Abstract
Varied evidence for a hypothesis confirms it more strongly than less varied evidence, ceteris paribus. This epistemological Variety of Evidence Thesis enjoys long-standing widespread intuitive support. Recent literature has raised serious doubts that the correlational approach of explicating the thesis can vindicate it. By contrast, the eliminative approach due to Horwich vindicates the Variety of Evidence Thesis but only within a relatively narrow domain. I investigate the prospects of extending the eliminative approach to a larger domain by considering a larger class of sensible explications of evidential variety. For a large subclass class of such explications I show how to construct cases in which the less diverse body of evidence for a hypothesis confirms more strongly. I hence argue that these prospects are dire since the eliminative approach widely fails to vindicate the thesis.
Similar content being viewed by others
Notes
Earman, Footnote 14, p. 240 dates his ideas back to “the mid 1970’s” pre-dating Franklin and Howson (1984) by several years.
Wayne also challenged the champions of the correlational approach to put forward an explication of evidential variety that incorporates more than just prior probabilities. Such explications are a topic of current discussion (Bovens and Hartmann 2003; Claveau 2013; Claveau and Grenier 2019; Landes 2019), see Section 3 for more details.
Complexity and its opposite, simplicity, of a hypothesis, H, are here understood in terms of the degree of the smallest family of polynomials containing H. The greater this degree, the greater the complexity of the hypothesis.
I think that Horwich could easily defend himself against the criticism that the hypotheses are non-exhaustive in realistic cases. Horwich could point out that every set of non-exhaustive hypotheses can always be made exhaustive by adding a catch-all complement hypothesis, H: all other hypotheses are false, cf. Wenmackers and Romeijn (2016). Another way to deflect this criticism is to point out that “knowing that one of the hypotheses is true” (Horwich 1982, p. 118) does not entail that the disjunction of these hypotheses is a logical tautology.
Intuitions supporting the “source variety” VET seem to be predicated on the assumption that more source variety must lead to more hypothesis confirmation. However, less source variety makes the evidence more informative about the source(s) of evidence. Learning that sources of confirmatory evidence are reliable also increases hypothesis confirmation. In some cases, hypothesis confirmation via learning about the source(s) of evidence trumps learning from independent sources.
Intuitions supporting the “consequence variety” VET seem to be predicated on the assumption that more consequence variety must lead to more hypothesis confirmation. However, lower consequence variety makes the evidence more informative about fewer consequences of the hypothesis. Learning that a probabilistic consequence of the hypothesis is likely true also increases hypothesis confirmation (in particular if the consequence is a priori relatively unlikely). In some cases, hypothesis confirmation via boosting the belief in fewer consequences trumps learning less about more consequences of the hypothesis.
Measuring diversity has recently become important to measure our progress towards a more diverse society. Measures of diversity for quantifying species diversity are of much interest in ecological settings (Tuomisto 2010). Quantitative measures of entropy rather than diversity are an area of active research in information studies, (formal) epistemology and psychology (Landes 2015; Crupi et al. 2018; Csiszár 2008).
I want the notion of (comparative) evidential variety to be widely applicable so that a (possible) vindication of the VET in terms of eliminating competing hypotheses employing this notion covers a large part of scientific inference. Parameter estimation is one key task of scientific and statistical inference. A widely applicable vindication of the VET should hence apply to a number of cases of parameter estimation.
I here take a Bayesian view of parameter estimation according to which the constraints on the parameter(s) of interest constrain rational beliefs.
I’m grateful to an anonymous reviewer for pressing me to clarify the relation between evidential variety and comparative evidential variety. On the one hand, any notion of evidential variety which allows – at least in some cases – the comparison of degrees of evidential variety induces a canonical notion of comparative evidential variety. On the other hand, every notion of comparative evidential variety is – in some shape or form – based on a notion of evidential variety. To formulate a comparative notion however there is (a priori) no need to specify the underlying notion in full detail, a full-blown explication is not required.
To streamline my presentation I will say that the VET is vindicated in case \(P(H_{1}|\mathcal {E}_{N})\leq P(H_{1}|\mathcal {E}_{D})\).
Mutatis mutandis, my formal analysis continues to apply for a hypothesis of interest which is more complicated than the competing hypotheses: replace f1 [defined below] by a very complicated function \(f^{*}_{1}\), which is a very good approximation of f1.
There is a natural sense of evidential diversity this explication c captures: The measurements in \(\mathcal {E}_{D}\) are more spread out, in other words, the measurements in \(\mathcal {E}_{N}\) are more clustered around their average. In this sense, the body of evidence \(\mathcal {E}_{D}\) is more diverse than the body of evidence \(\mathcal {E}_{N}\). Consequences of the existence of a case in which bD − aD < bN − aN are considered in Section 4.2. Note however that the evidential variety of a body of evidence for an explication c ∈ C depends, in general, also on factors other than the spread of measurements.
The condition that P(H5) is close to but different from zero is the only assumption I put on the prior probability function P. This condition is in agreement with the suggestion at Horwich (1982, pp. 70-71) that complex hypotheses have low prior probability.
ID ⊂ IN is the only exceptional case here since it is already covered by Horwich’s explication of evidential variety, \(P(\mathcal {E}_{D}|H_{i})\leq P(\mathcal {E}_{N}|H_{i})\) for all i follows from ID ⊂ IN and the axioms of (conditional) probability.
All fi are chosen such that the coefficient of x of the second highest order is three times the coefficient of x of the highest order. There is hence a natural sense in which the Hi are similar.
Clearly, \({\int \limits }_{-5}^{5}f_{5}(x)dx=1\) holds and f5 is the most complicated hypothesis and thus H1 remains the simplest hypothesis entertained.
References
Borm, G.F. , Lemmers, O., Fransen, J., Donders, R. (2009). The evidence provided by a single trial is less reliable than its statistical analysis suggests. Journal of Clinical Epidemiology, 62(7), 711–715. https://doi.org/10.1016/j.jclinepi.2008.09.013.
Bovens, L., & Hartmann, S. (2002). Bayesian networks and the problem of unreliable instruments. Philosophy of Science, 69(1), 29–72. https://doi.org/10.1086/338940.
Bovens, L., & Hartmann, S. (2003). Bayesian epistemology. Oxford: Oxford University Press .
Carnap, R. (1962). Logical foundations of probability, 2nd edn. Chicago: University of Chicago Press.
Claveau, F. (2013). The independence condition in the variety-of-evidence thesis. Philosophy of Science, 80(1), 94–118. https://doi.org/10.1086/668877.
Claveau, F., & Grenier, O. (2019). The variety-of-evidence thesis: a Bayesian exploration of its surprising failures. Synthese, 196, 3001–3028. https://doi.org/10.1007/s11229-017-1607-5.
Crupi, V., Nelson, J., Meder, B., Cevolani, G., Katya, T. (2018). Generalized information theory meets human cognition: Introducing a unified framework to model uncertainty and information search. Cognitive Science, 42, 1410–1456. https://doi.org/10.1111/cogs.12613.
Csiszár, I. (2008). Axiomatic characterizations of information measures. Entropy, 10(3), 261–273. https://doi.org/10.3390/e10030261.
Earman, J. (1992). Bayes or bust? Cambridge: MIT Press.
Fitelson, B. (1996). Wayne, Horwich, and evidential diversity. Philosophy of Science, 63(4), 652–660. https://doi.org/10.1086/289982.
Fitelson, B. (2001). A bayesian account of independent evidence with applications. Philosophy of Science, 68(3), S123–S140. https://doi.org/10.2307/3080940.
Franklin, A., & Howson, C. (1984). Why do scientists prefer to vary their experiments? Studies in History and Philosophy of Science Part A, 15(1), 51–62. https://doi.org/10.1016/0039-3681(84)90029-3.
Heit, E., Hahn, U., Feeney, A. (2005). Defending diversity. In W. -k. Ahn, R. L.Goldstone, B.C. Love, A. B. Markman, P. Wolff (Eds.) Categorization inside and outside the laboratory: Essays in honor of Douglas L. Medin (pp. 87–99). Washington: American Psychological Association, DOI https://doi.org/10.1037/11156-006, (to appear in print).
Hempel, C. (1966). Philosophy of natural science. Upper Saddle River: Prentice Hall.
Horwich, P. (1982). Probability and evidence. Cambridge: Cambridge University Press.
Horwich, P. (1993). Wittgensteinian bayesianism. Midwest Studies in Philosophy, 18(1), 62–77. https://doi.org/10.1111/j.1475-4975.1993.tb00257.x.
Horwich, P. (1998). Wittgensteinian Bayesianism. In M. Curd, & J. A. Cover (Eds.) Philosophy of science: the central issues (pp. 607–624). New York: W. W. Norton & Company.
Howson, C., & Urbach, P. (2006). Scientific reasoning, 3rd edn. Chicago and La Salle: Open Court.
Hüffmeier, J., Mazei, J., Schultze, T. (2016). Reconceptualizing replication as a sequence of different studies: a replication typology. Journal of Experimental Social Psychology, 66, 81–92. https://doi.org/10.1016/j.jesp.2015.09.009.
Keynes, J. M. (1921). A treatise on probability. London: MacMillan.
Landes, J. (2015). Probabilism, entropies and strictly proper scoring rules. International Journal of Approximate Reasoning, 63, 1–21. https://doi.org/10.1016/j.ijar.2015.05.007.
Landes, J. (2019). Variety of evidence. Erkenntnis. https://doi.org/10.1007/s10670-018-0024-6. p 41. Ahead of print.
Landes, J. (2021). The variety of evidence thesis and its independence of degrees of independence. Unpublished manuscript.
Landes, J., & Osimani, B. (2020). Varieties of error and varieties of evidence. Unpublished manuscript.
Lukoianova, T., & Rubin, V. (2014). Veracity roadmap: is big data objective, truthful and credible? Advances in Classification Research Online, 24(1), 4–15. https://doi.org/10.7152/acro.v24i1.14671.
Meehl, P.E. (1990). Appraising and amending theories: the strategy of Lakatosian defense and two principles that warrant it. Psychological Inquiry, 1(2), 108–141. https://doi.org/10.1207/s15327965pli0102_1.
Paris, J.B. (2014). What you see is what you get. Entropy, 16(11), 6186–6194. https://doi.org/10.3390/e16116186.
Sagiroglu, S., & Sinanc, D. (2013). Big data: a review. In Proceedings of CTS (pp. 42–47): IEEE, DOI https://doi.org/10.1109/CTS.2013.6567202, (to appear in print).
Schupbach, J.N. (2015). Robustness, diversity of evidence, and probabilistic independence. In U. Mäki, I. Votsis, S. Ruphy, G. Schurz (Eds.) Proceedings of EPSA13 (pp. 305–316). Cham: Springer, DOI https://doi.org/10.1007/978-3-319-23015-3_23, (to appear in print).
Schupbach, J. N., & Glass, D. H. (2017). Hypothesis competition beyond mutual exclusivity. Philosophy of Science, 84(5), 810–824. https://doi.org/10.1086/693928.
Steel, D. (1996). Bayesianism and the value of diverse evidence. Philosophy of Science, 63(4), 666–674. https://doi.org/10.1086/289984.
Tuomisto, H. (2010). A consistent terminology for quantifying species diversity? Yes, it does exist. Oecologia, 164 (4), 853–860. https://doi.org/10.1007/s00442-010-1812-0.
Wayne, A. (1995). Bayesianism and diverse evidence. Philosophy of Science, 62(1), 111–121. https://doi.org/10.1086/289842.
Wenmackers, S., & Romeijn, J.-W. (2016). New theory about old evidence. Synthese, 193(4), 1225–1250. https://doi.org/10.1007/s11229-014-0632-x.
Worrall, J. (2007). Evidence in medicine and Evidence-Based medicine. Philosophy Compass, 2(6), 981–1022. https://doi.org/10.1111/j.1747-9991.2007.00106.x.
Acknowledgments
Acknowledgments]I gratefully acknowledge funding from the European Research Council (‘PhilPharm’ grant 639276) and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)– 405961989 and 432308570. Furthermore, I want to thank Lorenzo Casini, Stephan Hartmann and Barbara Osimani for many helpful discussions and comments. I’m heavily indebted to Lorenzo Casini for continued feedback and proof-reading. Many thanks also to anonymous reviewers which helped me to improve my arguments which has in turn led to a much more readable presentation.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A:: Formal analysis
Appendix A:: Formal analysis
1.1 A.1 Varied evidence constraining less tightly
Here the formal analysis, in some detail. We set:
- 1.
\(\mathcal {E}_{N}:=[0,0.1]\) and \(\mathcal {E}_{D}:=[3-\sqrt {4.59},1]\approx [0.86,1]\).
- 2.
For the fi we have fi(y) := 0 for all y∉[0, 3] and for all x ∈ [0, 3]Footnote 18
$$ \begin{array}{@{}rcl@{}} f_{1}(x)&:=&2/3-2x/9\\ f_{2}(x)&:=&x\cdot f_{1}(x)=x\cdot (2/3-2x/9)=2x/3-2x^{2}/9\\ f_{3}(x)&:=&1/30+3x/5-x^{2}/5\\ f_{4}(x)&:=&1/12+x/2-x^{2}/6 . \end{array} $$The Hi are probabilistic hypotheses, \(P(X\in [0,3])={{\int \limits }_{0}^{3}} f_{i}(x)dx=1\) for all i ∈{1, 2, 3, 4}.
We now check that the assumptions are satisfied:
ID is more diverse than IN [from 1].
H1 is simpler than the other Hi: H1 is affine-linear while the other Hi are parabola [from 2].
The Ceteris Paribus clause (3) holds:
$$ \begin{array}{@{}rcl@{}} P_{1}(\mathcal{E}_{N}|H_{1})&=&{\int}_{0}^{0.1} f_{1}(x)dx=[2x/3-x^{2}/9]_{0}^{0.1}=2/30-1/900 = 59/900\\ P_{1}(\mathcal{E}_{D}|H_{1})&=&{\int}^{1}_{3-\sqrt{4.59}} f_{1}(x)dx=[2x/3-x^{2}/9]_{3-\sqrt{4.59}}^{1}\\ &=&2/3-1/9-(2-\frac{2}{3}\sqrt{4.59})+1/9(9-6\sqrt{4.59}+4.59)\\ &=&5/9-1+ 459/900\\ &=&-4/9+ 459/900 = 59/900 . \end{array} $$\(\mathcal {E}_{N},\mathcal {E}_{D}\) are both evidence for H1: To show that \(P(H_{1})<P(H_{1}|\mathcal E)\) for \(\mathcal E\in \{\mathcal {E}_{N},\mathcal {E}_{D}\}\) we first note that \(P(\mathcal E|H_{1})>P(\mathcal E|H_{i})\) for i ∈{2, 3, 4}, since for all x ∈ IN ∪ ID ∖{1} it holds that f1(x) > fi(x) > 0. We now find
$$ \begin{array}{@{}rcl@{}} P(H_{1}|\mathcal E)>P(H_{1})&\Leftrightarrow& \frac{P(\mathcal E|H_{1})}{P(\mathcal E)}>1\\ &\Leftrightarrow& P(\mathcal E|H_{1})>{\sum}_{i=1}^{4} P(\mathcal E|H_{i})P(H_{i})\\ &\Leftrightarrow& P(\mathcal E|H_{1})(1-P(H_{1}))>{\sum}_{i=2}^{4} P(\mathcal E|H_{i})P(H_{i}) . \end{array} $$
To show that \(\mathcal {E}_{N},\mathcal {E}_{D}\) are both evidence for H1, it now suffices to note that
It only remains show that the more diverse body of evidence confirms less strongly, \(P(H_{1}|\mathcal {E}_{D})<P(H_{1}|\mathcal {E}_{N})\):
Since for all i ∈{2, 3, 4} it holds that \(\sup _{x\in I_{N}}f_{i}(x)<\inf _{x\in I_{D}}f_{i}(x)\) and since |IN| < |ID| we indeed have
1.2 A.2 Varied Evidence constraining more tightly
The formal analysis for this case resembles the previous one and his hence given in a more condensed form. First put
- 1.
IN := [− 0.1, 0.1] and \(I_{D}:=[3-\sqrt {4.59},1]\approx [0.86,1]\).
- 2.
For i ∈{1, 2, 3, 4} the fi are as above and for f5 we have f5(y) = 0 for all y < − 5 and y > 5 and for all x ∈ [− 5, 5]Footnote 19
$$ \begin{array}{@{}rcl@{}} f_{5}(x)&:=\left\{\begin{array}{ll}\frac{1}{25}+\frac{\sin(x)}{25} & x\in[-5,0)\\ \frac{4}{25}+\frac{\sin(x)}{25}& x\in[0,5] . \end{array}\right. \end{array} $$ - 3.
Let P(H5) be tiny but non-zero, \(P(H_{5})=\frac {1}{10000}\) say.
Mutatis mutandis, the proof above (Appendix A1) applies here, too. After going through the calculations again we obtain \(P(H_{1}|\mathcal {E}_{D})<P(H_{1}|\mathcal {E}_{N})\), since the prior of H5 is tiny and all other hypotheses assign probability to zero to the points added to the left interval.
1.3 A.3 The general case
The two proofs above apply to a special case with fixed \(\mathcal {E}_{D},\mathcal {E}_{N}\). I now show how to prove the general case for varied evidence constraining less tightly, i.e., arbitrary \(\mathcal {E}_{D},\mathcal {E}_{N}\) satisfying |IN| < |ID|. This proof shows how to modify the functions fi for the general case. The case for varied evidence constraining more tightly is all but analogous and is only sketched here.
In the general case of varied evidence constraining less tightly we have that |IN| < |ID|. The ceteris paribus condition \(P_{1}(\mathcal {E}_{D})=P_{1}(\mathcal {E}_{N})\) is also assumed throughout.
Let us first assume that IN and IDare disjoint. By flipping the x-axis (fi(x)↦fi(−x)) if necessary, we may assume that ID is to the right of IN.
Shifting along the x-axis (fi(x)↦fi(x + shift)) if necessary, we may assume that IN = [0,α],ID = [β,γ] with 0 < α <β< γ and γ−β > α.
If γ ≤ 1, then define the functions fi for all i ∈{1, 2, 3, 4} as in the proof in Appendix A1 which then applies, mutatis mutandis. a and b hold. c holds by assumption, d holds since fi(x) < f1(x) for all x ∈ (0, 1) all i ∈{2, 3, 4}. \(P(H_{1}|\mathcal {E}_{N})>P(H_{1}|\mathcal {E}_{D})\) follows.
If γ > 1, then put gi(x) := γfi(γx). Geometrically, this shrinks the x-axis by the factor γ while the y-axis is dilated by the factor γ. Hence, all ordinal comparisons are preserved for all \(x\in \mathbb {R}\): fi(γx) < fj(γx), if and only if gi(x) < gj(x). It only remains to check that the gi define probability functions; we hence calculate for all i ∈{1, 2, 3, 4}
where Fi denotes a primitive function of fi (uniquely defined up to an inconsequential additive constant) characterised by \(\frac {d}{dx}F_{i}(x)=f_{i}(x)\).
IN and ID are not disjoint and IN is not a proper subset of ID, IN ∩ ID≠∅. We begin by putting \(I_{D}^{\prime }:=I_{D}\setminus (I_{D}\cap I_{N})\neq \emptyset \) and \(I_{N}^{\prime }:=I_{N}\setminus (I_{D}\cap I_{N})\neq \emptyset \). Now proceed with these bodies of evidence as above. The Ceteris Paribus Clause (3) again holds because
It again follows that \(P(H_{1}|I_{D}^{\prime })>P(H_{1})<P(H_{1}|I_{N}^{\prime })\) [because f1(x) > fi(x) on ID ∪ IN for all i ≥ 2] and eventually that \(P(H_{1}|I_{N}^{\prime })>P(H_{1}|I_{D}^{\prime })\). Thus, \(P(H_{1}|\mathcal {E}_{N}^{\prime })>P(H_{1}|\mathcal {E}_{D}^{\prime })\).
Finally, ifIN is a proper subset of ID, then by Ceteris Paribus Clause (3) it holds that P(ID ∖ IN|H1) = 0 (almost everywhere). Now simply devise functions fi for i ∈{2, 3, 4} which take large values on ID ∖ IN. Then, these Hi are more compatible with \(\mathcal {E}_{D}\) than with \(\mathcal {E}_{N}\). Hence, \(\mathcal {E}_{N}\) confirms H1 stronger than \(\mathcal {E}_{D}\), \(P(H_{1}|\mathcal {E}_{N})>P(H_{1}|\mathcal {E}_{D})\). Now proceed as above.
Here now the sketch for the case in which varied evidence constrains more tightly, |ID| < |IN|.
Again, first assume that IN and ID are disjoint. As above, by flipping and shifting, if necessary, we may assume that 0 is in the interior of IN and that ID is to the right of IN. Next, shift the intervals again such that for the lower and upper bounds of IN,ID, lN,lD,uN,uD, it holds that uN ⋅ 0.14 = (uD − lD) ⋅ 0.1. That is, the parts of the interval to the right of zero, stand in the same proportion as those in the example.
Now define a hypothesis H5 in terms of \(f_{5}:\mathbb {R}\to \mathbb {R}_{\geq 0}\) such that i) f5(x) > 0 for all x ∈ IN ∪ ID, ii) f5(x) is sufficiently small (so that we can ignore it in the remainder) relative to the other hypotheses where f5 is non zero and iii) such that f5 is a probability distribution, \({\int \limits }_{\mathbb {R}} f_{5}(x)dx=1\).
IN and ID are not disjoint and ID is not a proper subset of IN: The idea remains the same, pick functions fi which vanish on the intersection ID ∩ IN. Proceed as in the case with |IN| < |ID| with those intervals outside the intersection.
The only not analogous case for varied evidence constraining more tightly occurs for the proper subset case, if ID is a proper subset of IN. In this case, Eq. 1 holds for at least one j ≥ 2 – otherwise the current frame of investigations collapses IN to ID. For all other hypotheses Hj we have P(ID|Hj) ≤ P(IN|Hj) since ID ⊂ IN. We are hence in a case in which all assumptions of the eliminative, including the caveats, apply (see Eq. 5). So, this is not a case in which could extend the eliminative approach, since it is already covered by it.
Rights and permissions
About this article
Cite this article
Landes, J. Variety of evidence and the elimination of hypotheses. Euro Jnl Phil Sci 10, 12 (2020). https://doi.org/10.1007/s13194-019-0272-6
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s13194-019-0272-6