changing family distribution, but not as a part of a glm or regression model

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

changing family distribution, but not as a part of a glm or regression model

kkarr
Hi,
Thanks for reading and any suggestions.  I am slowly teaching myself R.  I am currently modeling critical thresholds or breakpoints/switchpoints in ecological systems.  I have using the strucchange package, but tweaking it for my questions at hand.

What I am trying to figure out is how to tell the model that my metric (dependent variable) is either binomial (logit-link), negative binomial (log-link) or multinomial (log-link), before I run the model in strucchange.

I have found several suggestion online, but they all refer to packages and a glm or gam is part of the syntax.  Am I missing something?  
I do have two options in strucchange to interpret the data, and change the assumptions about the distribution of the data, for both of these myrespones is the y-axis or dependent variable.

1. interp.data <- myresponse
# for (i in 1:length(interp.data))
# {if (is.na(interp.data[i]))
# interp.data[i] <- rnorm(1,mean(interp.data[mystart:myend],na.rm=T),
# sd(interp.data[mystart:myend],na.rm=T))}

2. myresponse <- (myresponse-mean(myresponse, na.rm=TRUE))/sd(myresponse, na.rm=TRUE)

 Since I am new to R, sorry if my terminology is not right, hopefully you can get the basics of my question.
thanks,
Kendra