Use of Iverson's language APL for statistical computing
Francis John Anscombe · NASA Technical Reports Server (NASA) · 1968
There has certainly been some progress toward realizing the potentialities of the computer i n s t a t i s t i c a l work.The main reason why more progress has not been made is, presumably, that camputers are hard t o communicate with.The old desk calculator had some aerits not found readily i n a large computer, operator of a desk calculator sometimes obtained unexpected insight, and could at any moment modify h i s plan of calculation.program i s liable t o seem uninformative and inflexible or, if informative and flexible, then confusing.Many new techniques have been tried.I n laboring over the figures, the By contrast a computer S t a t i s t i c a l analysis i s experimental.It has always been so t o some extent, it w i l l become more so i n the future.When a s t a t i s t i c i a n is presented w i t h some data, he may know what analytic steps he would like to see carried out first, but he cannot know u n t i l he has explored the consequences of preliminary analyses what he w i l l come t o consider a satisfactory final analysis of the data.must be adjusted t o f i t the data; the adequacy of theoretical descriptions or "models" must be assessed.w e l l defined a t the outset than the task of, say, tabulating a mathematical function. Good s t a t i s t i c a l analysis i s done i n steps. MethodsThe task of s t a t i s t i c a l analysis is much less Interaction between the investigator acrid the computer is required.The persons who have occasion t o use computing equipment for statist i c a l analysis are very diverse i n their knowledge of, and interest in, both s t a t i s t i c s and computing.intolerable t o another.For example: A f a c i l i t y that i s attractive t o one may be ( a ) Consider a researcher who works i n a f i e l d t h a t yields s t a t i s t i c a l problems, but who i s not himself primarily a statistician.He wishes t o use good s t a t i s t i c a l tools t o better h i s understanding of his field.interested i n statisticalmethcds f o r t h e i r own sake, and does net wish t o experiment with them (more than he has t o ) .The l e s s he must be cencerned with the techaicalities of s t & t i s t i c a l calculations the better.He is not (b) A t another extreme, consider a statistician, interested in the development of s t a t i s t i c a l procedures, interested i n relevant mathematics, interested i n particular bodies of data mainly as examples of statistical.prQblems.but only if he f u l l y understands what the Frogran does.with the computer i s easy enough, he generally prefers t o write his own programs, so that he may do exactly what he wishes, rather than what someone e l s e has wished.He may be happy t o use a computer program written by someone else, If communication Many The following matters have not been explained: things have not been said.how t o sign on and off, how t o move material i n and out of storage, how t o correct an error i n typing, how t o add to, emend or "edit" a function definition (program), how t o interpret error reports, mount of unused space remaining i n the active workspace, ( b ) users of the system currently signed on, ( c ) time used since sign-on.