Linear/Non-linear regression - Help

[spssconfused]

Hello all,

Hopefully somebody will be able to shed some light on my SPSS problems!

I have been given 65 values. 57 of these data values are quarterly results and 8 are the holdback data to be used.

I have to do:
- Regression with Dummy variables with a linear trend cycle component
- Dummy variables with a nonlinear trend cycle component

Does anyone know what to do as my results aren't making much sense?

For the first part - I obviously split the data into dummy variables for the relevant quarters (Q1-Q4).

I then performed regression analysis - linear. But all my values are extremely large and not significant. Also Q2 has been listed as 'excluded variables' in the results? I have followed the steps and I am unsure why this has happened.

Then I thought of removing Q4, due to multi-collinearity but again the values are still quite large (>.450).

Not sure if I am doing something wrong at the start (especially with the excluded variables aspect)

Anybody got any idea? This is driving me nuts.

[Maguin, Eugene]

Post your syntax, including all relevant computation commands. And, try out Curvefit. That might be helpful to you.
Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of spssconfused
Sent: Friday, November 22, 2013 2:42 PM
To: [hidden email]
Subject: Linear/Non-linear regression - Help

Hello all,

Hopefully somebody will be able to shed some light on my SPSS problems!

I have been given 65 values. 57 of these data values are quarterly results and 8 are the holdback data to be used.

I have to do:
- Regression with Dummy variables with a linear trend cycle component
- Dummy variables with a nonlinear trend cycle component

Does anyone know what to do as my results aren't making much sense?

For the first part - I obviously split the data into dummy variables for the relevant quarters (Q1-Q4).

I then performed regression analysis - linear. But all my values are extremely large and not significant. Also Q2 has been listed as 'excluded variables' in the results? I have followed the steps and I am unsure why this has happened.

Then I thought of removing Q4, due to multi-collinearity but again the values are still quite large (>.450).

Not sure if I am doing something wrong at the start (especially with the excluded variables aspect)

Anybody got any idea? This is driving me nuts.



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Linear-Non-linear-regression-Help-tp5723256.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

=====================
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

[bdates]

Could you also please post how Q1 to Q4 were coded?

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Maguin, Eugene
Sent: Friday, November 22, 2013 4:51 PM
To: [hidden email]
Subject: Re: Linear/Non-linear regression - Help

Post your syntax, including all relevant computation commands. And, try out Curvefit. That might be helpful to you.
Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of spssconfused
Sent: Friday, November 22, 2013 2:42 PM
To: [hidden email]
Subject: Linear/Non-linear regression - Help

Hello all,

Hopefully somebody will be able to shed some light on my SPSS problems!

I have been given 65 values. 57 of these data values are quarterly results and 8 are the holdback data to be used.

I have to do:
- Regression with Dummy variables with a linear trend cycle component
- Dummy variables with a nonlinear trend cycle component

Does anyone know what to do as my results aren't making much sense?

For the first part - I obviously split the data into dummy variables for the relevant quarters (Q1-Q4).

I then performed regression analysis - linear. But all my values are extremely large and not significant. Also Q2 has been listed as 'excluded variables' in the results? I have followed the steps and I am unsure why this has happened.

Then I thought of removing Q4, due to multi-collinearity but again the values are still quite large (>.450).

Not sure if I am doing something wrong at the start (especially with the excluded variables aspect)

Anybody got any idea? This is driving me nuts.



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Linear-Non-linear-regression-Help-tp5723256.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

=====================
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

=====================
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

[spssconfused]

We were given the quarterly data like:
16180 17425 17424 17240 18240 19880 20143 20545 22155 23344 23717 23467 25442 27278 27848 25704 28919 30280 32095 31041 33182 35067 35557 34420 35948 38643 39612 39185 40143 40056 41360 41343 43652 44554 47903 46460 49402 50254 50335 48763 51529 53481 53482 53882 55219 56180 56037 54106 54915 54641 53805 52179 52026 51522 51733 50672 50882 50878 52199 50261 49615 47995 45273 42836 43321

Then told to perform the linear and non-linear regression with dummy variables.

I do understand the stats aspect of it but have no clue how to do it in SPSS.

For non-linear:
I have done:
time
Time^2
Q1
Q2
Q3
(leave out Q4)

Then do regression - linear?

No clue how to build the non-linear properly in SPSS.

(the next steps are ARIMA analysis so it follows on)

[Art Kendall]

what was your syntax?
On 11/23/2013 8:43 AM, spssconfused [via SPSSX Discussion] wrote:

We were given the quarterly data like:
16180 17425 17424 17240 18240 19880 20143 20545 22155 23344 23717 23467 25442 27278 27848 25704 28919 30280 32095 31041 33182 35067 35557 34420 35948 38643 39612 39185 40143 40056 41360 41343 43652 44554 47903 46460 49402 50254 50335 48763 51529 53481 53482 53882 55219 56180 56037 54106 54915 54641 53805 52179 52026 51522 51733 50672 50882 50878 52199 50261 49615 47995 45273 42836 43321

Then told to perform the linear and non-linear regression with dummy variables.

I do understand the stats aspect of it but have no clue how to do it in SPSS.

For non-linear:
I have done:
time
Time^2
Q1
Q2
Q3
(leave out Q4)

Then do regression - linear?

No clue how to build the non-linear properly in SPSS.

(the next steps are ARIMA analysis so it follows on)

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/Linear-Non-linear-regression-Help-tp5723256p5723263.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

[David Marso]

Sounds suspiciously like a homework assignment?
Take home midterm exam?
--

spssconfused wrote

We were given the quarterly data like:
16180 17425 17424 17240 18240 19880 20143 20545 22155 23344 23717 23467 25442 27278 27848 25704 28919 30280 32095 31041 33182 35067 35557 34420 35948 38643 39612 39185 40143 40056 41360 41343 43652 44554 47903 46460 49402 50254 50335 48763 51529 53481 53482 53882 55219 56180 56037 54106 54915 54641 53805 52179 52026 51522 51733 50672 50882 50878 52199 50261 49615 47995 45273 42836 43321

Then told to perform the linear and non-linear regression with dummy variables.

I do understand the stats aspect of it but have no clue how to do it in SPSS.

For non-linear:
I have done:
time
Time^2
Q1
Q2
Q3
(leave out Q4)

Then do regression - linear?

No clue how to build the non-linear properly in SPSS.

(the next steps are ARIMA analysis so it follows on)

[spssconfused]

Not a homework assignment (well not completely)- thankfully finished with all that. Been sent on a training course by my firm and need to compute this all into a report (becoming the bane of my life at the moment)

All we were given was that data set, and told to keep 58 as modelling data and 8 as holdback data. Then to perform the relevant regression analysis' in SPSS.

I studied Math and Stats at uni so I have a firm understanding of the background and the results that get generated but fairly lost at what to do within SPSS.

[Anthony Babinec]

It is difficult to respond to your question because it is not clear
what the assignment is or what you have done. It appears that
you are doing an exercise in time series regression.

It sounds like
- you must build the model on 57 observations
and assess it on 8 holdouts.

- linear regression is the tool.

- you are building quarterly dummy variables to
represent the seasonal component.

- you are building Time and Timesq to represent trend.
Time is just sequential numbers 1,2,...65.
Timesq is Time^2.

If linear regression is the tool, you can fit a linear or "nonlinear"
trend by using regression. By default, regression has an intercept term,
so you will not be able to include all 4 quarterly dummies.

To model linear trend, you might include:

Q1, Q2, Q3
Time

To look for nonlinear trend, you might include:

Q1, Q2, Q3
Time, Timesq

By their very nature, the two Time terms are highly although
not perfectly correlated.

As for the quarterly dummies, if you try to include all 4, and also
include the intercept, Regression will not allow all 4 in.

Tony Babinec
[hidden email]

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
spssconfused
Sent: Friday, November 22, 2013 1:42 PM
To: [hidden email]
Subject: Linear/Non-linear regression - Help

Hello all,

Hopefully somebody will be able to shed some light on my SPSS problems!

I have been given 65 values. 57 of these data values are quarterly results
and 8 are the holdback data to be used.

I have to do:
- Regression with Dummy variables with a linear trend cycle component
- Dummy variables with a nonlinear trend cycle component

Does anyone know what to do as my results aren't making much sense?

For the first part - I obviously split the data into dummy variables for the
relevant quarters (Q1-Q4).

I then performed regression analysis - linear. But all my values are
extremely large and not significant. Also Q2 has been listed as 'excluded
variables' in the results? I have followed the steps and I am unsure why
this has happened.

Then I thought of removing Q4, due to multi-collinearity but again the
values are still quite large (>.450).

Not sure if I am doing something wrong at the start (especially with the
excluded variables aspect)

Anybody got any idea? This is driving me nuts.



--
View this message in context:
http://spssx-discussion.1045642.n5.nabble.com/Linear-Non-linear-regression-H
elp-tp5723256.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

=====================
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

[Rich Ulrich]

Terminology.
Did someone really ask for "nonlinear regression"?  Or did they
say "regression including nonlinear terms for time"?

As most (all?) statisticians use "nonlinear regression", it refers to a
regression that is nonlinear in the model parameters.  Including
(b1*time^2)  as a term is still a linear regression.  Including
(b1*time^b2)  would be a nonlinear regression.

Nonlinear regression would be unusual for business applications.

--
Rich Ulrich

> Date: Sat, 23 Nov 2013 07:51:51 -0800
> From: [hidden email]
> Subject: Re: Linear/Non-linear regression - Help
> To: [hidden email]
>
> Sounds suspiciously like a homework assignment?
> Take home midterm exam?
> --
...

> >
> > For non-linear:
> > I have done:
> > time
> > Time^2
> > Q1
> > Q2
> > Q3
> > (leave out Q4)
> >
> > Then do regression - linear?
> >
> > No clue how to build the non-linear properly in SPSS.

...

[spssconfused]

Thanks.

My data set is based on 65 quarterly pieces of data for credit lending.

I first did basic tests to check for trends, seasonality, etc.

Next step is to use the modelling data to do a regression analysis with dummy variables with a linear and also non-linear trend-cycle components. Then to perform both the decomposition method and the Box-Jenkins ARIMA approach.

The results I was generating for the non-linear and linear regression didn't seem to be correct so I think I am making a simple mistake somewhere.

Analyze - Regression - Linear - Within option 'statistics' select Durbin-Watson - Within option 'save' tick unstandardised for both predicted variables and residuals, and tick individual within prediction individuals.  Does that sound correct? :S

I posted the complete data set above. Any help is very much appreciated - first time using SPSS :(

[spssconfused]

I know, I was asked to perform 'regression with the dummy variables method with a linear trend cycle component' and then the same with a 'nonlinear trend cycle component'

For nonlinear, I was recommended to use a polynomial to fit nonlinear trend-cycle. I was thinking of building a model to accommodate the quadratic trendcycle with seasonality to predict the quarterly sales, but I am fairly lost at how to do that in SPSS

[Rich Ulrich]

Okay.  It is good to know that they were not asking for "nonlinear regression."

A dummy variable for time that goes (1, 2, 3, ... )  for sequential
quarters codes up the linear trend.

Any other dummying for time, by month or season, or even a single
high point for a special month, or a step-increase to represent the
introduction of  <whatever>, ...  will be a "non-linear" term.  You do
not need to think of "quadratic" here.

A simple and straightforward regression approach to accommodate
the season would be to use a dummy variable for each quarter that
is set equal to the average for that quarter. 

After you remove the linear trend and the seasonal trend, the residual
might show a bit of curvature if the "percent growth" has been constant,
since the latest terms are rather larger than the earliest ones.  To avoid
working with the logarithm of the outcome, I think that economics
data models may be written to use something like a quadratic in time,
so you could have an excuse to try a quadratic term, like (time-mean_time)^2.

--
Rich Ulrich

> Date: Sun, 24 Nov 2013 04:49:58 -0800
> From: [hidden email]
> Subject: Re: Linear/Non-linear regression - Help
> To: [hidden email]
>
> I know, I was asked to perform 'regression with the dummy variables method
> with a linear trend cycle component' and then the same with a 'nonlinear
> trend cycle component'
>
>
[Nabble shows me that "this post was updated" about an hour later.
The following was not in what I received from the list.]
> For nonlinear, I was recommended to use a polynomial to fit nonlinear
> trend-cycle. I was thinking of building a model to accommodate the
> quadratic trendcycle with seasonality to predict the quarterly sales,
> but I am fairly lost at how to do that in SPSS


>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Linear-Non-linear-regression-Help-tp5723256p5723277.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>