On Wed, 12 Mar 2003, VOLTOLINI wrote, inter alia: > Reading some stats textbooks I can find that.... > > 1 - An experiment is an experiment because there is an independent > variable (a factor or treatment) with diferent levels (a qualitative > variable)
That a variable "has different levels" does not imply that it is [inherently] a "qualitative" (by which I understand "categorical") variable: only that a finite, usually smallish, number of different levels have been selected and imposed by the experimenter. For example (I found data of this sort in a statistics text once, and used it in classes on multiple regression for exam questions on orthogonal polynomials and logarithmic transformation) an experimenter interested in the growth rate(s) of chick embryos might choose to sacrifice eggs at ages of 6, 7, 8, ..., 16 days, rather than at randomly chosen ages in the range of 6 to 16 days. Clearly these numbers have quantitative meaning, as well as defining 11 different "levels" of an independent variable called (say) "age". (Equally clearly, although a simple one-way ANOVA would reveal that the mean weights were not all equal, one would really rather analyze the data via multiple regression, in order to model the growth function: which for the data given was very close to an exponential growth curve.) > 2 - But...... following this reasoning....... people working with > regression are never working with experiments? Non sequitur. Both regression and analysis of variance are subsets of the general linear model. One's choice of style of analysis depends in part on the nature of the data (if the numerical codes don't have quantitative meaning -- if they identified religious preference, for example -- it would be silly to ascribe interval-scale (or even ordinal- scale) qualities to them), and in part on the nature of the question(s) one wants to ask of the data. (It would also be silly not to ascribe quantitative meaning to the "age" variable in the chick embryo example above, even though one might have equal sample sizes at all the selected ages.) > Then, what is precisely an experiment ???? In general, an enquiry in which the investigator (called an experimenter in this context) imposes particular values on one or more "independent" variables, either by manipulating the values of the variable (as in assigning some entities to one experimental condition and other entities to one or more other experimental conditions) or by selecting particular values of the variable (as in choosing to observe the dry weight of chick embryos at 11 different ages, and not at intermediate ages although in principle one could have chosen to do so). In brief, an experiment is an investigation in which the investigator CONTROLS the values of one or more independent variables. The control required may be exerted by manipulating the variable directly, or by selecting data with specific values of the variable. Some folks would extend the idea of "control" to measuring an independent variable that is not amenable to more direct control. > I would like to have your comments because I am feeling that many > people are using statistical theory without asking themselves if > they agree or believe in what they are reading, writing and > saying...... Yes, well, that's certainly worth worrying about. I hope this will have been helpful. -- DFB. ----------------------------------------------------------------------- Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
