7 Objectivity: A Bulwark against Bias
Despite what I’ve said thus far, worries about objectivity might remain.1 To address these worries, we need to consider what objectivity is and why it is valuable. On one interpretation, objective judgments are nonperspectival representations that carve the mind-independent world at its mind-independent joints. Clearly the epistemological position I have been developing cannot claim that the commitments it countenances display that sort of objectivity. This is just as well. I consider that sort of objectivity a chimera. It is not something we can achieve (or, I suspect, even make sense of). Nor is it something we need to achieve to serve our epistemic purposes. But it does not follow that objectivity is a chimera. I will argue that my position countenances a robust form of objectivity that is procedural, perspectival, and either impersonal or impartial.
Utter Objectivity
Some think that for a contention to be objective is for it to express, reflect, or represent what is there, regardless of what anyone or any group thinks or does. That is, it is to reflect what Williams (1978) calls the absolute conception of reality. Such objectivity is, Nagel (1986) maintains, what the view from nowhere delivers. Let us call contentions that satisfy this standard utterly objective. Products of and statements about attitudes, institutions, and artifacts reflect human interests and goals. They involve views from somewhere. They are excluded from the utterly objective realm, regardless of their truth-value or warrant. But, Williams, Nagel, and their followers maintain, natural science is different. It discovers and explains what is, in Williams’s term, there anyway, most of which would have been there even if rational organisms had never evolved. It and its findings are utterly objective. Utter objectivity is a consequentialist notion. What makes a procedure utterly objective is that its products represent aspects of the way the world is anyway.
Science has been stunningly effective at providing an understanding of nature. No other approach even comes close. Much of nature is as it is regardless of what, if anything, anyone thinks about it. Perhaps, then, it is reasonable to think that science represents the way the world is anyway, that at least in the ideal end of inquiry it will (would?) do so completely and without bias or distortion, and that current science approximates the ideal. If so, scientific representations are utterly objective: what they say or otherwise represent does not depend essentially on users or potential users with distinctive interests and points of view. The semantics of such a representation is thus independent of and logically prior to its pragmatics.
A representation is a putatively denoting symbol. Among the familiar sorts of representation are verbal descriptions, mathematical equations, pictures, diagrams, and maps. Such symbols need not actually denote; their semantic character is determined by their being the sort of symbol that would, if the world obliged, denote. ‘Phlogiston’ is as much a denoting symbol as ‘oxygen’ is. In one sense, any representation depends on users or potential users with distinctive interests and points of view. Something would not be a representation unless it was used or could be used to represent, and it would not be a representation of a particular thing unless it was used or could be used to represent that very thing. There would be no representation of a particular item if no one had interests that would be served by producing one. The sentence ‘Tibet is mountainous’ would not represent the topography of a particular geographical region if speakers did not call a particular land area ‘Tibet’ and did not call a particular sort of terrain ‘mountainous’. But the choice to correlate particular labels with particular referents is effectively arbitrary. The very same information would have been conveyed by the sentence ‘Murble is morbish’ if ‘Murble’ had been the English word for Tibet and ‘morbish’ the English word for mountainous. The user-independence required for utter objectivity is independence except for the making of such arbitrary choices. The contention that scientific representations are utterly objective is a contention that, beyond arbitrary choices, they exhibit no dependence on users or potential users.
Van Fraassen (2008) poses a challenge to this picture: if scientific representations—models, equations, maps, graphs, and the like—are to perform their scientific functions, they cannot be utterly objective. Cartwright (1983) makes similar arguments about particular scientific models. Like van Fraassen, she regards this feature not as a defect of models but as an insight into the sort of epistemic access they supply. Van Fraassen’s argument runs deeper. It is not just that this or that model, or models in this or that science, cannot be utterly objective. Rather, in order to afford the sort of understanding of nature that it does, science cannot consist exclusively or predominantly of utterly objective representations. Science as we know it must deploy perspectival representations. It ineliminably involves views from somewhere. For science is an epistemic practice. To make this out, we need to consider the nature of perspective.
Getting Perspective
Linear perspective is a miracle of Western art. First described by Alberti (2011), it is a method for representing spatial depth on a flat surface. Lines orthogonal to the picture plane converge at the vanishing point, and the depicted sizes of objects are proportional to their actual size and to their ostensible distance from the viewer. The pictorial effectiveness of linear perspective in works like Raphael’s School of Athens is obviously of great aesthetic interest. Van Fraassen maintains that linear perspective should be of equally great interest to philosophy of science. With the development of projective geometry, it became possible to prove that properly drawn linear perspective representations are rigorous geometric projections. Because cross-ratios are invariant across changes in orientation and origin, drawings in perspective convey objective information about constancies in the appearances items present from different points of view. They are not utterly objective, though, for they still depend on points of view. Perspective drawings are indexical: they represent how things look from here (for some value of ‘here’). Their indexicality does not, however, make them subjective. Cross-ratios are determinate matters of fact about projective relations, not mere matters of opinion. Van Fraassen concludes that a linear perspective drawing of an actual scene is a measurement, a mapping.
There are inherent limitations on what any single representation can represent. Every representation is a product of selection. The representer has to choose what to represent, what aspects of the represented item to capture in the representation, and what level of detail to represent. Despite the fact that these are matters of choice, when it comes to scientific representations, it is possible to make mistakes. Simpson’s paradox arises because fine-grained statistical regularities are obscured at a coarser grain. Although the pattern of graduate admissions across the university suggested that Berkeley discriminates against women, the admissions patterns of each graduate department indicated the contrary. Settling the issue requires knowing which statistics to use (Cartwright, 1983, 37). To use a coarse-grained representation to answer questions that require a finer grain is to make a mistake, even if all the features displayed in the coarse-grained representation actually obtain. Likewise, it is an error to use a fine grain when a coarse grain is needed.
Because they depend on a specific origin and orientation, perspective drawings have further limitations. From a given vantage point, one object may occlude another. If the horse is behind the barn from the picture’s point of view, then the picture’s representing the barn precludes its representing the horse. Nor is this always a matter of big things occluding little ones. If the mouse is close to the picture plane and the barn is far from it, the perspectival representation of the mouse may occlude the barn. A representation is implicitly noncommittal with respect to a property if it makes no commitment as to whether the represented object has that property. A stick figure is implicitly noncommittal with respect to its subject’s girth. It simply does not go into the matter. A representation is explicitly noncommittal with respect to a given property if its representing the having of one feature precludes its taking a stand on the having or lacking of another. A representation of a man wearing a hat is explicitly noncommittal with respect to whether he is bald, because representing him as hatted makes it impossible for the picture to commit itself on the question of his baldness. To be sure, the horse, the girth, and the hairline could be represented in a perspective drawing. But to represent the horse requires removing the barn or changing the perspective. To represent the girth requires a more rounded figure. To be committal with respect to the man’s baldness requires omitting the hat.
Indexicality, occlusion, and noncommitment do not either severally or jointly entail that there are things that cannot in principle be represented perspectivally. They entail limitations on what a single perspectival representation can represent. But because of these limitations, there cannot be a single, comprehensive perspectival representation that represents everything from a single point of view. The ‘God’s eye view’ cannot be a point of view.
The pictures painted by Renaissance artists are, of course, of (putatively) visible objects, and the space they (purport to) represent is (putatively) physical space.2 These restrictions are philosophically incidental. Early perspective drawings had only one vanishing point. But it is straightforward to create works with multiple vanishing points. Le Déjuner sur l’Herbe, for example, is in two-point perspective. Thus the number of vanishing points is optional. Features at higher levels of abstraction than size, shape, and distance can be represented perspectivally, if the requisite structural relations are preserved. So the restriction to visible features is optional as well. The space need not even be physical.
A logical space is a multidimensional array of possibilities open to the items that occupy the space. To locate an item in a logical space is to determine which of the possibilities defined by that space it realizes. To represent an item in a logical space is to represent it as having a particular position in the array of possibilities the space marks out. Representations in a logical space, like representations in a physical space, can be perspectival. They can show how occupants of that space appear from a certain vantage point. And they can do so with no loss of rigor. In what follows, I take the term ‘perspectival’ to refer to any representation that represents how things appear from a particular point of view.
With the restrictions to physical space and visible features lifted, it is evident that science could avail itself of perspectival representations. It could generate and deploy a host of perspectival drawings, diagrams, scale models, and maps. There is plenty of evidence that it does so. Still, one might think, only overtly pictorial or diagrammatic representations—the sketch of a harmonic oscillator, the diagram of the double helix, the tinker-toy model of the protein—could be perspectival. Many scientific representations are mathematical models—systems of equations. It may be hard to imagine how any extrapolation from linear perspective could characterize to them. This may suggest that the visual models that admit of a perspectival interpretation are mere heuristics. They are visual aids that help us imagine the phenomena, but not essential elements of science. The mathematical laws and models—the equations—are the true scientific representations, and they are utterly objective.
Things are not so simple. Analytic geometry provides the resources to interpret geometry algebraically. Geometric truths are provably equivalent to algebraic ones. Contemporary Cartesian geometry demonstrates that equivalence goes both ways (van Fraassen, 2008, 41). Algebraic truths are mathematically equivalent to geometric ones. If we define an appropriate space, pretty much everything we can characterize mathematically can be spatialized. Indeed, we do not even need to do the reduction. The equivalence shows that the equations themselves, whatever their ostensible form, are construable as spatial representations. To be sure, the space need not be three-dimensional physical space. All that is needed is that there be an n-dimensional space of alternatives that embeds the mathematical model. Nor, of course, does this show that the spatialized representations are perspectival. But the mere fact that they are presented as equations provides no reason construe mathematical models as nonperspectival. In principle, then, equations are as capable as pictures and diagrams of bearing a perspectival interpretation.
The View from Nowhere
Plainly, not all pictorial or diagrammatic representations are perspectival, even in the extended sense in which I am using the term. The Cartesian coordinate system provides a familiar and elegant framework for nonperspectival representations. Cartesian representations have no vanishing points; parallel lines in every dimension remain parallel. Although Cartesian representations locate represented objects by reference to an origin and a direction, the choice of origin and direction are arbitrary. “The chosen frames of reference, the co-ordinate systems, are inessential to the geometry taken in and by itself” (van Fraassen, 2008, 69). Such representations are not indexical. One need not locate oneself in the space of the representation to understand what and how it represents. Nor are the representational powers of Cartesian systems limited by occlusion or explicit noncommitment. Rousseau’s picture of a tiger “is explicitly non-committal about the [number of the] tiger’s stripes, because it represents the tiger as obscured by leaves, and this precludes it from representing all the tiger’s stripes” (Lopes, 1996, 118–119). A Cartesian representation, though it would not look like a tiger, could easily circumvent the difficulty. Let the x-axis represent stripes, and the y-axis represent leaves in the environment. The only available positions are integer values along either axis. Then where a stripe is not obscured by leaves, y = 0; where it is overlapped by leaves y > 0, with the value of y indicating the number of leaves overlapping a particular stripe. Where there are leaves but no stripes, x = 0. Such a graph represents the overlap of stripes by leaves, but nothing is occluded. It is expressly committed to the number of stripes on the tiger, regardless of their relation to leaves.
Why shouldn’t we think of the logical spaces of scientific representations as Cartesian spaces? Then, aside from the few scientific representations that are expressly perspectival, we could construe scientific representations as utterly objective. The mathematical models could be spatialized—we could represent them in a Cartesian coordinate system—but the results would still be utterly objective. There would be no reason to deny that they represent the way the world is anyway.
It is undeniable that science can generate Cartesian representations. It is undeniable that such representations are as likely to be accurate and adequate to their subject matters as any other representations we might produce. So perhaps we should consider them utterly objective. In the limit, if not now, they will embody aperspectival information about the way the world is anyway. If science’s function is merely to mirror nature, the claims of ideal science could bear an interpretation under which they reveal how things are anyway. But the fact that science could generate aperspectival representations gives us no reason to think that it does or that it should. As van Fraassen argues, such a construal does not do justice to science. Indeed, under such a construal we have no reason to trust the findings of science. Accuracy and adequacy to the subject matter are not enough.
A View from Somewhere
Science is not something that just happens to us; it is something we do. To do it, we need to use scientific representations. Use is a pragmatic matter; and to make use of anything an agent needs to properly locate herself with respect to it. This is evident in the relation between science and technology. To develop an alloy that resists metal fatigue, a metallurgist has to be able to recognize signs of metal fatigue. To do that, she must adopt a perspective from which she can discern metal fatigue if it is present. This will not, of course, be a matter of merely looking at distressed samples. It will involve measurements, many of them made with technologically complex measuring devices. It will involve subjecting samples to stress tests, many of them technologically mediated. The measurements and test results will exemplify features that put her in a position to say, ‘This is how metal fatigue (or its absence) looks from here’. Assuming she is scrupulous, she will run series of tests that yield different perspectives on the phenomena. She will check one appearance against another, drawing her conclusion from what she observes from a variety of points of view. She does not, because she cannot, solve her problem using the view from nowhere. For she has to know how metal fatigue is manifested.
This point must be conceded; but maybe it is just a point about technology, not about pure science. It is to be expected that when we put something to use we need to adopt a perspective. But, one might think, science is not technology. Pure science may be utterly objective even though deliverances of technology cannot be.
As van Fraassen points out, however, scientific results are established by testing and experimentation, and they are in principle always open to further testing. Testing is as indexical and perspectival as technological deliverances. If a scientist wants to ascertain whether semiconductors operate in a magnetic field, she must run a series of experiments and take measurements. To design the proper experiments and take the proper measurements, she must adopt a perspective on the phenomena—she must figure out how the phenomena would present themselves under various circumstances, how they would look from various points of view. Indeed, the distinction between testing and technology is spurious, since the testing devices are products of technology, and every technological application is in principle a test. Even as simple a measuring device as a tape measure owes its status to considered judgments about the appearances things present—for example, that the items a tape measure measures are not affected by the fact that they are being measured. Hence the fact that something that measures 32 cm is a reliable indication that it is ≈32 cm long.
Van Fraassen’s argument, then, is this: science affords epistemic access to nature. That access is achieved through experimentation and measurement. The only way for an observer to perform the experiments or make the measurements is to locate herself in the logical space of the phenomena. Thus, her stance is indexical and perspectival. She has to think, ‘If φ is going on, these are the appearances it will present under these test conditions’. That is, if φ is going on, this is how it will look from here. This locates the observer in the logical space she is evaluating; it affords a perspective on how things should look from here (for some value of ‘here’).
Although we could take the representations produced by science to be utterly objective, doing so would divorce them from the empirical methods that generate and confirm them. In that case, however, science would provide no reason to believe or accept them. To interpret the representations as utterly objective is to cease to consider them scientific. For science to do its epistemic job, it requires measurement. Measurement is always indexical and perspectival. Hence for science to do its epistemic job, it must involve indexical, perspectival representations.
To be testable, science must use perspectival representations. Those representations are objective in that they contain information that is invariant across representations of the same object. They are testable in that multiple representations of the same objects from the same perspective yield equivalent information, and in that that information can be accessed from multiple perspectives. But they cannot plausibly be construed as embodying the view from nowhere or the way the world is anyway. They are not utterly objective.
Procedural Objectivity
Still, it is clear that in some sense scientific claims are objective. Following Arthur Fine (1998) and Heather Douglas (2004), I suggest that the sort of objectivity we seek is procedural. In the first instance, then, it is procedures that are objective. Findings are objective because they result from or are confirmed by objective procedures. Objective procedures are epistemically valuable because they promote trustworthiness. They are devised, tested, and certified by epistemic communities, who understand their domains, their disciplines, and the available and appropriate means for investigating the phenomena. Objectivity emerges from the self-reflective activities of epistemic communities. It is neither mere correlation of an opinion with mind-independent facts nor a matter of pure consensus. Procedural objectivity is nonconsequentialist. The objectivity of procedures is logically prior to the objectivity of their objects. What justifies calling a particular result objective is that it is the product of an objective procedure.
Trustworthiness might seem to derive from a source. Being someone of great moral and intellectual integrity, Bill would neither lie nor intentionally mislead nor convey epistemically dubious information about an important topic. He would not assert that p unless he was and took himself to be in a position to back up his claim. Nevertheless, his integrity is not enough to make him worthy of confidence. For Joan to responsibly take his word that p, she needs to know or reasonably believe that he is a man of integrity. But we live in a world populated by strangers. We do not personally know everyone whose opinions we might—indeed, must—draw on. Some are trustworthy; some are dishonest; some are unduly cavalier in forming their views; some are careless in conveying them. Often we do not know whom to trust.
To alleviate this problem, we devise and deploy procedures whose outputs are worthy of confidence. If such procedures are properly carried out, their results are trustworthy, regardless of the moral and/or intellectual character of the performer. A postage scale affords excellent reason to think that the envelope weighs 0.5 oz. no matter who is doing the weighing. The results of Marsh’s test afford excellent reason to think that a substance contains arsenic so long as a competent chemist performs the test. Such procedures, rather than the particular individuals who perform them, are the sources of trustworthiness (Fine, 1998; Douglas, 2004). The point is not that the procedures are reliable; it is that we have good reason to trust them. They have been devised, calibrated, and validated to satisfy the standards of a realm of epistemic ends.
Objectivity, so understood, is not equivalent to and does not ensure truth. Hal’s jingoistic conviction that his American car gets better mileage than his neighbor’s foreign car may in fact be true; but his opinion is not objective, since his reasons are political rather than automotive. Nor is Fred’s true belief that his investments lost value today, since his conviction is grounded in his horoscope rather than up-to-date economic information. Nor is Maria’s belief that there are more Muslims than Christians, based as it is on a coin flip. In none of these cases does the procedure used properly align with the conclusion drawn. Procedural objectivity neither assures nor is assured by utter objectivity. There are objective procedures for determining whether Fred’s investments are down even though money is a human institution; whether Joe’s car gets better gas mileage than his neighbor’s, even though cars are human artifacts; whether there are more Muslims than Christians, even though religions are institutions, religious beliefs are attitudes, and religious observances are practices. Moreover, a conclusion arrived at objectively may nevertheless be false. A titer test for Lyme disease yields false positives in the relatively few cases where the patient’s blood contains antibodies that the test cannot distinguish from those standardly produced by the bacterium Borrelia burdorferi. The test has a margin of error. Still, since the margin of error is small, such a test is objective. Although it yields a few false positives, it aligns well enough with the phenomenon in question.
It might seem that trustworthiness is a matter of reliability. What validates the titer test and invalidates the coin flip is that the titer test is, and the coin flip is not, highly reliable. There are at least two difficulties with this position. The first is brought out by a variant on BonJour’s (1985, 38–40) example. Suppose Marie has ESP. She is subject to reliable extrasensory deliverances. In 96.2 percent of the cases where she has such a deliverance, it is accurate. She is also subject to the normal range of hunches and intimations, and her track record with these is no better than anyone else’s. Moreover, there is no phenomenologically salient difference between her deliverances of ESP and her hunches. Nothing in her experience enables her to tell them apart. Nevertheless, the etiology is different. The deliverances of ESP result from a reliable perceptual mechanism; the hunches result from unreliable mechanisms. (Possibly an fMRI could reveal the difference if only we knew where to look.) Marie is aware that scientists believe that ESP does not exist. So although her genuine extrasensory perceptions are reliable, she has no reason to trust them. She has no way to distinguish them from her unreliable hunches, and has good reason to suspect that there is no such thing as ESP. She does not and should not consider her ESP deliverances trustworthy. Given our available procedures, it is not procedurally objective. Evidently, mere reliability is not enough. Minimally, for the process to be trustworthy, the subject should have reason to think it is reliable.
Even that seems inadequate. A second worry brings this out. An alarming number of therapeutic regimens work for no known reason. An effective treatment for bladder cancer, for example, involves flooding tumors with live bacteria. Somehow this kills the cancer cells. The statistical evidence is strong; the therapy is effective. But, not unreasonably, patients feel epistemically insecure about trusting their lives to a course of treatment that sounds unsanitary and whose effectiveness cannot currently be explained. This suggests that knowing that a process is reliable is not enough. We want to understand (at least roughly) why and how it works. Otherwise any confidence we place in it is precarious.
Evidently, what we seek from objectivity is a stay against bias, idiosyncrasy, and chance. Objective procedures eliminate or control for the personal element and for randomness. Those that eliminate the personal element are impersonal; those that control for it are impartial. The assumption driving the quest for objectivity is that bias, idiosyncrasy, and chance are likely to skew results. That being so, they are untrustworthy.
My discussion does not purport to be a comprehensive account of the factors that can underwrite procedural objectivity. My goal is simply to describe aspects of practices we consider objective to show how communities of inquiry devise and validate them, and why the role of such communities does not impugn their claim to being objective.
Impersonality
Impersonal procedures and devices are, or can be made to be, automatic. They deliver what Porter (1995, 7) calls ‘mechanical objectivity’. A motion detector turns on the lights without anyone having to fumble for the switch. It could become a measuring device if a counter were attached to record the frequency with which it does so. A thermometer reflects changes in temperature whether or not anyone notices the changes. The impersonality of such devices and procedures is an achievement, not just a fact of nature. It pays to see how it comes about.
One area where mechanical objectivity is rampant is measurement. We time reactions, weigh objects, calculate areas, and so on. To do this, we require magnitudes—numbers that can be used as bases for comparison of members of a set. So we need to define a magnitude. This is largely a pragmatic matter; it depends on what we want to do with the information we glean. There is plenty of room for choice. But the choice is not completely arbitrary. A viable magnitude must be suitably invariant under the conditions of interest. No one would be happy with the definition of the inch as the length measured by an elastic tape measure. Its readings would be too variable to be worthwhile. But absolute invariance is not required either. Weight is a function of mass and gravity. An object whose mass is constant varies in weight depending on the force of gravity it is subject to. For interplanetary assessments, comparisons of weight are apt to be confusing or misleading; mass would be a preferable magnitude to use. Still, weight is a fine magnitude so long as we limit ourselves to a single gravitational field, and do not seek too much precision. Although there are small differences in the force of gravity across the Earth, for many purposes they are negligible. For mundane terrestrial uses and most scientific uses, weight is an acceptable magnitude. It is invariant enough.
Once we have a magnitude, we need a way to measure it. The units may be arbitrary. Perhaps the length we call an inch was originally chosen because it was the distance around a man’s thumb. That, if true, is a historical curiosity that in no way affects the acceptability of the unit. What matters is that it equips us to make acceptable measurements. An acceptable measurement procedure should insure sufficient interrater agreement, sufficient test–retest agreement, and its results should accord with other acceptable ways of measuring the same magnitude. That is, so long as the item being measured does not change, different competent measurers should obtain pretty much the same result, different measurement procedures should yield pretty much the same result, and a single measurer, making multiple measurements should obtain pretty much the same result. ‘Pretty much’, because measuring devices are never exact. There is always a small margin of error.
It might seem that once we have defined a magnitude, the procedures for measuring that magnitude are fixed, at least in cases that do not involve proxies. But that is not so. It is one thing to specify what magnitude is to be measured and another thing to specify how it is to be measured.
Town halls in eighteenth century Europe were likely to display a bushel vessel, valid for the region. If anybody questioned the accuracy of any particular bushel, its contents could be poured into the official one to see if they were equal. But this was by no means the end of the matter. Everybody knew that grain could be packed more densely by pouring it from a greater height, and for certain purposes the method of filling might be specified in contracts or by law. (Porter, 1995, 24)
Even where we have the magnitude, we need to specify how to properly perform the measurement.
Nor need the magnitude be defined prior to the development of the measuring procedure. Temperature as a magnitude emerged with the development of the thermometer. The process of development was iterative. It involved a series of steps, each refining and correcting previous ones. A crude thermoscope enabled scientists to improve on everyday sense impressions as of colder and hotter, yielding an ordinal scale. This enabled them to make further investigations and refinements, eventually concluding that water’s behavior was invariant enough that its freezing and boiling points could anchor the scale. They had a choice. They could have, and for a while some of them did, attempt to use other substances—such as alcohol—to anchor the temperature scale. The early ‘fixed points’ were only relatively fixed. It is not the case that water boils at the same temperature come what may. But the chosen points were relatively fixed, giving scientists a more stable platform to build on (Chang, 2004). The process of devising an adequate measurement device and a magnitude to be measured went hand in hand. They were informed both by what the community sought to do and by the opportunities and obstacles that the world presented. What resulted was a reflective equilibrium that balances, among other things, a magnitude, a measurement practice, a device, and an epistemic goal.
Things are even more complicated when proxies are involved. When it is impossible or impracticable to measure a variable of interest directly, investigators may resort to proxies: variables that correlate with the variables of interest. Climatologists cannot go back in time and measure earlier conditions directly. So they study such things as tree rings and ice cores, investigating preserved phenomena that, they believe, correlate with earlier climate conditions. A core sample of glacial ice, amassed over centuries of snowfalls, contains different isotopes of hydrogen and oxygen at different levels. Since these correlate with differences in the temperature when the snow originally fell, trends in the distribution of isotopes are indicative of changes in ambient temperature. Such proxies are only as good as the correlations they depend on. So the results of investigations that rely on proxies are only as objective as the procedures for establishing the correlations. And these in turn depend on the background theories of the communities devising and deploying the proxies.
The calibration of measuring instruments is similar. In some cases, to be sure, the process is simple, in theory at least. Two seemingly identical measuring devices should be tweaked until their readings are brought into accord. Since whatever the one measures, the other does too, and in exactly the same way, calibration is a matter of minor adjustments. But when it comes to the cross-calibration of instruments, things are different. A balance scale should yield (to a close approximation) the same weight for a given object as a spring scale. To bring this about requires considerable ingenuity, grounded in an understanding of how each of the devices is sensitive to the weight of an object.
Even the most mechanical of experimental results are not as impersonal as they might appear. A result may simply register on a measuring instrument. But for this to happen, someone had to devise an instrument capable of capturing such a result (see Chang, 2004). A scientist had to devise an experiment that would produce the result. This may have involved a good deal of configuring and calibrating apparatuses. Someone had to set the threshold of measurement, ensuring that the result is not statistical noise. None of these scientists works in epistemic isolation. All of their efforts must be justifiable to the community of inquiry, hence must satisfy community standards. Otherwise there would be no reason to think that the read-out registers anything, that the course of events that occurred in and around the apparatus constituted a test, that the result is worth taking seriously.
Once we have the raw data, we are apt to subject it to statistical analysis. Statistics is the science that controls for chance. It appeals to infinite populations to bracket potentially misleading peculiarities in actual (finite) distributions. It ignores outliers on the grounds that they are not representative of the phenomena of interest. It sets standards for significance and power to rein in the dangers posed by false positives and false negatives. It develops a variety of regressions to disclose subtle patterns in the data.
The impersonality of mechanical objectivity is an achievement. It is the fruit of considerable epistemic labor concerning the phenomena, the proper ways to investigate it, and the available resources for conducting investigations. Methods have to be validated, measures defined, proxies identified, instruments invented and calibrated, thresholds set. Each of these steps is answerable to the standards of a community of inquiry. When and only when this is satisfactorily done can the mechanical procedure yield an objective result.
We should not think, however, that mechanical objective is a mere product of consensus. The world has to oblige. Only some items display the sort of invariance required for measuring devices. Only some materials can be manipulated so as to make instruments that can take and hold their calibration. Only in some areas are there suitable proxies available for variables we cannot access directly. And so on. The products of mechanical procedures are not ‘out there’ to be read off nature. Nor are they just products of agreement. They are fruits of careful deliberation and manipulation, grounded in an understanding of the object of inquiry and the available resources for investigating it. (See, among others, Cartwright, 1983; Hacking, 1983; Douglas, 2004.) This is a familiar point, but it is worth emphasizing to make clear that even the most mechanical or impersonal of objective procedures is not a mere matter of mirroring nature.
Impartiality
Impersonality is achieved by off-loading judgment to rules, standards, and techniques whose proper application ensures the acceptability of results. That requires setting limits on precision and on the number of dimensions and fineness of distinctions to be considered. This means, of course, that the community of inquiry had to exercise a good deal of judgment in setting the standards, formulating the rules, and devising the techniques that produced the results. There remain areas where we are unwilling or unable to off-load judgment completely. We want or need to preserve the possibility of more complicated, fine-grained, nuanced descriptions and assessments than available mechanical procedures afford. At the same time, though, we do not want to open the floodgates to idiosyncrasy, bias, and chance.
We thus devise methods that are impartial but not impersonal. When a method is impartial but not impersonal, people—qualified people—apply it and generate results. What makes the method impartial is that it does not matter who in particular those qualified people are. Such methods ground their findings in interrater agreement, where the criterion for acceptability is agreement among members of the community rather than answering to some external standard.
An example of an impartial procedure is the judging of competitive diving. Each dive is to be assessed along multiple dimensions that are spelled out in advance. In each of these dimensions a dive can be done well or badly. And what it is to do well or badly in a given dimension is also spelled out in advance. These specifications remove a good deal of room for idiosyncrasy and bias. The judge is not permitted to ignore the extension of the diver’s legs. Nor is she permitted to give or take away points for the grace with which the diver climbs the ladder. But further, more nuanced assessments need to be made within the domains set out by the explicit standards. For example, the rules specify that entry to the water should be vertical, with the diver’s body straight and her toes pointed. When these conditions are not met, judges are instructed to deduct points. How many points is left to each judge’s discretion. There the judge is instructed to use her own judgment (USA Diving, 2013: 14). That discretion is needed is no surprise. Individual judges have to decide just how far from the specified ideal a dive was in each of several respects. Although the rules tell them what dimensions to attend to, they cannot and do not try to tell judges exactly how to assess each divergence from the ideal.
The judge is a trained judge. She is not someone randomly recruited off the street. The way she was trained involves having her initial verdicts calibrated against those of others who are already expert judges.3 The newly appointed diving judge acquires the tacit knowledge that guides more experienced judges. This is not the end of the story, though. A diving competition has multiple judges—typically seven or nine—and to obtain the overall rating for a dive, the officials drop the highest and lowest scores and average the rest. So a judge’s personal assessment has a role to play, but it is not an overriding role. If her assessment is out of sync with the assessments of her peers, it will simply be dropped.
Whether the results of diving competitions are objective may seem epistemically insignificant. But what goes on in judging such competitions is similar to what goes on in scientific coding—the process of classifying some observational outcomes as alike, others as unlike. This is often the first step in assessing evidence. Suppose an investigator wants to know whether infants recognize their fathers’ voices. A plausible experiment would expose infants to a variety of men’s voices, filming and coding their reactions. Multiple trained investigators would look at the films and code what they observe. A critical issue is to decide what sort of reaction indicates recognition. There will presumably be clear cases—ones where any trained observer would recognize that the infant plainly did or plainly did not react differentially to her father’s voice. For such cases, interrater agreement and test–retest agreement are easily achieved. But there are apt to be borderline cases. Should the baby’s turning slightly in the direction of her father’s voice count as a positive response or not? This is a judgment call in much the same way that the diving judge’s assessment of whether the diver’s toes were sufficiently pointed is a judgment call.
As in diving, what is important is that all assessors use the same standard. It would muddle the experiment if some assessors treated the borderline cases as positive and others treated the same cases as negative. Coders have to calibrate their judgments against each other. Everyone coding the same data should deliver the same verdicts on at least most of the cases. And a single coder should deliver the same verdicts when she looks at the film a second time. The procedure for classifying cases is impartial. Although epistemic agents are ineliminably involved in the assessment, we have devised methods to control for bias and idiosyncrasy.
The standards may take the form of rules. If so, assessors need criteria to determine how to tell whether the rules have been followed, and what margin of error there is for following the rules. They may also take the form of exemplars. Rather than an articulate or articulable rule, we may rely on cases that exemplify the features we seek to match. The novice diving judge and the novice infant-behavior coder might be better served by being shown clear cases. They would see what a perfect dive or a recognitional response is supposed to look like. They might be presented with foil exemplars to see what should count as a definite mismatch. This would enable them to figure out how far an instance can deviate from the exemplar without being defective. (Arguably, this is what Kuhn’s normal science exemplars do [Kuhn, 1970]. They show rather than say what is to count as acceptable normal science.)
The Value of Objectivity
Objectivity as I have construed it depends on the availability and the applicability of public standards of assessment. Despite my talk of magnitudes, measures, and the like, objectivity allows for loose and flexible standards. That students are to turn their papers in on time is an objective requirement, and one that can be objectively measured, even though ‘on time’ is less than precise. Whether half an hour late counts as being on time may be undecidable without stipulation. But three weeks late definitely does not satisfy the standard. Moreover, the standard may have an implicit ceteris paribus clause that allows for exceptions in the case of unforeseeable emergencies. If so, it would be an impartial but not an impersonal standard. But if there was widespread agreement about what did and did not count as an unforeseeable emergency, objectivity would not be compromised. We could, of course, be more precise if we wished. How precise should we be: on time to the minute? the second? the nanosecond? This way madness lies. For the more precise standards would be harder to apply but no more useful. In setting objective standards, we need to consider what they are supposed to accomplish. Insisting that they be objective goes but a little way toward answering that question.
Following van Fraassen, I have argued that epistemically useful standards need to be answerable to evidence. We should expand this slightly to allow for answerability to reasons more generally. This is necessary if we want a conception of objectivity that comprehends mathematics as well as empirical inquiry. The satisfaction of procedurally objective standards cannot purport to reflect the absolute conception of reality. They are always products of a view from somewhere. But they are products that stand up to scrutiny when viewed from somewhere else, so they are not purely subjective.
It might seem, then, that we should conclude that objective procedures are designed to secure consensus in a community of inquiry. Since everyone in the community has the requisite level of expertise, it might seem, they will all agree on what the result of an objective procedure is and what it portends. This is plausible, but even for science, it is not quite right. As we saw in chapter 5, science has reason to promote a measure of disagreement (Kitcher, 1990). Within limits, some scientists can responsibly accept a risky new theory while others are more conservative and stick to the old one. Objective procedures leave room for responsible agents to disagree. Rather than fostering agreement per se, I suggest, procedural objectivity constitutes a venue in which both agreement and disagreement are possible.
Notes
1. This chapter was made possible through the support of a grant from the Varieties of Understanding Project at Fordham University and the John Templeton Foundation. The opinions expressed here are those of the author and do not necessarily reflect the views of the Varieties of Understanding Project or the John Templeton Foundation.
2. I say ‘purportedly’ and ‘putatively’ because some of the pictures have religious subjects. I am not remotely qualified to say whether heaven, if it exists, is three dimensional, or whether angels, if they exist, are visible. But it is plain that the artists represented heaven as a three-dimensional physical space, and angels as visible in such a space.
3. Initially, I suppose, the several judges negotiated until their assessments converged.