[R] GLM result output..

[David Winsemius]

On Sep 13, 2013, at 9:38 AM, Lutfor Rahman wrote:

> Dear forum members,
>
> Please help me understanding significance value when GLM done in r.
>
> After doing minimal adequate model, I have found a number of  
> independent
> values  which are significant. But doing their anova significant  
> values are
> different. Please find my result following. Which significant values  
> should
> I use.
>
>
> glm(formula = richness ~ moistcont + orgmatter + baresoil + grass10 +
>    wood10 + rdnet10 + moistcont:orgmatter + moistcont:baresoil +
>    grass10:wood10 + grass10:rdnet10 + wood10:rdnet10, family =  
> poisson,
>    data = data)
>
> Deviance Residuals:
>     Min        1Q    Median        3Q       Max
> -1.19112  -0.33682   0.09813   0.32808   0.70509
>
> Coefficients:
>                     Estimate Std. Error z value Pr(>|z|)
> (Intercept)         11.384447   4.014170   2.836  0.00457 **
> moistcont           -0.095813   0.084995  -1.127  0.25962
> orgmatter           -1.810116   0.613688  -2.950  0.00318 **
> baresoil            -1.636707   0.559129  -2.927  0.00342 **
> grass10             -0.018979   0.065647  -0.289  0.77250
> wood10               0.150683   0.128386   1.174  0.24053
> rdnet10             -0.011448   0.068090  -0.168  0.86648
> moistcont:orgmatter  0.025698   0.011521   2.231  0.02571 *
> moistcont:baresoil   0.044110   0.015799   2.792  0.00524 **
> grass10:wood10       0.010740   0.006498   1.653  0.09838 .
> grass10:rdnet10      0.011013   0.004412   2.496  0.01255 *
> wood10:rdnet10      -0.088297   0.027120  -3.256  0.00113 **

The only p-value I would have expected to be the same would have been  
the last one in the avova output:

>                    Df Deviance Resid. Df Resid. Dev Pr(>Chi)
> .....
> wood10:rdnet10       1  10.7812         6      3.928 0.001025 **

And that particular p-value is not far off from the 0.00113 value  
reported in the model summary. The other p-values are not of the same  
sort. The p-values above are basically reporting the "significance" of  
removing single predictors or interactions from the full model. The  
anova reported below is perfoming sequential addition of terms to a  
NULL model as well as doing a different test:  LR tests instead of  
Wald statistics.

-- 
David.


> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> (Dispersion parameter for poisson family taken to be 1)
>
>    Null deviance: 36.1673  on 17  degrees of freedom
> Residual deviance:  3.9276  on  6  degrees of freedom
> AIC: 97.893
>
> Number of Fisher Scoring iterations: 4
>
>> anova(data1, test="Chisq")
> Analysis of Deviance Table
>
> Model: poisson, link: log
>
> Response: richness
>
> Terms added sequentially (first to last)
>
>
>                    Df Deviance Resid. Df Resid. Dev Pr(>Chi)
> NULL                                   17     36.167
> moistcont            1   8.6322        16     27.535 0.003303 **
> orgmatter            1   2.1244        15     25.411 0.144966
> baresoil             1   0.0029        14     25.408 0.956986
> grass10              1   1.5251        13     23.883 0.216842
> wood10               1   3.6952        12     20.187 0.054570 .
> rdnet10              1   0.0001        11     20.187 0.990564
> moistcont:orgmatter  1   2.0482        10     18.139 0.152381
> moistcont:baresoil   1   2.8730         9     15.266 0.090076 .
> grass10:wood10       1   0.1431         8     15.123 0.705247
> grass10:rdnet10      1   0.4141         7     14.709 0.519883
> wood10:rdnet10       1  10.7812         6      3.928 0.001025 **
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> 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.

David Winsemius, MD
Alameda, CA, USA