Levels of Explanation Reconceived

Transcription

1 Levels of Explanation Reconceived Angela Potochnik According to explanatory reductionism, the theories and laws of physics explain higherlevel phenomena better than the theories and laws that deal with these higher-level phenomena directly. Many arguments that have been formulated against explanatory reductionism amount to the idea that higher-level explanations are sometimes or always preferable because they are more general than the alternative lower-level explanations (Fodor, 1974; Putnam, 1975; Garfinkel, 1981; Kitcher, 1984; Sober, 1984; Jackson and Pettit, 1992; Sober, 1999). Here I challenge two basic assumptions that are needed for this argument to succeed. It cannot be assumed that higher-level explanations are more general than their lower-level alternatives, nor can it be assumed that higher-level explanations are general in the right way to be explanatory. Problems with both of these assumptions trace back to the fact that higher- and lower-level explanations are often not related by supervenience. I use explanations formulated in population biology to illustrate these points. Finally, I suggest a novel form of pluralism regarding levels of explanation that springs from this reconceptualization of the relationship among levels. In my view, different levels of explanation are preferable in different circumstances because explanations vary according to the type of generality they offer, and these different types of generality are appropriate in different circumstances of explanation. 1

2 1 Generality and Explanation The claim that higher-level explanations are preferable because they are more general than lower-level explanations relies upon the idea that generality is of explanatory worth. This idea is commonly appealed to and yet seldom defended, so it bears investigating here. Garfinkel s (1981) argument for higher-level explanation offers the beginning of a justification for the claim that generality benefits an explanation. Garfinkel argues that higher-level explanations are preferable because they focus on the causes that critically affect the outcome to be explained, whereas the alternative lower-level explanations cite overly specific causal details. Though he does not explicitly invoke the idea of generality in this argument, an explanation that avoids specific causal details that another includes is more general than the other. In Garfinkel s view, then, generality aids explanation by identifying the critical causal factors and excluding causal information that is in some sense unimportant. Garfinkel is on the right track regarding the explanatory value of generality. An explanation s generality is inversely related to the amount of causal information it includes. Increasing the amount of causal information in an explanation specifies additional features of the causal process, which decreases the explanation s generality. 1 By including fewer causal details, a general explanation focuses attention on the causal information it features. This causal information is made more perspicuous than it would be if the explanation also included other causal details. If the information featured in the explanation regards the causes that critically affect the outcome to be explained, as Garfinkel thinks is the case for higher-level explanations, then the perspicuity resulting from the explanation s generality improves the explanation. This is one argument for the idea that generality is explanatory: 1 Here I employ a comparative notion of generality, according to which an explanation E 1 is more general than an explanation E 2 if the set of possible systems to which E 2 applies is a proper subset of the set of possible systems to which E 1 applies. This definition of generality is similar to and influenced by Strevens s (2004; 2008). I prefer this formulation in terms of possible rather than actual systems, for this correlates the generality of an explanation with the amount of causal detail it incorporates (Weisberg, 2003; Strevens, 2004). 2

3 a general explanation makes the critical causal relationship perspicuous, thereby focusing attention on the causal information that is properly explanatory. The explanation advertises the unimportance of other features of the causal process by neglecting them. 2 Two claims are needed in order to use this argument for the explanatory worth of generality to defend higher-level explanation. First, it must be the case that higher-level explanations are more general than lower-level explanations. Otherwise, the explanatory value of generality would not translate into the value of higher-level explanations. Fodor (1974) and others appeal to the idea of multiple realizability to motivate this claim. Second, it must be the case that the information included in the higher-level explanation is important to the outcome to be explained, whereas the neglected information is unimportant to the outcome. According to the argument sketched above, the value of general explanations is that they focus attention on the important information. If neglected lower-level information is important, then the generality of a higher-level explanation does not improve the explanation. Both of these claims turn out to be problematic. The problems stem from the fact that the relationship between higher- and lower-level explanations often cannot be accurately described in terms of supervenience. 2 Supervenience and Levels of Explanation The belief that higher-level explanations are more general than their lower-level alternatives is a common one (Fodor, 1974; Putnam, 1975; Garfinkel, 1981; Jackson and Pettit, 1992; Sober, 1999; Strevens, 2004; Strevens, 2008). This belief is motivated by the idea that higher-level properties are multiply realized by lower-level properties that is, that there are many lower-level properties that individually suffice as supervenience bases for a single higher-level property. Suppose that a lower-level explanation cites causal properties that 2 Strevens (2004; 2008) offers a similar rationale for his endorsement of general explanations. 3

4 are supervenience bases for the causal properties cited in a higher-level explanation, and that these higher-level properties are multiply realizable. It follows that the higher-level explanation is more general i.e., applies to a wider range of possible systems than the lower-level explanation. This is the situation in Putnam s (1975) familiar example of the peg and the board. Putnam imagines that there are two holes in the board: both have a diameter of one inch, but one is circular while the other is square. The peg is a cube with sides that are a tiny bit shorter than one inch. The fact to be explained is that the peg passes through the square hole but not the round hole. Putnam considers two explanations for this fact. The higher-level explanation references the dimensions of the peg and the two holes in the board and cites geometrical information about the relationship between these dimensions to account for why the peg goes through the square hole but not the round hole. The lower-level explanation gives information about the atomic structure of the peg and board, states that the atoms form a rigid lattice, etc., and then computes all possible trajectories of the peg s atoms in order to deduce that the peg never passes through the round hole but does pass through the square hole on at least one trajectory. The micro-level properties cited in the lowerlevel explanation are only one way among many of realizing the geometrical relationship between the peg and the board, so the higher-level explanation is more general than the lower-level attempt. Putnam s conclusion is that this more general, higher-level explanation is preferable. Yet this relationship need not exist between explanations formulated at different levels. Higher-level explanations are at least sometimes and I suspect often not accurately described as more general than their lower-level alternatives. I do not contest the point that all higher-level properties supervene on and are potentially multiply realized by lower-level properties. But this relationship among properties at different levels need not translate into the predicted relationship between higher- and lower-level explanations. 4

5 The claim that higher-level explanations are more general than their lower-level competitors requires an additional premise, beyond the supervenience of higher-level properties on lower-level properties. It must also be the case that lower-level explanations are formulated in terms of the supervenience bases of their higher-level competitors. I see no reason to think that this is always or even often true. Fields of science that focus on lower-level phenomena develop explanations for their own ends. The lower-level properties that are of interest in these fields are not typically the same as the supervenience bases for higher-level properties of interest in other fields. An example of this is the relationship between an organism s genotype (genetic composition) and phenotype (observable characteristics). Population genetics focuses on changes in the distribution of genotypes over time, whereas evolutionary ecology charts changes in the distribution of phenotypes. Philosophers tend to think of genetic properties as occupying a lower-level of organization than phenotypic properties (e.g., Dupré, 1993; Rosenberg, 1994). Yet phenotypes do not supervene on genotypes. If anything, a phenotype supervenes on a combination of genotype, environmental factors, and historical conditions. So, in this case at least, the lower-level properties under investigation are not identical to the lower-level properties upon which the higher-level properties under investigation supervene. There is no reason to think that this example is atypical in this regard. Because different fields of science formulate explanations for different purposes, lower-level explanations and higher-level explanations are often not related by supervenience. This is so even for some of the examples that Garfinkel (1981) uses to defend higher-level explanation. For instance, Garfinkel examines different possible explanations for the death of a rabbit with the aim of demonstrating the superiority of the higher-level explanation. The lower-level explanation that he considers cites the presence of the rabbit in the capture-space of a particular fox to account for the rabbit s death. The competing higher-level explanation instead cites the high concentration of foxes in the area. These explanations are formulated at lower and higher levels respectively, in the sense that the first deals with an individual and 5

6 the second with a population of individuals. However, the property of high fox-concentration clearly does not supervene upon the property of the rabbit being located in the capture-space of one fox. This higher-level property instead supervenes on a complete description of the location of all foxes in the area, a description that need not even cite the location of the rabbit. The only relationship apparent between the higher-level property and the lower-level property referenced in Garfinkel s example is that the high concentration of foxes makes it more likely that the rabbit is in the capture-space of some fox or other. This relationship bears no resemblance to supervenience. The same is also true for many real examples of competing explanations formulated in fields of science that employ different levels of description. Evolutionary explanation offers a ready example of this. Recall from above that genotypes and phenotypes are commonly considered to be lower- and higher-level properties of organisms, respectively. Approaches to modeling natural selection include genetic models, which represent generational change in the distribution of genotypes, and phenotypic models, which represent how fitness-effects of the environment change the distribution of phenotypes. Both genetic and phenotypic models can be used to explain, e.g., a particular phenotypic trait becoming fixed in a population. The genetic model provides a lower-level explanation, for it focuses on genotypes in order to explain a phenotypic outcome. The phenotypic model provides a higher-level explanation, for it focuses directly on changes in phenotypes and is silent regarding changes at the lower (genetic) level. Yet, as I pointed out above, phenotypes do not supervene on genotypes, but on a much more inclusive combination of lower-level properties. The relationship between genotype and phenotype is causal rather than supervenient in nature: an organism s genotype is a cause that partially determines that organism s phenotype. Genetic and phenotypic explanations for the fixation of a phenotypic trait are thus not related via supervenience. When higher- and lower-level explanations are not linked by supervenience relations, nothing ensures that higher-level explanations are more general than their lower-level com- 6

7 petitors. If a lower-level explanation does not give full information about the supervenience bases of the properties cited by a higher-level explanation, the lower-level explanation may apply to a range of possible systems that is not a proper subset of those to which the higher-level explanation applies. The possibility is introduced that the lower- and higherlevel explanations apply to sets of systems that only partially overlap. If so, the two levels of explanation will be general in different ways. This is the case for genetic and phenotypic explanations of evolutionary phenomena. Explanations provided by phenotypic models are not more general than those provided by genetic models; instead, the two generalize in different ways. A phenotypic explanation is applicable whenever the represented fitness relationship obtains, even if the genetic details are different. So, e.g., a single phenotypic model explains the latitudinal cline in clutch-size, a trait widely shared among bird populations (Ashmole, 1963; Ricklefs, 1980), despite the nearcertainty that this trait has different genetic causes in different populations. Because genetic explanations represent genetic transmission, they do not generalize in this way. In contrast, a genetic explanation is applicable whenever the represented combination of selection and transmission obtains, regardless of the nature of the environmental sources of fitness. A single genetic model can explain, e.g., various instances of heterozygote superiority, where genes bear similar fitness relationships to one another, regardless of what environmental factors result in the heterozygote being the fittest alternative. 3 Because phenotypic explanations represent the sources of fitness, they do not generalize in this way. 3 Heterozygote superiority occurs when the heterozygote (the genotype with two different alleles at the locus in question) is fitter than the alternative homozygotes (the genotypes with identical copies of one allele at that locus). 7

8 3 Contexts and Explanatory Importance In order to defend higher-level explanation on the basis of the explanatory worth of generality, it also must be the case that the information included in the higher-level explanation is important to the outcome to be explained, whereas the neglected information is unimportant to the outcome. According to the argument I outlined in Section 1, general explanations are valuable because they focus attention on the information that is explanatorily important. Recall that an explanation s generality is inversely related to the amount of causal information it includes. If an explanation purchases greater generality by omitting causal information that is important to the explanation, then the increased generality impoverishes rather than improves the explanation. Even granting that a higher-level explanation is more general than its lower-level competitor, this generality only improves explanation if it is of the appropriate type. That is, the information omitted for the sake of generality must be unimportant to the explanation. It is not correct simply to say that generality improves explanation, full stop. That more generality is not always better is demonstrated by taking this idea to the extreme. Rigorously applying the precept of maximizing generality would result in explanations that are vast disjunctions, involving all of the possible causes of the event to be explained. So, e.g., the most general explanation of a room getting dark is: Either the light was turned off, or it is twilight, or someone drew the curtains, or a storm is gathering outside, or... Though this sort of disjunctive explanation is quite general, it provides little if any causal information. The argument is that, since expansive disjunctions like this surely cannot be the preferable form of explanation, the generality of an explanation should not be maximized (Sober, conversation). If generality is indeed of explanatory worth, then the trick is to find an alternate way to express this, for simply maximizing generality clearly has unhappy results. 8

9 There is also a more subtle formulation of this concern. The causal chain leading up to an event traces far back in time; its links can be individuated more and more finely; and each of these links has countless aspects. No explanation can give full information about all parts of this causal chain, and the generality of an explanation is increased by neglecting some of the information about the causal chain. This can be accomplished in countless ways. Something thus must be said regarding which parts of the causal chain should be neglected from an explanation for the sake of generality, and which parts of the causal chain the explanation should reference. As radically disjunctive explanations illustrate, simply neglecting as much causal information as possible is not a workable solution. The need to specify the causal information that should be neglected for the sake of generality is readily apparent with evolutionary explanations. A vast number of causal factors are involved in bringing about a particular evolutionary outcome e.g., that Homo sapiens have enlarged brains relative to other primates. The causal chain includes various genetic factors, developmental processes, environmental conditions, perhaps other human evolutionary events, phylogenetic features, etc. It is readily apparent that some degree of generality is desirable in an evolutionary explanation: neglecting some of these many causal factors increases the perspicuity of the causal relationship(s) featured in the explanation. Yet some ways of increasing the generality of an explanation would be detrimental. For instance, an explanation for humans large brains that outlines the developmental causes and neglects all more distal, evolutionary events would be general in a way that makes it ill-suited to be an evolutionary explanation. Despite the explanatory worth of generality, then, some attempted explanations are actually too general, or general in the wrong way. An exhaustive disjunction of all possible causes of the room getting dark is an explanation that is too general. And citing only the developmental causes of humans large brains is general in the wrong way to be an evolutionary explanation. This fails as an evolutionary explanation because it neglects the very causal 9

10 information that is of explanatory import. It thus must be resolved what degree and type of generality an explanation should have. Simply put, generality that is gained from leaving out unimportant information is good, whereas generality that is gained from leaving out important information is bad. The question thus becomes: what type of causal information is important? 4 What parts of the vast amount of information about a causal process are important to explaining the resulting event depends upon which causal relationships are explanatory, and this depends upon the role the explanation is meant to play. Prima facie it seems that every causal factor leading to an event is potentially relevant to the explanation of that event. Alternately, if not all causal factors qualify as difference-makers (Strevens, 2004; Strevens, 2008), then at least every difference-maker is potentially of explanatory relevance. From causal or difference-making relevance, explanatory relevance seems to follow (assuming a broadly causal notion of explanation). And yet there are innumerable causal/differencemaking factors for any event, for causal processes extend far back in time and many steps involve several interacting causes. This potential explanatory relevance of many causes, coupled with the explanatory value of generality, makes it so that different causal information is valuable in different circumstances. Different circumstances thus lead to explanations that make different causal relationships perspicuous that is, that offer different types of generality. 5 In other words, the determination of what causal information is important to an explanation is context-sensitive. 4 Garfinkel (1981) thinks that lower-level causal information is unimportant to explaining higher-level events because the lower-level causes are not critical to the outcome. He calls this redundant causality : the lower-level causal history could have been different in many respects without affecting the higher-level event to be explained. But this solution amounts to maximizing generality and faces the same problems as that idea. If disjunctive explanations are not the ideal, then some of the information about redundant causes is important. 5 When I say that explanations offer different types of generality, I do not mean that different explanations qualify as more general according to different definitions of generality. Instead, the idea is that different explanations are general because they neglect different causal information, so they generalize to different sets of possible systems. 10

11 For instance, there are several different contexts in which an explanation might be sought for why humans have large brains relative to other primates. Imagine a research program that focuses on the role natural selection played in major events of hominid evolution. In this context, the causal relationship that is of particular importance to explaining humans large brains is the relationship between certain environmental conditions and the resultant selective advantage for large-brained humans. In contrast, imagine a research program directed at uncovering the developmental pathways that produce the human nervous system. Here the causal relationship of note is between the nature of certain developmental pathways in humans and the creation of big brains. Both of these causal relationships and many, many more besides influence the event to be explained. Which causal relationship is important to the explanation depends on context, i.e., on the role that the explanation is designed to play in scientific investigation. 6 Those who want to resist this move to context-sensitive explanation might argue that different contexts actually result in distinct explananda. Regarding the above example in particular, one might argue that in the first context, one explains the evolution of humans large brains, whereas in the second context, one explains the development of humans large brains. But individuating explananda in this manner is simply a way of implicitly specifying the context of the explanation, i.e., the causal relationship of interest. In both contexts, the explanandum is the fact that humans evolved to develop large brains. Depending on research interests i.e., on context this explanandum might be expressed alternately as (a) the fact that humans evolved (to develop) large brains, or (b) the fact that humans (evolved to) develop large brains. Both regard a single evolutionary/developmental event, resulting from a single causal process. The alternate expressions result from singling out a distal part of this causal process and 6 I develop the idea of context-sensitive explanatory inclusion more fully in Potochnik (forthcoming). 11

12 from singling out a proximate part of this causal process, respectively (Mayr, 1961). 7 In this manner, the context of the explanation directs attention toward one part of the causal process and leads to bracketing other parts of the process. Another way to resist the move to context-sensitivity is to claim that the role of context is captured by specifying the contrast class of the explanandum. Contrastive explanation is the idea that an explanation is in part determined by what one considers to be the alternatives to the event to be explained (Dretske, 1972; van Fraassen, 1980; Garfinkel, 1981). So, e.g., one might wonder why humans have large brains (rather than smaller brains) or, alternately, why humans have large brains (rather than chimps having large brains). But although specifying different contrast classes may very well lead to different explanations, it does not capture the full relevance of context. Consider explanations for why humans have large brains (rather than smaller brains). It is still the case that research interests in development result in an explanation that focuses on developmental pathways why it is that a certain genotype causes large brains rather than smaller brains whereas research interests in selection result in an explanation that focuses on selection s role why it is that large brains were selected for rather than smaller brains. Specifying the contrast class does not fully account for the relevance of context. 8 The idea that which causal information is important to an explanation depends upon the context in which the explanation is formulated casts doubt on the assumption that more general, higher-level explanations are always general in the right way. Even if a lowerlevel explanation is less general than its higher-level competitor, the causal relationships the lower-level explanation focuses upon are likely to be explanatorily important in some 7 Developmental systems theorists maintain that the parts of the evolutionary/developmental process often cannot be prised apart in this way, that is, that it is not possible to adequately answer evolutionary questions while bracketing development. If this is right, then in contexts favoring evolutionary information, the explanandum nonetheless must be expressed as the evolution of the developmental process leading to large brains. 8 On this, I agree with van Fraassen (1980), who argues that context determines both the contrast class and the relevance relation governing an explanation. 12

13 context or other. The force of this point is only increased when one also appreciates that the relationship between higher- and lower-level explanations often cannot be captured in terms of supervenience. Many lower-level explanations offer information about causal relationships that the alternative higher-level explanations do not. In these cases, explanations formulated at different levels offer different types of generality; i.e., they generalize to different sets of phenomena. In some contexts, the type of generality provided by the lower-level explanation is preferable. 4 New Reasons for Explanatory Pluralism In the previous two sections I argued that, despite common assumptions about the relationship between higher and lower levels of explanation, different levels of explanation offer different types of generality, and different types of generality are valuable in different contexts of explanation. These points suggest an alternate conception of the relationship among levels of explanation. The result is a new form of pluralism regarding levels of explanation, one that is sensitive to the complexity of real-world causal processes and to the ways in which science grapples with this complexity. Generality (of some type) is valuable to explanation because grouping similar systems is a ubiquitous goal of scientific explanation. Causal information that applies to every member of a set of similar systems yields insight into the source of the similarities. In other words, this information highlights causal regularities or laws. A match fails to light due to the absence of oxygen; a window breaks because a rock over a certain mass was hurled at it with a great enough velocity. Information about causal regularities is the right sort of information to be featured in an explanation; it gets at why the event in question came about. 9 The 9 This suggests that there is something to the unification account of explanation. Indeed, I think that my arguments in this paper are compatible with both the causal and the unification approaches to explanation. I couched this discussion in terms of causal explanation simply because this is common in the debate over levels of explanation. 13

14 type of generality of an explanation determines the set of similar systems into which the event to be explained is grouped. More specific explanations pick out a narrower grouping; more general explanations pick out a broader grouping. Explanations that offer different types of generality that is, provide different causal details pick out incommensurate sets of comparison systems. Different sets of comparison systems, and so different types of generality, are appropriate in different circumstances. Depending on the explanation-seekers interests, attention may be directed at various different causal relationships. Which causal relationships potentially belong in an explanation is, of course, constrained by what event is to be explained and what causal process in fact leads to the occurrence of this event. But which of these potentially explanatory causal relationships actually deserves a place in a particular explanation is decided by the context in which the explanation is formulated. This context is determined by the role to be played by the explanation, that is, the research program in which the explanation is formulated. Explanations formulated on different levels are often not related via supervenience, and so they offer different types of generality. These different types of generality make for good explanations in different circumstances. This is why different levels of explanation are preferable in different contexts, even for a single explanandum. This accounts for, e.g., distinct evolutionary and developmental explanations of large brains in humans. The evolutionary explanation will provide insight into causal regularities involving certain selection pressures, range of genotypes, and earlier evolutionary history. Alternately, the developmental explanation will provide insight into causal regularities involving certain developmental processes, given the population s genotypic composition. The evolutionary and developmental explanations offer different types of generality, and by doing so, they highlight different causal regularities in the process that results in humans having large brains. Each explanation thus captures one aspect of why humans have big brains. 14

15 A different form of pluralism regarding levels of explanation has been defended with the argument that some explanations are improved by greater generality, whereas others are improved by greater specificity, and that higher-level explanations and lower-level explanations respectively satisfy these two roles (Jackson and Pettit, 1992; Sober, 1999). This version of pluralism misidentifies the value of multiple levels of explanation. I have shown that the implicit assumption that higher-level explanations are more general is wrong. Lower- and higher-level explanations offer their own, distinct forms of causal information and their own types of generality. Explanations formulated at different levels are valuable in different circumstances not because some explanations should be general and others should be specific, but because different circumstances call for different types of generality. This conception of the relationship linking generality and context to level of explanation grounds a rich explanatory pluralism that accommodates science s many diverse projects. The idea that some explanations should be maximally general and others maximally specific allows for only two forms and thus levels of explanation. My view instead introduces the possibility of many different explanations for a single explanandum, each explanation potentially formulated at a different level of description. This version of pluralism is corroborated by scientific practice. The world is full of complex phenomena, phenomena that are studied in multiple fields of science. Different fields tend to focus on different kinds of patterns. Consider the various fields of biology that investigate the traits of living organisms: ecologists describe the effects of organism-environment interactions; population geneticists describe patterns of genetic inheritance; developmental biologists describe the interactions among genetic and other factors. These different focal points often result in different explananda. Yet because the patterns studied in distinct fields are intertwined and overlapping, sometimes the same questions are asked. When this happens, different fields develop different explanations for a single explanandum. Each explanation describes some part of the phenomenon s causal history and omits information that is not relevant to the causal relationship of interest. For 15

16 this reason, different explanations oftentimes formulated at different levels are best suited to different projects. References Ashmole, N. (1963). The regulation of numbers of tropical oceanic birds. Ibis, 103b: Dretske, F. I. (1972). Contrastive statements. The Philosophical Review, 81(4): Dupré, J. (1993). The Disorder of Things: Metaphysical Foundations of the Disunity of Science. Harvard University Press, Cambridge, MA. Fodor, J. (1974). Special sciences: The disunity of science as a working hypothesis. Synthese, 28: Garfinkel, A. (1981). Forms of Explanation: Rethinking the Questions in Social Theory. Yale University Press. Jackson, F. and Pettit, P. (1992). In defense of explanatory ecumenism. Economics and Philosophy, 8:1 21. Kitcher, P. (1984) and all that: A tale of two sciences. Philosophical Review, 93: Mayr, E. (1961). Cause and effect in biology. Science, 134: Reprinted in E. Mayr, Towards a New Philosophy of Biology. Harvard Univesity Press, Potochnik, A. (forthcoming). Optimality modeling and explanatory generality. Philosophy of Science, Proceedings. Putnam, H. (1975). Philosophy and our Mental Life, volume 2 of Philosophical Papers, chapter 14, pages Cambridge University Press, London, UK. Ricklefs, R. (1980). Geographical variation in clutch size among passerine birds: Ashmole s hypothesis. Auk, 97: Rosenberg, A. (1994). Instrumental Biology or the Disunity of Science. University of Chicago 16

17 Press, Chicago, IL. Sober, E. (1984). The Nature of Selection. MIT Press, Cambridge, MA. Sober, E. (1999). The multiple realizability argument against reduction. Philosophy of Science, 66: Strevens, M. (2004). The causal and unification accounts of explanation unified causally. Nous, 38: Strevens, M. (2008). Depth: An Account of Scientific Explanation. Harvard University Press, Cambridge, MA. van Fraassen, B. C. (1980). The Scientific Image. Clarendon Press, Oxford. Weisberg, M. (2003). When Less is More: Tradeoffs and Idealization in Model Building. Philosophy Dissertation, Stanford University, Stanford, CA. 17