Global and Local Pessimistic Meta-inductions

Abstract

The global pessimistic meta-induction argues from the falsity of scientific theories accepted in the past to the likely falsity of currently accepted scientific theories. I contend that this argument commits a statistical error previously unmentioned in the literature and is self-undermining. I then compare the global pessimistic meta-induction to a local pessimistic meta-induction based on recent negative assessments of the reliability of medical research. If there is any future in drawing pessimistic conclusions from the history of science, it lies in local meta-inductions, but these meta-inductions will not result in global distrust of the results of science.

Acknowledgements

I am grateful for helpful feedback from Ludwig Fahrbach, Nicholas Horton, Brian Kierland, Bradley Monton, Aaron Smith, Bob Snyder, David Soergel, the philosophy department at Boise State University, and the anonymous referees of this journal.

Notes

[1] There is not agreement among philosophers about how to use the phrase ‘pessimistic induction’ (Wray 2013, 1723n). Some discussions use the phrase to refer both to PI and PMI (Worrall 1994; Psillos 1999; Lewis 2001; Lange 2002; Newman 2005; Stanford 2006; Doppelt 2007; Fahrbach 2011a); some to PI only (Hardin and Rosenberg 1982; Kitcher 1993; Elsamahi 2005; Saatsi 2005; Lyons 2006; Enfield 2008; Ritchie 2008; Fahrbach 2011b); some to PMI only (Hobbs 1994; Leplin 1997; Bishop 2003; Magnus and Callender 2004; Magnus 2010; Park 2011); and others do not specify (Chang 2003; Chakravartty 2004).

[2] B2 is from the explication of PI in Lewis (2001, 373), following Psillos (1999, 102–103).

[3] There are statistical ways of handling not being able to sample from future members of populations, a phenomenon known as right-censoring. More on this in section 5.

[4] For more on eliminative inference, see Earman (1992), Kitcher (1993), and Forber (2011).

[5] I am grateful to an anonymous referee for stressing the importance of this distinction. Although the general idea of strong PMI is easily stated—the population is expanded to include future as well as past and current theories—it is difficult to explicate. For example, the reductio in B1–B4 does not work because it does not make sense to assume that current and future theories are true. I overlook this difficulty as my main concern lies with modest PMI.

[6] It is unclear whether Stanford thinks there are enough unconceived alternatives for modest or strong pessimism (see Ruhmkorff 2011, 879–880).

[7] See below for a caveat to this claim.

[8] In my calculations, I leave out theories unchallenged by subsequent studies. Ioannidis (2005a, 225) points out that sometimes studies are not challenged because they are seen as definitive, and ethical and other considerations bar attempts at replication. My percentages may be slightly more pessimistic than is warranted, as I am excluding some studies that are more likely to be reliable than a random member of the sample.

[9] Things are not always so tidy; for example, ‘diminished effects’ also includes shorter duration of the effect rather than the effect size. I do not think this detail matters for my point here.

[10] I am grateful to an anonymous reviewer for recommending a fuller exposition of PMI-M and the differences between it and global PMI.

[11] Mizrahi (2013) sketches one way PMI might look if it were to use random sampling. He samples theories from those labelled as such in texts like the Oxford Dictionary of Biology. The motivation to develop PMI in a more sophisticated manner is admirable, but the result that most theories in his sample remain unrefuted does not tell against PMI because three of his four sources are not primarily historical in their focus. The agendas of scientific reference works may not include presenting theories accepted in the past as comprehensively as currently accepted theories, and so the latter may be overrepresented.