Hello,
I am trying to do a BACI analysis on count data, and I am having trouble
interpreting the output from multcomp::glht. I don't understand how
the contrast's coefficients are related to effect size (if at all??).
I have 5 treatment conditions (one is a control), and I have counts from
before the treatments were applied and after. Let's say that my model
form is this ("Period" is the 'before' vs 'after' factor):
m.pois <- glm(Y_count ~ Treatment + Period + Treatment:Period,
data = df.temp,
family = "poisson")
As in all BACI designs, I am interested in the interaction term, i.e.,
the differences of the differences. For example, I'd like to test
whether TreatmentVR30 changed more than the Control did:
(TreatmentVR30Later - TreatmentVR30Before) - (ControlLater -
ControlBefore).
I have done the math, and I created all my planned contrasts, run them
through the multcomp::glht, and I am struggling to interpret the
output. As an example of my confusion, I ran the same contrast in two
directions (A-B and B-A), which should give the same result (one
positive, one negative):
contr <- rbind(
"VR30 vs Control" = c(0, 0, 0, 0, 0, 0, -1, 0, 0, 1),
"Control vs VR30" = c(0, 0, 0, 0, 0, 0, 1, 0, 0, -1) )
m.pois.contr <- summary(glht(m.pois, contr))
which works perfectly, returning one positive and one negative estimate,
as expected:
Linear Hypotheses:
Estimate Std. Error z value Pr(>|z|)
VR30 vs Control == 0 0.7354 0.5621 1.308 0.191
Control vs VR30 == 0 -0.7354 0.5621 -1.308 0.191
(Adjusted p values reported -- single-step method)
Understanding that the estimates are in log space (due to the link
function of the poisson family in the glm), I back transformed using
exp(coef(m.pois.contr) to get:
VR30 vs Control Control vs VR30
2.0862414 0.4793309
So, which is it? Did Control change more than VR30, or did VR30 change
more than Control, and for both questions, by how much?
Clearly I am missing something here. I expect that this will be a
simple fix, but surprisingly, I cannot find it anywhere online.
Thanks in advance to anyone who can help,
David Robichaud, Victoria, BC, Canada
Interpreting GLM Interaction Contrasts in R (using glht)
3 messages · David Robichaud, Thierry Onkelinx
Dear David, You have missed the fact that exp(-a) = 1/exp(a). Additive effects on the log scale are multiplicative effects on the original scale. Best regards, ir. Thierry Onkelinx Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance Kliniekstraat 25 1070 Anderlecht Belgium To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of. ~ Sir Ronald Aylmer Fisher The plural of anecdote is not data. ~ Roger Brinner The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data. ~ John Tukey 2015-05-15 0:53 GMT+02:00 David Robichaud <drobichaud at lgl.com>:
Hello,
I am trying to do a BACI analysis on count data, and I am having trouble
interpreting the output from multcomp::glht. I don't understand how the
contrast's coefficients are related to effect size (if at all??).
I have 5 treatment conditions (one is a control), and I have counts from
before the treatments were applied and after. Let's say that my model form
is this ("Period" is the 'before' vs 'after' factor):
m.pois <- glm(Y_count ~ Treatment + Period + Treatment:Period,
data = df.temp,
family = "poisson")
As in all BACI designs, I am interested in the interaction term, i.e., the
differences of the differences. For example, I'd like to test whether
TreatmentVR30 changed more than the Control did:
(TreatmentVR30Later - TreatmentVR30Before) - (ControlLater -
ControlBefore).
I have done the math, and I created all my planned contrasts, run them
through the multcomp::glht, and I am struggling to interpret the output.
As an example of my confusion, I ran the same contrast in two directions
(A-B and B-A), which should give the same result (one positive, one
negative):
contr <- rbind(
"VR30 vs Control" = c(0, 0, 0, 0, 0, 0, -1, 0, 0, 1),
"Control vs VR30" = c(0, 0, 0, 0, 0, 0, 1, 0, 0, -1) )
m.pois.contr <- summary(glht(m.pois, contr))
which works perfectly, returning one positive and one negative estimate,
as expected:
Linear Hypotheses:
Estimate Std. Error z value Pr(>|z|)
VR30 vs Control == 0 0.7354 0.5621 1.308 0.191
Control vs VR30 == 0 -0.7354 0.5621 -1.308 0.191
(Adjusted p values reported -- single-step method)
Understanding that the estimates are in log space (due to the link
function of the poisson family in the glm), I back transformed using
exp(coef(m.pois.contr) to get:
VR30 vs Control Control vs VR30
2.0862414 0.4793309
So, which is it? Did Control change more than VR30, or did VR30 change
more than Control, and for both questions, by how much?
Clearly I am missing something here. I expect that this will be a simple
fix, but surprisingly, I cannot find it anywhere online.
Thanks in advance to anyone who can help,
David Robichaud, Victoria, BC, Canada
______________________________________________ R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
Hi Thierry and everyone else, This makes sense! I knew it would be something totally straightforward. Thanks for taking the time to let me know, Dave