[R-sig-eco] manova repeated measures + Error
Steve Brewer
jbrewer at olemiss.edu
Fri Feb 13 22:51:49 CET 2015
Marina,
Eu falo um pouco de Português.
I don't think my suggestion will result in pseudoreplication. Assuming there
is only one adult individual per plot, plot is the random source of error.
When you specify "plot" as the error term, I believe R will sum (or average)
the responses across all times within each plot and use this sum to test the
between-subjects effects of nurse, target, and nurse:target interaction.
I think it would be pseudoreplicaton if you specified no error term or
specifically used an error term that was equivalent to the observation
number. In that case, you would not get separate between-subjects and
within-subjects tests. It's easy to check though. Give it a try and check
the between subjects and within subjects error df.
It wasn't clear to me what the between-subjects anova table was. Time should
not be one of the between-subjects effects.
I'm also not clear as to what the "target" treatment is. Can I assume the
same three seedling species are used in each plot? Are you averaging the
seedling responses within each plot? The large residuals error df (which
seems too large) suggests that you're calculating and analyzing a separate
RII for each seedling species and not averaging them. If so, THAT is
pseudoreplication. Target species should be treated as a split-plot effect.
This complicates the design and analysis greatly. The resulting anova table
would need to be partitioned into four parts!
Between subjects – whole plot
Nurse____________________________________________df = 20-1 = 19
Whole plot bs error = Plot(Nurse)_______________________df = (5-1)*20 = 80
Within subjects – whole plot
Time_____________________________________________df = 8-1 = 7
Nurse*Time_______________________________________df = (20-1)*(8-1) = 133
whole plot ws error =Time*Plot(Nurse)__________________df = (8-1)*(5-1)*20 =
560
Between subjects – split plot
Target____________________________________________df = (3-1) = 2
Nurse*Target______________________________________df = (20-1)*(3-1) = 38
Whole-plot bs error = Target*Plot(Nurse)________________df= (3-1)*(5-1)*20 =
160
Within subjects – split plot
Time*Target_______________________________________df = (8-1)*(3-1) = 14
Time:Nurse:Target__________________________________df = (8-1)*(20-1)*(3-1) =
266
Split-plot ws error = Time:Target:Plot(Nurse)______________df =
(8-1)*(3-1)*(5-1)*(20) = 1120
Total df = 2399 or 2400 – 1, 20 nurse species x 5 plots x 3 target species x
8 weeks minus 1.
I'm not sure how to advise you on how to get R to do this with manova.
Do you really need to do a repeated measures analysis? Why not just use the
first and last measurements of height and leaf number, log transform them,
and obtain relative height growth rate and a relative leaf production rate.
You could then take the difference between these relative rates between the
two quadrants (one with the target species and one without) to obtain a
measure of interaction intensity. That would eliminate the time variable and
make it a simple split-plot analysis.
Good luck,
Steve
J. Stephen Brewer
Professor
Department of Biology
PO Box 1848
University of Mississippi
University, Mississippi 38677-1848
Brewer web page - http://home.olemiss.edu/~jbrewer/
FAX - 662-915-5144 Phone - 662-202-5877
From: Marina Vergara Fagundes <marvefa at gmail.com>
Date: Fri, 13 Feb 2015 15:47:06 -0300
To: Pyro Maniac <jbrewer at olemiss.edu>
Subject: Re: [R-sig-eco] manova repeated measures + Error
Hi!
Thanks for you reply.
Yes, tempo means time (sorry for that), and my variables plot and time are
factors.
I think if I don't put 'time' on Error, wont be pseudoreplication ?
I have 8 measures of the same individuals, because i want to understand how
nurses affect targets through time.
What more intrigues me its the error message "No error degrees of freedom"
for time and everything else in anova's table are fine.
Thanks again for your time and patience!
Marina.
2015-02-13 15:00 GMT-03:00 Steve Brewer <jbrewer at olemiss.edu>:
> Marina,
>
> I'm a little confused by your description. What is "tempo"? Is it just the
> Portuguese word for the time variable?
>
> From what I gather, "plot" should be nested within "nurse" and "target",
> but not within time. So, I think your model should be manova(Y ~
> nurse*target*time + Error(plot). Not sure why tempo (assuming it is the
> same as time) would need to be in parentheses.
>
> Make sure that the time and plot variables are not being treated as
> numeric continuous variables. They should be categorical variables.
>
> Finally, a comment about RII. I would be wary of using this index. I don't
> think it is necessary, and, like proportions, it usually is NOT normally
> distributed. Just because many people use it doesn't mean it is a good
> idea. I think just log transforming the data and taking the difference
> between the nurse-present and nurse-absent plot halves would provide you
> what you want (it is equivalent to a log response ratio, LRR, which is
> superior to RII).
>
> Good luck,
> Steve
>
>
>
>
> J. Stephen Brewer
> Professor
> Department of Biology
> PO Box 1848
> University of Mississippi
> University, Mississippi 38677-1848
> Brewer web page - http://home.olemiss.edu/~jbrewer/
> FAX - 662-915-5144 <tel:662-915-5144>
> Phone - 662-202-5877 <tel:662-202-5877>
>
>
>
>
> On 2/13/15 9:58 AM, "Marina Vergara Fagundes" <marvefa at gmail.com> wrote:
>
>> >Dear all,
>> >
>> >I conducted a interaction plant experiment, with 20 adult plant species (5
>> >individuals each) and 3 seedling species, and a 100 plots in total. Each
>> >plot consist of two 2m² quadrants, the first contains one adult individual
>> >as a potential “nurse” and 3 seedling species below their canopy, and the
>> >second just with 3 seedling species with no canopy influence (in an open
>> >area).
>> >I measured growth variables of the seedlings (leaves number and height)
>> >over 8 weaks (times).
>> >For each growth variable I calculated a index (rii) with the difference of
>> >growth below and far from canopy.
>> >
>> >I'm analyzing this data with MANOVA with repeated measures. Where Y axes
>> >are the index of growth measurements (for leafs and height), and the X
>> >axes
>> >are the 20 nurse species and interactions with seedlings species and time.
>> >
>> >This is how my data frame is organized:
>> >plot nurse target time height_rii leaves_rii
>> >113 algodao angico 1 0.03703704 0.00000000
>> >113 algodao aroeira 2 0.05660377 0.09090909
>> >113 algodao cating 3 0.03846154 0.00000000
>> >115 algodao angico 1 0.01818182 0.11111111
>> >115 algodao aroeira 2 0.05660377 0.50000000
>> >115 algodao cating 3 0.03703704 0.00000000
>> >
>> >This is the code for the MANOVA:
>> >Y <- with(tabela3, cbind(altura_rii, folhas_rii))
>> >m1 <- manova( Y ~ nurse*target*time+Error(tempo/plot),data=tabela3)
>> >summary(m1)
>> >
>> >When i ask for anova table i got this specific message:
>> >
>> >
>> >
>> >
>> >
>> >
>> >
>> >*Error: timeNo error degrees of freedom Dftempo 7*Error: time:plot
>> > Df Pillai approx F num Df den Df Pr(>F)
>> >nurse 19 0.17070 3.14316 38 1280 8.633e-10 ***
>> >nurse:time 133 0.24253 0.66405 266 1280 1
>> >Residuals 640
>> >---
>> >Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>> >
>> >Error: Within
>> > Df Pillai approx F num Df den Df Pr(>F)
>> >target 2 0.013439 4.3297 4 2560 0.001715 **
>> >nurse:target 38 0.178898 3.3090 76 2560 < 2.2e-16 ***
>> >target:time 14 0.027914 1.2941 28 2560 0.138459
>> >nurse:target:time 266 0.174314 0.4594 532 2560 1.000000
>> >Residuals 1280
>> >---
>> >Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>> >
>> >Any idea what is it about?
>> >Thanks a lot for help!
>> >Marina
>> >
>> > [[alternative HTML version deleted]]
>> >
>> >_______________________________________________
>> >R-sig-ecology mailing list
>> >R-sig-ecology at r-project.org
>> >https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>
>
[[alternative HTML version deleted]]
More information about the R-sig-ecology
mailing list