Dear Sam, 
You should consider mean sums of squares rather than rough sums of squares.... 
(F stat is computed on the ratio of these not on SS directly). 
If you design was replicating measurement per individual, and if mean sums of 
squares for error is greater than mean sums of squares for individual 
variation, it means that the expected trace for individual variances is 
negative; and that your percentage of measurement error (or intraclass 
coefficient) is likely greater than what you are measuring. 
You can try to provide the percentage of measurement error by computing 
expected traces... I think some of us have been using that strategy in the past 
and this can give you an idea (see for instance how I made it in the hystrix 
volume script; it is referring to Yzerinac et al in syst biology, 1992).. 

let's call 
m: number of replicates per individual 
SSi: trace for inter-individual sum of squares and crossproduct 
SSe: trace for intra-individual sum of squares and crossproduct 
n: numer of individuals 

Your averaged sums of squares are obtained: 
MSSi<-SSi/(n-1) 
MSSe<-SSe/(nm - n) 
your percentage of measurement error should be: 
MSSe / [MSSe+ (MSSi- MSSe)/m] 

In your case, F being greater than 1, then MSSe> MSSi and the last denominator 
will exceed the numerator; so your percentage of measurement error is higher 
than 100%. You are then trying to measure something that is not measurable 
because you do not have enough precision. Individual shape variation is 
sometime very subtle and can not be estimated, but maybe you made a super big 
error measurement for one or two individuals...To avoid that, you can try to 
increase replicates, or to see whether there is not one or a set of individuals 
on which error was extreme and correct your digitization. If you are lucky, the 
percentage will become < 100%, and then it will mean that you will be able to 
measure something from your data. 

Julien 

> De: "ying yi" <[email protected]>
> À: "Morphmet" <[email protected]>
> Envoyé: Jeudi 3 Novembre 2022 15:19:18
> Objet: [MORPHMET2] Measurement error in geometric morphometrics

> Dear all,
> I used the “procD.lm” function in the geomorph package to test the measurement
> error. I was surprised to find that the within-groups ANOVA sum of squares I
> got was greater than the among-groups ANOVA sum of squares. I wonder if
> something went wrong. What does it mean for “procD.lm” function to get an F
> value <1?
> I would be very happy if someone could help me.
> Yours,
> Sam

> References are as follows:

> --
> You received this message because you are subscribed to the Google Groups
> "Morphmet" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email
> to [ mailto:[email protected] |
> [email protected] ] .
> To view this discussion on the web visit [
> https://groups.google.com/d/msgid/morphmet2/06065841-c42e-4a58-a5d3-a96eb3c5787dn%40googlegroups.com?utm_medium=email&utm_source=footer
> |
> https://groups.google.com/d/msgid/morphmet2/06065841-c42e-4a58-a5d3-a96eb3c5787dn%40googlegroups.com
> ] .

-- 
You received this message because you are subscribed to the Google Groups 
"Morphmet" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/morphmet2/518085623.3329405.1667531417996.JavaMail.zimbra%40umontpellier.fr.

Reply via email to