The science wars and the past's quiet voices

Social Studies of Science

Volumn 29, Issue 2, 1999, Pages 199-213

  MacKenzie, Donald ^a  

^a UNIVERSITY OF EDINBURGH   (United Kingdom)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 0033423759     PISSN: 03063127     EISSN: None     Source Type: Journal    
DOI: 10.1177/030631299029002002     Document Type: Review

References (55)

1. Philip A. Sullivan, 'An Engineer Dissects Two Case Studies: Hayles on Fluid Mechanics and MacKenzie on Statistics', in Noretta Koertge (ed.), A House Built on Sand: Exposing Postmodernist Myths about Science (New York: Oxford University Press, 1998), 71-98 [hereafter PS]; Donald MacKenzie, 'Statistical Theory and Social Interests: A Case-Study', Social Studies of Science, Vol. 8, No. 1 (February 1978), 35-83 [hereafter DM].
2. Philip A. Sullivan, 'An Engineer Dissects Two Case Studies: Hayles on Fluid Mechanics and MacKenzie on Statistics', in Noretta Koertge (ed.), A House Built on Sand: Exposing Postmodernist Myths about Science (New York: Oxford University Press, 1998), 71-98 [hereafter PS]; Donald MacKenzie, 'Statistical Theory and Social Interests: A Case-Study', Social Studies of Science, Vol. 8, No. 1 (February 1978), 35-83 [hereafter DM].
3. Donald MacKenzie, Statistics in Britain, 1865-1930: The Social Construction of Scientific Knowledge (Edinburgh: Edinburgh University Press, 1981), 225-26.
4. t from his critique.
5. t from his critique.
6. t from his critique.
7. DM, 47 and passim; David Heron, 'The Danger of Certain Formulae Suggested as Substitutes for the Correlation Coefficient', Biometrika,Vol. 8 (1911-12), 109-22, at 110.
8. G. Udny Yule, 'Karl Pearson, 1857-1936', Obituary Notices of the Royal Society of London, Vol. 2 (1936-38), 73-104, at 84.
9. N. Koertge, 'Part II: Myths, Metaphors, and Readings' (Introduction), in Koertge (ed.), op. cit. note 1, 57-58, at 58.
10. Karl Pearson, 'On the Inheritance of the Mental and Moral Characters in Man, and its Comparison with the Inheritance of the Physical Characters', Journal of the Anthropological Institute of Great Britain and Ireland, Vol. 33 (1903), 179-237, at 204 and 206-07 (emphasis in original).
11. Karl Pearson, 'Mathematical Contributions to the Theory of Evolution, III: Regression, Heredity, and Panmixia', Philosophical Transactions of the Royal Society of London, Series A, Vol. 187 (1896), 253-318, at 259.
12. G. Udny Yule to Major Greenwood, 3 April 1912 and 17 August 1920; Yule, op. cit. note 3, 150. In 1978 the Yule-Greenwood correspondence was in the possession of the latter's son, George B. Greenwood.
13. Thus the book that served as the 'bible' for many quantitatively-minded sociologists described Q as 'commonly used': Hubert M. Blalock, Jr, Social Statistics (Tokyo: McGraw-Hill, 2nd edn, 1972), 306. H.T. Reynolds, The Analysis of Cross-Classifications (New York: Free Press, 1977), 39, describes Q as 'One of the best-known measures of association in the social sciences', though Reynolds (40) is critical of the criterion of perfect association (either b = 0 or c = o, but not necessarily both) embodied in Q. As Sullivan acknowledges (op. cit. note 1, 90), Q is also the 2X2 case of a more modern coefficient, Goodman and Kruskal's γ. see Leo A. Goodman and William H. Kruskal, 'Measures of Association for Cross Classifications', Journal of the American Statistical Association, Vol. 49 (1954), 732-64, at 750. In addition, Q is a simple, monotone function of the maximum likelihood estimator (ad/bc) of the odds or cross-product ratio, which is fundamental to loglinear modelling. See A.W.F. Edwards, 'The Measure of Association in a 2 X 2 Table', Journal of the Royal Statistical Society, Series A, Vol. 126 (1963), 109-14, for a statement of the view that a measure should be a function of the observed cross-product ratio, and that this provides a criterion for distinguishing between what Edwards regards as acceptable coefficients, such as Q, and unacceptable ones, such as φ (see below). Of course, the question of how prevalent Q was in the 1970s among sociologists analyzing contingency tables could only be settled properly by a systematic survey, but I am afraid the intrinsic interest of the question does not seem to me to merit the necessary effort.
14. Thus the book that served as the 'bible' for many quantitatively-minded sociologists described Q as 'commonly used': Hubert M. Blalock, Jr, Social Statistics (Tokyo: McGraw-Hill, 2nd edn, 1972), 306. H.T. Reynolds, The Analysis of Cross-Classifications (New York: Free Press, 1977), 39, describes Q as 'One of the best-known measures of association in the social sciences', though Reynolds (40) is critical of the criterion of perfect association (either b = 0 or c = o, but not necessarily both) embodied in Q. As Sullivan acknowledges (op. cit. note 1, 90), Q is also the 2X2 case of a more modern coefficient, Goodman and Kruskal's γ. see Leo A. Goodman and William H. Kruskal, 'Measures of Association for Cross Classifications', Journal of the American Statistical Association, Vol. 49 (1954), 732-64, at 750. In addition, Q is a simple, monotone function of the maximum likelihood estimator (ad/bc) of the odds or cross-product ratio, which is fundamental to loglinear modelling. See A.W.F. Edwards, 'The Measure of Association in a 2 X 2 Table', Journal of the Royal Statistical Society, Series A, Vol. 126 (1963), 109-14, for a statement of the view that a measure should be a function of the observed cross-product ratio, and that this provides a criterion for distinguishing between what Edwards regards as acceptable coefficients, such as Q, and unacceptable ones, such as φ (see below). Of course, the question of how prevalent Q was in the 1970s among sociologists analyzing contingency tables could only be settled properly by a systematic survey, but I am afraid the intrinsic interest of the question does not seem to me to merit the necessary effort.
15. Thus the book that served as the 'bible' for many quantitatively-minded sociologists described Q as 'commonly used': Hubert M. Blalock, Jr, Social Statistics (Tokyo: McGraw-Hill, 2nd edn, 1972), 306. H.T. Reynolds, The Analysis of Cross-Classifications (New York: Free Press, 1977), 39, describes Q as 'One of the best-known measures of association in the social sciences', though Reynolds (40) is critical of the criterion of perfect association (either b = 0 or c = o, but not necessarily both) embodied in Q. As Sullivan acknowledges (op. cit. note 1, 90), Q is also the 2X2 case of a more modern coefficient, Goodman and Kruskal's γ. see Leo A. Goodman and William H. Kruskal, 'Measures of Association for Cross Classifications', Journal of the American Statistical Association, Vol. 49 (1954), 732-64, at 750. In addition, Q is a simple, monotone function of the maximum likelihood estimator (ad/bc) of the odds or cross-product ratio, which is fundamental to loglinear modelling. See A.W.F. Edwards, 'The Measure of Association in a 2 X 2 Table', Journal of the Royal Statistical Society, Series A, Vol. 126 (1963), 109-14, for a statement of the view that a measure should be a function of the observed cross-product ratio, and that this provides a criterion for distinguishing between what Edwards regards as acceptable coefficients, such as Q, and unacceptable ones, such as φ (see below). Of course, the question of how prevalent Q was in the 1970s among sociologists analyzing contingency tables could only be settled properly by a systematic survey, but I am afraid the intrinsic interest of the question does not seem to me to merit the necessary effort.
16. Thus the book that served as the 'bible' for many quantitatively-minded sociologists described Q as 'commonly used': Hubert M. Blalock, Jr, Social Statistics (Tokyo: McGraw-Hill, 2nd edn, 1972), 306. H.T. Reynolds, The Analysis of Cross-Classifications (New York: Free Press, 1977), 39, describes Q as 'One of the best-known measures of association in the social sciences', though Reynolds (40) is critical of the criterion of perfect association (either b = 0 or c = o, but not necessarily both) embodied in Q. As Sullivan acknowledges (op. cit. note 1, 90), Q is also the 2X2 case of a more modern coefficient, Goodman and Kruskal's γ. see Leo A. Goodman and William H. Kruskal, 'Measures of Association for Cross Classifications', Journal of the American Statistical Association, Vol. 49 (1954), 732-64, at 750. In addition, Q is a simple, monotone function of the maximum likelihood estimator (ad/bc) of the odds or cross-product ratio, which is fundamental to loglinear modelling. See A.W.F. Edwards, 'The Measure of Association in a 2 X 2 Table', Journal of the Royal Statistical Society, Series A, Vol. 126 (1963), 109-14, for a statement of the view that a measure should be a function of the observed cross-product ratio, and that this provides a criterion for distinguishing between what Edwards regards as acceptable coefficients, such as Q, and unacceptable ones, such as φ (see below). Of course, the question of how prevalent Q was in the 1970s among sociologists analyzing contingency tables could only be settled properly by a systematic survey, but I am afraid the intrinsic interest of the question does not seem to me to merit the necessary effort.
17. Maurice G. Kendall, The Advanced Theory of Statistics (London: Griffin, 4th edn, 1948), Vol. 1, 322. Kendall was a collaborator of Yule's, updating the latter's textbook: see G. Udny Yule and M. G. Kendall, An Introduction to the Theory of Statistics (London: Griffin, 14th edn, 1950). As Sullivan quite rightly points out [PS, 89], Yule became critical of his own coefficients. What is crucial, though, is that his grounds for this were different from those of Pearson and Heron: see Major Greenwood, Jr and G. Udny Yule, 'The Statistics of Anti-typhoid and Anti-cholera Inoculations, and the Interpretation of such Statistics in General', Proceedings of the Royal Society of Medicine (Epidemiology), Vol. 8 (1915), 113-90, as reprinted in Stuart & Kendall (eds), op. cit. note 3, 171-248, esp. 240-47.
18. Maurice G. Kendall, The Advanced Theory of Statistics (London: Griffin, 4th edn, 1948), Vol. 1, 322. Kendall was a collaborator of Yule's, updating the latter's textbook: see G. Udny Yule and M. G. Kendall, An Introduction to the Theory of Statistics (London: Griffin, 14th edn, 1950). As Sullivan quite rightly points out [PS, 89], Yule became critical of his own coefficients. What is crucial, though, is that his grounds for this were different from those of Pearson and Heron: see Major Greenwood, Jr and G. Udny Yule, 'The Statistics of Anti-typhoid and Anti-cholera Inoculations, and the Interpretation of such Statistics in General', Proceedings of the Royal Society of Medicine (Epidemiology), Vol. 8 (1915), 113-90, as reprinted in Stuart & Kendall (eds), op. cit. note 3, 171-248, esp. 240-47.
19. Maurice G. Kendall, The Advanced Theory of Statistics (London: Griffin, 4th edn, 1948), Vol. 1, 322. Kendall was a collaborator of Yule's, updating the latter's textbook: see G. Udny Yule and M. G. Kendall, An Introduction to the Theory of Statistics (London: Griffin, 14th edn, 1950). As Sullivan quite rightly points out [PS, 89], Yule became critical of his own coefficients. What is crucial, though, is that his grounds for this were different from those of Pearson and Heron: see Major Greenwood, Jr and G. Udny Yule, 'The Statistics of Anti-typhoid and Anti-cholera Inoculations, and the Interpretation of such Statistics in General', Proceedings of the Royal Society of Medicine (Epidemiology), Vol. 8 (1915), 113-90, as reprinted in Stuart & Kendall (eds), op. cit. note 3, 171-248, esp. 240-47.
20. Maurice G. Kendall, The Advanced Theory of Statistics (London: Griffin, 4th edn, 1948), Vol. 1, 322. Kendall was a collaborator of Yule's, updating the latter's textbook: see G. Udny Yule and M. G. Kendall, An Introduction to the Theory of Statistics (London: Griffin, 14th edn, 1950). As Sullivan quite rightly points out [PS, 89], Yule became critical of his own coefficients. What is crucial, though, is that his grounds for this were different from those of Pearson and Heron: see Major Greenwood, Jr and G. Udny Yule, 'The Statistics of Anti-typhoid and Anti-cholera Inoculations, and the Interpretation of such Statistics in General', Proceedings of the Royal Society of Medicine (Epidemiology), Vol. 8 (1915), 113-90, as reprinted in Stuart & Kendall (eds), op. cit. note 3, 171-248, esp. 240-47.
21. 2 is not a single distribution, but a family of distributions, and the question is which member of the family, which number of 'degrees of freedom', to use in the contingency tables discussed here: see MacKenzie, op. cit. note 2, 209-10.
22. 2 given in DM, 44 and 75, note 9. In larger tables, φ has the disadvantage that its maximum value can exceed 1.
23. Karl Pearson, 'Mathematical Contributions to the Theory of Evolution, XIII: On the Theory of Contingency and its Relation to Association and Normal Correlation', Drapers' Company Research Memoirs: Biometrie Series I (London: Dulau, 1904).
24. Ibid., 21; Karl Pearson, 'Mathematical Contributions to the Theory of Evolution, VII: On the Correlation of Characters not Quantitatively Measurable,' Philosophical Transactions of the Royal Society, Series A, Vol. 195 (1900), 1-47, at 12.
25. Ibid., 21; Karl Pearson, 'Mathematical Contributions to the Theory of Evolution, VII: On the Correlation of Characters not Quantitatively Measurable,' Philosophical Transactions of the Royal Society, Series A, Vol. 195 (1900), 1-47, at 12.
26. Pearson, op. cit. note 15, 7.
27. hk, but since the formulae for the two are identical, Sullivan's interpolation is not seriously misleading.
28. See Pearson & Heron, op. cit. note 18, 166, 176, 179-81, 183, 214-16, 230-32, 238-41, 247-50, 258 and 277.
29. Pearson & Heron, op. cit. note 18, 210 and 300.
30. Ibid., 207 and 212 (emphases in original deleted). For Pearson's ambivalent attitude to Mendelian theory, see M. Eileen Magnello, 'Karl Pearson's Mathematization of Inheritance: From Ancestral Heredity to Mendelian Genetics (1895-1909)', Annals of Science, Vol. 55 (1998), 35-94.
31. Ibid., 207 and 212 (emphases in original deleted). For Pearson's ambivalent attitude to Mendelian theory, see M. Eileen Magnello, 'Karl Pearson's Mathematization of Inheritance: From Ancestral Heredity to Mendelian Genetics (1895-1909)', Annals of Science, Vol. 55 (1998), 35-94.
32. ps (which involved treating individuals in a given category as in a certain sense identical)' [DM, 63-64]. Pearson rejected not the coefficient in itself (as noted in the text, he used it before Yule did), but the assumptions involved in what he believed to be its indiscriminate use.
33. Stephen M. Stigler, The History of Statistics: The Measurement of Uncertainty before 1900 (Cambridge, MA: Harvard University Press, 1986), Chapter 10.
34. M. Eileen Magnello, 'The Non-Correlation of Biometrics and Eugenics: Rival Forms of Laboratory Work in Karl Pearson's Career at University College London', History of Science (forthcoming).
35. Koertge, op. cit. note 6, 57-58.
36. Sullivan cites Murray R. Spiegel, Schaum's Outline of Theory and Problems in Statistics (New York: McGraw-Hill, 2nd edn, 1988), 246-48.
37. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
38. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
39. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
40. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
41. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
42. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
43. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
44. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
45. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
46. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
47. 2 (see, for example, Agresti, op. cit., 75, problem 3.24, and 442, problem 12.13), and some modern statisticians have to be reminded what φ is before they can express a view on it! From the viewpoint of the modelling approaches that now dominate, φ seems a 'dead end': Fienberg, email message to D. MacKenzie (18 November 1998). I am grateful to Professors Fienberg, Agresti and Leonard, and to Fionntán Hurley, for helpful comments on current thinking.
48. Goodman & Kruskal, op. cit. note 10, 740 and 763.
49. MacKenzie, op. cit. note 2, 225-26.
50. Of course, excellent courses can be designed that cover both a subject's historical evolution and its current status. Time restrictions in a curriculum that has to cover the full range of sociological research methods make that approach to the teaching of statistics impractical in my case.
51. MacKenzie, op. cit. note 2, 117-19 and 225-26. Note that though 'the statistical literature of the past 25 years on this topic would indicate that Yule's position now dominates' (Fienberg, Analysis of Cross-Classified Categorical Data, op. cit. note 27, 5), interest in Pearson's underlying continuous variable model is not dead (ibid., 4), and efforts are still made to produce models that in some sense reconcile the approaches of Pearson and Yule. Thus Alan Agresti (appropriately for a student of historian of statistics Stephen Stigler!) has produced an adaptation of loglinear models that 'fits well when there is an underlying continuous normal distribution' (op. cit. note 27, 27 and 261-97). See also Leo A. Goodman, 'A Single General Method for the Analysis of Cross-Classified Data: Reconciliation and Synthesis of Some Methods of Pearson, Yule, and Fisher, and also some Methods of Correspondence Analysis and Association Analysis', Journal of the American Statistical Association, Vol. 91 (1996), 408-28.
52. MacKenzie, op. cit. note 2, 117-19 and 225-26. Note that though 'the statistical literature of the past 25 years on this topic would indicate that Yule's position now dominates' (Fienberg, Analysis of Cross-Classified Categorical Data, op. cit. note 27, 5), interest in Pearson's underlying continuous variable model is not dead (ibid., 4), and efforts are still made to produce models that in some sense reconcile the approaches of Pearson and Yule. Thus Alan Agresti (appropriately for a student of historian of statistics Stephen Stigler!) has produced an adaptation of loglinear models that 'fits well when there is an underlying continuous normal distribution' (op. cit. note 27, 27 and 261-97). See also Leo A. Goodman, 'A Single General Method for the Analysis of Cross-Classified Data: Reconciliation and Synthesis of Some Methods of Pearson, Yule, and Fisher, and also some Methods of Correspondence Analysis and Association Analysis', Journal of the American Statistical Association, Vol. 91 (1996), 408-28.
53. MacKenzie, op. cit. note 2, 117-19 and 225-26. Note that though 'the statistical literature of the past 25 years on this topic would indicate that Yule's position now dominates' (Fienberg, Analysis of Cross-Classified Categorical Data, op. cit. note 27, 5), interest in Pearson's underlying continuous variable model is not dead (ibid., 4), and efforts are still made to produce models that in some sense reconcile the approaches of Pearson and Yule. Thus Alan Agresti (appropriately for a student of historian of statistics Stephen Stigler!) has produced an adaptation of loglinear models that 'fits well when there is an underlying continuous normal distribution' (op. cit. note 27, 27 and 261-97). See also Leo A. Goodman, 'A Single General Method for the Analysis of Cross-Classified Data: Reconciliation and Synthesis of Some Methods of Pearson, Yule, and Fisher, and also some Methods of Correspondence Analysis and Association Analysis', Journal of the American Statistical Association, Vol. 91 (1996), 408-28.
54. MacKenzie, op. cit. note 2, 117-19 and 225-26. Note that though 'the statistical literature of the past 25 years on this topic would indicate that Yule's position now dominates' (Fienberg, Analysis of Cross-Classified Categorical Data, op. cit. note 27, 5), interest in Pearson's underlying continuous variable model is not dead (ibid., 4), and efforts are still made to produce models that in some sense reconcile the approaches of Pearson and Yule. Thus Alan Agresti (appropriately for a student of historian of statistics Stephen Stigler!) has produced an adaptation of loglinear models that 'fits well when there is an underlying continuous normal distribution' (op. cit. note 27, 27 and 261-97). See also Leo A. Goodman, 'A Single General Method for the Analysis of Cross-Classified Data: Reconciliation and Synthesis of Some Methods of Pearson, Yule, and Fisher, and also some Methods of Correspondence Analysis and Association Analysis', Journal of the American Statistical Association, Vol. 91 (1996), 408-28.
55. MacKenzie, op. cit. note 2, 117-19 and 225-26. Note that though 'the statistical literature of the past 25 years on this topic would indicate that Yule's position now dominates' (Fienberg, Analysis of Cross-Classified Categorical Data, op. cit. note 27, 5), interest in Pearson's underlying continuous variable model is not dead (ibid., 4), and efforts are still made to produce models that in some sense reconcile the approaches of Pearson and Yule. Thus Alan Agresti (appropriately for a student of historian of statistics Stephen Stigler!) has produced an adaptation of loglinear models that 'fits well when there is an underlying continuous normal distribution' (op. cit. note 27, 27 and 261-97). See also Leo A. Goodman, 'A Single General Method for the Analysis of Cross-Classified Data: Reconciliation and Synthesis of Some Methods of Pearson, Yule, and Fisher, and also some Methods of Correspondence Analysis and Association Analysis', Journal of the American Statistical Association, Vol. 91 (1996), 408-28.