Abstract

In this paper I analyze and discuss Collins’ Experimenters’ Regress (Collins 1992) and offer an alternative explanation of how to break out of it. The strategy of the paper goes as follows. I will first show that two different, albeit related epistemic problems are confused in the Experimenters’ Regress: (i) the replication regress that consists in the occurrence of an infinite regress when judging whether or not a proper replication of an experiment has been carried out, and (ii) general reciprocity, according to which the determination of the proper functioning of an experiment and the correctness of an experimental outcome are determined reciprocally. I claim that: (1) the soundness of the replication regress requires the soundness of the general reciprocity argument, so by showing the unsoundness of the second we also show the unsoundness of the first. (2) Reciprocity is not problematic on its own; what is problematic is Collins’ explanation of how it is overcome, i.e., his claim that non-scientific criteria are required in order to break the circularity. Finally, I will show that (3) there is another possible explanation of the way out from general reciprocity, one that is extra-experimental but intra-scientific. In order to provide this alternative account, an analysis of how experimental outputs acquire representational content is required. The paper thus offers such an account, the semantics of experimental results.

1. Introduction

Despite the fact that reproduction of experiments by peers has traditionally been regarded as of the utmost importance in enabling the intersubjectivity of scientific practice, reproductions may yield discordant results and deciding which result should be favored may not be an easy task. According to Harry Collins (1992), experimental disagreement is resolved by the action of social, political and economic factors, but not by means of epistemic and scientific, or so-called internal reasons. His motivation for such a claim is the presence of an infinite regress at the core of experimental activity that, according to him, cannot be stopped with scientific resources alone: the experimenters’ regress. The goal of this paper is to offer an alternative to Collins’ account.

Let us begin by considering the possible situations that can arise when reproducing an experiment. Given an experiment designed to test whether x is the case, and two research groups, there are eight possible scenarios: if x is the case, either both research groups got the result right, or both got the result wrong, or one of them got it right and the other got it wrong. The same possibilities also appear if x is not the case, of course. Table 1 illustrates the possible reproduction scenarios. Possibilities 1, 4, 5, and 8 represent instances of confirmation (insofar as Team 2 confirms Team 1’s findings) and hence, of agreement between the researchers; but, as we can see, it is possible to confirm an experimental finding only to later find out, for example, that an error occurred in both experimental setups. Those situations (represented by cases 4 and 5) in which we can be justified in believing false empirical claims, are testimony to the fallibility of scientific knowledge (and, particularly, to the fallibility of experimental practice). The problem of induction is also represented in the table. Consider, for example, cases 1 and 8. Despite the fact that, on the one hand, they confirm what happens to be the case and also confirm each other’s results, the legitimacy of projecting those findings is, to say the least, problematic. I will not, however, dwell on these time-honored problems here. Instead, I will focus on a more restricted problem. The problem I will address can be summarized by the following questions: How are situations like 2, 3, 6, and 7 resolved? Which elements help researchers to overcome disagreement or, if consensus among researchers is not reached, what helps the scientific community to decide which experiments yielded the correct result?

Table 1. 

Alternative scenarios when reproducing an experiment.1

Is it the case that x?Team 1Team 2
1 x is the case x/right x/right 
2 x is the case x/right −x/wrong 
3 x is the case −x/wrong x/right 
4 x is the case −x/wrong −x/wrong 
5 x is not the case x/wrong x/wrong 
6 x is not the case x/wrong −x/right 
7 x is not the case −x/right x/wrong 
8 x is not the case −x/right −x/right 
Is it the case that x?Team 1Team 2
1 x is the case x/right x/right 
2 x is the case x/right −x/wrong 
3 x is the case −x/wrong x/right 
4 x is the case −x/wrong −x/wrong 
5 x is not the case x/wrong x/wrong 
6 x is not the case x/wrong −x/right 
7 x is not the case −x/right x/wrong 
8 x is not the case −x/right −x/right 

Not only are there several possible scenarios when reproducing an experiment but there are various ways of reproduction. In fact, the reproduction of an experiment can be performed in several ways: 1) Repeating a procedure with the original device (I will call it I-repetition, where I stands for instrument, which remains constant in this kind of reproduction). 2) By performing a replication: which means creating a carbon copy of the original arrangement in which the experimental design and the theoretical presuppositions remain constant. 3) By performing a T-repetition, that is to say, developing a more sensitive version, so that the new experiment shares all the theoretical presuppositions with the original but varies some of the features of the experimental design. 4) By performing an independent test, which consists of devising an experiment that relies on independent theoretical presuppositions and that operates on the basis of different physical processes.2

However, according to Collins (1992), these different reproduction modalities diverge with respect to their testing power. This divergence constrains, he argues, the epistemic legitimacy of the reproduction strategy chosen in a given scenario (Collins 1992, pp. 35–6).3 Confirming an experimental result, he claims, requires independent testing; disconfirming it requires replication. Given this, let us now suppose that a research group reproduced an experiment but failed to confirm the original findings (they are in a scenario of either type 2, 3, 6, or 7). Then, according to Collins, for the results to count as a legitimate test of the original experiment, the research team should have performed a replication. Collins’ strategy of argumentation consists, on the one hand, in showing that the only epistemically admissible reproduction method for judging the correctness of an experimental result is replication; on the other, he points out that differences with respect to the results obtained can be attributed to an unsuccessful replication rather than to having obtained an incorrect result. The consequence is that checking the adequacy of the replication would lead to an infinite regress: the experimenters’ regress. Since, however, it is an empirical fact that science has a way out of this regress, an explanation of how it is achieved is required. The explanation that Collins provides relies on the role of external factors in reaching consensus and settling the controversy.4

My challenge is to offer an alternative and epistemic explanation of how disputes regarding discordant results may be overcome.5 In order to do so, I will introduce Collins’ stance on the debate, presenting and discussing the experimenters’ regress, the main reason why he advocates for an external resolution of disagreements in experimental activity. Afterwards, I will make explicit the two problems that are conflated under the experimenters’ regress and the relations between them. The first is peculiar to experimental practice, and I will call it the replication regress. The second is a more general problem, and it is not specific to experimental practice, but to establishing any empirical claim whatsoever. I will call it general reciprocity.6 I will then examine three episodes in the history of science and propose a semantics for experimental results, which is to say, for the way in which an experimental outcome acquires representational content. I will finally apply this analysis to offer an internal explanation of how general reciprocity is overcome, the most dramatic way in which an experiment may regress.

2. Exposing the Two Different Problems Hidden behind Collins’ Experimenters’ Regress

In his book, Changing Order, Collins introduced the experimenters’ regress in the following way:

This is a paradox which arises for those who want to use replication as a test for the truth of scientific knowledge claims. The problem is that, since experimentation is a matter of skilful practice, it can never be clear whether a second experiment has been done sufficiently well to count as a check on the results of a first. Some further test is needed to test the quality of the experiment, and so forth. (Collins 1992, p. 2)

In a later chapter he provided what he considered to be an alternative and equivalent characterization of the regress:

What the correct outcome is depends upon whether there are gravity waves hitting the Earth in detectable fluxes. To find this out we must build a good gravity wave detector and have a look. But we won’t know if we have built a good detector until we have tried and obtained the correct outcome! But we don’t know what the correct outcome is until…and so on ad infinitum. (Collins 1992, p. 84)

Given the occurrence of the experimenters’ regress in experimental activity, Collins claims that it would be normal to find scientists arguing endlessly against each other about the quality of their findings. But, at the same time, from the fact that scientists eventually reach agreement about which result is correct and which experiment was properly performed, he concludes that scientific resources alone are not sufficient to establish the correctness of an experimental result, and that external factors are required to break the circle. He declares: “There is then, no set of ‘scientific’ criteria which can establish the validity of findings in the field. The experimenters’ regress leads scientists to reach for other criteria of quality” (Collins 1992, p. 88). I understand that each of the characterizations of the regress single out different problems. I take the first to represent what I will call the replication regress, while the second stands for what I will call general reciprocity. Subsections 2.1 and 2.2 are devoted to analyzing them and showing the relations between them. Sub-section 2.3 connects the critical with the constructive parts of this paper.

2.1. Examining the Replication Regress (RR)

Let us begin by disclosing and articulating the premises that lead to the regress.

  • (1) 

    Reproducibility thesis: in order to be scientifically relevant, an experimental result has to be robust. In order to find if this is the case, several procedures are available, amongst them, Collins considers the following:

    • a-I-Repeating the experiment.

    • b-Replicating the experiment.

    • c-Independently testing the original results.7

  • (2) 

    Epistemic inequality of the reproduction procedures thesis: the alternatives introduced in (1) are not equivalent with respect to the degree of confirmation or disconfirmation of the original findings.

    • a-I-repetition do not increase the degree of confirmation of the original result. (Collins 1992, p. 34)

    • b-Replication is problematic when the possible sources of error that could cause a false positive or negative are unknown. (Collins 1992, p. 35)

    • c-While independent testing could be a way to confirm a result, it is not a legitimate way of disconfirming one, since it could omit relevant elements that led to the result. (Collins 1992, p. 36)

  • (3) 

    Goal-oriented choice of the reproduction method thesis: if not every testing method provides the same degree of confirmation, then it is necessary to decide which method to use considering the purpose of the reproduction. If the reproduction is aimed at confirming the original finding, then an independent test should be performed. If it is aimed at disconfirming the original finding, then a replication is required (Collins 1992, p. 34).

  • (4) 

    From (1), (2), and (3): disconfirmation of an experimental result requires replicating the original experiment (Collins 1992, p. 36).

  • (5) 

    Impact of tacit knowledge and invisibility of its transference thesis: replication requires tacit knowledge transference (Collins 1992, pp. 73–4).

  • (6) 

    Experimental device soundness and experimenter’s expertise criteria thesis: the evaluation of the proper functioning of the experimental device and the experimenter’s expertise are determined by their ability to produce the correct experimental result (Collins 1992, p. 75)

  • (7) 

    Reciprocal determination of the proper functioning of the experiment and of the correct experimental result thesis: The only way to determine whether an experiment is being properly performed is by obtaining a correct result; but what the correct result is can only be known if the experiment is properly performed (Collins 1992, p. 147).

The conclusion Collins extracts from the above premises is the following: (CRR) Replication, which is the only legitimate way in which to disconfirm an experimental result, cannot settle a disagreement. But, as a matter of fact, even in the face of disagreement, scientists reach consensus (scenarios like 2, 3, 6, or 7 listed in table 1 are bound to be settled.) Therefore, if there were a regress, it would be necessary to explain how it is overcome. The explanation that Collins provides is that the way out from the experimenters’ regress is achieved by means of applying non-scientific strategies. He claims: “Some non-scientific tactics must be employed because the resources of the experiment alone are insufficient” (Collins 1992, p. 143).

However, we are not forced to accept such an explanation unless we presuppose that to scientifically resolve an experimental disagreement is to experimentally resolve it; an assumption that is playing a crucial role in Collins’s proposal. Let us assume that premises (1), (2), (3), (4), (5), (6), and (7) of Collins’ argument are true. The only thing we can conclude is that it is not always possible to determine the correctness of an experimental result by means of replication. Given this, the way out from the infinite regress has to be explained by appealing to an extra-experimental criterion. But for Collins, as we have seen, extra-experimental equals external, contingent, social, political, economic, etc. This would be the case if all scientific criteria were experimental, which is clearly not the case.

Although I believe that several of the theses of the RR are open to debate, in this paper I will only examine the truth-value of premise (7). If we can show its falsity then it will became clear that an extra-experimental criterion can still be a scientific and epistemic criterion. While (7) is necessary to run the replication regress, it is also the core assumption of general reciprocity (GR). Therefore, the denial of (7) requires the analysis of GR, to which next subsection is devoted.

2.2. Examining General Reciprocity (GR)

As I have claimed, given that Collins presents the experimenters’ regress alternatively as RR or as GR, it is probable that he considers them to be equivalent. However, the RR contains a premise that is the core of the GR: (7) the reciprocal determination of the proper functioning of the experiment and of the correct experimental result. It is quite curious that Collins does not seem to acknowledge that the scope of this thesis is far more general than the argument in which it appears. As it will become clear from our reconstruction, GR does not stem from the practical and conceptual difficulties that arise in attempts to determine whether or not a replication was successful; in fact, it is a more general problem. If (7) were true, Collins would be able to show that, not only replication, but any reproduction modality is problematic. Contrary to Collins, I believe that the problem of determining which result is the correct result is not an exclusive problem of the replication of experiments,8 but it is a problem that haunts any form of reproduction when no particular result is anticipated. (7) is the conclusion of what I labeled general reciprocity, GR, which can be cashed out as a reciprocal argument:

  • (1) 

    Experimental device soundness criterion: This is one of the conjuncts that constitutes (6) in the previous argument. According to this criterion, we accept that a good x-detector is one that indicates that x is the case if x is indeed the case and that x is not the case if x is not the case. I do not think there is anything to criticize in this thesis.

  • (2) 

    Experimental introduction of the result: With this thesis it is claimed that the determination of x is achieved by means of a good x-detector. According to this thesis, the only way to know if x is or is not the case is by means of our detection instruments. To grasp correctly the scope of this premise is of the utmost importance since it will be the target of our critique. Although this thesis is central to Collins’ work, he does not provide us with any compelling reason to accept it.

He claims:

“But what is the correct outcome? To find this out we must build a good detector and have a look.” (Collins 1992, p. 84)

Collins should have justified (2). It is quite problematic to assert that the only criterion to determine whether the reference of a theoretical entity exists is the outcome of an experiment. In saying so, he seems to assume that in an experiment the reference of a theoretical entity can be detected directly. I will claim that this premise rests on a mistaken understanding of what an experimental result is. It assumes that experimental results lack a theoretical component. If this assumption were false, then there would be a reasonable and scientific way out of the reciprocity. The conclusion of this argument would be: (CGR) The proper functioning of the experimental device and the correct result of an experiment are reciprocally determined.

CGR is, as the reader probably noticed, premise (7) of the previous argument, RR. (7) is derived from (1) and (2) in GR and for it to be a valid and sound argument, (2) should be true. But is it? The reciprocity this argument portrays is a problem with which any foundationalist stance regarding the justification of empirical knowledge has to deal. Accordingly, the denial of (2) requires the adoption of a minimal coherentist approach. Therefore, I will claim that an experimental result is a complex entity that is not introduced in a purely experimental way, but possesses theoretical content. If that is the case, the reciprocity would not take place, because there is an epistemic criterion, independent from the judgment regarding the proper functioning of the experimental device, that helps us to determine what the correct result is or at least to narrow the set of acceptable results given the accepted scientific knowledge available. For example, if we consider the gravity wave detection episode―which is Collins’ favorite example to argue for the external closure of the experimenters’ regress―Weber’s findings were highly improbable in the context of existing physical and cosmological knowledge and, moreover, had the results been correct, they would have been accompanied by observable effects which were not in fact observed; these, when reconsidered under the light of several negative results, made it quite reasonable to consider that Weber’s experimental results were incorrect (Levine 2004). In other words, the reciprocity was broken with the aid of theoretical considerations.9

2.3. The Plan: How to Show that General Reciprocity Can Be Overcome by Scientific Resources

So far, I have explained the two senses of the experimenters’ regress and the relation between these two problems. The RR confronted us with the problem of judging whether a replication had been performed correctly. To this challenge, Collins replied by invoking reasons external to science itself. I suggested, on the contrary, that it was possible to deny the truth of premise (7), which postulates that the determination of the proper functioning of an experimental arrangement is reciprocal with determining the correctness of an experimental result. In section 2.2 I explicated the relation between (7) and the GR and claimed both that the GR has a far more general scope than RR and that its conclusion is a premise of the RR. The next step is to show why (7) is false, and I will do so by denying that the introduction of the correct result is necessarily experimental, hence rejecting premise (2) in my reconstruction of GR. With this aim in mind, the next section will present and analyze three paradigmatic experiments that will serve as the empirical basis for studying the content of an experimental result. After this presentation, in section four, I will be in a position to show how the representational content of experimental results is acquired, and how the semantics of experimental results supports the view that in conflictive situations, an experimental result may be theoretically introduced. If that is the case, then there is a way out from GR that does not require non-epistemic explanations.

3. The Empirical Basis

In what follows I will briefly describe three kinds of experiments in physics, which differ regarding the ontological status of what they purport to detect. I will call them quantitative, qualitative, and existential experiments.10 A brief characterization of each of them will precede an analysis of a paradigmatic case. This section presents the cases studies needed for understanding how to bridge the gap between what is in fact perceived as the outcome of the material realization of an experiment11 and what is claimed to be its result.

3.1. A Quantitative Experiment: Michelson and the Speed of Light

I will consider as quantitative experiments such as Michelson's measurements of the speed of light and Cavendish's determination of the gravitational constant. All these experiments assume the existence of a kind of entity, a process-type or an event-type, and they also assume that the entity, etc. under investigation possesses a certain property. The goal of these experiments is to determine precisely a magnitude for a quantity that is already introduced by a theory. They yield a magnitude—with an associated error—as the final result of the experiment.

I will briefly portray here one of the first experiments with which Albert Michelson measured the velocity of light (Michelson 1880). During 1877 Michelson found a way to improve Foucault's revolving mirror device so as to provide a more accurate measurement of the velocity of light. As is well known, Kepler's optical investigations show that the intensity of a light source decreases with the square of the distance, so the further the distance that light travels through the experimental arrangement, the less distinct the output is. But the shorter the distance traveled, the harder it is to measure the output. Michelson avoided both problems by using a spherical lens of great focal length (L), placing the revolving mirror (R) within the principal focus of L and replacing the original spherical fixed mirror with a plane one (M). See figure 1 for a schema of the experimental design he proposed.

Figure 1. 

A schema of the experimental arrangement. Taken from Michelson (1880).

Figure 1. 

A schema of the experimental arrangement. Taken from Michelson (1880).

Let me now describe the principles that govern the design of the experiment. Consider a ray of light traveling from the source (S)12 to the revolving mirror (R) through the lens (L). If R is at rest, the ray will form an image at M, and because of the law of reflection, it will return to R and finally to S. But if R were to rotate on its axis, a new light spot, deflected in the direction of rotation of R, will be formed. Figure 2 exemplifies how R’s change in position can cause a second bright spot on S.

Figure 2. 

Θ, the deflection angle.

Figure 2. 

Θ, the deflection angle.

S′R S, (from now on, θ) is the deflection angle. It is subtended by the rays of light originating in the revolving mirror, R. θ is half the angle through which the mirror has turned since the departure of the ray of light from S to M and finally back to R. Calculating the value of θ is crucial for this experiment. For if we measure the distance between S and S′ and we measure the length of the segment R-S we can then calculate the tangent of θ and its inverse function, which gives us the value of θ in radians. Once this variable is known, together with the number of revolutions per second that R performed,13 if velocity is the ratio between distance and time, and if the distance is 2RM and time is represented by: (θ/2)/n.360,14 then we have that:
formula

So, we can calculate the time that light took to travel the distance considered. Now that we have at least a rough idea of how the experiment works, we are able to single out different aspects of the process of producing the result. To begin: what is the output in this experimental arrangement? Figure 3 is a magnification of the output. It is doubtful that this image, on its own, can tell us anything regarding the nature of light and its celerity. It is only when it is interpreted as something else that it can be informative, and this requires theoretical interpretation. Michelson uses a micrometer in order to measure the distance between the bright spots. In doing so he is no longer concerned with the dots themselves, but with a specific relationship between them: distance. In this example, the distance is the salient feature to consider because it will allow the researcher to calculate the tangent of θ. And with this, it will be possible to relate the displacement of the mirror to the time required for the formation of the second image, S′. This is a first step in the output’s acquisition of representational content, in which it is laden with two theories: a measurement theory (which establishes the length of the distance between S and S′) and a branch of geometry: trigonometry (which entails that the segment S-S′ is the tangent of θ).

Figure 3. 

The output of the experiment. Taken from Michelson (1880).

Figure 3. 

The output of the experiment. Taken from Michelson (1880).

Later on, every magnitude for the measured length is introduced in an equation that relates the different variables in the experiment and that enables us to calculate the velocity of light between the two mirrors of the experimental arrangement. This requires a new interpretative step, this time provided by empirical, particularly, physical theories, such as kinematics. Afterwards, the data collected is reduced by means of the application of a statistical method. This reduction implies further theoretical interpretation. In the experiment just considered, Michelson studies the different sources of error and calculates the mean of the measurements and its standard deviation. He announces the final result to be: V = 299944 ± 51 km/s (Michelson 1880, p. 141).15

3.2. A Qualitative Experiment: Newton and the Composition of White Light

The aim of a qualitative experiment is to determine the properties of a previously detected entity-type. The purpose of these experiments is to gain knowledge about these entities by means of detecting the different properties that they may have. For instance, once the detection of neutrinos took place, efforts were devoted to discovering whether they oscillate or not. The experiment conducted in the Japanese observatory Superkamiokande is a good example.16 The prism experiment carried out by Newton, in order to analyze the composition of white light, also belongs to this category. The first experiment will show whether neutrinos oscillate or not; the prism experiment will allow us to state that white light is composed of light of different colors.17 The result of these experiments is the attribution of a new property to an entity or process.18 In what follows, I will analyze Newton’s experiment on the composition of white light as a paradigmatic example of this category.

Among his several optical studies, Newton devoted himself to providing an account of the phenomenon of color and of the nature of white light. Here I would like to briefly consider one of the experiments he presents in the letter he wrote to the Royal Society of London in 1671 according to which he demonstrated the composition of white light and the differential refraction of the simple rays that constitute it.19 There he claims:

Sir, to perform my late promise to you, I shall without further ceremony acquaint you, that in the beginning of year 1666, […], I procured me a triangular glass prism, to try therewith the celebrated phenomena of colours. And in order thereto having darkened my chamber, and made a small hole in my window shuts, to let in a convenient quantity of the Sun’s light, I placed my prism at his entrance, that it might be thereby refracted to the opposite wall. It was at first a very pleasing divertissement, to view the vivid and intense colors produced thereby; but after a while applying myself to consider them more circumspectly, I became surprised to see them in an oblong form; which, according to the received laws of Refraction, I expected should have been circular. (Newton 1671, pp. 3075–76)

Notice in figure 4 the schema he offers. It should be read from right to left. It shows how a ray of sunlight passes through a prism in a minimum deviation position. The image projected in the opposite wall is oblong.

Figure 4. 

A representation of the first refraction. Taken from Newton (1704).

Figure 4. 

A representation of the first refraction. Taken from Newton (1704).

Why did Newton express surprise? Because according to the received laws of refraction, the angle of refraction of a ray of light depends only on the angle of incidence and on the variation of the refraction index of the media. Therefore, if the different rays of light pass through the same medium, the prism, and if the medium is isotropic, then they should be equally refracted, and hence, they should produce a circular image on the opposite wall. If we follow Fig. 4 from right to left, then we will notice that there is a refraction that corresponds to the surprising oblong image that Newton reported in the letter I quoted.20

After checking that the oblong image could not be an artifact, he considered a second refraction to understand what these images suggested regarding the nature of light. By slightly rotating the prism on its horizontal axis, Newton was able to selectively project, on a second panel, regions of the spectrum formed on the first panel, and to study the behavior of the rays when undergoing a second refraction. The conclusion that Newton reaches is that the rays that undergo the most extreme deviation during the first refraction are those that also experience the most extreme deviation during the second. These rays can be individuated by means of their color, and, when separated, their images are circular, as was expected. Newton would show, then, that white light is a compound of rays of different colors that manifest a particular refrangibility.

Figure 5 displays a representation of the experimentum crucis’ experimental arrangement that can be helpful to keep in mind. Again, it should be read from right to left. It represents how a ray of sunlight passes through an orifice and undergoes a prism induced refraction. The refracted ray of light casts an image in the wall DGE that, as Newton states, is oblong, instead of circular. A portion of the refracted ray of light (individuated by its color) passes through two more panels before undergoing a further refraction. Rotating the first prism allows Newton to select which part of the spectrum will undergo a second refraction. He notices that those parts of the spectrum that were most deviated during the first refraction are those that are most deviated in the second one. This filtered ray of light does indeed produce a circular image in the third panel as expected.

Figure 5. 

An Experimentum crucis’ schema. Taken from Newton (1704).

Figure 5. 

An Experimentum crucis’ schema. Taken from Newton (1704).

Let me now consider the different interpretative steps involved in this experiment. To begin with, we can find two images that will constitute the output of the experiment: the multicolored oblong spectrum on the first panel and the monochromatic circular image cast on the third panel. A comparison of the shape, the position, and the color of each image seem to be what is required, in this experiment, for conceptualizing the output. Again, the relevance of these features arises from the theoretical background assumed in the experiment. For instance, for conceptualizing the output, geometrical optics is presupposed. As I said before, according to Snell’s laws, given that the rays of light under examination go through the same medium, they should be refracted with the same angle, if the prism is in its minimum deviation position.

In this experiment, in contrast to the previous case I presented, there isn’t any statistical analysis, but an extra experiment to determine whether the oblong image could be an artifact. It consisted in making the ray undergo a second refraction through a prism in the inverse position to see if an image of the original source of light could be obtained. Finally, in order to explain the different angle with which each ray of light refracts on the second panel, Newton introduces the property of differential refrangibility, a dispositional property of the simple rays that constitute white light.

3.3. An Existential Experiment: Weber and the Detection of Gravity Waves

Existential experiments are conducted to find out whether an entity-type exists, or whether a process-type or an event-type takes place or does not take place. They usually involve searching for the referent of a concept introduced by a theory. Examples of this kind of experiment include Weber’s attempt to detect gravity waves, predicted by general theory of relativity, and Reines and Cowan’s attempt to detect Neutrinos, those ghostly particles that Pauli posited in 1930 to account for the energy missing during beta decay, and hence, to preserve the principle of conservation of energy. The results of this type of experiment can be stated as an affirmative or negative existential statement about a class of entities or processes. The kind of result expected consists either in 1) a divergence from the null hypothesis to a certain standard deviation if the entity under search exists, or 2) a confirmation of the null hypothesis if the entity does not exist. As an example, I will consider Weber’s gravity wave detection episode, which happens to be Collins’ favorite example in support of the external resolution of the experimenters’ regress: the first attempt to detect gravity waves.

It is accepted that if gravitational radiation exists it would have several sources. Taking into account the properties of the sources, we could know what kind of radiation to expect and at what frequency we could detect it. Sources could be either discrete or continuous: nova and supernova explosions, the creation of black holes, and the collisions between black holes give rise to the first kind of discrete radiation. Binary pulsars, neutron stars and the creation of space-time itself are continuous sources of radiation. Determining the radiation source is of the utmost importance in building an adequate antenna that is suitably tuned to detecting the range of frequencies that are expected and to properly analyzing and modeling the data obtained (Davies 1980, p. 96).

Joseph Weber, who was not only a physicist but also an engineer, designed one of the first gravity wave antennas. In fact, today, this kind of detector is still known as the “Weber antenna.”21 It is not a very sophisticated apparatus. In fact, it is a compact aluminum cylinder (an aluminum bar), 1.53 m long, 0.66 m in diameter and weighing 1.4 tons. The principle behind its functioning is that it will behave as a harmonic oscillator that will respond to frequencies close to 1660 hertz that would allow the detection of the emission of gravitational radiation from the collapse of a Supernova (Weber 1969, p. 1320). This resonant mass antenna is linked to transducers that transform the oscillations of the aluminum bar into electrical impulses that in turn have to be amplified and recorded.

In Weber’s experiment, gravity waves are to be detected by means of a Newtonian kind of apparatus: a harmonic oscillator. When a gravity wave impacts on the detector its energy is absorbed by the antenna and converted into sound waves. Since the oscillation of the bar will be transduced by means of the piezoelectric components attached to the apparatus, the outcome of the experiment will be a voltage. Despite the fact that gravity waves are supposed to carry enormous amounts of energy, the signal that they cause in an antenna is quite weak, and consequently, the antenna has to be very sensitive. But, in turn, and as a result of such sensitivity, the antenna will receive lots of extra signals that raise enormously the chances of getting a false positive. Hence, a threshold has to be established, and several noise reduction techniques have to be taken into account. In this experiment, the problem of noise was taken into consideration and vibrations that would stimulate the oscillator were detected independently. For instance, Weber used several independent detectors to record seismic, acoustic, and electromagnetic inputs. The antenna was also kept in a vacuum chamber at a very low temperature, so as to reduce, as far as possible, thermal noise. However, this last source of noise could not be avoided completely, since the antenna could not be maintained at 0° Kelvin, which is the temperature at which molecular movement ceases. The data was also modeled by means of applying a Fourier transform to the voltages obtained.

Only after these precautions were taken, and only after comparing the outputs of two resonant bars located at different places and taking into account as data only the coincident marks between both detectors,22 did Weber announce his results, in a communication submitted to Physical Review Letters (Weber 1969). According to his records, there were seven coincident pulses per day that could not be a result of anything other than gravity waves from a frequency close to 1660 hz. The rest of this interesting story is, unfortunately, outside the scope of the paper. All I shall say here is that these results, while not incompatible with the then current cosmological and astrophysical knowledge, were highly improbable. For this reason, they were received with skepticism within the physics community (Levine 2004).23

To sum up: gravity waves are measured by one of the consequences of their impact on a resonant bar. The impact of a gravity wave on the antenna modifies the mode of vibration of the bar and this can be transduced―by means of piezoelectric crystals―into a voltage. So what is actually being measured in this experiment is the variation of voltage over time. Second, the voltages recorded are analyzed and modeled so as to take into account only those that cannot be due to any of the other causes of the vibration of the antenna. This, in turn, is converted into a variation of amplitudes and to make this conversion a signal processor is needed. After that, statistical analysis is done to determine whether the deviation from what is expected if the null hypothesis is true is significant enough to permit rejection of the null hypothesis.

4. The Representational Content of Experimental Results and the Experimenters’ Regress

4.1. The Semantics of Experimental Results and How Experimental Results Gain Their Representational Content

Collins claimed that the determination of the correctness of an experimental result is coextensive with the determination of the proper functioning of the experimental device (premise (7) in RR and GR’s conclusion) and that breaking such reciprocity requires the appeal to external factors. But is this necessarily so? Can we find an alternative way of introducing the correct result (premise (2) in GR) that helps us to break the reciprocity without requiring an externalist explanation? I believe so. To put forward this alternative answer I will have to partially address what Marcel Weber (Weber 2012) refers to as the problem of the representational content of data. He claims:

[R]eliable data are correct representations of an underlying reality, whereas so-called artifacts are incorrect representations. This characterization assumes that data have some sort of representational content; they represent an object as instantiating some property or properties […]. The question of what it means for data to represent their object correctly has not been much discussed in this context. (Weber 2012, his emphasis)

I will offer an approach that may help us understand how an experimental result acquires its representational content. In order to do so I will now draw some general conclusions about how the representational content of experimental results is gained. I will do so by means of proposing a semantics of experimental results.

One of the central problems regarding the content of experimental results consists in determining the process through which raw data is transformed into information about the natural world, for example, how a click in a Geiger counter is linked to the flux of solar neutrinos. If we consider the output to be the observable outcome of the experimental apparatus, it is necessary to provide it with meaning. Indeed, the final product of an experimental run is a directly observable event, such as the movement of a needle in a voltmeter, the sound of a Geiger counter, or a line in a cloud chamber. However, output and final result seem to be quite different. Depending on the experiment, the position of the needle would allow us to claim that we have detected gravity waves, whereas the sounds emitted by the Geiger counter would, in turn, lead us to count solar neutrinos.24 This can be understood as the process by which outcomes acquire representational content, representing an object as instantiating a certain property or set of properties; this is what I call a semantics for experimental results. Outcomes, I claim, gain their representational content by means of a sequence of interpretative steps. I will now show how this is the case by drawing some general conclusions from the experiments presented in the previous section.

Recall, for instance, Michelson’s experiment. The output is an indirect indicator of the velocity of light insofar it is related to the laws and the theoretical assumptions that govern the experimental apparatus. Data are obtained when actually interpreting the relevant features of the output (the distance between the dots) under these laws and assumptions. Moreover, given that the measurement of a quantitative property that can be instantiated in different degrees is at stake, a suitable statistical analysis has to be applied to the data. The theoretical construct obtained after data reduction can be called an e-result, where the e stands for experimental. I take this to be, strictly speaking, the final contribution to the content of an experimental result that the experiment can provide. Finally, given that the attribution of a finite velocity to light is not compatible with every theory regarding its nature, the experimental knowledge obtained has to be subsumed under a theory that can accommodate the result. I call this last interpretative step external interpretation, and its product is what I will call a t-result. In this case, the t stands for theoretical. I take this to be the theoretical explanation of the e-result, and one of the layers of an experimental result that habilitates the question regarding whether or not it is a correct or an incorrect representation of the phenomena that the experiment aims at detecting.

More generally, the output produced in each material realization will undergo several interpretative steps. It will acquire part of its meaning when subsumed under a concept, whether classificatory or metrical, by means of a process that I will call internal interpretation. As a consequence, we will get data that will undergo statistical analysis or be subjected to the controls that are required for the kind of experiment in question. As a result, we will obtain what I call an e-result, that is the final result of the experiment, and, finally, this result will be subsumed under an explanatory theory.25 External interpretation can be thought of as an explanation, in the sense that the e-result of an experiment can be thought of as the experiment’s explanandum while the theories that contribute to the meaning of the t-result explain why we measured what we had measured or why we observed what we observed (i.e., they provide the explanans for the e-result).26

My proposal, therefore, is that experimental results comprise four elements: output, data, e-result and t-result, and that they are so formed by means of two interpretative procedures: internal and external. Each of these theoretical constructs can be revised and may obviously undergo a change in interpretation and therefore in meaning, given a change in theory.27Table 2 is a general schema that can be abstracted from the analysis of the experiments I have offered. While table 3 consists of an abridged presentation of the elements in each of the examples provided above.

Table 2. 

A model for the semantics of an experimental result

graphic
 
graphic
 
Table 3. 

The elements of the experimental results considered.

Velocity of LightDifferential RefrangibilityGravity Waves
Output Image of two bright dots. Polychrome oblong spectrum + monochrome round images + their positions. Traces in strip chart paper. 
Datum Length- Tangent of angle S′RS, relation between the variables in the experimental arrangement. Shape variation + color variation + same position equals same color. Voltage (or power variation as a function of time) 
E-result Average of the velocities measured. Control: the oblong image is not an artifact. Data modeling to filter noise. 7 signals per day over noise. 
T-result The velocity of light is 299944 ± 51 km/s Simple rays are differentially refrangible. Gravity waves have been detected. 
Velocity of LightDifferential RefrangibilityGravity Waves
Output Image of two bright dots. Polychrome oblong spectrum + monochrome round images + their positions. Traces in strip chart paper. 
Datum Length- Tangent of angle S′RS, relation between the variables in the experimental arrangement. Shape variation + color variation + same position equals same color. Voltage (or power variation as a function of time) 
E-result Average of the velocities measured. Control: the oblong image is not an artifact. Data modeling to filter noise. 7 signals per day over noise. 
T-result The velocity of light is 299944 ± 51 km/s Simple rays are differentially refrangible. Gravity waves have been detected. 

4.2. The Semantics of Experimental Results and GR

What is the relevance of the semantics of an experimental result when trying to offer an account of the way out from the GR? Let us recall the second premise of the reciprocal argument, which was also a necessary premise to run the infinite regress argument, i.e., the RR. It stated that in order to know what the correct experimental result was, the only information available was the one that a good experimental arrangement could provide us with. However, as I showed in the previous section, an experimental result is not merely the outcome of an experiment, but it possesses representational content. I take that Collins has to agree with this claim since he makes explicit reference to interpreted experimental results. Therefore, he has to concede that for an outcome to correctly represent a phenomenon it has to possess representational content. Acquiring representational content requires, as I showed in section 3, theoretically interpreting the outcome. Furthermore, he has to concede that general reciprocity occurs at the level of the t-results. If that is the case, then the correctness of the representational content of an experimental result cannot be judged merely by evaluating the correct functioning of the experimental arrangement, but by taking into account the adjustment between our theoretical expectations and the available relevant accepted knowledge with the e-result. Hence, even if we have to accept that in certain situations (especially in those where qualitative properties, attributions, or existential claims are at stake), establishing an experimental result is not done solely by the experiment, this does not mean that this is accomplished by extra-scientific criteria as Collins claims. It just means that experiments may require theoretical considerations so as to break the regress, which are precisely those considerations required for providing an external interpretation of the experimental result. This fits nicely with Kuhn’s approach towards the reliability of experimental techniques:

When measurement is insecure, one of the tests for reliability of existing instruments and manipulative techniques must inevitably be their ability to give results that compare favorably with existing theory. In some parts of natural science, the adequacy of experimental technique can be judged only in this way. When that occurs, one may not even speak of “insecure” instrumentation or technique, implying that this could be improved without recourse to an external theoretical standard. (Kuhn 1977, pp. 194–5).

Therefore, the determination of which is the correct result, especially in problematic cases, can be done by appealing to the coherence between one of the theories that would provide the external interpretation of the result, a theory that has to be compatible with current accepted knowledge.28 Consequently, we would be justified in claiming that the experiment is functioning properly if it provides outcomes that are compatible with accepted knowledge, in what can be considered a theoretical calibration of an experiment.29

At this point, the interested reader may well wonder why I appeal to a calibration of an experiment by a theory but do not highlight the relevance of empirical calibration when overcoming the GR. Let me explain why this is the case. Calibration is the use of a surrogate signal to standardize an instrument (Franklin 1999, p. 237). In order to perform a traditional calibration, the researchers must know the signal that the experimental arrangement purports to detect and they must know how to manipulate it. But this presupposes that general reciprocity is already broken! A different kind of calibration, typical for the sort of experiments that are more prone to disagreement (those that seek to detect a hitherto unobserved phenomenon), is one in which a device is calibrated against a surrogate signal that is presumably similar in relevant respects to the signal the experiment is aimed to detect. Nevertheless, even Franklin, for whom calibration is an epistemological strategy that can help us to validate experimental results, considers that in cases such as Weber’s attempts to detect gravity waves, calibration cannot be decisive (Franklin 2002, p. 64).

I do believe, nevertheless, that there is a special kind of calibration that may play an important role in determining when two instruments are replicas: I am referring to reciprocal calibration, which consists in using a surrogate signal (other than the signal that the arrangements purport to detect, but assumed to be similar in the relevant respects) to check if two experimental arrangements respond in the same way to it and therefore, to infer functional identity between them (see Zuppone 2010). However, I do not think that a coincidence in this respect is sufficient for claiming that the experiments are working properly. What a reciprocal calibration can show is that two experiments are offering the same output given the same input, but it does not suffice for showing that the arrangements are functioning properly and are yielding correct outputs. Identity of outputs (identity to a certain extent, of course) is perfectly compatible with two experimental arrangements malfunctioning. Experimental replicability is not a sufficient condition for establishing the correctness of an experimental result, for, as I have shown, general reciprocity is a broader phenomenon than the problem of replication.

Another worry that may arise is related to the general scope of my strategy. Is it helpful to overcoming the experimenters’ regress by internal means in any kind of experimental situation? For example, imagine an episode in which two experiments yielded discordant e-results yet both were compatible with accepted theories. In such a scenario it is not clear up to what extent the t-result may help to decide between both values in order to overcome a possible disagreement. Such a situation could arise in quantitative experiments, where determining the value of a parameter is at stake. Insofar as the value with which a certain parameter instantiates cannot be inferred from a theory or group of theories, it is true that the external interpretation in quantitative experiments do not constrain e-results as much as it does in qualitative or existential experiments. It will, however, restrict the general features that e-results can take. In the measurement of the speed of light, for example, the most the external interpretation can contribute to the determination of the e-result concerns surface and qualitative features of the property, such as its finite character and the constancy of the value. However, since I believe that the values of certain parameters are brute facts that have to be determined empirically, I do not think that I would like to endorse a proposal that does not allow this situation to be the case. One option in such situations is to appeal not to a theoretical calibration but to a triangulation with indirect determinations that can contribute to the calibration of the experimental techniques. This would enable a non-experimental, yet internal introduction of the experimental result, which would be compatible with the denial of premise 2 of the reconstruction of the general reciprocity that I presented in section 2.2.30

4.3. Assessing My Answer vis a vis Radder’s and Franklin’s

It may be convenient to separate my proposal from two main answers available in the literature: those of Hans Radder and of Allan Franklin. I will do so in the remainder of this section.31 In “Experimental Reproducibility and the Experimenters’ Regress” (1992), Hans Radder differentiated the two presentations of the regress and offered a solution to what I labelled the replication regress. However, he did not mention the differences between the logical structure of the arguments (an infinite regress and a reciprocal argument) or the relations between them (namely, that since GR is a premise of RR, its truth is required for the RR to run). We both claim that the proper working of an experiment has to be judged by means of theoretical knowledge. In this sense, my paper could be considered in line with Radder’s approach while aiming to provide a conceptually more detailed and empirically wider illustrated answer to the problem with which Collins is confronting us. Radder’s strategy involves two steps. He first offers an elucidation of the concept of reproducibility. He later shows how a phenomenon can be stabilized (“delocalized”) by means of different material realizations. However, Radder’s concept of replication is different from the concept used by both Collins and I. In Radder’s paper, replication means obtaining the same experimental outcome regardless of the material realization. For instance, he claims:

Next, remarkably enough, it is the very procedure of replication of a result q by means of different experimental processes that restricts its dependence on specific, local skills. A well-known example is the replication of experiments to test Avogadro’s hypothesis […]. This claim was tested in the early decades of this century by means of a large number of very different experimental replications, viz. through Brownian motion, alpha decay, X-ray diffraction, black body radiation, and electrochemical processes, among others. (Radder 1992, p. 70)

But the change of meaning of this central concept requires that Radder justifies the dismissal of one of Collins’ thesis, namely, premise (2) of RR. Notice that Collins would agree with Radder about the relevance of delocalization for confirming an experimental finding, but he would still disagree with him about the proper way of settling an experimental disagreement, i.e., in disconfirmation situations. In those circumstances, to replicate in Radder’s sense (which is much the same as to independently test, in my sense), would be to beg the question against Collins, who precisely denies that this is an epistemically legitimate strategy in disconfirmation situations.

Comparing my approach with that of Franklin’s, it is important to notice that my proposal focuses first in philosophically analyzing Collins’ arguments and, later on, in exploring case studies to develop a semantics of experimental results. Whereas, Franklin’s approach is fundamentally descriptive, in that with the aid of case studies he shows how Collins’ answer to the experimenter’s regress is not empirically adequate. His explanation of how the regress was overcome in the gravity wave episode shows that an internal reading is possible and the epistemological strategies he extracts from the scientific practice helps him to argue in favor of science being a rational enterprise insofar these strategies are deployed in experimental activity. And indeed I agree with Franklin. However, while his strategy blocks the sociological explanation of the way out of the experimenters’ regress, my proposal seeks to enable a conceptual discussion of Collins’ arguments. Also, as far as I can see, it is also immune to Giora Hon’s criticism, according to which Franklin’s epistemology of experiment is eclectic and ad hoc (Hon 2003). In fact, the analysis of the representational content of experimental results that I presented can shed some light on how to organize, in a principled way, Franklin’s epistemological strategies. But this is something that can only be suggested here. In any case, I take my approach and Franklin’s to be complementary.

5. Conclusion

In this paper I examined the experimenters’ regress and claimed that, contrary to what Collins believes, it conflates two related epistemic problems that could arise in experimental practice. I claimed that the experimenters’ regress does not arise because of the problems that replication pose for the experimenter, but because of the possibility of a circular (reciprocal) determination of the correct result and the proper functioning of the experimental apparatus. I claimed that the circularity is broken through means which are internal to scientific practice, since an experimental result is not merely the output of a material realization, but requires theory in order to be produced. If that is the case, general reciprocity can be overcome by means that are internal to scientific practice, without the need to appeal to any external explanation. This, of course, is compatible with a global circularity regarding empirical knowledge, but yet, I believe we have good reasons to accept that there is, in each particular experiment, a local answer to it.

Notes

1. 

Two remarks regarding the schema: first, I take the content displayed in the second column to be, strictu sensu, unknowable and unreachable. It represents an external point of view, or, as Putnam would say: a god’s eye view of reality. Second, the schema portrays a scenario in which the experimental question has a categorical answer such as x exists/ x does not exist; x has the property y/ x does not have the property y; x is effective for treating y/ x is not effective for treating y.

2. 

Let me illustrate these different reproduction strategies by considering some of the experiments that aimed to detect solar neutrinos. The Homestake mine experimental arrangement, when used on subsequent occasions, exemplifies an instance of I-repetition (1) Had there been an identical copy of his arrangement elsewhere, it would have counted as a replication (2) The fact that Gallex and Sage use a different chemical as a reactant, varying, therefore, the original experimental design, make these experiments qualify as T-repetitions, (3) according to my classification. Finally, the experiment performed at the Sundbury Neutrino Observatory is an instance of independent test (4), where the theoretical assumptions that guide the research and the physical processes that take place in the experimental setup are different. While, for example, Homestake experiment relied on the weak interaction of neutrinos, the Sundbury Neutrino observatory depended on the Cerenkov effect. For details concerning different experiments related to the solar neutrinos puzzle I recommend Allan Franklin’s book on the topic (Franklin 2004).

3. 

Unfortunately, Collins omits T-repetition from his analysis.

4. 

By external factors, Collins understands non-scientific reasons. The decision that one result is correct rather than another has to do with, according to him, the persuasive skills of the actors, their influence and renown in the scientific community, etc., but not with scientific reasons. (Collins 1992, cps. 2, 6).

5. 

There are several responses to Collins’ challenge that I will not discuss here. Larry Laudan (1982) points out the inconsistency between Collins’ empiricism and his dismissive attitude towards the role that empirical evidence plays in belief formation. Nancy Cartwright (1991) suggests distinguishing between two varieties of repeatability: (i) replication, which aims at showing that the phenomenon under study is stable and (ii) reproduction which tends to secure what she calls the theory of the instrument, guarding against errors in our instruments. Sylvia Culp (1995) offers a proposal that relies on the generation of robust bodies of data, achieved via independent testing. Godin and Gingras (2002) trace Collins’ challenge back to Ancient skepticism. Recently, David Teira (2012) offered an answer to the regress when using debiasing procedures in biomedical research.

6. 

To be more precise, this reciprocity concerns not only empirical knowledge but formal knowledge as well, logical reasoning being a case in point.

7. 

Since I am following Collins’ arguments, I will omit here what I called T-repetition, since it is a way of reproduction that he, himself, omits.

8. 

That he takes this to be the case becomes evident when considering the quotation I used to present the RR, in which Collins explicitly claims that the experimenters’ regress is a paradox that appears if replication is used to test an empirical claim.

9. 

In Changing Order, as well as in The Golem series, Collins presents and analyses several case studies. While the TEA laser case represents the good, purely empirical experimental situation in which there is a clear way of determining the proper functioning of the apparatus, the correct experimental outcome and the acquisition of the experimental relevant skills, the rest of the cases he deals with are such that they are subject to the experimenters’ regress. As such, in the case of disagreement, they would require a non-experimental resolution. However, Collins claims (personal communication) that he believes that some experimental disagreement can be solved by appealing to theoretical arguments. If this is the case, and if experimental disputes can be sometimes settled by theoretical considerations, then the scope of the regress thesis and the associated arguments should be restricted. In order to be sufficiently interesting, however, this scope should be quite wide, i.e. the regress thesis should not be restricted to very rare and exceptional cases. If this is so and his thesis aims to apply to enough interesting cases, then my criticism would still hold, for it claims that in most cases such a resolution of the regress does not apply. In any event, I deny that the external resolution of the regress applies to many cases in which Collins claims the opposite, such as, for example, the gravity wave detection episode discussed here and the case of Vitamin C as a cure for cancer that I discuss in Zuppone 2014.

10. 

In this paper, I restrict my claims regarding the representational content of experimental results to physics. As far as I can see, something along the lines of what I will defend can also be shown for several biological experiments, by taking into consideration the knowledge of mechanisms when interpreting outputs. This excludes biomedical research, which, I believe, deserves an independent investigation.

11. 

I borrow this apt expression from Hans Radder. See for example his (1992) and (2003).

12. 

S is not only the source of light but also an observatory of the output of the experiment.

13. 

Michelson calculated this using a stroboscope.

14. 

This equation is obtained by considering:

  • ν = n = cicles per second;

  • 360 ν = angle/time

  • T = angle/ 360 ν

If as was said before the angle of interest is θ/2, then:
  • T = (θ/2)/ 360 n

15. 

This value contemplates the correction for vacuum which was theoretically calculated (Michelson 1880, p. 141).

16. 

For the problem of the missing neutrinos and an analysis of the interpretation of the experimental results and its epistemological consequences see Pinch (1993) and Shapere (1982).

17. 

By stating Newton’s result in these terms, I am trying to avoid both anachronism and the appeal to concepts belonging to wave-theory to explain his results.

18. 

One might ask if the distinction between quantitative and qualitative experiments is properly justified. However, I consider that for epistemic reasons it is worth emphasizing the difference between measuring a property and claiming that a new property can be predicated of a system or entity. As the kind of errors associated with each kind of experiment may differ, the kind of change expected in the interpretations of the results of each experiment given theoretical change may also differ, it is, therefore, important to preserve the distinction. Moreover, because not every property that a system can possess will be a gradable property, and when an absolute property is discovered it may not involve a measurement, as is the case with Newton’s prism experiment, preserving the distinction is also relevant for conceptual reasons.

19. 

Newton’s experimentum crucis generated a lot of controversy among scholars. Here I will merely offer a possible analysis of the experiment in order to show how experimental results gain their representational content.

20. 

This will hold only if the prism is the position of minimum deviation. I cannot argue for this here, but I refer the interested reader to Westfall (1962) for a detailed explanation.

21. 

For Weber’s papers regarding the detection of gravity waves see Weber (1960, 1967, 1968a, 1968b, 1969, 1970, 1972).

22. 

One of them was placed in his laboratory at Maryland University and the other was situated at the Argonne National Laboratory, in Illinois.

23. 

Unfortunately, Weber made several mistakes: 1) in establishing the time-correlations between the detectors, 2) in claiming that a sidereal correlation was taking place, 3) in the signal processing electronic system, and 4) in the analysis procedure he chose (Franklin 1998, 2005). In this respect, Collins (2004, 2011) are worth reading.

24. 

For an externalist proposal regarding the interpretation of experimental results, see Pinch 1993.

25. 

In this paper I would like to remain neutral regarding any specific account on what a scientific theory is. I take my proposal to be compatible both with syntactic and with semantic approaches. However, it has been influenced by the Structuralist conception of scientific theories developed by Balzer, Moulines, and Sneed (1987), and Jose Díez’s research on the content of scientific concepts (2002).

26. 

I believe this proposal fits nicely with Giora Hon’s analysis of experimental errors in terms of idols of experiment. See for example, his 2003.

27. 

This, in turn, invites us to reconsider the scope of Hacking’s catchy slogan according to which experiments have life of their own (see Zuppone 2011).

28. 

It would be interesting to determine whether (im)proper working of the experimental devices and (in)compatibility with accepted theoretical knowledge will always have the same epistemic weight. Indeed empiricist philosophers and empirically minded scientists will tend to say that, ultimately, empirical arguments should count more than theoretical ones. However, I do not have, at least not yet, a positive proposal regarding this particular aspect and my proposal does not require that such an answer be offered. I would like to thank an anonymous referee for pointing this out to me.

29. 

As an alternative strategy to deny (2) in our reconstruction of GR, namely, the experimental introduction of the result, it may also be interesting to consider the triangulation between indirect (non-experimental) determinations and the results of the experiments under scrutiny, see below how this may be relevant, especially with regards to quantitative experiments.

30. 

Thanks to the anonymous referee for pointing this out to me.

31. 

The literature regarding the philosophy of experiment is now vast. I will refer to just two scholars who dealt with the problem that motivates my paper. For addressing the thesis of the autonomy of experiments within the New Experimentalists see Zuppone 2011.

References

Balzer
,
Wolfgang
,
Ulises
Moulines
, and
Joseph
Sneed
.
1987
.
An Architectonic for Science
.
Dordretch
:
Reidel
.
Cartwright
,
Nancy
.
1991
. “
Replicability, Reproducibility and Robustness: Comments on Harry Collins
.”
History of Political Economy
23
(
1
):
143
155
.
Collins
,
Harry
.
[1985] 1992
.
Changing Order: Replication and Induction in Scientific Practice
.
Chicago
:
University of Chicago Press
.
Collins
,
Harry
.
2004
.
Gravity’s Shadow: The Search for Gravitational Waves
.
Chicago and London
:
University of Chicago Press
.
Collins
,
Harry
.
2011
.
Gravity’s Ghost: Scientific Discovery in the Twenty First Century
.
Chicago
:
University of Chicago Press
.
Collins
,
Harry
and
Trevor
Pinch
.
2005
.
Dr. Golem, How to Think about Medicine
,
Chicago
:
The University of Chicago Press
.
Culp
,
Sylvia
.
1995
. “
Objectivity in Experimental Enquiry: Breaking Data-Technique Circles
.”
Philosophy of Science
62
:
430
50
.
Davies
,
Paul
.
1980
.
The Search for Gravity Waves
.
Cambridge
:
Cambridge University Press
.
Díez-Calzada
,
José Antonio
.
2002
. “
A Program for the Individuation for Scientific Concepts
.”
Synthese
130
:
13
48
.
Franklin
,
Allan
.
1998
.
Avoiding the Experimenters’ Regress
. Pp.
151
65
in
A House Built on Sand: Exposing Postmodernist Myths about Science
. Edited by
Noretta
Koertge
.
Oxford
:
Oxford University Press
.
Franklin
,
Allan
.
1999
.
Can That Be Right? Essays on Experiments, Evidence and Science
.
Dordretch, Boston and London
:
Kluwer Academic Publisher
.
Franklin
,
Allan
.
2002
.
Selectivity and Discord: Two Problems of Experiments
.
Pittsburgh
:
University of Pittsburgh Press
.
Franklin
,
Allan
.
2004
.
Are There Really Neutrinos? An Evidential History
.
Cambridge
:
WestView Press
.
Franklin
,
Allan
.
2005
.
No Easy Answers: Science and the Pursuit of Knowledge
.
Pittsburgh
:
University of Pittsburgh Press
.
Godin
,
Benoit
, and
Yves
Gingras
.
2002
. “
The Experimenter Regress: From Skepticism to Argumentation
.”
Studies in History and Philosophy of Science Part A
33
(
1
):
133
48
.
Hon
,
Giora
.
2003
.
The Idols of Experiment: Transcending the ‘etc. List’
. Pp.
174
197
in
The Philosophy of Scientific Experimentation
. Edited by
Hans
Radder
.
Pittsburgh
:
University of Pittsburgh Press
.
Kuhn
,
Thomas
.
1977
.
The Essential Tension: Selected Studies in Scientific Tradition and Change
.
Chicago
:
University of Chicago Press
.
Laudan
,
Larry
.
1982
. “
A Note on Collins’ Blend of Relativism and Empiricism
.”
Social Studies of Science
12
:
131
132
.
Levine
,
James
.
2004
. “
Early Gravity-Wave Detection Experiments, 1960–1975
.”
Physics in Perspective
6
:
42
75
.
Michelson
,
Albert
.
1880
. “
Experimental Determination of the Velocity of Light Made at the United States Naval Academy Annapolis
.”
Astronomical Papers
1
:
109
46
.
Newton
,
Isaac
.
1671
. “
Letter to the Royal Society
.”
Philosophical Transactions
80
:
3075
87
.
Newton
,
Isaac
.
1704
.
Optics
.
Chicago-London
:
Enciclopaedia Britannica
.
Pinch
,
Trevor
.
1993
. “
Testing, One, Two, Three, Testing!: Towards an Epistemology of Testing
.”
Science, Technology and Values
18
:
25
41
.
Radder
,
Hans
.
1992
. “
Experimental Reproducibility and the Experimenters’ Regress
.”
Philosophy of Science Association, Proceedings
1
:
63
73
.
Radder
,
Hans
.
2003
.
Technology and Theory in Experimental Science
. Pp.
152
73
in
The Philosophy of Scientific Experimentation
. Edited by
Hans
Radder
.
Pittsburgh
:
The University of Pittsburgh Press
.
Shapere
,
Dudley
.
1982
. “
The Concept of Observation in Science and Philosophy
.”
Philosophy of Science
49
:
485
525
.
Teira
,
David
.
2012
. “
A Contractarian Solution to the Experimenters’ Regress
.”
Philosophy of Science
80
(
5
):
709
20
.
Weber
,
Joseph
.
1960
. “
Detection and Generation of Gravitational Waves
.”
Physical Review
117
(
1
):
306
313
.
Weber
,
Joseph
.
1967
. “
Gravitational Radiation
.”
Physical Review Letters
18
(
13
):
498
501
.
Weber
,
Joseph
.
1968a
. “
Gravitational Radiation from the Pulsars
.”
Physical Review Letters
21
(
6
):
395
6
.
Weber
,
Joseph
.
1968b
. “
Gravitational Wave Detector Events
.”
Physical Review Letters
20
(
23
):
1307
8
.
Weber
,
Joseph
.
1969
. “
Evidence for Discovery of Gravitational Radiation
.”
Physical Review Letters
22
(
24
):
1320
4
.
Weber
,
Joseph
.
1970
. “
Anisotropy and Polarization in the Gravitational-Radiation Experiments
.”
Physical Review Letters
25
(
3
):
180
4
.
Weber
,
Joseph
.
1972
. “
Computer Analyses of Gravitational Detector Coincidences
.”
Nature
240
:
28
30
.
Weber
,
Marcel
.
2012
. “
Experiment in Biology
.” In
The Stanford Encyclopaedia of Philosophy
. Edited by
Edward N.
Zalta
.
Westfall
,
Richard
.
1962
. “
The Development of Newton’s Theory of Color
.”
Isis
53
(
3
):
339
358
.
Zuppone
,
Romina
.
2010
. “
El argumento del regreso del experimentador y la replicación de experimentos
.”
Scientiae Studia
8
(
2
):
243
71
.
Zuppone
,
Romina
.
2011
. “
La vida propia del experimento: un análisis crítico de la autonomía de la experimentación
.”
Revista Latinoamericana de Filosofía
37
:
213
238
.
Zuppone
,
Romina
.
2014
. “
Overcoming the Experimenters’ Regress in Biomedical Research
.”
Philosophical Writings
43
(
1
):
45
56
.

Author notes

Special thanks to Alejandro Cassini and José Antonio Díez Calzada. Thanks also to the participants of LOGOS GRG for fruitful discussions of previous versions of this paper, especially to Aurélien Darbellay, David Rey, Giovanni Merlo, Stefan Reining, and Ljubomir Stevanovic. I am also grateful for the helpful comments I received from the attendees at the 2nd Barcelona-Urbino Meeting on History and Philosophy of Science. I am indebted to Pablo Couto for helping me to process the images that appear in the paper. This work was supported by PERSP Project, funded by the Consolider-Ingenio 2010 Scheme CSD2009-00056, Spanish Ministry of Science and Innovation.