[R] Combining two ANOVA outputs of different lengths

[Peter Dalgaard]

Christoph Scherber wrote:
> Dear R users,
>
> I have been trying to combine two anova outputs into one single table 
> (for later publication). The outputs are of different length, and share 
> only some common explanatory variables.
>
> Using merge() or melt() (from the reshape package) did not work out.
>
> Here are the model outputs and what I would like to have:
>
> anova(model1)
>              numDF denDF  F-value p-value
> (Intercept)     1    74 0.063446  0.8018
> days     	1    74 6.613997  0.0121
> logdiv          1    74 1.587983  0.2116
> leg             1    74 4.425843  0.0388
>
> anova(model2)
>               numDF denDF   F-value p-value
> (Intercept)      1    73 165.94569  <.0001
> funcgr           1    73   7.91999  0.0063
> grass            1    73  42.16909  <.0001
> leg              1    73   4.72108  0.0330
> funcgr:grass     1    73   8.49068  0.0047
>
> #"merge(anova(model1),anova(model2),...)"
>
> 		F-value 1	p-val1  F-value 2	p-value 2
> (Intercept)	0.063446	0.8018	165.94569	<.0001
> days		6.613997	0.0121	NA		NA
> logdiv		1.587983	0.2116	NA		NA
> leg		4.425843	0.0388	4.72108		0.033
> funcgr		NA		NA	7.91999	  	0.0063
> grass		NA		NA	42.16909	<.0001
> funcgr:grass	NA		NA	8.49068		0.0047
>
>
> I would be glad if someone would have an idea of how to do this in 
> principle.
>   
The main problems are that the merge key is the rownames and that you 
want to keep entries that are missing in one of the analysis. There are 
ways to deal with that:

 > example(anova.lm)
.....
 > merge(anova(fit2), anova(fit4), by=0, all=T)
  Row.names Df.x  Sum Sq.x Mean Sq.x F value.x    Pr(>F).x Df.y  Sum Sq.y
1      ddpi   NA        NA        NA        NA          NA    1  63.05403
2       dpi   NA        NA        NA        NA          NA    1  12.40095
3     pop15    1 204.11757 204.11757 13.211166 0.000687868    1 204.11757
4     pop75    1  53.34271  53.34271  3.452517 0.069425385    1  53.34271
5 Residuals   47 726.16797  15.45038        NA          NA   45 650.71300
  Mean Sq.y  F value.y     Pr(>F).y
1  63.05403  4.3604959 0.0424711387
2  12.40095  0.8575863 0.3593550848
3 204.11757 14.1157322 0.0004921955
4  53.34271  3.6889104 0.0611254598
5  14.46029         NA           NA



Presumably, you can take it from here.