rolling median summary

["Leeds, Mark" <mleeds@mlp.com>]

to all the people who replied to my rolling median

question : I really appreciate all the useful

and quick responses.

I don't have all the people at my fingertips but

off of the top of my head were bill dunlap,

andy liaw and gernot schmidt. i apologize because

i think i left some people out.

 

i will summarize very briefly.

 

1) some people sent me code of their own

which was very kind of them to write.

 

2) other people recommended the

aggregateseries function in Splus.

 

3) other people referred me to the gregmisc

package in R which has a function called running.

 

I started to use 2) but that required that

your series be a signalseries object or

a timeseries object and, although

it's probably not terribly didfficult,

I did not want to get involved in this

because I am very unfamilar with the

time series facilities in Splus.

i tend to use raw data and keep

the time dates separate and

then put them in at the end.

a brutal way of working but

it avoids dealing with timeseries stuff.

 

I ended up copying the running function

from gregmisc and using it in Splus.

It works very well and it's

general in that you can send in any function

into the function.

 

for example,  you can send the median function into

running to get the rolling median

and the mad function in to get the rolling

robust standard deviation.

 

thanks to all.

 

                                        mark