The 'segmented' package in R doesn't seem to work with mixed linear models, and since I repeatedly used the same animals, I need to include 'Animal ID' as a random factor, so I need to do a mixed effects model which doesn't seem possible using 'segmented' in R. Does anyone know if there is a way I can do this in R?
These might be relevant:
http://stackoverflow.com/questions/22313728/interpretation-of-piecewise-mixed-effects-output
http://stats.stackexchange.com/questions/19772/estimating-the-break-point-in-a-broken-stick-piecewise-linear-model-with-rando
What do you mean by using the same animals many times? Usually, if you use the same subject multiple times, you need to use Repeated Measures, Longitudinal Analysis, Panel Data Analysis, etc.
Andrew, I measured metabolic rate in the same group of birds, over a variety of ambient temperatures. But due to the body heat of the bird heating up their immediate environment, each ambient temperature was slightly different. So that rules out a repeated measures ANOVA. Also, since some birds don't have readings at certain temperatures, the power of repeated measures in this case is very low. Mixed effects analyses on the other hand can account for missing values, which is why I wanted to do piecewise regression using mixed effects. I'll have a look at the methods you suggested, many thanks.
I got it to work:
mod.1 = lmer(VO2WA ~ Sex + BodyMass + I(Ta * (Ta < 20)) + I(Ta* (Ta>= 20)) + (1 | BirdID), data = dat)
However, I needed to write a bunch of models, each with a different breakpoint (in this case I've specified a breakpoint at 20, i.e. temperature 'Ta' = 20C), and then compare models using AICc.
For finding the breakpoint in piecewise regression, I used the code given here:
http://stats.stackexchange.com/questions/19772/estimating-the-break-point-in-a-broken-stick-piecewise-linear-model-with-rando/19777#19777
where the random factor (BirdID in my case), should just be in numbers, not a combination of letters and numerals.
Thanks so much Gregory, it took a while, but eventually it worked, many thanks.
To calculate the breakpoint in R with piecewise method, you can use this code:
require(SiZer)
model.pwl = piecewise.linear(x, y, middle = 1, CI = FALSE, bootstrap.samples = 1000, sig.level = 0.05)
plot(model.pwl)
print(model.pwl)
in the call:
the Threshold alpha is the breakpoint
a = Intercept
slope1(b1) = x
slope2(b2) = (x+w)
Thanks Carlos, here is some other code for doing piecewise regression in R
rm(list=ls())
dat
To fit segmented mixed models (with random changepoints) have a look to
https://www.researchgate.net/publication/292629179
best,
vito
Working Paper Segmented mixed models with random changepoints in R
So useful! Thank you! This was exactly what I was looking for for my data.
I was recently asked for a paper where I cited the above method, here it is in case it is useful to anyone, please let me know if you want a copy.
Thompson, L.J., Brown, M. and Downs, C.T., 2015. Seasonal metabolic variation over two years in an Afrotropical passerine bird. Journal of thermal biology, 52, pp.58-66.
Good day, I would like a copy of this article please as I will be needing to do regressive analysis in my thesis. Thank you
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R Statistical Package