R: log-normal distribution and shapiro test
Hi: Thanks for your answer. Do you know how to test whether the data would fit to a gamma-distribution? How can I call fBasics? Note: I installed R-language on my Macintosh today; I have used the binary -- pre compiled -- package. Some of the R-help facilties do not function on my Mac. Again to my data: How can I compute the skew? I think I lack some basic packages - right? The curious things actually is that the median and the mean are quite similar, e.g. 0.19 and 0.2 respectively; the skew is about 1.0 (I calculated the skew by my own computer code in Bigloo). The problem actually is: my boss expects from me that I make some tests; personally I am a bit generous and everything is a Gaussian or log-Gaussian distribution, because how can I be sure that the underlying data to not have any serious flaws? Statistics is black art - right? Regards, S. Gonzi
Vito Ricci wrote:
Hi, from what you're writing: "The logaritmic transformation "shapiro.test(log10(y))" says: W=0.9773, p-value= 2.512e-05." it seems the log-values are not distributed normally and so original data are not distributed like a log-normal: the p-value is extremally small! Other tests for normality are available in package: nortest compare the log-transformation of your ecdf with normal cdf: see ? ecdf use qqnorm and qqplot did you calculate skewness and kurtosis? see in package fBasics. I remember to you that the log-normal distribution as three parameters: shape parameter, location parameter and scale parameter. Transfroming by the simple log, you are missing the location parameter, or implicitely you assuming is =0. See: http://www.itl.nist.gov/div898/handbook/eda/section3/eda3669.htm for more news about log-normal distribution. I hope I give you a little help. Best Vito you wrote: Hello: Yes I know that sort of questions comes up quite often. But with all due respect I din't find how to perform what I want. I am searching archives and bowsing manuals but it isn't there, though, it is a ridiculous simple task for the experienced R user. I have data and can do the following with them: == hist(y, prob=TRUE) lines(density(y,bw=0.03) == The result actually is a nice histogram superimposed by a line plot. The histogram is a bit skewed to the left. My assumption actually is that a log-normal transformation would cure the problem. But how the hell can one plot such a density function or Gaussian function which has logarithmic scales on x axis. For example I tried: == plot(hist(y),log="x") or plot(hist(log10(y)),log="x") == But with no avail. I want my axis like: 1,10,100 What would be other methods to test whether the data are logaritmically distributed. A last question to the Shapiro-Wilk test. Were can I get critical parameters? I mean I get for my distribution: W=0.9686, p-value=6.887e-07. What does that mean? Yes I have got some books about statics, but none of them says what one should do with the values then. The logaritmic transformation "shapiro.test(log10(y))" says: W=0.9773, p-value= 2.512e-05. Sorry for disturbing you. Although, it is really no homework. I need it for my Phd in physics; after a lengthy computation on the computer I would like to go to see whether the outputs are log-normal or normal distributed. Regards, Siegfried Gonzi == University of Graz Institute for Physics Tel.: ++43-316-380-8620 ===== Diventare costruttori di soluzioni Became solutions' constructors "The business of the statistician is to catalyze the scientific learning process." George E. P. Box Visitate il portale http://www.modugno.it/ e in particolare la sezione su Palese http://www.modugno.it/archivio/cat_palese.shtml
___________________________________