1
$\begingroup$

I have done an experiment of how different animal species affect nutrient fluxes in sediments. I had a number of experimental units (sediment boxes) to which different animal species were added. I then measured nutrient fluxes in these units. Each unit was measured five times. I have used the ezANOVA package for R to do a repeated measures ANOVA with two factors (between subjects factor “Species” and within subject factor “Time”). Here's an example of my input:

> ezANOVA(data=NoP_3_5,dv=.(AcPO4),wid=.(Subject),within=.(Time),between=.(Species),return_aov=T)
$ANOVA
    Effect DFn DFd           F            p p<.05        ges
2      Species   2   6 10.60384830 1.072453e-02     * 0.50875722
3         Time   2  12 27.88590570 3.081718e-05     * 0.76667542
4 Species:Time   4  12  0.08093179 9.867291e-01       0.01871588

$`Mauchly's Test for Sphericity`
        Effect         W           p p<.05
3         Time 0.1439855 0.007866779     *
4 Species:Time 0.1439855 0.007866779     *

$`Sphericity Corrections`
        Effect       GGe       p[GG] p[GG]<.05       HFe       p[HF] p[HF]<.05
3         Time 0.5387889 0.001349263         * 0.5630403 0.001103048         *
4 Species:Time 0.5387889 0.933492515           0.5630403 0.939138882          

$aov

Call:
aov(formula = formula(aov_formula), data = data)

Grand Mean: -0.004276663 

Stratum 1: Subject

Terms:
                     Species    Residuals
Sum of Squares  0.0004410742 0.0001247870
Deg. of Freedom            2            6

Residual standard error: 0.004560464 
Estimated effects may be unbalanced

Stratum 2: Subject:Time

Terms:
                        Time Species:Time    Residuals
Sum of Squares  0.0013994206 0.0000081229 0.0003011028
Deg. of Freedom            2            4           12

Residual standard error: 0.005009181 
Estimated effects may be unbalanced

I want to use the $aov output to do a Tukey HSD post hoc test of the between factor ("Species"). However, it does not work:

mod <- ezANOVA(data=NoP_3_5,dv=.(AcPO4),wid=.(Subject),within=.(Time),between=.(Species),return_aov=T)
    > TukeyHSD(mod$aov)
Error in UseMethod("TukeyHSD") : 
  no applicable method for 'TukeyHSD' applied to an object of class "c('aovlist', 'listof')"

I have also tried to use the aov-command directly:

> mod2 <- aov(AcPO4~(Species)+Error(Subject/Time)+(Species),data=NoP_3_5)
> TukeyHSD(mod2)
Error in UseMethod("TukeyHSD") : 
  no applicable method for 'TukeyHSD' applied to an object of class "c('aovlist', 'listof')"

I have tried to use lme() to specify the anova-model with the aim to produce something that TukeyHSD() will accept without success.

Any help would be greatly appreciated!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

In R when you have any ANOVA (other than simple one-factor), you have to provide to TukeyHSD the variables to have intervals calculated. 

summary(cc<-aov(weight~as.factor(Diet)+as.factor(Chick),data=ChickWeight))
TukeyHSD(cc,"as.factor(Diet)",data=ChickWeight)

Use ?TukeyHSD to get the detailed help.

Improve this answer
$\endgroup$
5
  • $\begingroup$ Thanks for your reply. I tried specifying the variable and data. TukeyHSD(mod$aov,"Species",data=NoP_3_5) and TukeyHSD(mod$aov,Species,data=NoP_3_5) and TukeyHSD(mod$aov,"as.factor(Species)",data=NoP_3_5) but keep getting the same error message as in the above question. $\endgroup$
    – Nils Ekeroth
    Mar 18, 2013 at 14:39
  • $\begingroup$ I just went to the helps for the ez package where you are getting ezANOVA from. From my quick read, it appears that this package does not support TukeyHSD. However, you need to look at the provided ezPlot and ezStats functions for your problem. They both do post-hoc comparisons for you, among other things. $\endgroup$
    – R. Schumacher
    Mar 18, 2013 at 15:20
  • $\begingroup$ The problem with ezStats and ezPlot is that the post-hoc testing is by Fisher's Least Significant Difference (FLSD). I dont think it is appropriate to compare between-subject effects (i.e. Species in my case) with FLSD when the anova contains also within-subject effets (i.e. Time). As the interaction is not significant, would it be possible omit "Time", re-run the anova with only "Species" and then do the post-hoc comparisons based on FLSD? $\endgroup$
    – Nils Ekeroth
    Mar 18, 2013 at 16:26
  • $\begingroup$ One needs to be careful with n-way Anova. If there is not interaction and the effects are strictly additive, then I think you can do that. (but I'm not sure of the theory support without a careful literature search). $\endgroup$
    – R. Schumacher
    Mar 18, 2013 at 19:14
  • 1
    $\begingroup$ Second idea. Send the author of the ez package a description of what you doing and what you want. I personally found that the majority of package writers are extremely helpful on assisting users of their packages. The author's address in the documentation for ez. (??ez and look toward the bottom of the documentation) $\endgroup$
    – R. Schumacher
    Mar 18, 2013 at 19:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.