasoning about wet
and dry food.
The Later Peter Winch
• “the understanding we already have is
expressed in the concepts which
constitute the form of the subject matter
we are concerned with. These concepts
on the other hand also express certain
aspects of the life characteristic of those
who apply them.”
So concepts are the dead end
• They happen to contain the tacit
knowledge that grounds explanation. The
tacit grounds are then the real dead end,
but they are an inaccessible mystery, so
the de facto dead end is the concepts,
which are explanatory.
Lots of problems
• Alien concepts– are they understandable?
• How are concepts “understood”? Or are
they just “grasped”?
• Does conceptual relativism imply no
understanding, or is it via something like
similarities in underlying practices– we
understand Zande beliefs because there is
a similarity to our religious practices, which
we already understand/
Is there an alternative?
• Empathy. This is direct understanding of
action and beliefs of others.
• It is the end of the line.
• “Concepts” themselves can be omitted.
• Quinean ostensive definition is an
example of empathy in action.
• Translation can account for variant beliefs.