I would like to thank Prof. Brush for his helpful comments on an eariler version of this paper.
 I myself use the expression 'the plate tectonics revolution' to describe the change, but I do not intend to commit to the Kuhnian sense of 'revolution.' This is supposed to be a neutral expression, and used just for convenience.
 For a more truly Whiggish account of this history, see Allègre 1988; for a criticism of that book, see Leveson 1991.
 Hallam 1983 shows a little change in his position. He still maintains that the Kuhnian view explains this revolution well, but he admits that Lakatos' approach or T.C. Chamberlin's method of multiple working hypotheses provide better account for the Ice Age controversy (Hallam 1983, 165-168).
 Curiously, Frankel does not mention L.Laudan's name in this paper, even though obviously his argument is inspired by Laudan's three categories of problems, i.e. solved, unsolved, and anomalous problems.
6 It seems to me that this is a perfect case for Whewell's 'consilience of inductions' mentioned by Ruse. If so, Frankel undermines his own objection to Ruse's argument for methodological continuity.
 The expantion theory was proposed in the 1950s to explain the peculier structure of sea-floor (oceanic ridges etc.) within the fixist tradition by the expansion of the earth. See LeGrand 1988, 193-195 for details.
 Instead, Mareschal proposes another set of programs, namely the geophysical program, the geological program and the drift program (Mareschal 1987, 195). This is an interesting proposal, but I do not have the space to analyze it.
 Prof. Brush pointed out to me that Oreskes can be classified as a historian of geology with strong geological backgound.
 Formally put, this means that for the given hypothesis T and fact F, P(F/T) is great enough, while for every other rival hypothesis Ti, P(F/Ti) is very small. When someone subscribes to a rival hypothesis Ti in a rival research programme, The person' s prior probability must be something like P(T)<<P(Ti). To convert the person with the fact F, i.e. to be P(T/F)>P(Ti/F), something like the above condition is sufficient. Frankel's account, which does not require the smallness of P(F/Ti) of Ti's in other research programmes, does not provide enough reason for conversion for the subscriber of other research programmes. See Nunan 1984, 279 n.18.
 Stewart is not a social constructivist, though he expresses sympathy with the position. He is also influenced by Mertonian sociolgy. See Stewart 1990, 189-204.
 R. Laudan 1987 questioned this result doubting the existence of some hidden variable, and Stewart 1987 answered very effectively. So this result seems to stand the criticism.