The ordered probit seems open to two
objections:
1. month (really season) is a circular scale, not an
ordered one. Clearly January follows December just
as December follows January, but an ordered probit
model would take no account of that fact. (That's
assuming a conventional numbering 1 ... 12 but
any other start to the year provokes a similar
comment.
2. regarding season as a response seems the wrong
way round to me.
As you have you have a temperature variable as well, it
is not clear that month as such is needed.
In some other problems with circular predictors, using
sine and cosine terms works well. On the whole, however,
this looks more like a fairly standard regression problem with
temperature as presumably the main forcing covariate,
except that you may well need to try to capture various
production and consumption influences as well; and
the whole has a time series flavour.
Nick
[email protected]
Harald Scheule, PhD
> I am currently trying to model the monthly frequency distribution of
> electricity consumption (e.g., 20% of the annual consumption
> takes place in
> February) based on explanatory variables such as temperature,
> oil prices
> etc..
>
> I was thinking that an ordered probit model might work with
> the month being
> the dependent variable. The problem is that my dataset
> includes historic
> relative frequencies (e.g., January: 25%, February: 20%,
> March: 10%, ...,
> December:10%) rather than individual observations. Can I use
> a weighting
> function such as pweight to take the different frequencies
> into account? Do
> you think another model type might be more suitable?
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
