Re: time series with binary data
- From: glenbarnett@xxxxxxxxxxxxx
- Date: 15 Nov 2005 13:07:46 -0800
You're asking for an index that weights periods of not working?
If the number of days in the gaps where it doesn't work are g(1), g(2),
.... g(k)
what about something like index = 1 - [Sum_i
{g(i)^(1+m)}/90^(1+m)}^{1/(1+m)}
where m is some small positive number (between 0 and say 3, most likely
between 0 and 1).
for example, if we have two gaps of 4 days each and take m=1, the index
is
1 - sqrt[(4^2 + 4^2)/90^2] = 0.937
If we have 7 two day gaps, and m=1 index = 1 - sqrt[(7*2^2)/90^2] =
0.941
(so two 4-day gaps is slightly worse than 7 two-day gaps)
Larger m penalises long gaps more.
If you give more of an idea of what should produce some of the
intermediate values, more could be done to help you construct a
function.
Does this help?
Glen
You can tune m so as to give the sort of ordering you ask for in the
single four day gap vs every third day.
There are many easy-to-write-down functions that might suit your
purpose. It might be easier to make suggestions if you gave an example
of what you'd like to come out somewhere near 0.5, and maybe 0.1, 0.25,
0.75 and 0.9.
.
- References:
- time series with binary data
- From: gauger
- time series with binary data
- Prev by Date: stata
- Next by Date: Re: Multiple regression coefficients
- Previous by thread: time series with binary data
- Next by thread: stata
- Index(es):
Relevant Pages
|
Loading