Gottfried, The problem with your response is that you conclude an interesting point by refusing to go deeper. This is exactly the attitude that may have produced a highly ineffective and unscientific discipline of statistics. Think about the mileage example. Those cars were not thrown together randomly. Engineers who knew exactly what they were doing designed the cars to have targeted gasoline conservation. The decision to create better cars is probably the real cause of the automobile specifications. Because the engineers knew what they were doing, it is very reasonable to expect they could produced what they wished. Thus, gas mileage is very likely to be confounded with the decision to make better cars.
This issue is nothing new. Aristotle described final causation as the intention that produces designs which bring about results for some purpose. When it comes to engineering automobiles, the final cause is likely to be extremely highly correlated with the result. This might not be the case with other things. For example, I did an experiment years ago in which I asked students to rate their willinginess to work various summer jobs and their expectations of getting such work. CR suggested that the expected outcomes were partially caused by the willingness to pursue the jobs. Here is a case where willingness to pursue and expectations of work would not be confounded, since people try for jobs that they probably will not get. Thus, we see the final cause (willingness) determing the expected job. If the expectations were certainties, then the willingness and outcome would be perfectly correlated/confounded... as it probably is with engineers and automobiles. You are willing to make comments but you do not show the dedication of a real scientist. Tell us why CR works so well for simulations but not for convenience data? If you doubt it is convenience data, explain how the bladder cancer data has had the many socioeconomic, dietary, and other factors that might be confounded with bladder cancer removed? These things have not been screened out of Steve's data. With experiments we would be a little more convinced. The confounding has not been such a big issue to statisticians in the past who use correlations, sem, etc because there was no way of detecting such odd things as bladder cancer apparently causing smoking. But this could be explained by confounding, if we only had the tenacity to bother to ask just how confounded our data is. Statisticians get away with being mediocre to down right worthless at measurement. The very idea of normally distributed cell sizes in ANOVA is so strange to the people on this and other lists, that you can not even conceive of its parallel with what people do in convenience samples all the time. This blindness points to incompetence in the measurement field. Its not just you. Its the way we are educated. Its most of us in the social/beahvioral sciences. It was me before CR forced me to ask questions that the standard dogma glossed over. But once you see it, it is obvious! What kind of statistician would sample so that the CELL SIZES (not dv values) across a one way ANOVA were normally distributed???? No one with any intelleigence and an undergraduate education would do this honestly. TENACITY. CURIOSITY. The willingness to explore. Sufficient temporary tolerance of ambiguity to allow more precise study and clarification... versus satifycing. These are the things that make scientists and creative statisticians. Snap out of your negativity. Tell us why CR does not work. Saying it is not robust is too vague. Such nonrobustness could be a function of sloppy data and confounding. Where in the continuum between logic, simulations, convenience samples and tea leaf reading does CR or the data break down? Then, from a logical/mathematical perspective tell us WHY? Bill p.s. (I have to leave town for a few days later tonight and may not answer until I get back. But the issues will remain, for those who have the courage and the scientific temperament to embrace them. ----- Original Message ----- From: "Gottfried Helms" <[EMAIL PROTECTED]> Newsgroups: sci.stat.edu Sent: Thursday, September 19, 2002 10:15 AM Subject: Re: Applying Corresponding Regressions across five data sets > "Simon, Steve, PhD" schrieb: > > > Hi Steve, > > one more, maybe small, addendum. From the question: "what > causes what", when you correlate the amount of shopping > over the monthes and the coming of christmas, one can > think of an analogy. One could consider, whether the > expectation to be able to have great mileage could cause > the decision to buy a big car. > Then there is a sampling-effect. > > (But I'm not planning to go deeper into this matter...) > > Regards- > > Gottfried. "Gottfried Helms" <[EMAIL PROTECTED]> wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > "Simon, Steve, PhD" schrieb: > > > Hi Steve, > > one more, maybe small, addendum. From the question: "what > causes what", when you correlate the amount of shopping > over the monthes and the coming of christmas, one can > think of an analogy. One could consider, whether the > expectation to be able to have great mileage could cause > the decision to buy a big car. > Then there is a sampling-effect. > > (But I'm not planning to go deeper into this matter...) > > Regards- > > Gottfried. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
