Prediction against target (stats question, not SPSS)

[Spot Fire]

Dear All,

I am trying to predict the number of teenage pregnancies in 2015 based on
values for 2004-2009. I want to see if we will hit our target of 15 per 1000
by that year.

A simple linear regression in spss 17 predicts a value of 47.5, with a 95%
range of 25.8 to 67.4. So it it safe to say that we won't hit our target. My
question though was what if our target was 35. How you intercept the results
then? The target would be inside your 95% interval, but a long way off from
47.5.

thanks for any thoughts...

Regards,
Jimbo

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Hector Maletta]

Jimbo,
The confidence interval is not about evaluating reality ("Have we meet or
will we meet our target") but about the predictive ability of your data and
your model. It means that you have confidence that if your five measurements
are random samples of the many (inexact) measurements that you may have for
the 2004-2009 period, then the true value for any of those five years will
be within the confidence interval 95% of the time. In your case, you extend
this belief to the NEXT year (2010), which is in fact outside your range of
observation and thus may have a still larger margin of possible error.
However, a 95% confidence level means that 95% of the time the true value
will be within the interval, with 2.5% chances of being below, and 2.5% of
being above.
Of all the values in the interval, your "best prediction" is the midpoint,
but in this uncertain world things may be different. If the observed value
is actually within the interval, then you may conclude that YOUR DATA AND
MODEL behave reasonably well. But this is not about your target: it is about
your data.
Whether the predicted value is or is not within a reasonable distance of
your target is a totally different matter. As you suggest, if your target is
15, you're way off the mark: there are (if your model is correct and your
data are good) 97.5% chances that the true value will be above 25.8, well
above your target. Of course, if your target was 35, this value is within
the interval (from 25.8 to 67.4) emerging from your model, and you may say
that you have some chances of hitting the target but will probably be above
it (since your best prediction is 47, not 35). If your target was 47, your
best prediction is that you will hit this target, though there is a wide
margin of error in that prediction, so do not call a press conference just
as yet.

Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de Spot
Fire
Enviado el: Sunday, August 15, 2010 11:42 PM
Para: [hidden email]
Asunto: Prediction against target (stats question, not SPSS)

Dear All,

I am trying to predict the number of teenage pregnancies in 2015 based on
values for 2004-2009. I want to see if we will hit our target of 15 per 1000
by that year.

A simple linear regression in spss 17 predicts a value of 47.5, with a 95%
range of 25.8 to 67.4. So it it safe to say that we won't hit our target. My
question though was what if our target was 35. How you intercept the results
then? The target would be inside your 95% interval, but a long way off from
47.5.

thanks for any thoughts...

Regards,
Jimbo

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Se certificó que el correo entrante no contiene virus.
Comprobada por AVG - www.avg.es
Versión: 8.5.441 / Base de datos de virus: 271.1.1/3064 - Fecha de la
versión: 08/15/10 18:35:00

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Granaas, Michael]

It isn't completely clear what is going on here, but you have far too little data for good prediction, ignoring the issue of problems with predicting the future.  Even if you had enough data it isn't clear whether you have a trend to model or not.

 

Was there an intervention implemented in 2003/2004 that you are trying to evaluate?  Was there an intervention implemented in 2009 that you are trying to evaluate?  

 

If there was no intervention until 2009 then you have no trend to model from 2004 -2009 data. 

 

Michael

 

 

****************************************************
Michael Granaas             [hidden email]
Assoc. Prof.                Phone: [phone]
Dept. of Psychology         FAX:  [phone]
University of South Dakota
414 E. Clark St.
Vermillion, SD 57069
*****************************************************

---

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Spot Fire [[hidden email]]
Sent: Sunday, August 15, 2010 9:41 PM
To: [hidden email]
Subject: Prediction against target (stats question, not SPSS)

Dear All,

I am trying to predict the number of teenage pregnancies in 2015 based on
values for 2004-2009. I want to see if we will hit our target of 15 per 1000
by that year.

A simple linear regression in spss 17 predicts a value of 47.5, with a 95%
range of 25.8 to 67.4. So it it safe to say that we won't hit our target. My
question though was what if our target was 35. How you intercept the results
then? The target would be inside your 95% interval, but a long way off from
47.5.

thanks for any thoughts...

Regards,
Jimbo

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD