**Date:** Thu, 22 Aug 2002 13:32:03 +0100
**Reply-To:** David Hitchin <D.H.Hitchin@sussex.ac.uk>
**Sender:** "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
**From:** David Hitchin <D.H.Hitchin@sussex.ac.uk>
**Subject:** Re: Newbie question
**In-Reply-To:** <009c01c249d4$0f0a8fc0$8f70010a@ifucengl02>
**Content-Type:** text/plain; charset=us-ascii
The question isn't so much whether you can predict next winter using
multiple regression as whether you can predict it at all by any method.

For a start, winters aren't all the same. You might have collected your
data in a particularly good or a particularly bad winter, but as far as the
regression model is concerned all the winters are the same as the one that
you sampled. So, you would need to record data over a number of winters.

You might reckon that date and time were important variables in fitting the
model for last winter, and that you could just plug a date and a time into
your equation to get an estimate, but that is likely to be widely out
unless you fit data over a very long period - and we aren't even sure of
the extent of global warming yet.

Now you might have tried building an explanatory model, for example
relating independent variables such as temperature, humidity, traffic flow
and other measured variables to the dependent variable which is the amount
of pollution. With this, at any time of day you could observe the
independents and from them calculate an estimate of the dependent. That
might explain the values at the time quite well.

However, you referred to "prediction" and if you are to predict next
winter, you need to know the values of the independent variables next
winter. That may be difficult (impossible), although you could use
estimates of them.

That takes us to an old adage, that interpolation is usually pretty safe,
but extrapolation is risky. As Dan Quayle (I think) said, its hard to
predict anything, especially the future.

David Hitchin

--On 22 August 2002 14:01 +0200 Sunrise <sunrisevn@lycos.co.uk> wrote:

> Hi all
> Anyone tell me whether I can use multiple regression for predicting. For
> instance, I have data set of hourly pollutant concentrations in winter
> time (3 months). After regression analysis, I get regression
> coefficients. My question is that this regression model is capable of
> predicting pollutant concentrations for next winter times. If not, how I
> can do to predict pollution with regression model. Thank you very much
> Sunrise