Section 1

Ontological pluralism is committed to the notion that properties exist within multiple levels of organization. Higher level organization, consisting of higher level properties, is brought about by the appropriate structure of lower level properties. Higher level properties are to an extent relational properties. These higher levels of organization are autonomous from lower levels because the higher level properties are often multiply realizable. Any characterization of the phenomena at a lower level will fail to account for the higher level properties that supervene on the lower level properties.
By saying that higher levels of organization are autonomous, the level’s actualization is not solely dependent on the specific lower level properties that bring it to bear. Coincidentally, the higher levels of organization are specific important features of the world. This is not to claim that higher level properties are not dependent on their underlying constituency. These levels of organization can be reduced to lower levels of organization if and only if we fix the context that the lower level properties occur. In absence of recognizing the local context of lower level properties, the multiply realizability of properties within the higher levels dismiss any reductive attempt to account for their activity within complex systems.
The relationship between higher and lower levels of organization is functional and may have its scope shifted to span a numerous amount of nested levels. For instance, organisms are a higher level property brought about by the lower level property of organs and whether an organ is artificial or natural does not necessarily affect the phenomena of the organism. Furthermore, we could shift down to view organs as a higher level of organization instantiated by organic material or view these molecules as a higher level of organization that atoms may be configured. These various levels of organization are ubiquitous within any complexly ordered system.

Recognizing that each level of organization is autonomous, pluralists face two problems. Often conventionalist argue that such a preference for a level is pragmatice or arbitrary. Further they argue that this preference is not necessary because the prefered level is a matter of convience and not fact [Sterelny and Kitcher, 1988; Waters, 1991]. However, even conventionism needs to determine a reasonable method for preferring a given level of organization when discussing a particular phenomenon. Without such a method for identifying how to choose an appropriate level of explanation, we cannot adjudicate the possible intra-level explanatory conflicts that may arise. Pluralism then rapidly degenerates into vapidly valuing all levels of organization as equivalent, i.e., complete ecumenicalism.
Second, how do higher level properties exist and affect lower level levels of organization without appealing to occult powers. For example, often accounts of social phenomena (higher levels of organization) require social groups to be formulated by adding, “some sort of occult social fluid to individual and their interactions [Sober, 1984a, p. 185].
This paper will address both of the above problems. It will review higher level causation and offer a plausible account for how it may exist. Then I shall critically examine a method for selecting a preferred level of organization for describing occurrences. Finally I will briefly suggest how this method can apply to higher level phenomena.

Section 2.

A step toward avoiding total ecumenicalism is to distinguish between two types of higher level properties. First some of these properties could be epiphenomenal, i.e., they are mere by-products of lower level properties. To illustrate, the sound made by you automobile’s engine is epiphenomenal. With the possible exception of Harley Davidson products, internal combustion engines are not designed to harness, utilize, or produce noise. They produce sound as an accidental feature of their design, i.e., the internal combustion engine is not 100% efficient. Epiphenomenal properties occur in virtue of other properties of an event. They do not directly contribute in a causal manner to the behavior of the system in question.
Alternatively higher level properties may causally affect the system in which they belong. Putnam [1973] captures the notion of higher level causation with his example of a one inch square peg being incapable to penetrate a one inch diameter circular hole. While the micro-properties determine the higher level properties of rigidity and shape, the macro-feature of the peg’s geometry essentially explains why the peg will not pass through the circular hole. Simply, the Macro-properties of shape and rigidity of the peg explains this situation. The benefit of the macro-level explanation, unlike the micro-properties of the peg, is that the higher level properties specify a general and interesting class of natural regularities. It is in virtue of this class of properties that causes the peg to resist passing through the hole.
Accepting that higher level properties can either be epiphenomenal or causal, determining which depends on the nature of the complex system. Sober [1984b] supplies a good Gedanken experiment for discerning causal properties from epiphenomenal ones. Imagine a toy consisting of a sealed tube with balls of various sizes inside it. Each different size ball has a unique color distinguishing its size. The tube is sectioned off with filters so that when the tube is placed upright, it sorts the balls. Each different level contains a particular ball size. Thus, when the tube is upright, the balls rest in each section -- descending by size.

Since each ball size has a specific color that identifies it, the sections of the tube demarcate the balls by their color. While the layers each contain a different type of colored sphere, color is an epiphenomenal feature of the system. The balls do not come to rest at their respective sections in virtue of their color. The correlation between color and the balls' position is spurious. However, shape is a casually potent feature that determines where the balls will land within the toys partitioning levels of the toy. For this reason, this higher level property best explains the balls’ position. When identifying a preferred level of organization for explanation, we need to somehow capture these causal cases and not the epiphenomenal ones.

Section 3.

Reichenbach’s [1956] theory of conjunctive forks and the concept of screening off offers a good beginning for developing techniques for selecting preferred levels of explanation. These notions introduce key aspects for identifying deciding features that are causally relevant within a systems’ behavior. I will argue that this account is flawed. However, using the idea of screening-off levels in light of particular modifications, we can have principles that guide us in selecting preferred levels of organization.

Conjunctive forks occur when A and B are correlated, but an asymmetry exists between A and B, P(C|A) ? P(C|B) [Reichenbach, 1956]. If A or B is irrelevant if the other is fixed and not vice-versa, the fork is causal. In these instances the cause can be screened-off [Salmon, 1971, 1984; Brandon 1984, 1990]. More precisely, A screens off B from C when P(C|A&B) = P(C|A) ? P(C|B).

Screening-off causes intuitively plays to our notions of causality. A screens-off B from C because B and C are correlations in virtue of their relation to A. This screening-off relation recognizes a directional aspect of a causal chain [Salmon, 1984]. Likewise, it captures the proximity of causation by screening-off distal causes from more proximal ones.
Brandon [1984, 1990] also employs the idea of screening-off for offering insight on picking an appropriate level of selection in biology. He claims,

‘selection occurs at a given level (within a common selective environment) if and only if (1) there is differential reproduction among the entities at that level; and (2) the “phenotypes” of the entities at that level screen-off properties of entities at every other level from reproductive values at the given level [Brandon, 1990, p. 88].

Brandon uses this criterion to argue, “in standard cases of organismic selection the mechanism of selection, or differential reproduction, is best explained in terms of differences in organismic phenotypes”[p85]. Hence, organismic phenotypes standardly screen-off their respective genotypes when looking for the appropriate unit of selection.

It is apparent if Brandon’s use of screening-off is successful at selecting appropriate levels of organization in general. Screening-off recognizes that phenotypes are a preferred level for most instances of natural selection. Nevertheless, this could be an accidental correlation. Standardly phenotypical properties are more proximate and genotypical properties are generally more distal causes of reproductive success. This instance can be subsumed by our notion that proximal causes should screen-off more distal causes. The collelation cannot add any additional support for Brandon’s view that the phenotypes are the prefered level of selective explanation. Screening-off may only identify the causally relavent levels of organization when that level coincides with having a greater degree of proximity to the effect. This does not show that screening-off manages to identify the casually relavent level for any given debate over the primacy of a level of organization of a complex system.

Central to this problem is an ambiguity that rests in the term “proximate’. The term could refer to the temporal relationship a cause has to an event or it could identify that causal efficacy the given cause has on an effect. It is not disputed that organismic phenotypes are temporally more proximate to reproduction. The debate is whether the phenotypical level of organization is casually potent for issues of natural selection.

Sober [1992] appears to indirectly address this issue with his example of a phone call. “My dialing your number causes your phone to ring: your phone’s ringing causes you to answer it" [Sober, 1992, p. 142]. In this case, R (your phone’s ringing) screens-off the more distal cause D (my dialing the phone) from A (your answering the phone). Technically, R screens-off D from A if and only if:

P(A|P&D) = (P(A|R) ? P(A|D)).
This is compatible with Salmon and Brandon’s account of screening-off. Sober continues,

“...the screening-off requirement takes one closer and closer to the effect itself. Between the phone’s ringing and your answering it, there was your forming the intention to answer the phone. Are we compelled to say that your having the intention to answer the phone is a better explanation of why you did so that wither the fact that the phone was ringing or that I dialed the number? I see no reason to think this...[1992, p. 149].

First, Sober’s objection to screening-off is obviously similar to Russell’s criticism of causation. Russell [1918] argued that given event E1, and event E2, and that E1 caused E2, we are faced with a problem. Since both E1 and E2 occurred at different points in time, we can find an event E1’ which occurs between the two events. Thus E1’ actually caused E2 and not E1. This process can occur ad infinitum. Likewise, Sober criticizes screening-off proximal causes because the compact nature of time allows for an infinite number of more proximal, and thus more explanatory, causes.

Brandon’s reply is that such a criticism rests on an ontology of events and not one of causal process and causal interactions [Brandon et. al, 1994]. A causal process persists and transmits its own structure through space and time. This structure is not a linear progression of discrete events. A causal interaction is a space-time intersection of more than one causal process that modifies properties of both of these processes [Salmon, 1984]. To this effect, Brandon dismisses the Russellian objection.

What Brandon does not explicitly address, and what Sober did not lucidly express, is a more subtle objection within the criticism of the screening-off of causes. Sober is implicitly asking if the more proximate, temporal, cause constitutes a better explanation. His example is asking whether: my dialing the your number, your phone ringing, your intention to answer the phone, your walking towards the phone, or some other cause best explain why you answered the phone? Screening-off specifies that the more proximal cause better explains the phenomena. However, when selecting a preferred level of organization, it is not perspicuous that distal causes cannot offer better explanations.

In addressing Sober, Brandon [1994] gave the example of an explanation of a baseball’s flight pattern. The ball will interact with a countless number of atomic particles multifariously. Brandon argues that these micro-properties are unnecessary for explaining the ball’s flight path and that,

The countless interactions, each having only a minute effect on the ball, can be treated statistically where the relevant macro-properties are wind speed and direction, barometric pressure, temperature, and relative humidity...Thus we can explain the trajectory of the ball after it is hit in terms of that major interaction and prevailing conditions’ [1994, p. 478, italics are his].

Consequently, Brandon dismisses the Russellian objection by appealing to a causal model where causally potent properties are those which best explain the system’s behavior. These relavent causes are not necessarily the forces that are more temporally immediate.

Again we face the ambiguity of ‘proximate’. Brandon wants to argue that screening-off identifies the casually potent, yet screening-off only identifies the more immidiate. In the baseball example, immediate temporal causes, micro-properties, that act on the ball just after it is struck by a bat do not vary the ball’s trajectory. Simply using the standard case of screening-off one can fix the conditions of the event of the ball’s impact on the bat and all the counterfactual causal paths that the following micro-properties create remain constant. These micro-properties will thus be screened-off. However, micro-properties comprise the macro-properties that Brandon beleives explain the ball’s flight path. If we are to atemporally look at the moment that the ball was struck by the bat, wouldn’t the micro-properties more immediately explain what caused the force behind the ball than a general of macro-properties?

To Brandon, a causally potent explanation screens-off more distal causes from proximal ones. This analysis breaks down in the more interesting cases of token identical relationships within complex systems. In these cases the levels of properties in question are properties of the same physical structure. For example, temperature is exactly what its lower level properties instantiate. Because of the token identity, one cannot take advantage of the temporal difference to identify the immediacy of either level.

Brandon’s use of screening-off as a criterion for distinguishing groups from individuals as units of selection displays this problem. He suggests that group selection occurs when there exists, “Some group property (the group phenotype) that screens-off all other properties from the group reproductive success” [Brandon, 1990, p. 87]. Consequently, the expected number of propagule groups that a group produces bears an asymmetry between the group phenotype and the organisms within the group. However, this asymmetry may never occur because the unary and relational properties of individuals determine the group properties [Sober and Wilson, 1994]. The properties of the individuals will specify what the group will possess and coincidentally will never be screened-off.

Sober and Wilson identified this interesting case where neither level is closer to the effect under scrutiny. In their example, the group is token identical to the individuals that create it. Here, unlike the case of genotypical and phenotypical properties of organisms, the levels in question are properties of the same physical structure and occur at the same time.
Brandon’s account of screening-off captures our intutitions about identifying causally properties by distinguishing tthe property’s proximity to the effect. WE have seen that this account is either falsely favors more immediatet temoral properties or recognizes properties that are more causally responsible for the effect in question. Brandon has confirmed that he belives the latter [Brandon, et. al., 1994]. However, his position breaks down at this point. He does not have a mechanism that can discern relevant causes thtat are token identical within a complex system. Here atemporality bars issues of a properties’ proximity since the properties describe the same physical structure.

4. Multiple Realizability and Context Dependance

Brandon’s account of screening-off can be supplemented to account for token identity. Multiple realizability and context sensitivity may act as a substitute for spatio-temporal isolation that is generally used when screening-off [McClamrock, 1995a]. In this section I will suggest how multiple realizability and context sensitivity can identify causally potent factors from epiphenomenal ones and suggest that this supplement to Brandon’s theory, but the addition will also conflict with Brandon’s notion that proximity of a cause makes the cause efficacious.

Multiple realizability allows variations in lower level phenomena to occur without changing the higher level properties of the complex system. For example, the higher level property of a progestin is realizable by both levonorgestrel or medroxyprogesterone. Contrarily, context dependence allows local lower level properties to remain constant while fluctuations occur in the higher level properties, e.g., a screwdriver can either be a pry, a driver of screws or a weapon depending on its surrounding context.

Multiple realizability can make causally potent higher level properties perspicuous within a complex system. If we can counterfactually determine the level that directly affects the systematicity of a complex structure, the that level is causally potent. By fixing the higher level property within a structure, if it is causally potent, the lower level properties that instantiate this property should be independent from the higher level property. They are not independent by virtue of their causal history, but by the irrelevance of the local lower level implementation. There higher level properties then screen-off the independent lower level properties.

Conversely, we can identify the causal immediacy of certain lower level properties using context dependence. The higher-level properties of a system may vary, depending on context, when the lower level properties are fixed. To illustrate, if we fix the chemical structure of deoxyribonucleic acid (DNA) and modify its position in the genome, the genetic properties will greatly vary. These instances of context dependence allow lower level properties of structures to screen-off higher-level properties from some class of effects by virtue of he higher-level causal independence.

The t allele in Mus musculus demonstrates how both context sensitivity and multiple realizability can recognize a preferred level of organization. Homozygotes for the t mutant allele are either unconditionally lethals or male steriles [Dunn, 1957]. So homozygote t/t males have a radical phenotypic variation from t heterozygotes and non-t males; the homozygous t sperm males are sterile (or dead). If we fix the phenotypic property we can observe that the variation in the fitness of mice is screened-off by phenotype. Homozygous male are fatally worse than other males.

A difference in fitness also occurs between heterozygous t-sperm males and non-t males. A powerful organelle selection favors the t allele; 85%-99% of all males carry t-sperm [Lewontin, 1968]. Regardless of the ill effects of homogeneity, at a lower level of organization t-alleles are selected for. Heterozygous males risk producing homozygous offspring when procreating with heterozygous females. Non-t males do not have this hazard. Intuitively we should expect non-t males to have a better fitness and suggest that non—t males should have a significantly better fitness than heterozygous t-sperm males. However, his is not the case. If we fix the genotypical property and not the phenotypical property of the group behavior, we can counterfactually determine that the fitness of these two different genotypes depends on their surrounding context, i.e., the number of heterozygous females within the environment. In a significantly large enough population where heterozygous males can travel to different populations of females, their fitness variation from non-t males is insignificant. In this case, contrary to Dawkins [1984] and in more agreement with Mayr [1963] and Gould [1980], the selectional force seems to favor the phenotype instead of the genotype. Further, because the group dynamic within the population determines the success of the t-allele, we may suggest that a higher-level property at the group level determines the t-alleles fitness.

Using context sensitivity and multiple realizability we can help discern how to screen-off the causal behavior. While powerful organelle selection favors the t-allele, if we fix the properties at the genetic level, we will note that the fitness will greatly fluctuate given changes in the environment. Taken individually non-t males and male t-heterozygotes will have various fitness values across different environments. The t-allele within groups, given fixed population or environmental features will remain stable, while this property of the population is multiply realized. Since the fitness and degree of the t-allele is fixed given the context of the population and is multiply realized throughout the population, it best ‘screens-off’ the selectional phenomena even though technically the organelle level of organization is more proximate to the male mice carrying the allele.

Returning to Brandon’s ball example of explaining a ball’s trajectory with screening-off we can observe how context dependence and multiple realizability can determine causal primacy in obvious token identical situations. Brandon argues that those ‘relevant properties’ are macro-properties, such as wind speed and temperature. Meanwhile the micro-properties are either events that are causally irrelevant or he does not address them. Nevertheless, the macro-properties are precisely the type occurrence of the token the micro-properties instantiate. Further, Brandon wants to argue that his theory is not relying on pragmatic concerns, such as conventionalism, but his theory helps, “point out an objective feature of our world” [Brandon, et. al. 1994]. If we can fix the macro-properties and the micro-properties become independent, then these macro-properties will be more directly responsible for the effect (the ball’s trajectory). They do not screen-off the lower level properties in because they are more pragmatic, they are more conventional, or that micro-properties are characteristically irrelevant. The macro-properties are relevant because the ball’s trajectory is best explained, more directly caused, in virtue of the general class of properties these macro-properties specify.

Observing that multiply realizability can screen-off lower level properties conflicts with Brandon’s theory. The immediate cause of the ball’s path is brought about by the particular token instantiation of the macro-properties that he mentioned. If the macro-properties are multiply realizable, then no lower level property, a more proximal cause, could supply the generality required to informatively explain the ball's trajectory. The ball’s path is fixed, regardless of the lower level properties because the variations in the natural class the macro-properties, distal causes, is fixed.

In Brandon’s example, the ‘relevant macro-properties’ are relevant because these distal causes best explain the behavior in virtue of the class these macro-properties describe. In these instances the distal causes ‘screen-off’ the more proximal causes. They do this by creating a taxonomy that better accounts for the systematicity of the complex structure.
The context dependence and multiple realizability thesis conflicts with Brandon’s notion of screening-off distal causes for proximal ones, but this is not necessarily a bad thing. It gives us insight on how the higher level properties may be responsible for causation at a distance and offers a more consistent account of higher level causation. Coincidentally, screening-off by using context sensitivity and multiple realizability can accommodate higher level causation, such as the shape of Putnam’s peg penetrating a hole or possibly how intentional behavior may have causal efficacy.

References

Brandon, R. [1984] ‘The Levels of Selection’ in Genes, Organisms, and Populations: Controversies Over the Units of Selection, R. Brandon and R Burian (eds.) Cambridge: MIT Press.
Brandon, R. [1990] Adaptation and Environment, Princeton: Princeton University Press.
Brandon, R, et. al. [1994] Discussion: Sober on Brandon on Screening-off and the Levels of Selection’ Philosophy of Science, 61:475- 486.
McClamrock, R. [1995a] ‘Screening-off and the Levels of Selection,” Erkenntnis, 42: 107-112.
McClamrock, R. [1995b] Existential Cognition Chicago: University of Chicago Press.
Putnam, H. [1973] ‘Reductionism and the Nature of Psychology’ Cognition, 2: 131-146.
Reichenbach, H. [1956] The Direction of Time Ed. Reichenbach, L.A.: University of California Press.
Russell, B. [1918] ‘On the Notion of Cause’ in Mysticism and Logic and Other Essays. London: Longmans Green.
Salmon, W. [1971] Statistical Explanation and Statistical Relevance, Pittsburgh: University of Pittsburgh Press.
Salmon, W. [1984] Scientific Explanation and the Causal Structure of The World, Princeton: Princeton University Press.
Sober, E. [1984a] ‘Holism, Individualism, and Units of Selection’ in Conceptual Issues in Evolutionary Biology, Ed. E. Sober, Cambridge: MIT Press.
Sober, E. [1984b] The Nature of Selection, Cambridge: MIT Press.
Sober, E. [1992] ‘Screening-off and the Units of Selection’ Philosophy of Science, 59: 92-106.
Sober, E. and D. S. Wilson [1994] ‘A Critical Review of the Philosophical Work on the Units of Selection Problem’ Philosophy of Science, 61: 534-555.