Goodness of fit
in [Matlab]
|
Prev: gradient
Next: Extract all non zero rows from array
From: Frank on 8 Mar 2010 10:40 My technique is to do 1) plot the data on a probability plot for a particular distribution (normal, gamma, weibull, etc.) and 2) perform a GOF test using methods mentioned above (chi-square, ks test, anderson-darling, etc.). In 1), the data should lie on a straight line. If not, I stop there and go to the next distribution. If the data is more or less straight for several distributions, I go to 2) to see if there are any significant differences in the fits. Majority of the time, it's just a judgment call to which distribution to use. I recommend D'Agostino and Stephens, "Goodness-of-Fit Techniques".
From: dpb on 8 Mar 2010 16:18 Frank wrote: > My technique is to do 1) plot the data on a probability plot for a > particular distribution (normal, gamma, weibull, etc.) and 2) perform > a GOF test using methods mentioned above (chi-square, ks test, > anderson-darling, etc.). In 1), the data should lie on a straight > line. If not, I stop there and go to the next distribution. If the > data is more or less straight for several distributions, I go to 2) to > see if there are any significant differences in the fits. Majority of > the time, it's just a judgment call to which distribution to use. That's pretty much a description of process H&S discuss... > I recommend D'Agostino and Stephens, "Goodness-of-Fit Techniques". Ah, good recommendation--don't have it on shelf and didn't think of it earlier...more covered than are in H&S (which has a bent towards distributions used in reliability models other than the ubiquitous normal). --
From: Gulcin Tekin on 16 Mar 2010 04:39 Thank u very much for your answers... Best regards; Gülçin. dpb <none(a)non.net> wrote in message <hn3pl2$bmn$1(a)news.eternal-september.org>... > Frank wrote: > > My technique is to do 1) plot the data on a probability plot for a > > particular distribution (normal, gamma, weibull, etc.) and 2) perform > > a GOF test using methods mentioned above (chi-square, ks test, > > anderson-darling, etc.). In 1), the data should lie on a straight > > line. If not, I stop there and go to the next distribution. If the > > data is more or less straight for several distributions, I go to 2) to > > see if there are any significant differences in the fits. Majority of > > the time, it's just a judgment call to which distribution to use. > > That's pretty much a description of process H&S discuss... > > > I recommend D'Agostino and Stephens, "Goodness-of-Fit Techniques". > > Ah, good recommendation--don't have it on shelf and didn't think of it > earlier...more covered than are in H&S (which has a bent towards > distributions used in reliability models other than the ubiquitous normal). > > --
From: dpb on 16 Mar 2010 12:31
Gulcin Tekin wrote: > Thank u very much for your answers... > Best regards; > Gülçin. > .... >> Frank wrote: >> > My technique is to do 1) plot the data on a probability plot for a >> > particular distribution (normal, gamma, weibull, etc.) ... >> > ... In 1), the data should lie on a straight >> > line. If not, I stop there and go to the next distribution. If the >> > data is more or less straight for several distributions, I go to 2) ... I'd again stress the above and particularly look (visually) at the deviation(s) towards the tails--there's where the brunt of what really distinguishes one distribution from another is. Of course, there's where you have less data, undoubtedly, as well which is why it's such a difficult question to answer unequivocally. -- |