R help - May 2003 - Forward Stepwise regression with stepAIC and step

Dear all,

I cannot seem to get the R functions step or stepAIC to perform forward
or stepwise regression as I expect. I have enclosed the example data in
a dataframe at the end of this mail. Note rubbish is and rnorm(17) variable
which I have deliberately added to the data to test the stepwise procedure.

I have used 

wateruse.lm<-lm(waterusage~.,data=wateruse)   # Fit full model
wateruse.lm.back<-stepAIC(wateruse.lm,trace=FALSE)

in which rubbish is removed (calling wateruse.lm.back$anova gives      

Step         Df Deviance Resid. Df Resid. Dev      AIC
1           NA       NA        11   743720.8 193.6655
2 - rubbish  1 76.67448        12   743797.5 191.6673

However if I run

wateruse.lm.forward<-stepAIC(wateruse.lm,trace=FALSE,direction="forward")

wateruse.lm.forward$anova gives 

Stepwise Model Path 
Analysis of Deviance Table

Initial Model:
waterusage ~ avetemp + product + days + payroll + rubbish

Final Model:
waterusage ~ avetemp + product + days + payroll + rubbish


  Step Df Deviance Resid. Df Resid. Dev      AIC
1      NA       NA        11   743720.8 193.6655


In other words no forward procedure has been run (similar results apply
if I use direction="step")!!

What am I doing wrong?? Thanks in advance for any help.

Thanks, Paul.






> wateruse
   waterusage avetemp product days payroll    rubbish
1        3067    58.8    7107   21     129 -0.8393514
2        2828    65.2    6373   22     141 -0.8781807
3        2891    70.9    6796   22     153  1.1939278
4        2994    77.4    9208   20     166  0.2230589
5        3082    79.3   14792   25     193  1.7165612
6        3898    81.0   14564   23     189 -0.3810258
7        3502    71.9   11964   20     175 -0.1891337
8        3060    63.9   13526   23     186  1.1204482
9        3211    54.5   12656   20     190  2.6239803
10       3286    39.5   14119   20     187  0.9282131
11       3542    44.5   16691   22     195 -1.7371026
12       3125    43.6   14571   19     206 -0.1368085
13       3022    56.0   13619   22     198  1.3408773
14       2922    64.7   14575   22     192 -0.1125558
15       3950    73.0   14556   21     191  0.4668919
16       4488    78.9   18573   21     200  0.6309489
17       3295    79.4   15618   22     200  1.0918370

On Thu, 8 May 2003 pfm401 at lineone.net wrote:
> I cannot seem to get the R functions step or stepAIC to perform forward
> or stepwise regression as I expect.
[...]
>
wateruse.lm.forward<-stepAIC(wateruse.lm,trace=FALSE,direction="forward")
[...]
> In other words no forward procedure has been run (similar results apply
> if I use direction="step")!!
That's not a valid value for `direction': see the help page.
> What am I doing wrong?? Thanks in advance for any help.
You not given a `scope' argument, so there are no variables to add.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Paul,
Try this
l1<-lm(waterusage~product,data=wateruse)
water.forw<-step(l1,scope=~.+avetemp+days+payroll+rubbish,direction="forward")

Good luck

Edgar
University of Puerto Rico

On Thu, 8 May 2003 pfm401 at lineone.net wrote:
> Dear all,
>
> I cannot seem to get the R functions step or stepAIC to perform forward
> or stepwise regression as I expect. I have enclosed the example data in
> a dataframe at the end of this mail. Note rubbish is and rnorm(17) variable
> which I have deliberately added to the data to test the stepwise procedure.
>
> I have used
>
> wateruse.lm<-lm(waterusage~.,data=wateruse)   # Fit full model
> wateruse.lm.back<-stepAIC(wateruse.lm,trace=FALSE)
>
> in which rubbish is removed (calling wateruse.lm.back$anova gives
>
> Step         Df Deviance Resid. Df Resid. Dev      AIC
> 1           NA       NA        11   743720.8 193.6655
> 2 - rubbish  1 76.67448        12   743797.5 191.6673
>
> However if I run
>
>
wateruse.lm.forward<-stepAIC(wateruse.lm,trace=FALSE,direction="forward")
>
> wateruse.lm.forward$anova gives
>
> Stepwise Model Path
> Analysis of Deviance Table
>
> Initial Model:
> waterusage ~ avetemp + product + days + payroll + rubbish
>
> Final Model:
> waterusage ~ avetemp + product + days + payroll + rubbish
>
>
>   Step Df Deviance Resid. Df Resid. Dev      AIC
> 1      NA       NA        11   743720.8 193.6655
>
>
> In other words no forward procedure has been run (similar results apply
> if I use direction="step")!!
>
> What am I doing wrong?? Thanks in advance for any help.
>
> Thanks, Paul.
>
>
>
>
>
>
>
> > wateruse
>    waterusage avetemp product days payroll    rubbish
> 1        3067    58.8    7107   21     129 -0.8393514
> 2        2828    65.2    6373   22     141 -0.8781807
> 3        2891    70.9    6796   22     153  1.1939278
> 4        2994    77.4    9208   20     166  0.2230589
> 5        3082    79.3   14792   25     193  1.7165612
> 6        3898    81.0   14564   23     189 -0.3810258
> 7        3502    71.9   11964   20     175 -0.1891337
> 8        3060    63.9   13526   23     186  1.1204482
> 9        3211    54.5   12656   20     190  2.6239803
> 10       3286    39.5   14119   20     187  0.9282131
> 11       3542    44.5   16691   22     195 -1.7371026
> 12       3125    43.6   14571   19     206 -0.1368085
> 13       3022    56.0   13619   22     198  1.3408773
> 14       2922    64.7   14575   22     192 -0.1125558
> 15       3950    73.0   14556   21     191  0.4668919
> 16       4488    78.9   18573   21     200  0.6309489
> 17       3295    79.4   15618   22     200  1.0918370
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>