[R] Re lative novice: Working with fitdistr(MASS): 3 questions

Ted Byers r.ted.byers at gmail.com
Mon Sep 22 17:48:18 CEST 2008


OK, I am now at the point where I can use fitdistr to obtain a fit of one of
the standard distributions to mydata.

It is quite remarkable how different the parameters are for different
samples through from the same system.  Clearly the system itself is not
stationary.

Anyway, question 1:  I require a visual perspective of the fit I get.  I can
use hist.scott to get a hisogram (and just have to figure out how to get
finer granularity from it - my samples are taken weekly, but the histogram
bars cover two weeks of data and the most interesting changes happen in the
first three to four weeks - after that things slow down tremendously), but
how would I overlay a plot of the best distribution I get from fitdistr over
it?

Second question: I don't see anything in the documentation for fitdistr that
says anything about using the distribution obtained to integrate the
distribution over some range of values.  I get weekly sampled, and for each
sample I get a certain number of events each week for about three months.  I
need to be able to use the distribution to estimate the number of such
events next week or the week after, and how long it will be that the
probability of such an event is so low that no more of them are likely to be
observed from that sample ever.  What package or functions should I be
looking at here to get this done?

Third question: I see nothing in the docs about non-central distributions. 
The distribution most likely to fit is cauchy, but we know that there is
skew that depends on the magnitude: large positive deviates are more common
that large negative deviates, but extremely large positive deviates are less
common that extremely large negative deviates.  What we don't know is how
significant such skewness is for the overall distribution.  How can I assess
this, or can I assess this, using fitdistr (or some other function I haven't
found yet)?

Thanks

Ted
-- 
View this message in context: http://www.nabble.com/Relative-novice%3A-Working-with-fitdistr%28MASS%29%3A-3-questions-tp19610812p19610812.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list