Hi all,
Where f(x) is a logistic function, I have data that follow:
g(x) = f(x)*.5 + .5
How would you suggest I modify the standard glm(..., family='binomial')
function to fit this? Here's an example of a clearly ill-advised attempt to
simply use the standard glm(..., family='binomial') approach:
########
# First generate some data
########
#define the scale and location of the modified logistic to be fit
location = 20
scale = 30
# choose some x values
x = runif(200,-200,200)
# generate some random noise to add to x in order to
# simulate real-word measurement and avoid perfect fits
x.noise = runif(length(x),-10,10)
# define the probability of success for each x given the modified logistic
prob.success = plogis(x+x.noise,location,scale)*.5 + .5
# obtain y, the observed success/failure at each x
y = rep(NA,length(x))
for(i in 1:length(x)){
y[i] = sample(
x = c(1,0)
, size = 1
, prob = c(prob.success[i], 1-prob.success[i])
)
}
#show the data and the source modified logistic
plot(x,y)
curve(plogis(x,location,scale)*.5 + .5,add=T)
########
# Now try to fit the data
########
#fit using standard glm and plot the result
fit = glm(y~x,family='binomial')
curve(plogis(fit$coefficients[1]+x*fit$coefficients[2])*.5+.5,add=T,col='red',lty=2)
# It's clear that it's inappropriate to use the standard
"glm(y~x,family='binomial')"
# method to fit the modified logistic data, but what is the alternative?
--
Mike Lawrence
Graduate Student
Department of Psychology
Dalhousie University
www.thatmike.com
Looking to arrange a meeting? Check my public calendar and find a time that
is mutually free:
http://tinyurl.com/mikescalendar
(tiny url redirects to google calendar)
~ Certainty is folly... I think. ~
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
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.
