Re: stepwise regression, interpretation of results

While you've included a lot of numbers in yokur post, they are of
little value in interpreting a regression analysis of your data.

It's a sure bet that the "independent" variables are correlated among
themselves. Some may be very strongly correlated. Personally, I will
not attempt an analysis of multivariable data without access to the
correlation matrix.

It appears that you have created some "interactions" by cross-
multiplying variables. Are you aware that when you do this those
variables must be "centered" before cross-multiplying? If not, that
alone is sufficient to negate the analysis.

Certain transformations of the data may be needed.

They may be large outliers in the fitted model.

The only way I could possibly help with this would be to have access
to the data and to a description of how the data came to be in the
first place.

If you wish, contact me directly via e-mail by clicking on "Send to
Author". OMU



On Feb 8, 1:42 pm, "jf" <j...@xxxxxxxxxxxx> wrote:
I need help interpreting the results of stepwise regression with
SPlus. I used forward and backward stepwise regression and below is
the last entry of the stepwise process before it spit out a linear
model. My questions are:

1. Why does the final linear model include variables that were
not in the last step, the model with the lowest AIC?

2. How can I figure out how much variation each of the variables
contributes to the total variation?

Step: AIC= 9.594

ExpDec ~ Ck + Yr + JD + JD2 + QNxtDy + QAve3 + WTDy + WTAve3 + DDCum
+
L + Yr:JD + QNxtDy:QAve3 + QAve3:JD + WTAve3:JD + L:JD

Single term deletions

Model:

ExpDec ~ Ck + Yr + JD + JD2 + QNxtDy + QAve3 + WTDy +WTAve3 + DDCum +
L + Yr:JD + QNxtDy:QAve3 +
QAve3:JD + WTAve3:JD + L:JD

scale: 0.009112796

Df Sum of Sq RSS Cp

<none> 9.30242 9.59403

Ck 1 1.239157 10.54158 10.81496
JD2 1 0.389247 9.69167 9.96505
WTDy 1 0.085947 9.38837 9.66175
DDCum 1 2.607181 11.90960 12.18299
Yr:JD 1 0.112556 9.41498 9.68836
QNxtDy:QAve3 1 1.081182 10.38360 10.65699
QAve3:JD 1 0.038569 9.34099 9.61438
WTAve3:JD 1 0.517758 9.82018 10.09357
L:JD 1 0.116229 9.41865 9.69204

*** Linear Model ***

Call: lm(formula = ExpDec ~ Ck + Yr + JD + JD2 + QNxtDy +QAve3 + WTDy
+ WTAve3 + DDCum + L + Yr:JD +
QNxtDy:QAve3 + QAve3:JD + WTAve3:JD + L:JD,

data = global.02.03.07, na.action = na.exclude)

Residuals:

Min 1Q Median 3Q Max

-0.7327 -0.04916 0.001962 0.05493 0.2442

Coefficients:

Value Std. Error t value Pr(>|t|)

(Intercept) -0.6042 0.1279 -4.7227 0.0000
Ck -0.0472 0.0040 -11.6735 0.0000
Yr 0.0566 0.0115 4.9032 0.0000
JD 0.0099 0.0013 7.9236 0.0000
JD2 0.0000 0.0000 -6.5426 0.0000
QNxtDy -1.0587 0.0593 -17.8450 0.0000
QAve3 0.4737 0.0976 4.8542 0.0000
WTDy 0.0212 0.0069 3.0744 0.0022
WTAve3 -0.1433 0.0198 -7.2379 0.0000
DDCum -0.0012 0.0001 -16.9327 0.0000
L -0.0344 0.0111 -3.0999 0.0020
Yr:JD -0.0004 0.0001 -3.5182 0.0005
QNxtDy:QAve3 -0.6734 0.0618 -10.9041 0.0000
QAve3:JD 0.0015 0.0007 2.0595 0.0397
WTAve3:JD 0.0013 0.0002 7.5458 0.0000
L:JD 0.0004 0.0001 3.5752 0.0004

Residual standard error: 0.09536 on 1023 degrees of freedom
Multiple R-Squared: 0.8761
F-statistic: 482.3 on 15 and 1023 degrees of freedom, the p-value


.

Relevant Pages
Loading