Causation in biology: stability, specificity, and the choice of levels of explanation


This paper attempts to elucidate three characteristics of causal relationships that are important in biological contexts. Stability has to do with whether a causal relationship continues to hold under changes in background conditions. Proportionality has to do with whether changes in the state of the cause “line up” in the right way with changes in the state of the effect and with whether the cause and effect are characterized in a way that contains irrelevant detail. Specificity is connected both to David Lewis’ notion of “influence” and also with the extent to which a causal relation approximates to the ideal of one cause–one effect. Interrelations among these notions and their possible biological significance are also discussed.

This is a preview of subscription content, log in to check access.


  1. 1.

    A related point is that for ease of exposition, I generally discus what it is for a causal relationship linking a (single) factor C to an effect E to be stable, specific etc. But my discussion should be understood as applying also to the stability, specificity etc. of relationships linking combinations of causal factors, C 1, C 2 etc. to effects—these too can be more or less stable etc. In particular, it should be kept in mind that even if the individual relationships between C 1 and E and between C 2 and E are by themselves relatively unstable, non-specific etc., it is entirely possible for relationships linking different combinations of values of C 1 and C 2 to E, to be much more stable and specific.

  2. 2.

    Another way of describing the project is in terms of the development of a vocabulary and framework for describing features of causal relationships that are often of biological interest; a framework that (I would claim) is more nuanced and illuminating than more traditional treatments of causation in terms of laws, necessary and sufficient conditions and so on.

  3. 3.

    Philosophers often focus on causal claims relating types of events. We can represent this with a framework employing variables, by thinking of X and Y as two-valued, with the values in question corresponding to the presence or absence of instances of the event types.

  4. 4.

    A more precise and detailed characterization of this notion is given in Woodward (2003, p. 98).

  5. 5.

    For more detailed discussion, see Woodward (2006).

  6. 6.

    Relatedly, it is no part of my argument that relatively stable gene → gross phenotypical traits relationships are common. Arguably (e.g., Greenspan 2001) they are not, but if so, we still require the notions of stability/instability to express this fact.

  7. 7.

    This second condition is not redundant; even if each individual link in the chain satisfies M, there may be no overall counterfactual dependence between X n and X 1. See Woodward (2003, pp. 57ff).

  8. 8.

    As Kendler has pointed out to me, this is essentially the logic behind looking for so-called endophenotypes in psychiatric genetics, when these are construed as common pathway variables that are causally intermediate between genotype and phenotype—see, e.g., Gottesman and Gould (2003). Ideally, relationships between endophenotype and phenotype will be more stable than genotype—phenotype relationships and also perhaps more causally specific in the 1–1 sense described in Sect. 5.

  9. 9.

    Some macro-level relationships may be highly stable (under, say, some range of changes in features of their components) and may better satisfy other conditions like proportionality described below. Relationships among thermodynamic variables provide examples. Whether stable relationships are to be found at more micro or more macro levels is thus always an empirical question.

  10. 10.

    With respect to a set of variables like {wish for victim’s death, firing of gun, victim’s death}, the relationship between the second and third variables will be “direct” or “proximal”. With respect to an expanded more fine grained set of variables {wish for victim’s death, firing of gun, penetration of victim’s heart by bullet, loss of blood supply to brain, victim’s death} the relationship between firing and death is mediated or distal. But the overall stability of the firing → death relationship does not depend on whether we employ a representation with these intermediate variables.

  11. 11.

    Suppose one has a network of interacting causal structures or units, with, e.g., C 1 causing C 2, C 2 in turn influencing both C 3 and C 4 and so on. I have elsewhere (Hausman and Woodward 1999; Woodward 1999, 2003) characterized such a structure as modular to the extent that various of these causal relationships can be changed or disrupted while leaving others intact—that is, a relatively modular structure is one in which, e.g., it is possible to change the causal relationship between C 1 and C 2 while leaving the causal relationship between C 2 and C 3 intact. When modularity is so understood, it is one kind or aspect of stability—it involves stability of one causal relationship under changes in other causal relationships (which we can think of as one kind of background condition). Like stability, modularity comes in degrees and relative modularity is a feature of some sets of causal relationships, not all. (As recognized in Woodward 1999). Hausman and Woodward (1999) contains some mistaken assertions to the contrary, appropriately criticized in Mitchell (2009). Notions of modularity figure importantly in recent discussions of genetic regulatory networks and other structures involved in development and in evolutionary change—see, e.g., Davidson (2001). Obviously, it is an empirical question to what extent any particular example of such a structure is modular (see Mitchell 2009 for additional discussion.) My claim is simply that modularity (and its absence), like stability more generally, is a feature of causal relationships and their representation that is of considerable biological interest.

  12. 12.

    That is, there is a change in the condition cited in (3.1) (from scarlet to non-red) which is associated with a change in pecking, so that M judges that (3.1) is true; hence requires revision if (3.1) is false.

  13. 13.

    Another way of understanding proportionality is in terms of employing variables that allow for the parsimonious maximization of predictive accuracy. When P fails there will either be a characterization of the cause such that variation in it could be exploited for predictive purposes but is not so used or else “superfluous” variation in the cause which does not add to the predictability of the effect.

  14. 14.

    A point recognized by many writers. Greenspan (2001) writes, “specificity has been the shibboleth of modern biology” (383) and Sarkar (2005) that “specificity was one of the major themes of twentieth century biology” (263).

  15. 15.

    Waters speaks in this passage of DNA as “the” causally specific actual difference maker for RNA molecules “first synthesized” in eukaryotic cells (i.e., presumably pre-mRNA) but he goes onto note that in eukaryotes different varieties of RNA polymerase and different splicing agents are involved in the synthesis of mature RNA, with different splicing agents also acting as causally specific actual difference makers for this mature RNA. Thus, according to Waters, while DNA is causally specific actual different maker for mature RNA in eukaryotes it is not the only such causally specific agent. As previously emphasized, this will not affect my discussion below, which focuses on what it might mean to say that DNA is causally specific with respect to RNA and not on whether other causes are also present that act in a causally specific way. Also the DNA that acts as a causally specific actual difference maker is of course activated DNA.

  16. 16.

    See Kvart (2001) for examples.

  17. 17.

    A mapping F from X to Y is a function iff F(x 1) = y 1 and F(x 1) = y 2 implies y 1 = y 2. A function F is 1–1 iff F(x 1) = F(x 2) implies x 1 = x 2. F is onto iff for every y in Y, there exists an x in X such that F(x) = y. This characterization may be compared with the characterizations in and Weber (2006) and in Sarkar (2005), which I discovered only after formulating the ideas above. I believe that Sarkar’s intent is to capture notions that are very similar to mine, but have some difficulty in understanding how the mechanics of his definitions work. In particular his use of “equivalence classes” seems to make his condition on “differential specificity” redundant; satisfaction of this condition is insured just by the assumption that different elements in the domain of the mapping, a and a’, belong to different equivalence classes. In other respects there is close parallelism: Sarkar’s condition (ii) that B be “exhausted” is (I assume) just the assumption that F is onto and the intent of his “reverse differential specificity” condition seems to be captured by the assumption that F is 1–1.

    Weber (2006) suggests that “causal specificity is nothing but the obtaining of a Woodward-invariance for two sets of discrete variables”. Weber’s paper is highly illuminating about the role of specificity in Crick’s central dogma, but his characterization of specificity is very different than mine: a functional relationship might be invariant and involve discrete variables but not be 1–1 or onto, might relate only two-valued variables (in violation of the “many different states” requirement in INF) and might violate the one cause one effect condition described below. Weber’s condition seems to me to have more to do with stability than specificity.


  18. 18.

    This way of formulating matters makes it clear that Proportionality and specificity in the sense of INF are related notions. To the extent that, e.g., there are states of E that cannot be reached by realizing states of C, there will be a failure of proportionality.

  19. 19.

    This one-cause-one-effect notion of specificity is also closely intertwined with the notion of an intervention, as discussed in Woodward (2003). One wants the relationship between an intervention I and the variable C intervened onto be “targeted” or surgical in the sense that I affects C but does not indiscriminately affect other variables—in particular, those that may affect the candidate effect E via a route that does not go through C. A manipulation lacking this feature is not properly regarded as an intervention on C with respect to E. Thus, to use an example from Campbell’s (2006), derived originally from Locke, pounding an almond into paste is not a good candidate for an intervention on its color because this operation alters so many other properties of the almond. Often, as this example illustrates, the most causally significant variables in a system will be those we can manipulate specifically. Moreover, in many cases, these will be “mechanical” variables like position, density etc.

  20. 20.

    Referring back to Kendler’s discussion, recall he describes muteness as a “nonspecific consequence” of the hypothetical gene X (which causes mental retardation) in the first of his scenarios. Prima-facie, this may seem puzzling. After all muteness seems, if anything, more specific in the sense of being less abstract and a “narrower” category than mental retardation. The sense in which muteness, in comparison with mental retardation, a non-specific consequence of X seems to be that muteness is one of many effects of X, in contravention of the one cause-one effect ideal of specificity.

  21. 21.

    Compare Crick’s sequence hypothesis: “the specificity of a piece of nucleic acid is expressed solely by the sequence of its bases, and […] this sequence is a (simple) code for the amino acid sequence of a particular protein” and his association, in his statement of the Central Dogma, of both specificity and “information” with the precise determination of sequence, either of bases in the nucleic acid or of amino acid residues in the protein” (Crick 1958, 152, 153). The ideas of causal specificity and information are obviously closely linked; as this example illustrates, biologists tend to think of structures as carrying information when they are involved in causally specific relationships. I regret that I lack the space to explore this connection in more detail.

  22. 22.

    Here, though, we should keep in mind the caveat in footnote 1: it may be that specific stable control is achieved through the interaction of a number of different agents which taken individually have a much less stable and specific effect on the outcome of interest.

  23. 23.

    As a pre-cautionary move, let me try to head off some possible misunderstandings of this argument. When the issue is control by a human agent, whether a relationship is useful or not for that agent of course depends on (among other considerations) the agent’s purposes and values. In some cases, potential manipulators may not care that some cause has non-specific effects on many other variables (because they regard those effects as neutral) or may even think of this as making the cause a particular good target for intervention, as when these various non-specific effects are all regarded as undesirable and the cause provides a handle for affecting all of them. For example, smoking and childhood sexual abuse have many non-specific effects, virtually all of which are bad and this provides strong reason for trying to intervene to reduce the incidence of both causes. My discussion above is not intended to deny this obvious point. Rather my claim is simply that causal relationships that are stable, specific etc. have control-related features that distinguish them from relationships that are unstable, non-specific etc. Second, and relatedly, I emphasize that my aim has been the modest one of suggesting some reasons why the distinctions between stable and unstable relationships, specific and non-specific relationships and so on is biologically significant. Obviously nature contains (or at least our representations represent nature as containing) stable, specific etc. and unstable, non-specific relationships. I do not claim that the former are always more “important”, fundamental, valuable, or more worthwhile targets of research than the latter. One can coherently claim that the distinctions I have described are real and have biological significance without endorsing such contentions about importance. Thanks to Ken Kendler for helpful discussion of this point.

  24. 24.

    I don’t claim that these are the only considerations relevant to the classification of a factor as an enabler.


  1. Campbell J (2006) Manipulating color: pounding an almond. In: Gendler T, Hawthorne J (eds) Perceptual experience. Oxford University Press, Oxford, pp 31–48

    Google Scholar 

  2. Crick F (1958) On protein synthesis. Symp Soc Exp Biol 12:138–163

    Google Scholar 

  3. Davidson E (2001) Genomic regulatory systems: development and evolution. Academic Press, San Diego

    Google Scholar 

  4. Dawkins R (1982) The extended phenotype: the long reach of the gene. Oxford University Press, Oxford

    Google Scholar 

  5. Gottesman I, Gould T (2003) The endophenotype concept in psychiatry: etymology and strategic intentions. Am J Psychiatry 160:636–645

    Article  Google Scholar 

  6. Greenspan R (2001) The flexible genome. Nat Rev Genet 2:383–387

    Article  Google Scholar 

  7. Griffiths P, Gray R (1994) Developmental systems and evolutionary explanation. J Phil 91:277–304

    Article  Google Scholar 

  8. Hausman D, Woodward J (1999) Independence, invariance and the causal Markov condition. The Br J Philos Sci 50:521–583

    Article  Google Scholar 

  9. Hill A (1965) The environment and disease: association or causation? Proc R Soc Med 58:295–300

    Google Scholar 

  10. Kendler K (2005) A gene for…: the nature of gene action in psychiatric disorders. Am J Psychiatry 162:1243–1252

    Article  Google Scholar 

  11. Kvart I (2001) Lewis’ ‘causation as influence’. Australas J Philos 79:409–421

    Article  Google Scholar 

  12. Lewis D (1986) Postscript c to ‘causation’: (insensitive causation). In: Philosophical papers, vol 2. Oxford University Press, Oxford, pp 184–188

  13. Lewis D (2000) Causation as influence. J Phil 97:182–197

    Article  Google Scholar 

  14. Mitchell S (2000) Dimensions of scientific law. Phil Sci 67:242–265

    Article  Google Scholar 

  15. Mitchell S (2009) Unsimple truths: science, complexity, and policy. University of Chicago Press, Chicago

    Google Scholar 

  16. Oyama S (2000) Causal contributions and causal democracy in developmental systems theory. Phil Sci 67:S332–S347

    Article  Google Scholar 

  17. Rieke F, Warland D, van Steveninck R, Bialek W (1997) Spikes: exploring the nature of the neural code. MIT Press, Cambridge

    Google Scholar 

  18. Sarkar S (2005) How genes encode information for phenotypic traits. In: Sarkar S (ed) Molecular models of life. MIT Press, Cambridge

    Google Scholar 

  19. Susser M (1977) Judgment and causal inference: criteria in epidemiologic studies. Am J Epidemiol 105:1–15

    Google Scholar 

  20. Thompson J (2003) Causation: omissions. Phil Phenomenol Res 66:81–103

    Article  Google Scholar 

  21. Waters K (2007) Causes that make a difference. J Phil CIV:551–579

    Google Scholar 

  22. Weber M (2006) The central dogma as a thesis of causal specificity. Hist Philos Life Sci 28:595–609

    Google Scholar 

  23. Woodward J (1999) Causal interpretation in systems of equations. Synthese 121:199–257

    Article  Google Scholar 

  24. Woodward J (2003) Making things happen: a theory of causal explanation. Oxford University Press, New York

    Google Scholar 

  25. Woodward J (2006) Sensitive and insensitive causation. Phil Rev 115:1–50

    Article  Google Scholar 

  26. Yablo S (1992) Mental causation. Phil Rev 101:245–280

    Article  Google Scholar 

Download references


Versions of this paper were given as talks at a Boston Studies in Philosophy of Science Colloquium on causation in biology and physics, October, 2006, a University of Maryland conference on causation and mechanisms in April, 2007, at the University of Pittsburgh, October, 2007 and at meetings of the SPSP and the Behavioral Genetics Association in June, 2009. Particular thanks to James Bogen, Lindley Darden, Peter Machamer, Sandra Mitchell, Ken Schaffner, Ken Waters, Marcel Weber, and especially Ken Kendler for helpful discussion.

Author information



Corresponding author

Correspondence to James Woodward.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Woodward, J. Causation in biology: stability, specificity, and the choice of levels of explanation. Biol Philos 25, 287–318 (2010).

Download citation


  • Cause
  • Stability
  • Levels of explanation
  • Specificity