Why worry about theory‐dependence? Circularity, minimal empiricality and reliability

Abstract

It is a widely shared view among philosophers of science that the theory‐dependence (or theory‐ladenness) of observations is worrying, because it can bias empirical tests in favour of the tested theories. These doubts are taken to be dispelled if an observation is influenced by a theory independent of the tested theory and thus circularity is avoided, while (partially) circular tests are taken to require special attention. Contrary to this consensus, it is argued that the epistemic value of theory‐dependent tests has nothing to do with the circularity or non‐circularity of the test, but is instead based on the minimal empiricality and reliability of observations. Since theory‐dependence does not in general prevent observations fulfilling these requirements, it should not be regarded as a phenomenon that is basically detrimental, but as neutral with respect to successful scientific knowledge gathering.

Notes

Correspondence: Department of Philosophy, Bielefeld University, Universita¨tsstrasse 25, 33615 Bielefeld, Germany. E‐mail: [email protected]‐bielefeld.de

See, for example, Franklin et al. (1989) and Kosso (1989). See Longino (2002, p. 126) for a reference to the claim as current philosophical consensus.

See Popper (1972, para. 30, p. 107, footnote 3): ‘[O]bservations, and even more so observation statements and statements of experimental results, are always interpretations of the facts observed; … they are always interpretations in the light of theories. This is one of the main reasons why it is always deceptively easy to find verifications of a theory, and why we have to adopt a highly critical attitude towards our theories if we do not wish to argue in circles: the attitude of trying to refute them’ (Popper's emphases).

See Brown (1993) for more details. In order to keep the example transparent, I abstract from the many complexities of the actual investigations and calculations. For example, one has to account for the relativistic time dilatation (see Porcas, 1987, p. 13) and for the signal delay that results from different signal travel times, due to the different distances of the quasars for the different measurements of their position (see Falla & Floyd, 2002).

See Shogenji (2000). To be precise, Shogenji is primarily concerned with naturalized epistemology. He claims that empirical justifications of the reliability of empirical processes of belief formation are not circular and are thus in general permissible. Where such justifications do not work, this is due to predictive irrelevance. However, he also proposes extending this analysis to theory‐dependent empirical tests, in particular to Brown's example. In the following, I work out his proposal in more detail. In his Bayesian reconstruction, Shogenji refers to the relation between hypothesis and observation as one of prediction (and not merely as one of probabilistic relevance) since he treats Bayesian confirmation as a generalization of hypothetico‐deductive confirmation (Shogenji, 2000, pp. 288–289). By endorsing this specification, the results of the following discussion can be transferred to an analogous hypothetico‐deductive reconstruction of theory‐dependent tests. See notes 8 and 9.

• Writing Bayes' theorem as

•        Prob(H/O&B*)=Prob(H/B*) × [Prob(O/H&B*)/Prob(O/B*)]

•    and given that

•        Prob(O/H&B*)=Prob(O/B*),

•    it follows that

•        Prob(H/O&B^* )=Prob(H/B*).

•    When we make observation O and conditionalize according to

•        Prob′(H/B*)=Prob(H/O&B*),

•    this gives

•        Prob′(H/B*)=Prob(H/B*),

•    with Prob′(…) denoting the probabilities posterior to the observation. This means that the observation leaves the probability of the hypothesis unaltered (cf. Shogenji, 2000, p. 289). For a comprehensive exposition of the Bayesian approach to confirmation, see Howson and Urbach (1993).

Shogenji probably wanted his proposal to be understood in this way right from the start, for according to his reconstruction, the theory to be tested plays a single role, namely in the prediction of an observational result, but does not contribute to generation of the result. This is only possible if the relevant observational result does not lie at the level of the recession velocities (cf. Shogenji, 2000, pp. 292–293).

Again, the refinement of section 2 applies here. In so far as the recession velocities come out below c, the red shifts do not contribute anything empirical. The precise values of the recession velocities are empirically determined and depend on the precise red shifts.

The proposal as such does not presuppose the Bayesian account of confirmation. John Worrall argues within an effectively hypothetico‐deductive framework that theory‐dependent observations can be reconstructed as holistic tests jointly of the tested and the influencing theories (see Worrall, 1982, pp. 162–166).

This difficulty is not confined to the Bayesian account of confirmation. It also applies to hypothetico‐deductive confirmation, where the epistemic value of the observational result is unclear if the various hypotheses are incompatible.

Compare Chalmers (2003, p. 495) for the discussion of a closely related example.

See Hacking (1983, p. 201) (cf. Culp, 1994). But see Hudson (1999) for reservations about route independence.