Subject: Re C++, Linet study probabilities...... Date: Sun, 14 Jun 1998 003316 -0500 (CDT) From: "Roy L. Beavers" <rbeavers@llion.org> To: emfguru@hotmail.com -------------------------------------------------- ---------- Forwarded message ---------- Date: Sun, 14 Jun 1998 05:10:00 +0200 From: Christoph ReussTo: "Roy L. Beavers" Cc: Edward Maxey Subject: Re: C++ "Linet" study probabilities...... Edward Maxey wrote: > Several months ago results from a BASIC study showing that the > probability of 60 Hz EMFs being causal to childhood leukemia (the Linet > study) was 200,000 to one. Recently questions have come up about the > integrity of the random number function in MICROSOFT's BASIC. > Accordingly the program was written in C++ using the "rand" function. > This function "uses a multiplicative congruential random number generator" > with a period of 2^32. It is available on UNIX and is defined in ANSI C. [snip] > ***This run showed a 5,408 to one probability that EMFs are causal > to childhood acute lymphoblastic leukemia.*** Computation Theory tells us that no discrete algorithm run on a finite state machine (without external true random sources) can provide _true_ random numbers, but only pseudo-random numbers (whether in BASIC or in C++). And even if the numbers produced by the BASIC or C++ 'random'- generator were truly random, one instance (one 'run') of your program would only provide one possible outcome of the random experiment, and to calculate the true probability would require an infinite number of runs. Thus, these programs can only provide approximations of the true probability. That's why we prefer theoretical calculation in simple statistical problems such as your "heads and baskets" example: The probability of 267 or less heads appearing after a basket of 624 coins was overturned is: P [H<=267] = 1/2^624 * SUM(k=0..267) ( 624! / (624-k)! / k! ) = 0.0001795451715 = 1 / 5569.629033 Thus, the actual odds are 5,570 to one. Your C++ program came pretty close with 5,408. :-) But the problem with epidemiological studies like the Linet study is not to calculate probabilities according to known simple statistical models, but to find those complex statistical models that 'model' (fit) the even more complex reality best. (And to model this reality ***correctly***, not misleadingly like the Linet study did !!). In other words, the problem is not to find the correct mapping "model-->result" (that's easy), but to find the correct mapping "reality-->model". No BASIC and no C++ program can do this for us either... Cheerio, Chris ___________________________________________________________________________ "The history of cancer research has been a history of curing [?? causing!!] cancer in the mouse. We have cured mice of cancer for decades -- and it simply didn't work in humans." -- Dr. Richard Klausner, NCI director (1998) Archive provided courtesy of WaveGuide, http://www.wave-guide.org Reprinted with permission of Roy Beavers, http://www.feb.se/EMF-L/EMF-L.html