Linguistic Analysis of a Statistical Controversy
Irwin D. J. Bross · The American Statistician · 1963
About a year ago (THE AMERICAN STATISTICIAN, June 1961) 1 proposed that the Neo-Bayesian vs. Orthodox controversy be put to an empirical test by having representatives of both sides play a with an electronic computer serving as a referee. Although the challenge game has attracted no players, I have received some replies both in correspondence and in print (John W. Pratt, issue of April 1962) which explained at some length why the respondent didn't want to play the game. By using the replies as text samples for a linguistic experiment it will be possible, I believe, to go a long way toward clarifying the issues in the controversy. Since I would like to enable you to carry out the experiment for yourself and to draw your own conclusions, I will make only a few informal remarks about linguistic analysis at this point. In reading the forthcoming text sample, focus on the specialized (i.e. jargon) rather than the content of the message. The question is not whether you agree with the opinions expressed but rather whether you would express the opinion in the way that the author did. In other words are your linguistic habits (speech patterns) similar to those of the author? The linguists have a useful descriptive term for a dissimilarity of speech patterns; they call it an incongruence. The incongruences here might appear as an isolated word, the turn of a phrase, or the tone of the statement-anything about the test sample that sounds odd or strange to you. In this informal analysis focus on the sound-don't worry if the incongruence appears trivial, if you cannot pin down just where the incongruence lies, of if your judgment seems and subjective. Underline anything that doesn't sound like something that you would say. Here is a text sample from Pratt's letter: (2) Now that I have ceased pretending to be impartial, I may point out that no connected argument leading to orthodox methods has ever been advanced. Neyman and Pearson contributed vitally to our understanding by their formulation of statistical problems, but they have never claimed their methods were more than ad hoc procedures with some pleasant properties. Their methods, while extremely ingenious and useful, are not completely satisfactory, let alone uniquely objective and scientific. After you have underlined by sound try rereading the sample for context. In other words, the text sample criticises the work of Neyman and Pearson. What seems to be the grounds (or basis) for the criticism? What is the nature of the objections? Apart from content, would you criticise statistical methods along the lines used in the text sample? Do the comments seem meaninlgful to you or do you find them vague or hard to see? The point of the little experiment is simply this: If you have underlined the passage in several places or find the context somewhat confusing then you are not a 'native speaker of the jargon that is used in this text sample. If, in your daily work, you actually use the significance tests and confidence intervals that are the target of this criticism, then I venture to predict that you will have noted incongruences in this sample of NeoBayesian jargon. When the people on opposite sides of a controversy don't speak the same language then it will usually turn out that a large part of the argument arises from basic linguistic difficulties. Here, for instance, we are supposedly dealing with a methodological controversy. What I am suggesting is that methodology plays only a minor role in the controversy-that most of the argument arises from drastic differences in the meanings of key words (e.g. scientific) in the respective jargons. To clarify these matters of meaning, we can take advantage of the approach and findings of the scientific studies of living spoken languages that have been carried out in Modern Linguistics.