Curve estimation

[Marleen]

I am analyzing a dataset and I would like to test if the trends in the
dataset are linear or quadratic. I tried to use the function 'curve
estimation' in spss 12.0.1 an example of a result is shown below. Does
someone understand why spss doen not include all variables in the
equation, even they are significant? And does someone know what the notes
(g tolereance limits reached, etc.)mean? I don't understand it, so I hope
you do!

ependent variable.. arcALOPEPRA          Method.. LINEAR

Listwise Deletion of Missing Data

Multiple R           .73805
R Square             .54472
Adjusted R Square    .53793
Standard Error       .37715

            Analysis of Variance:

               DF   Sum of Squares      Mean Square

Regression      1        11.402803        11.402803
Residuals      67         9.530374          .142244

F =      80.16346       Signif F =  .0000

-------------------- Variables in the Equation --------------------

Variable                  B        SE B       Beta         T  Sig T

jaar               -.029916     .003341   -.738054    -8.953  .0000
(Constant)        60.419832    6.622506                9.123  .0000




Dependent variable.. arcALOPEPRA          Method.. QUADRATI

Listwise Deletion of Missing Data

Multiple R           .73918
R Square             .54638
Adjusted R Square    .53961
Standard Error       .37646

            Analysis of Variance:

               DF   Sum of Squares      Mean Square

Regression      1        11.437567        11.437567
Residuals      67         9.495609          .141726

F =      80.70225       Signif F =  .0000

-------------------- Variables in the Equation --------------------

Variable                  B        SE B       Beta         T  Sig T

jaar**2 -7.564305943467E-06  8.4203E-07   -.739178    -8.983  .0000
(Constant)        30.842384    3.308078                9.323  .0000

--------------- Variables not in the Equation ---------------

Variable         Beta In  Partial  Min Toler         T  Sig T

jaar          113.965488  .530601  9.833E-06     5.086  .0000

Notes:
9  Tolerance limits reached; some dependent variables were not entered.

Notes:
9  Tolerance limits reached; some dependent variables were not entered.

[Richard Ristow]

At 08:29 AM 10/10/2006, Marleen wrote:

>I am analyzing a dataset and I would like to test if the trends in the
>dataset are linear or quadratic. I tried to use the function 'curve
>estimation' in spss 12.0.1 an example of a result is shown below. Does
>someone understand why spss doen not include all variables in the
>equation, even they are significant? And does someone know what the
>notes
>(g tolereance limits reached, etc.) mean?

Hard to tell for sure without seeing your data, but a likely problem is
that variable 'jaar' is very highly correlated with its own square. The
problems you're seeing are reasonable for highly correlated predictors.

Many variables are highly correlated with their own squares, especially
of their coefficient of variation (standard deviation divided by mean)
is low. To make a guess: you variable 'jaar' might be a date in years,
with values like 1995, 1996, ..., 2006. That seems particularly likely
given the tiny estimated B for jaar**2:

>Variable                  B        SE B       Beta         T  Sig T
>
>jaar**2 -7.564305943467E-06  8.4203E-07   -.739178    -8.983  .0000
>(Constant)        30.842384    3.308078                9.323  .0000

It that's the case, try

COMPUTE jaar2 = jaar-2000.

and then do 'curve estimation' again, but with 'jaar2' as independent
variable.

-Onward, and good luck,
  Richard