Help establishing mixed model equation for split, plot design
Hello again everyone. Some weeks ago I wrote asking for help for establishing the adequate mixed effects model in a split plot design I'm working on. After many weeks of reading and surfing the web, this list gave me the direction I desperately needed. I'm facing now another challenge and I hope you will be able to help me (or point me in the right direction). The original mail is at the end of this one, I'll just copy in here the experiments description before asking you my questions. "The experiment consists in meassuring the germination percentage of one species in variable conditions of salinity, immersion time and presence of other species. I sow seeds in soil core's, that are inside plastic boxes that are filled periodically with saline water. There are three factors: salinity (3 levels: 0, 5 and 18), immersion time (3 levels: 0, 20, 40%), species treatment (2 levels: Baccharis, Baccharis + Juncus). Inside each box there are 6 cores that combine in a complete random design the factors of immersion and species treatment. The box is filled with water at one of the levels of salinity. Thus, I'm using a split plot design with salinity as the whole plot factor, and immersion and species treatment as the subplot factors. All factors are considered fixed. Each box is repeated 5 times. Thus, there are 15 boxes and 90 soil cores. The dependent variable is the percentage of germination of the species Baccharis in each core." For another dependent variable I measured, the conditions of normality and homoestacity are not fulfilled, so I'm thinking about using a non-parametric test to analyze the data. The thing I don't know which one is the best suited for my design and what R packages could be useful. I've seen options in the nparLD (https://cran.r-project.org/web/packages/nparLD/nparLD.pdf), the WRS2 package (https://cran.r-project.org/web/packages/WRS2/vignettes/WRS2.pdf), and the rlme package (https://www.rdocumentation.org/packages/rlme/versions/0.5/topics/rlme), but with none of this I can clear my head on how to make them work, and specially if they are what I need. This time, for simplifying the array, I'm interest in making bivariate tests using only the whole plot factor Salinity, and the sub-plot factor Immersion, using only the cores that had one species sown (removing the competition element). Could you give some guidance on what type of non-parametric tests I should use (if there's any!), and which package you consider to be the best suited for an array like this? Thanks a lot in advance, any tip is very well appreciated. Best regards, Felipe Calleja Ap?stegui Predoctoral researcher Instituto de Hidr?ulica Ambiental "IH Cantabria" C/ Isabel Torres, N? 15 Parque Cient?fico y Tecnol?gico de Cantabria 39011 Santander (Espa?a) www.ihcantabria.es<http://www.ihcantabria.es/> Tel: +34 942 20 16 16 Ext. 1153 Fax: +34 942 26 63 61 e-mail: felipe-francisco.calleja at alumnos.unican.es ------------------------------------------------------------------------------------------------------
On 03/04/18 19:16, r-sig-mixed-models-request at r-project.org wrote:
Message: 1
Date: Tue, 3 Apr 2018 07:53:15 +0000
From: "CALLEJA APESTEGUI, FELIPE FRANCISCO"
<felipe-francisco.calleja at alumnos.unican.es>
To: "r-sig-mixed-models at r-project.org"
<r-sig-mixed-models at r-project.org>
Subject: [R-sig-ME] Help establishing mixed model equation for split
plot design
Message-ID:
<DB6P191MB0088D09AF911107B33C1DE5C87A50 at DB6P191MB0088.EURP191.PROD.OUTLOOK.COM>
Content-Type: text/plain; charset="iso-8859-1"
Hello,
I'm looking for some help establishing a mixed model ANOVA using R, for a split plot design I've made for an experiment of saltmarsh germination. I'll explain as clear as possible the experimental design and afterwards what I've done and my doubts. Hope you can help me. I haven't worked very much with R so many of my doubts are about what I'm "telling" it to do with one or other command.
The experiment consists in meassuring the germination percentage of one species in variable conditions of salinity, immersion time and presence of other species. I sow seeds in soil core's, that are inside plastic boxes that are filled periodically with saline water. There are three factors: salinity (3 levels: 0, 5 and 18), immersion time (3 levels: 0, 20, 40%), species treatment (2 levels: Baccharis, Baccharis + Juncus). Inside each box there are 6 cores that combine in a complete random design the factors of immersion and species treatment. The box is filled with water at one of the levels of salinity. Thus, I'm using a split plot design with salinity as the whole plot factor, and immersion and species treatment as the subplot factors. All factors are considered fixed. Each box is repeated 5 times. Thus, there are 15 boxes and 90 soil cores. The dependent variable is the percentage of germination of the species Baccharis in each core.
I have several doubts about how to analyze the array.
1 - As far as I understand, although I'm treating all my factors as fixed, this is a mixed model because of the interaction between subjects (the boxes I believe), and the "split plot nature" of the array, right? In that sense, which function would be better to analyze this, the aov of the stats package, the lme of the nlme package, the lmer of lme4?
2 - I've had trouble calculating the degrees of freedom for the residuals. The only reference I have is that the error of the whole plot part should have 12 df's, and the within error should have 60 df's. With that reference I've established two possible R commands:
fit.aov2 <- aov(Plantsurvival ~ salinityF*immersionF*SpecTF + Error(rep:salinityF/immersionF:SpecTF), data=sp.datos)
With rep being the number of repetition. This last one gives the 12 and 60 df's for the error terms.
The other option is:
fit.okay <- lme(Plantsurvival ~ salinityF*immersionF*SpecTF, random= ~1|rep/salinityF, data=sp.datos)
But in this last case, the df's are 8 and 60, which makes me suspect maybe there is something wrong. But as I said, I haven't cleared my head on which should be the correct df's.
Questions: Is the aov line solving a mixed model adequate for my design?,
Is the lme line considering salinityF as a random factor? If it is, how can I tell it to consider all factors as fixed, but put salinity at the "higher level" of the whole plot and the other ones in the "lower level" of the subplot?
I hope the questions and the experimental array are clear. If there is any doubt or need more information please let me know. I attach a csv file with the data in case you want to see it.
Finally, if is not too much to ask, I'm fairly new to the splitplot anova's and R, so I would really appreciate if you could answer be with as much detail as possible, to fully understand what's going on and where to continue.
Thanks a lot,
Felipe Calleja Ap?stegui
Predoctoral researcher
Instituto de Hidr?ulica Ambiental "IH Cantabria"
C/ Isabel Torres, N? 15
Parque Cient?fico y Tecnol?gico de Cantabria
39011 Santander (Espa?a)
www.ihcantabria.es<http://www.ihcantabria.es/>
Tel: +34 942 20 16 16 Ext. 1153
Fax: +34 942 26 63 61
e-mail: felipe-francisco.calleja at alumnos.unican.es