[R] help interpreting a model summary

David Winsemius dwinsemius at comcast.net
Sun Sep 19 22:54:19 CEST 2010


That parameter is the difference between the estimated parameter for  
the product of reflection and angleNoise in regions where reflection  
was < Break(xMin) compared with the same product's parameter in those  
regions of 3-space where reflection was not.  In general it is at best  
merely speculative (and generally rather dangerous) to interpret the  
meanings of individual parameters which apply to variables that are  
modeled with interactions. It is in particular a fool's errand to look  
at the std errors of such parameters. Anova tables compared across  
nested models are much less misleading.

You certainly _cannot_ say that there is more "importance" in the  
region where reflection is < Break[min]. The parameter is measuring  
differences between both regions. If you had instead constructed the  
model with the reversed inequality, the parameter would have been of  
the same magnitude but reverse sign and would have had the same  
standard error.

It is usually much more informative to examine the predictions that  
result from the models, and this may be greatly aided by plotting  
across a range of values, in this case perhaps with persp() or  
contour(). Dealing with three-way interactions can be particularly  
messy, so I think it's fair to inquire why you are adding terms to  
models when you are not prepared to interpret them? Throwing terms  
into a model with no physical basis can be amusing but rarely good  
science. You are the domain expert, after all. There should be a  
design and rationale behind this process.

You will also note that the third and fifth of your five terms were  
superfluous because all of their estimates were NA. The other three  
terms covered all the possibilities, since the cases where reflection  
 >= Break[xMin] would be covered by their contribution to the  
angleNoise*reflction (with (reflection < Break[xMin]) ==0 )


-- 
David (son of a son of an engineer)


On Sep 19, 2010, at 3:54 PM, zozio32 wrote:

>
> Hello, I am all new here.
> Thanks for the job done, R really helped me in my thesis lately.  
> However, I
> am kind of new in statistics, coming from mecanical engineering, and I
> mostly teached myself with "The R Book", so I may do silly things  
> some time.
> PLease tell me if you think so.
>
> Anyway, I've just build up a piecewise linear model to fit some data,
> including some interaction and i am not sure of how to interpret the
> summary:.
> here it is:
>
> --------------------------------------------------------------------------------
> Call:
> lm(formula = weightedDiff ~ angleNoise +

>                            (reflection < Break[xMin]) *  reflection +

>                            (reflection >= Break[xMin]) * reflection +

>                             angleNoise:(reflection < Break[xMin]) *  
> reflection +

>                            angleNoise:(reflection >= Break[xMin]) *  
> reflection)
>
> Residuals:
>       Min         1Q     Median         3Q        Max
> -1.073e-03 -1.749e-04 -5.913e-06  1.650e-04  1.311e-03
>
> Coefficients: (4 not defined because of singularities)
>                                                      Estimate Std.  
> Error
> (Intercept)                                          0.0134798   
> 0.0001086
> angleNoise                                           0.0004658   
> 0.0002245
> reflection < Break[xMin]TRUE                        -0.0028766   
> 0.0001236
> reflection                                           0.0316122   
> 0.0014741
> reflection >= Break[xMin]TRUE                                
> NA         NA
> reflection < Break[xMin]TRUE:reflection              0.0683631   
> 0.0027668
> reflection:reflection >= Break[xMin]TRUE                     
> NA         NA
> angleNoise:reflection < Break[xMin]TRUE              0.0011158   
> 0.0002548
> angleNoise:reflection >= Break[xMin]TRUE                     
> NA         NA
> angleNoise:reflection < Break[xMin]FALSE:reflection -0.0055751   
> 0.0030620
> angleNoise:reflection < Break[xMin]TRUE:reflection  -0.0343745   
> 0.0049164
> angleNoise:reflection:reflection >= Break[xMin]TRUE          
> NA         NA
>
> t value       Pr(>|t|)
> (Intercept)
> 124.079  < 2e-16   ***
> angleNoise
> 2.075       0.0384    *
> reflection < Break[xMin]TRUE
> -23.265   < 2e-16  ***
> reflection
> 21.445    < 2e-16  ***
> reflection >= Break[xMin]TRUE
> NA             NA
> reflection < Break[xMin]TRUE:reflection
> 24.708   < 2e-16    ***
> reflection:reflection >=  
> Break[xMin]TRUE                               NA
> NA
> angleNoise:reflection < Break[xMin]TRUE                              
> 4.379
> 1.41e-05  ***
> angleNoise:reflection >= Break[xMin]TRUE                           NA
> NA
> angleNoise:reflection < Break[xMin]FALSE:reflection          -1.821
> 0.0692    .
> angleNoise:reflection < Break[xMin]TRUE:reflection            -6.992
> 7.35e-12  ***
> angleNoise:reflection:reflection >= Break[xMin]TRUE          NA
> NA
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> Residual standard error: 0.0002885 on 592 degrees of freedom
> Multiple R-squared: 0.9666,	Adjusted R-squared: 0.9662
> F-statistic:  2450 on 7 and 592 DF,  p-value: < 2.2e-16
>
> --------------------------------------------------------------------------------------------
> Basically, I am really not sure of the meaning of this parameter:
> angleNoise:reflection < Break[xMin]FALSE:reflection
>
> Overall, my interpretation is that reflection is important , angle  
> Noise
> also but specially when reflection is below the breaking point. Is  
> that
> correct?
>
> well, sorry for the first long post
>
> thanks in advance
>
> -- 
> View this message in context: http://r.789695.n4.nabble.com/help-interpreting-a-model-summary-tp2546161p2546161.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.



More information about the R-help mailing list