Modeling Reality

Christopher Pincock, Purdue University (pincock@[Link])1

October 20, 2008

Abstract: My aim in this paper is to articulate an account of scientific modeling

that reconciles pluralism about modeling with a modest form of scientific realism.
The central claim of this approach is that the models of a given physical
phenomenon can present different aspects of the phenomenon. This allows us, in
certain special circumstances, to be confident that we are capturing genuine
features of the world, even when our modeling occurs independently of a wholly
theoretical motivation. This framework is illustrated using a recent debate from
meteorology.

I.

Traditional scientific realism is the view that science aims at truth and that we

have some reason to believe that our most successful scientific theories are true or

approximately true. Realists typically appeal to the predictive success of these theories

when challenged to say exactly why we should place such confidence in these theories.

These realist arguments can take the crude form of asking why you would get on an

airplane if you did not believe the theories underlying its construction. But, more often,

realists develop sophisticated explanatory arguments for their position. For example,

Psillos argues that the best explanation of the success of these theories is that they are

true or approximately true, and that we have a reason, in this case at least, to believe that

the best explanation of this phenomena is true (Psillos 1999).

Explanatory arguments for scientific realism have been challenged by two quite

different groups. On the one hand, there are anti-realists who rest at the level of

generality of their realist opponents, and who consequently argue that we never have any

1
An earlier version of this paper was presented at Models and Simulations 2, Tilburg University, Oct.
2007. I would like to thank the members of the audience for their helpful comments and criticisms,
especially Rafaela Hillerbrand, Wendy Parker, Jan Sprenger and Mauricio Suárez. The final version was
greatly improved by suggestions from Anjan Chakravartty and an anonymous referee.

1
reason to believe that our most successful scientific theories are true or approximately

true. Global anti-realists offer alternative and supposedly more tractable goals for

science, as with van Fraassen’s empirical adequacy (van Fraassen 1980). But, on the

other hand, there are anti-realists who descend to the messy details of scientific practice

and use this local perspective to undermine the arguments for global realism. It is here

that we find most of the work on models and simulations, from Cartwright’s How the

Laws of Physics Lie (Cartwright 1983) through Giere’s recent Scientific Perspectivism

(Giere 2006). While not all of these authors draw the same conclusions about the

resulting conception of scientific knowledge, the challenge to global scientific realism is

relatively consistent. It takes the form of what I will call “the argument from modeling”.

Perhaps nobody has presented the argument from modeling in the exact way that I will

here, but hopefully it is close enough to the concerns about realism that modeling practice

raises.

The argument from modeling emphasizes the limitations that our successful

scientific theories face in motivating the details of the scientific models that are used in

deriving conclusions about physical systems. These limitations are manifest in the

widespread idealization or otherwise seemingly ad hoc techniques of model construction

that invariably appear whenever we turn to the details of some scientific practice. If these

moves are not motivated by the theory in question, the argument continues, then a crucial

part of the success of the theory in prediction and testing is exposed as unrelated to the

truth of the theory. In particular, it becomes less plausible to claim that the truth of the

theory would explain the success of the theory because so many steps in the successful

2
application of the theory depend on non-theoretical assumptions.2 There are two kinds of

cases to be distinguished here. First, the assumptions used may be inconsistent with the

theory. In this case, the content of the model seems inconsistent with the claims of the

theory, and there is no link between the success of the model and the truth of the theory.

Second, the assumptions are consistent with the theory but are applied in a given case

even though the theory itself does not dictate this. Here things might be more congenial

to a realistic interpretation, but there is still the residual worry that the assumptions have

been applied in an ad hoc manner. As a result, the anti-realist will refuse to attribute the

success of the model to the theory.

What results from this argument is a kind of limited anti-realism which concludes

that the presence of non-theoretical assumptions in modeling practice should undermine

our confidence that the theory yields true claims about the situation modeled. This limited

anti-realism takes several forms. In Cartwright, for example, we are told that we should

doubt the scope of the regularities observed in our successful modeling contexts

(Cartwright 1999).3 Still, for a group that takes scientific practice so seriously, there is a

strange disconnect between these pessimistic conclusions and the optimism found in the

works of the scientists themselves. Scientists are in many, though by no means all, cases

confident that, despite whatever idealizations or ad hoc adjustments they may have made

in their modeling practice, they have justified a claim about a genuine physical

phenomenon whose scope vastly exceeds the limits that someone like Cartwright would

impose. The challenge, then, is to uncover the reasons that scientists have for making

2
This is one way to make sense of the emphasis on the autonomy of models with respect to theories. See
Suárez 1999 for a clear case of an argument from the autonomy of models to the rejection of scientific
realism.
3
See also Morrison 2005 and Suárez & Cartwright 2008.

3
these limited realist claims, and to see to what extent these reasons can be grounded in

reality.

My strategy is to segment the aspects of a situation that a model represents and to

argue that in some cases we can have a good reason to think that the aspect of the

situation that the model gets right agrees with a model motivated entirely by the theory. I

start with a schematic outline of how such an argument might go and link it only to a

simple example involving velocity. In order to show the viability of the approach I turn,

in the final section, to a more involved case study concerning hurricane intensity and

global warming. Here my original schema proves too simple as several different theories

and models come together to motivate the conclusion. I reconstruct this case as a repeated

application of the schematic argument.

II.

Here I want to lay out a template for arguments for the conclusion that we know

some aspect A of some system (or type of system) S. The premises of instances of this

template, when well justified, will give a scientist a good reason to believe the instances

of the conclusion. In line with the limited realist conclusions that I wish to draw here, I

will be happy to grant that these conditions are not always met in scientific practice. But

they are met in many cases where models exceed the scope of a scientific theory. The

basic line of attack that I will develop is to draw a distinction between the different parts

of a scientific model and what these parts represent. For a model with parts P1 through Pn,

we can have a good reason to think that P1 accurately represents some aspect A1 of the

system even when we lack a reason to think that the remaining parts accurately represent

other aspects of the system. This division will allow us to place conditions on when an

4
idealization or non-theoretical assumption can coexist with a limited realist conclusion. It

will not always be clear that these sorts of distinctions are feasible, but when they are we

can vindicate a limited form of realism.

To fix some terminology, I will think of a model as a wholly mathematical entity

combined with a series of propositions about how the parts of the mathematical entity

correspond to the physical features of a system.4 For example, in a configuration space

model of some system of n physical particles, the mathematical entity might be some

subset of the set of R3n+1 (3n+1-tuples of real numbers). The propositions that are part of

this model will then relate the first three entries in each triple to spatial coordinates of a

particular physical particle and the last entry to the time at which these particles are at

that spatial position. Taken together the mathematical entity, these propositions impose a

vast array of conditions on the system. These conditions are the content of the model.

When these conditions are met, we say that the model accurately represents the system.

In the example given it might seem that all the features of the mathematical entity

are paired up with features of the physical system. However, a bit of reflection shows that

there are all sorts of mathematical features of this mathematical entity that have no

impact on the content of the model. For example, it may be that there is an equation in

3n+1 variables that is satisfied by all and only the 3n+1-tuples in the model. But this is

irrelevant to whether or not the model accurately represents the system because the

propositions that relate the mathematical entity to the system make no appeal to this

mathematical property. For any model, there will always be some surplus mathematical

4
Along with Chakravartty forthcoming I distinguish between what a model is and the means by which we
come to think about a model. In making this proposal for what a model is, I do not wish to rule out
contextual factors in how we come to think of a given model, e.g. where the propositions that fix the
content of the model come from.

5
structure, i.e. mathematical properties of the mathematical entity that do not figure into

the representational content of the model.

For the purposes of this discussion I will think of a theory as a collection of

claims about a kind of physical system. On this approach, a theory will often pick out a

large class of models. This class might include all mathematical entities satisfying a given

equation or series of equations, along with the propositions relating some of the features

of these entities to aspects of a physical system. A theory of classical physics, for

example, might pick out a series of subsets of R3n+1 that satisfy the equations of the

theory. Each model will then represent a system with a number of admissible trajectories,

i.e. trajectories consistent with the laws of the theory. This class is completely impractical

to work with, and so some measures must be taken to isolate some smaller collection of

models whose accuracy we have some chance of assessing. It is precisely here that steps

related to idealization and other kinds of non-theoretical assumptions threaten our

confidence that we end up with an accurate model. For suppose I add a condition on my

models that is not tied to the theory. Now it looks like I have simply shifted my attention

to a new model M that is unrelated to the models that I began with. And whatever success

I might have working with M will have no implications for the truth of the conclusions

that I draw from that model’s success. If the imposed condition is otherwise unmotivated,

then I seem to have no reason to think the model I have ended up with is accurate.

The way out of this difficulty is to recognize that we can adopt a more nuanced

conception of the accuracy of a given model. On this new approach, we will say that a

model is accurate with respect to aspect A of the system when its content concerning A is

correct. As we have seen, the content of a model is a product of two features: the

6
mathematical entity and the propositions relating features of the entity to aspects of the

system. Clearly a model can be accurate with respect to aspect A without being accurate

with respect to aspect B. Given this independence, if we move from model M1 to model

M2 while preserving the claims about aspect A, then we can assess M1’s accuracy with

respect to aspect A by checking M2’s accuracy with respect to aspect A. This is so even if

model M1 and model M2 differ in dramatic ways. In particular, model M2 may involve

non-theoretical assumptions. As long as these non-theoretical assumptions leave the

claims about aspect A untouched, model M2 can be a reliable guide to the A-aspect

accuracy of model M1.

To illustrate the consequences of this approach, let’s see how it can be used to

defend the widespread idealization of treating a discrete system as continuous. Suppose

that our best theory of gasses represents some gas like air as composed of discrete

particles moving rapidly, colliding with each other and preserved together by some

complicated nexus of gravitational and chemical forces. In many cases, we use a model

that represents the gas as continuous, i.e. as being composed of point-like particles whose

motions are not due to chemical forces. Consider a case where we make further

simplifications so that our continuous model obeys the Euler equations for irrotational,

inviscid fluid flow. My claim is that even though the content of this model disagrees with

any model that fits with the theory, it can still be the case that the model agrees with

some models that fit the theory in some aspects. For example, the two models might

agree on the velocity of the air past some object like an airplane wing. If we could track

the differences between the two models and ensure that the two models agreed in this

respect, then the success of the idealized model in predicting that the velocity would

7
exceed some threshold is a reason to think that the non-idealized model is accurate with

respect to this aspect of the gas.

It would be foolish, of course, to try to extend this argument into an argument that

the non-idealized model is accurate with respect to all aspects of the gas that it represents.

So, the template that we have developed cannot by itself vindicate our hope that there is a

model associated with the theory that is completely accurate. At the same time, there is

every reason to think that other idealized models could be developed that would match

our original model in some other respects. If this is repeated enough times, we may have

reason to think that the model that accords with the theory is accurate in a wide range of

respects.

Here, then, is a rough outline of the sort of argument that can provide a scientist

with a reason to think there is a model that accurately represents some aspect of some

system under consideration:

(i) There is a successful theory T whose models include a model M that represents
aspects A1, …, An of system S.
(ii) There is a model M’ that agrees with M with respect to its representation of
aspect A1 of S.
(iii) M’ is accurate with respect to aspect A1.
Therefore, M is accurate with respect to aspect A1.

The success involved in (i) is the feature of our best scientific theories that scientific

realists have typically focused on. But in contrast to the view suggested by most scientific

realists, the model M’ that figures in (ii) and (iii) need not be motivated by theoretical

assumptions. Still, its success serves to support the conclusion that we have a model M

that is tied to the theory that is accurate in some respects. An important thing to notice

about this schema is that the appeal to (i) is strictly speaking not required for the

8
conclusion to be supported by (ii) and (iii).5 This allows for cases where a model M is not

tied to any particular theory or where models from different theories are combined to

support a claim about the physical world. A case of the latter will be developed in section

V.

III.

Establishing instances of premises (i)-(iii) in any particular case can be extremely

challenging, and scientists have developed any number of different techniques to

convince themselves and their colleagues of such premises. I will focus on three different

modeling techniques which correspond roughly to three different goals that a scientist

might have at a given stage of inquiry. These different goals will further clarify the

different kinds of aspects of a system that I have been alluding to so far. The first concern

that a scientist might have is to uncover the ultimate causal mechanisms responsible for a

phenomenon of interest. A second concern would be to make accurate predictions or

retrodictions about observable features of the system. The third concern I will discuss is

to isolate structures that recur in a type of system and that may or may not track unique

underlying ultimate causal mechanisms. With some scientific problems, some of these

concerns may come together. For example, with the continuous model discussed above,

we were interested both in making an accurate prediction (the velocity would exceed

some threshold) and in isolating a recurring structure of some gas-wing systems (velocity

patterns). But our idealization erased the representation of the ultimate causal mechanism

responsible for the velocity and replaced it with point-like elements that work differently.

In this respect, then, our continuous model is inaccurate and would be a poor choice to try

to understand the causal mechanisms responsible for the velocity pattern.

5
I am indebted to Jan Sprenger for emphasizing this point.

9
 As this case illustrates, accurate prediction of observable aspects of a system does

not require a model that represents underlying causal mechanisms. More generally, our

best predictive model for some observable phenomenon may not represent the causes of

that phenomenon. Our best causal model may be very bad at making accurate predictions.

Finally, we may have a model that accurately isolates a recurring structure without

accurately representing the causes or making accurate predictions about observable

features. Consider, for example, a model of hurricane formation. The causes may be so

complex that they cannot be accurately represented in the model. For similar reasons, we

may not be able to use this model to predict when a hurricane will form in a given case.

Still, the hurricane formation model may lead to important and accurate claims about

hurricanes, e.g. that hurricanes can only form when sea surface temperature is above 26.5

degrees Celsius.6

Examples of cases where different models cooperate to generate accurate

representations of different aspects of a system are not too difficult to come by if we

consider a system organized on several different spatial and temporal scales.7 An

ecological system, say, will have a large number of very small physical parts. According

to our best physical theory, these parts are responsible for the ultimate causal mechanisms

that give rise to the observable features of the system, e.g. how many organisms there are.

But our best predictive model of how many organisms there will be over time will not

make reference to these causal mechanisms. Instead, it may consider predator-prey

relationships that erase the internal physical complexity of each organism. Finally, we

6
This classification is similar in many respects to the notion of a “construal” of a model as developed in
Godfrey-Smith 2006 and Weisberg 2007. I am in general agreement with their call to focus on a special
kind of model-based science.
7
Here I have been strongly influenced by Batterman 2002 and Wimsatt 2007. Wimsatt notes Levins 1966
as a major inspiration.

10
may develop a model that seeks to account for a recurring feature of ecosystems of some

kind. As explained by Weisberg, for example, biologists noted an asymmetric response to

“a general biocide” across predators and prey (Weisberg 2006). This phenomenon was

robust in the sense that it did not depend on many of the particular features of the system.

For our purposes, the key point is that the models used to explain this recurring

phenomenon do not accurately represent the ultimate causal mechanisms at work in the

system.8 At the same time, they fail to provide predictions about the behavior of any

particular ecosystem because the whole purpose of developing this model was to

understand a phenomenon that recurred across ecosystems. So, we have a case where

three different kinds of models can be accurate in different respects. As long as we are

clear on what kind of modeling purpose is in play, each kind of model can be used to

support the conclusion that we are getting this or that aspect right about the system.9

IV.

Up to now I have been somewhat cavalier in talking about the articulation of the

models of our best theory through appeal to non-theoretical assumptions like

idealizations. A reasonable worry about the limited realism so far presented is that my

optimism may depend simply on laziness. That is, if I actually worked through even one

example of how this works in any detail, I would quickly see that there is no reason to

think that the instances of the premises of my argument template are exemplified in

actual scientific practice. The charge of lazy optimism is indeed a charge that has been

8
Weisberg sometimes speaks of revealing “causal structure” (Weisberg 2006, 739) through robustness
analysis. This goes beyond what I would count as an ultimate causal mechanism, although it is not clear if
this difference is anything more than terminological.
9
Here I follow Parker 2007, but her conception of model pluralism may differ in some respects from the
view defended here.

11
leveled by Mark Wilson against a view similar to my own (Wilson 2006). Let’s see how

this objection works and if it can be countered.

One dimension of Wilson’s concerns turns on the need for boundary conditions

for a determinate model to be selected. For a wide class of theories these conditions will

always be non-theoretical assumptions according to the way I have been using the term.

This is because they do not follow from a theory whose scope includes all systems of a

given kind. Boundary conditions not only impose a spatial boundary around a system

which may be more or less artificial, but generally involve unrealistic claims about

exchanges across the boundary. For example, we may treat our system as completely

isolated. No actual systems are completely isolated and so the need for boundary

conditions raises a special problem for my proposal. If a theory purports to have universal

scope over its subject-matter, then typically it will not include a well-motivated recipe for

generating these boundary conditions.

An initial response to Wilson’s challenge based on boundary conditions is to

claim that there still remain cases where the false or otherwise unmotivated assumptions

about the boundary leave unaffected the content of the models concerning other aspects

of the system. If this is ever the case, and scientists can be in a position to establish that it

is the case, then the premises of my argument template can be justified and some form of

limited realism remains possible. To return to our gas case, in our continuous model we

assumed that air was inviscid. As a result we tacitly failed to impose the boundary

conditions for the flow in the region near the airplane wing. This omission leads to

inaccuracy as there is actually a thin boundary layer where viscosity produces a ‘no-slip’

condition, i.e. the velocity goes to zero. But getting this aspect of the system wrong does

12
not necessarily undermine its accuracy with respect to the aspect of the system that we

are concerned with, i.e. the velocity away from the wing.10 Adopting a model with these

boundary conditions is perfectly consistent with showing agreement with some

theoretical model in some respects. It would be a different story if we could not divide up

the content of the model in the way that I have been assuming or if the inaccurate

boundary conditions related directly to the aspect of the system we were concerned with.

But in those cases where we can partition the content in this way, no barrier to limited

realism arises.

Wilson has a more fundamental worry about this approach, however, which

brings in more general problems concerning concept possession and reference to physical

properties. Wilson uses the patchwork character of our modeling practices to question

what he calls a classical approach to concepts and their referents. According to this

approach, it is unproblematic to assume that some of the concepts that we employ

univocally track physical properties. By ostension or some other direct means, the

classicist aims to attach our concepts to the world in such a way that they pick out the

same physical property across time and scientific context. This “classical gluing” then

generates univocal contents for the claims made by scientists, independently of the

particular representational practices that they are then engaging in. Wilson rejects

classicism based on the disconnect between this picture of concept-property pairing and

the involved techniques that working scientists have developed to actually understand

physical properties. In one of his most convincing examples he explains how the concept

10
The appearance of a thin boundary layer is surprisingly a crucial part of the explanation of how an
airplane could take off. Acceptable models of how airplanes actually fly thus raise additional complications
that I hope to pursue elsewhere. I make a start on this in Pincock forthcoming. I am indebted to Rafaela
Hillerbrand for emphasizing the subtleties here.

13
of hardness picks out different physical properties for different materials (Wilson 2006,

335-355). For example, diamonds are said to be harder than steel because diamonds can

scratch steel and not vice versa (Mohs scale). At the same time, different materials are

ranked in terms of hardness by the degree to which they resist indentation. A little

reflection reveals that these tests are not associated with any single underlying physical

property. The classical dogma that these sorts of concepts pick out a single property is

thus exposed.

Now it is precisely this classical picture that I have assumed so far in my

description of models and my argument for limited realism. To see why, notice that I

identified models with the combination of a mathematical entity and a series of

propositions relating the parts of these models to physical properties. Without these

propositions, the model would just be a mathematical entity. So far there is no explicit

reliance on the classical picture, although my failure to spend any time explaining what

these propositions are or how they work suggests classical optimism. But classicism is

presupposed when I assumed that models motivated by widely divergent assumptions

overlapped in content with respect to some aspects of the system. That is, the models

ended up representing the same physical properties even though they work quite

differently. To take what might seem to be a trivial example, but in fact involves

considerable complexity, I assumed that both the discrete and the continuous model could

be models of velocity. If velocity turns out like hardness, then these models do not

represent the same aspect of the same system. If this happens, then the argument for

limited realism that I have sketched breaks down.11

11
See Butterfield 2006 for a survey of debates about velocity and a convincing proposal for how to resolve
them.

14
 While a full reply to Wilson’s rejection of classicism would be quite involved, a

defense of limited realism is easier to envision. The basic point of disagreement for any

particular case is whether or not the propositions that are constitutive of one model can be

identified with the propositions that are constitutive of the other model. In the simplistic

example of a discrete and continuous model of a gas, I think it is unproblematic to say

that the two models overlap in content to some degree. In other cases, there may be more

debate, and resolving this debate would be a necessary step in justifying the premises of

the argument for that instance of limited realism. It may happen, for example, that the

hardness claims of the two models are so divergent that there is no univocal concept of

hardness that figures in both models. When this happens, we cannot be confident that the

two models are equally accurate with respect to hardness. We may wind up with fewer

justified cases of limited realism, but I do not see why such problems should lead us to

abandon a properly qualified defense of scientific realism. In fact, it seems that Wilson

allows the point I am making here when he rejects an “indiscriminate holism” (Wilson

2006, 69) that would insist that the content of two models can never overlap.12

V.

To bolster my proposal I conclude with a brief case study concerning a recent

debate at the intersection of meteorology, climatology and public policy. This is the

controversial question of whether climate change (or global warming) causes an increase

in hurricane intensity. Here we see a combination of the three modeling purposes that I

have so far emphasized. To begin with, there is a question about causation, which

presumably should bottom out in the fundamental particles of the system and their

physical properties. Clearly there are also predictive aspects to the debate as, despite our
12
Here I aim in part to correct my previous discussion of Wilson in Pincock 2005.

15
understanding of the dynamics of hurricane formation, it is not currently possible to

predict hurricane intensity using these dynamical models. Instead, different models are

employed in hurricane forecasting. Finally, we have a case of a recurring structure when

we consider hurricanes. Each hurricane is different, but hurricanes are a well-defined

collection of meteorological phenomena that arise across a wide range of meteorological

systems.

The argument that I will reconstruct is more complicated than the argument

template given earlier. We can think of it as a repeated application of the argument

template, where a different conclusion about reality is drawn at each stage. The ultimate

conclusion that global warming causes an increase in hurricane intensity then follows

from a combination of these three subconclusions. As we will see, it is not a trivial matter

to determine if each of these subconclusions is supported by the evidence so far

assembled or if the subconclusions support the overall conclusion in question. But I

believe that these sorts of arguments are the sorts of things that could vindicate the

limited scientific realist position I am defending if enough evidence was assembled and

certain conceptual issues were resolved. The point I want to make with this example is

that there is no single theory and no single model that is used to justify the authors’

central hypothesis. Instead, different theories and different models are used to handle

each of the three subhypotheses. This will not be a problem if the scientific arguments for

each subhypothesis can be interpreted in terms of the different aspects of the system in

line with my argument template from section II. We begin to see with this case, then, that

the prospects for limited realism are somewhat wider than they might initially appear to

be.

16
 I will take as my primary focus a recent article by Curry, Webster and Holland

called “Mixing Politics and Science in Testing the Hypothesis that Greenhouse Warming

is Causing a Global Increase in Hurricane Intensity” (Curry et. al. 2006). The article

summarizes the objections that the authors received to an earlier Science article that

argued for a genuine causal connection (Webster et. al. 2005).13 Beyond this, they

provide some methodological reflections on how such disputes should be resolved as well

as some practical suggestions for how scientists can communicate productively with the

media.

Curry et. al. begin by presenting their “central hypothesis”: “greenhouse warming

is causing an increase in global hurricane intensity” (Curry et. al. 1026). This hypothesis

is then analyzed into three “subhypotheses”, which are then considered separately:

1) the frequency of the most intense hurricanes is increasing globally;

2) average hurricane intensity increases with increasing tropical SST [sea surface
temperature];
3) global tropical SST is increasing as a result of greenhouse warming (Curry et.
al. 1026).

They continue by noting that “The central hypothesis implies a causal chain 3
2
1 and

therefore depends upon the validity of each of the three subhypotheses” (Curry et. al.

1026). That is, greenhouse warming causes global tropical SST to increase, which in turn

increases average hurricane intensity. As a result, the observed increase in hurricane

intensity can be attributed (in part) to global warming. An additional conclusion is that

further greenhouse warming will produce further increases in hurricane intensity.

The theories at play in this example range from the fundamental theory of fluid

dynamics to more specialized meteorological theories concerning the origin and

propagation of hurricanes through to the theory of global warming as a byproduct of

13
H. R. Chang was a co-author in Webster et. al. 2005 but not in Curry et. al. 2006.

17
human activity. The challenges posed by this issue can be traced in part to the

involvement of phenomena from dramatically different scales. On the microscale, there

are the molecular interactions between water and air. On some medium scale, we have

the conditions necessary for the formation of hurricanes from tropical depressions.

Finally, on the largest scale, we have the global increase in SST temperature that is

attributed to human activity. The way in which all these scales are interrelated is tracked

in the different sorts of models that are used to resolve the issue.

Unfortunately, it is far from clear that the scientists have gotten things right here

and the issue remains a topic of ongoing debate. To begin with, there is an unfortunate

equivocation concerning “hurricane intensity” which raises concerns analogous to the

ones raised by Wilson about “hardness”. It is not clear how our records of hurricanes in

terms of the familiar Saffir-Simpson “Category n” classification drawn on in 1) relate to

the sorts of measures of hurricane intensity treated in 2). More interestingly, there is not

even agreement among scientists about how to estimate hurricane intensity from

accessible meteorological data. In fact, additional support to the claim that hurricane

intensity is increasing due to global warming has been offered by Emanuel 2005 and

Sriver and Huber 2006 based on two different representations of hurricane strength. The

latter paper is especially interesting from a methodological point of view as the authors

“test more robustly the hypothesis” (Sriver and Huber 2006, p. 2) by employing different

models than Emanuel 2005.14 The lack of agreement on hurricane intensity may,

somewhat paradoxically, turn out to give additional support to the hypothesis in question.

14
The independence of these models is questioned by Maue and Hart 2007, but Sriver and Huber 2007
offers a convincing reply.

18
Whether we have this optimistic scenario, or a simple case of argument by equivocation,

must be determined elsewhere.

Turning back to our main paper, Curry et. al. support subhypothesis 3) that global

tropical SST is increasing as a result of greenhouse warming by appeal to the 2001

Intergovernmental Panel on Climate Change Third Assessment Report along with

“subsequent climate modeling studies” that appeared in 2004 and 2005. These

publications proceed by describing highly complicated models that are meant to simulate

the global atmosphere-ocean evolution given some specified initial and boundary

conditions. Typically these models represent the Earth as a grid, where each cube in the

grid is treated as unstructured except for meteorological magnitudes like temperature and

pressure. When these models are evaluated with respect to the 20th century, it turns out

that only the models that include some factor tied to human activity can reproduce the

known data. Thus because “the global surface temperature since 1970 (including the

trend in tropical SSTs) cannot be reproduced in climate models without the inclusion of

anthropogenic greenhouse gases”, 3) is to be preferred over the “null hypothesis” that

global tropical SST is not increasing as a result of greenhouse warming. The important

thing to note here is that the models in question here do not represent any hurricane

activity, let alone hurricane intensity. So, it is not possible to directly justify the central

hypothesis using just these sorts of models.

This shows that a different collection of models is needed to justify hypothesis 2)

that average hurricane intensity increases with increasing SST. Here Curry et. al. appeal

to a theory of hurricane intensity that relates SST to the potential energy of a hurricane

that forms, as well as a 2006 paper which “clarified the relationship between seasonally

19
averaged hurricane intensity and seasonally averaged tropical SST on an individual ocean

basis” (Curry et. al. 1029). A review of this latter paper indicates that this conclusion

depends on a model that delineates the different contributions to the intensity of the

hurricanes that formed. Our best theory of hurricane intensity is used to isolate four

factors that contribute to increased hurricane intensity, including SST increase.15 The

meteorological data are then analyzed to see which factor is most responsible for the

long-term trend in hurricane intensity. While short-term variability in the other three

factors is part of the explanation of the short-term variability in hurricane intensity, the

authors conclude that only the long-term trend of SST increase is responsible for the

long-term trend of hurricane intensity increase.

Finally, hypothesis 1) that the frequency of the most intense hurricanes is

increasing globally is supported primarily by consulting records of hurricane activity

since 1970. Here there are potentially troubling uncertainties in the data as we would

expect that our ability to find hurricanes has increased since 1970, and so the upward

trend in the number of the most intense hurricanes may be merely due to our better

technology. The authors thus provide what we could call a model of the data that they

have at their disposal and conclude that there is no good reason to suspect that this

systematic error has been committed (Curry et. al. 1028-1029). Clearly the models

deployed in this analysis of the data are distinct from the models used to justify

hypotheses 2) and 3).

15
The other three factors are “increasing specific humidity, minimal vertical wind shear, and negative
stretching deformation” (Hoyos et. al. 2006, 94).

20
 Here, then, we see that different theories and sorts of models are used to justify a

scientific claim with broad scope.16 The argument can be made to fit my argument

schema by viewing the sub-arguments for (1)-(3) as each having their own instances of

premises (i)-(iii):

(3-i) There is a successful theory of global climate whose models include a

model M that represents the carbon-temperature relationship.
(3-ii) There is a model M’ that agrees with M with respect to its
representation of the carbon-temperature relationship.
(3-iii) M’ is accurate with respect to the carbon-temperature relationship
(because only it can account for the observed temperatures).
Therefore, M is accurate with respect to the carbon-temperature
relationship on Earth and (3) global tropical SST is increasing as a result of
greenhouse warming.
(2-i) There is a successful theory of hurricanes whose models include a
model M’’ that represents the link between SST and intensity.
(2-ii) There is a model M’’’ that agrees with M’’ with respect to its
representation of the link between SST and intensity.
(2-iii) M’’’ is accurate with respect to the link between SST and intensity
(based on an analysis of long-term and short-term trends in hurricane records).
Therefore, M’’ is accurate with respect to the link between SST and
intensity and (2) average hurricane intensity increases with increasing tropical
SST.
(1-i) The records of hurricanes indicate a relationship between time and
hurricane intensity.
(1-ii) There is a model M’’’’ of the records that agrees with the records
with respect to the long-term trend of increasing intensity.
(1-iii) M’’’’ accurately represents the long-term trend of increasing
intensity.
Therefore, the records accurately represent the long-term trend of
increasing intensity and (1) the frequency of the most intense hurricanes is
increasing globally.

One notable feature of the last argument is that (1-i) does not involve any theory

in a standard sense. Still, models of the data are needed to vindicate our confidence in (1).

As we see, if each model is considered using its entire content, then the steps in this

argument are contradictory and presumably prove nothing. For example, we do not take

16
Compare the apparent pessimism about these sorts of conclusions in Oreskes et. al. 1994 and Oreskes &
Belitz 2001.

21
the model in (2-ii) to be relevant to assessing the accuracy of the model in (1-ii). This

shows that the cogency of the scientific argument for the ultimate hypothesis that global

warming is responsible for the increase in hurricane intensity depends on the content of

each model being exploited only in a very limited fashion. There is no overarching theory

that could be used to motivate each of these steps, but taken individually there is a

detachable claim from each part that shows something about reality. When these claims

are successfully supported and combined, we can have a compelling conclusion about the

real world. Attention must always be paid, however, to the dangers of equivocation if we

are to vindicate limited scientific realism. At the same time, there is reason for optimism

that a segmented approach to the content of scientific models can help us to reconstruct

the scientific knowledge that the success of science suggests we have.

References

Batterman, Robert (2002), The Devil in the Details: Asymptotic Reasoning in

Explanation, Reduction, and Emergence. New York: Oxford University Press.

Butterfield, Jeremy (2006), “Against Pointillisme about Mechanics”, British Journal for

the Philosophy of Science 57: 709-753.

Cartwright, Nancy (1983), How the Laws of Physics Lie. New York: Oxford University

Press.

Cartwright, Nancy (1999), The Dappled World: A Study of the Boundaries of Science.

New York: Cambridge University Press.

Cat, Jordi (2005), “Modeling Cracks and Cracking Models”, Synthese 146: 447-487.

22
Chakravartty, Anjan (forthcoming), “Informational vs. Functional Theories of Scientific

Representation”, Synthese.

Curry, J.A., P.J. Webster and G.J. Holland (2006), “Mixing Politics and Science in

Testing the Hypothesis That Greenhouse Warming is Causing a Global Increase in

Hurricane Intensity”, Bulletin of the American Meteorological Society 87: 1025-1037.

Emanuel, K. (2005), “Increasing destructiveness of tropical cyclones over the past 30

years”, Nature 436: 686-688.

Flatow, Ira (2007), “Why an Airplane Flies: Debunking the Myth”, in Present at the

Future, Collins, 303-309.

Giere, Ronald (2006), Scientific Perspectivism. University of Chicago Press.

Godfrey-Smith, Peter (2006), “The Strategy of Model-based Science”, Biology and

Philosophy 21: 725-740.

Hoyos, C.D., P.A. Agudelo, P.J. Webster and J.A. Curry (2006), “Deconvolution of the

Factors Contributing to the Increase in Global Hurricane Intensity”, Science 312: 94-97.

Jones, Martin R. & Nancy Cartwright (eds.) (2005), Idealization XII: Correcting the

Model. Idealization and Abstraction in the Sciences. Amsterdam: Rodopi.

Levins, Richard (1966), “The Strategy of Model Building in Population Biology”,

American Scientist 54: 421-431.

Maue, R.N. and R.E. Hart, “Comment on R. Sriver and M. Huber (2006)”, Geophysical

Research Letters 34: L11703.

Morrison, Margaret (2005), “Approximating the Real: The Role of Idealizations in

Physical Theory”, in Jones and Cartwright (eds.) (2005), 145-172.

23
Oreskes, N., K. Schrader-Frechette and K. Belitz (1994), “Verification, Validation and

Confirmation of Numerical Models in the Earth Sciences”, Science 263: 641-646.

Oreskes, N. and K. Belitz (2001), “Philosophical Issues in Model Assessment”, in M.G.

Anderson and P.D. Bates (eds.), Model Validation: Perspectives in Hydrological Science.

Wiley, 23-41.

Parker, Wendy (2007), “Understanding Pluralism in Climate Modeling”, Foundations of

Science 11: 349-368.

Pincock, Christopher (2005), “Conditions on the Use of the One-dimensional Heat

Equation”, in G. Sica (ed.), Essays on the Foundations of Mathematics and Logic, Vol. 2,

Polimetrica, 67-79.

Pincock, Christopher (forthcoming), “Towards a Philosophy of Applied Mathematics”, in

O. Bueno and Ø. Linnebo (eds.), New Waves in Philosophy of Mathematics, Palgrave

Macmillan.

Psillos, Stathis (1999), Scientific Realism. New York: Routledge.

Sriver, R. and M. Huber (2006), “Low frequency variability in globally integrated

tropical cyclone power dissipation”, Geophysical Research Letters 33: L11705.

Sriver, R. and M. Huber (2007), “Reply to comment by R.N. Maue and R.E. Hart”,

Geophysical Research Letters 34: L11704.

Suárez, M. (1999), “The Role of Models in the Application of Scientific Theories:

Epistemological Implications”, in M. Morgan & M. Morrison (eds.), Models as

Mediators, Cambridge University Press, 168-196.

Suárez, M. and N. Cartwright (2008), “Theories: Tools versus Models”, Studies in the

History and Philosophy of Modern Physics 39: 62-81.

24
Van Fraassen, Bas (1980), The Scientific Image. Oxford University Press.

Webster, P.J., G.J. Holland, J.A. Curry and H.R. Chang (2005), “Changes in Tropical

Cyclone Number, Duration, and Intensity in a Warming Environment”, Science 309:

1844-1846.

Weisberg, Michael (2006), “Robustness Analysis”, Philosophy of Science (Proceedings)

73: 730-743.

Weisberg, Michael (2007), “Who is a Modeler?”, British Journal for the Philosophy of

Science 58: 207-233.

Wilson, Mark (2006), Wandering Significance: An Essay on Conceptual Behavior.

Oxford University Press.

Wimsatt, William (2007), Re-Engineering Philosophy for Limited Beings: Piecewise

Approximations to Reality. Harvard University Press.

25