12.07.2015 Views

1 Studies in the History of Statistics and Probability ... - Sheynin, Oscar

1 Studies in the History of Statistics and Probability ... - Sheynin, Oscar

1 Studies in the History of Statistics and Probability ... - Sheynin, Oscar

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>of</strong> various groups <strong>of</strong> mach<strong>in</strong>es, <strong>and</strong> this provides us a test forestimat<strong>in</strong>g <strong>the</strong> reliability <strong>of</strong> <strong>the</strong> <strong>in</strong>sulation. Purely statistical methodscerta<strong>in</strong>ly do not concern <strong>the</strong> improvement <strong>of</strong> that reliability which is atechnological problem. But at least we may say whe<strong>the</strong>r <strong>the</strong> reliability<strong>of</strong> <strong>in</strong>sulation had changed <strong>and</strong> <strong>in</strong> which direction or that it rema<strong>in</strong>ed asit was previously. This is <strong>the</strong> practical significance <strong>of</strong> <strong>the</strong> work donewhich would not be so important had <strong>the</strong> comparatively high statisticalhomogeneity <strong>of</strong> <strong>the</strong> <strong>in</strong>sulation not been established. [...]2.4. The naked eye study. We had assumed that smooth<strong>in</strong>g bypolynomials is usually successful because <strong>the</strong> data for that treatment isselected beforeh<strong>and</strong> by naked eye. It would have been improper to failto mention that physicists <strong>and</strong> eng<strong>in</strong>eers also <strong>of</strong>ten perform <strong>the</strong>smooth<strong>in</strong>g itself by naked eye without apply<strong>in</strong>g <strong>the</strong> method <strong>of</strong> leastsquares. And how do we decide that <strong>the</strong> smooth<strong>in</strong>g <strong>in</strong> a given case wassuccessful? Perhaps because <strong>the</strong> curve derived by least squares passesexactly where it would have been if drawn without apply<strong>in</strong>g thatmethod?An experimental smooth<strong>in</strong>g by naked eye <strong>of</strong> <strong>the</strong> broken l<strong>in</strong>e <strong>in</strong> § 2.2was carried out. Participants were ma<strong>the</strong>maticians, workers at astatistical laboratory, <strong>and</strong> eng<strong>in</strong>eers. Each received a list <strong>of</strong> paper withonly that l<strong>in</strong>e shown [...]. The results achieved by an overwhelm<strong>in</strong>gmajority were very good. Fifteen out <strong>of</strong> sixteen <strong>of</strong> those participat<strong>in</strong>ghad almost completely drawn <strong>the</strong>ir curves between <strong>the</strong> two curves,p 1 (t) <strong>and</strong> p 3 (t) as shown on Fig. 2. [...]I. V. Girsanov, <strong>the</strong> chief <strong>of</strong> one <strong>of</strong> <strong>the</strong> sections <strong>of</strong> <strong>the</strong> statisticallaboratory, achieved <strong>the</strong> best result; he unfortunately perished <strong>in</strong> a latertourist mounta<strong>in</strong> tour. [...] In general, <strong>the</strong> results <strong>of</strong> smooth<strong>in</strong>g bynaked eye are quite comparable <strong>in</strong> precision with <strong>the</strong> method <strong>of</strong> leastsquares. Had we been only <strong>in</strong>terested <strong>in</strong> curve p 2 (t), we could havewell drawn it without any calculations. However, a thorough statisticaltreatment dem<strong>and</strong>s an estimation <strong>of</strong> precision as well for which astatistical model <strong>and</strong> science <strong>in</strong> general are necessary.Thus, when estimat<strong>in</strong>g <strong>the</strong> probability <strong>of</strong> success <strong>in</strong> Bernoulli trials,we turn to frequencies, but for underst<strong>and</strong><strong>in</strong>g how large can <strong>the</strong>deviations <strong>of</strong> frequency from probability be, we should, first, consider<strong>the</strong> trials <strong>in</strong>dependent (<strong>the</strong> statistical model) <strong>and</strong> second, apply <strong>the</strong> DeMoivre – Laplace <strong>the</strong>orem which (however done) is proven <strong>in</strong> acomplicated manner [<strong>in</strong> essence, by <strong>the</strong> former <strong>in</strong> 1733] <strong>and</strong> this isundoubtedly science.When smooth<strong>in</strong>g a broken l<strong>in</strong>e by naked eye, we do not even haveto know <strong>the</strong> number <strong>of</strong> observations used for calculat<strong>in</strong>g its po<strong>in</strong>ts [...]<strong>and</strong> anyway it is impossible to <strong>in</strong>dicate <strong>the</strong> confidence region for <strong>the</strong>curve sought. Here, we need all <strong>the</strong> science connected with <strong>the</strong> method<strong>of</strong> least squares <strong>and</strong> still <strong>the</strong> almost complete co<strong>in</strong>cidence <strong>of</strong> <strong>the</strong> areashown on Fig.2 with that between <strong>the</strong> curves p 1 (t) <strong>and</strong> p 3 (t) dem<strong>and</strong>s tobe somehow expla<strong>in</strong>ed.Note, however, that for small values <strong>of</strong> t that first area is somewhatnarrower than <strong>the</strong> second one whereas that latter, as shown bycalculation, is 1.5 – 2 times narrower <strong>the</strong>re than a thoroughlyconstructed confidence region with <strong>the</strong> usual confidence coefficient <strong>of</strong>0.70 – 0. 95. This means that <strong>the</strong> <strong>in</strong>def<strong>in</strong>iteness <strong>of</strong> <strong>the</strong> naked eye63

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!