Towards an Objective Theory of Rationality

4. An Objectivist Epistemology

4.3 Criteria for Theory-Independence

Book cover: Essays in Memory of Imre Lakatos, Boston Studies in the Philosophy of Science, Volume 39 by R. S. Cohen, P. K. Feyerabend and M. W. Wartofsky

There is an important caveat here. A theory may have evidence in its favour in the sense specified by the three minimal requirements introduced above. However, we should not accept it as true on that basis alone. This is for two reasons. Firstly, the argument for the theory will simply beg the question if the truth of the evidence-statement is determined by presupposing the truth of the theory under consideration. Consider, for example, the simple cosmology, 'Every entity is causally related to every other entity', and the evidence proposed in support of it, 'Percepts are causally related to every other entity'. If the truth of the evidence-statement was determined by deriving it from the statements, 'Every entity is causally related to every other entity' and 'Percepts are entities', then the argument is viciously circular. Cosmologies logically inconsistent with the above, such as, 'No entity is causally related to any other entity', can just as easily be supported by such circular arguments. What is required is a condition of theory appraisal that specifies that the epistemic status of the evidence-statement be determinable independently of the theory under consideration.

Secondly, for any given body of data, any number of mutually inconsistent explanatory theories, conjoined with auxiliary hypotheses and entailing the data, may be adduced. This is the well-known problem of the underdetermination of theory choice.[9] On a superficial analysis, it will appear that the data equally constitute evidence for all of these mutually inconsistent theories. In addition, the appearance of new data inconsistent with a theory and its auxiliary hypotheses can always be accommodated by post hoc revisions to the theory or to the auxiliary hypotheses. The anomalous data itself can be modified by means of a newly introduced auxiliary theory explaining how the data is to be revised.[10] (Ontology-cosmologies are so all-encompassing that I shall henceforth refer to the metaphysical core of the theory, the auxiliary hypotheses and any newly introduced auxiliary theories as all part of the one ontological-cosmological theory.)

In order to evaluate such a plethora of seemingly equally evidenced theories, an added criterion for objective theory appraisal is required. It must stipulate that the evidence be independent of that which equally supports all of these mutually inconsistent theories. That is, the evidence must be independent of the data used to construct the original theory and used in its later revisions.

Now, evidence-statements that are independent of theory construction and revision may be classified as belonging to either of two mutually exclusive categories. The evidence-statement is either a 'novel prediction' or a 'novelly derived fact', where a 'fact' is a true contingent sentence. I shall mean by 'novel prediction' (relative to a particular theoretical development) a statement that is derivable from the theory but was not known to be true by the constructors and revisers of the theory during construction and revision.[11] And by 'novelly derived fact' (relative to a particular theoretical development), I mean a statement that is derivable from the theory and was known to be true to the constructors or revisers of the theory during construction or revision, but was not used to construct or revise the theory.[12]

This composite class of evidence-statements that are independent of theory construction and revision (that is, the class of novel predictions and novelly derived facts) may be divided into three sub-classes:

  1. those statements that are novel predictions or novelly derived facts for one or more rival theories;
  2. those statements that are derivable from one or more rival theories but are neither novel predictions nor novelly derived facts for such rivals;
  3. those statements that are either not derivable, or their negations derivable, from one or more rival theories.
Book cover: Beyond the Edge of Certainty by Robert Garland Colodny

If an evidence-statement for a theory is of type i), the statement is equally strong evidence for all rival theories for which it is also a novel prediction or a novelly derived fact. If it is of type ii), the statement is strong evidence for the theory and weak evidence for the rivals from which it can be derived. If it is of type iii), it is once again strong evidence for the theory, but is either no evidence or counterevidence for the rivals. Type iii) evidence-statements give the strongest relative weight to a theory because statements of this type can explain what the rivals cannot explain. Type ii) evidence-statements give a weaker, but still very significant, relative weight to the theory. This is because statements of this type explain independently that which the rivals can only explain dependently. Type i) evidence-statements give no relative weight to the theory in comparison with rivals that also give an independent explanation.

Recognising the crucial epistemic value of novel predictions and novelly derived facts helps to solve the problem of the underdetermination of theory choice. We are now in a position to supplement my three previously stated criteria for rational theory choice with two new conditions. We may roughly formalize these as follows.

An evidence-statement is strong evidence for a theory if:

Criterion 4:
the epistemic status of the evidence-statement is determinable independently of the theory;
(requirement of epistemic independence from theory)
Criterion 5:
the evidence-statement is independent of the data used to construct and modify the theory.
(requirement of independence from theory development)

This completes my formulation of the fundamental criteria for rational theory choice. Criterion 1, 2 and 3 specify the necessary and sufficient conditions that must be satisfied for a statement to be regarded as evidence at all. That a theory be supported by evidence satisfying these three minimal conditions does not warrant automatic rational assent to its truth. As mentioned earlier, it is not too difficult to construct theories logically inconsistent with the one in mind that are equally supported by such evidence. A theory is rationally justified, then, only if it is supported by evidence of the type stipulated in Criterion 4 and 5. Even then, the theory in question is supported only if there is no rival theory supported to a greater extent, or nearly to the same extent, by evidence of this type. With these five criteria for objective theory choice, we have here the basic ingredients of Lakatos' MSRP (with Zahar's important modification), with its emphasis on the epistemic value of novel facts. Quite neatly, Lakatos' basic schema turns out to be a consequence of the demands of an objectivist epistemology.


  1. [9] See, for example Quine and Ullian [1970] and Stanford [2013].
  2. [10] For simple examples of post hoc revisions, see Lakatos [1978a: 16f, 97f].
  3. [11] This is not exactly the same as Lakatos' use of the term 'novel prediction'. For Lakatos, a prediction is not novel if it is also predicted by a rival theory. See Lakatos [1978a: 32, 1978b: 170f]. Contrary to Lakatos, 'novelty' is relative to people and not to pairs of theories. For example, if a new theory predicts a novel fact of which we were hitherto unaware, and it is later discovered that the older rival theory had also, unbeknownst to us, predicted the same fact, it would be more natural to continue describing the newly discovered fact as 'novel', even though it could also be deduced from the rival theory. Fortunately, in the end, this terminological difference makes no difference to the theory of theory appraisal advocated here.
  4. [12] This is akin to Zahar's concept of 'novel fact'. See Zahar [1973] and Lakatos [1978a: ch. 4].

Copyright © 2016

You will be interested in
Book cover: The Logic of Scientific Discovery by Karl Popper
Book cover: From a Logical Point of View by Willard Van Orman Quine
Book cover: The Invention of Science: A New History of the Scientific Revolution by David Wootton
Book cover: The Web of Belief by W. V. Quine and J. S. Ullian
Book cover: From Copernicus to Einstein by Hans Reichenbach
Book cover: The Foundations of Scientific Inference by Wesley Salmon

Share This

  • twitter
  • facebook
  • linkedin
  • googleplus
  • gmail
  • delicious
  • reddit
  • digg
  • newsvine
  • posterous
  • friendfeed
  • googlebookmarks
  • yahoobookmarks
  • yahoobuzz
  • orkut
  • stumbleupon
  • diigo
  • mixx
  • technorati
  • netvibes
  • myspace
  • slashdot
  • blogger
  • tumblr
  • email
Short URL: