Hi All,
I am using the package 'mlogit'
I am having trouble constructing a nested logit model. I am headed down
this path as I am violating the IIA assumption when running a multinomial
logistic regression.
I am analyzing the choice a seabird floating on the surface of the water
makes when approached by a ship. I have three choices defined for the
dependent variable 'act2'; fly, dive, or remain on the water (none).
Independent variables I hypothesize might affect the decision are
perpendicular distance to the ship's path, group size, location, distance
to shore, speed of the ship, number of ships in the area and wind speed.
All the variables are individual specific.
A non-nested mutinomial logit model will run fine using the following code
(data converted to long format already).
> mod.1 <- mlogit(act2 ~ 0 | dist + as.factor(grp.bin) + as.factor(ship)
+
as.factor(sea) + avgknots + shore + as.factor(location),long_perp ,
reflevel = 'none')
After assessing that I am violating the IIA assumption I attempted to run a
nested logit model with the following code.
> nested.1 <- mlogit(act2 ~ 0 | dist + as.factor(grp.bin) +
as.factor(ship) +
as.factor(sea) + avgknots + shore + as.factor(location),long_perp ,
reflevel = 'none',
nests = list(react = c("dive","fly"), remain =
"none"), unscaled=TRUE)
I get the following error code.
"Error in solve.default(crossprod(attr(x, "gradi")[, !fixed])) :
Lapack routine dgesv: system is exactly singular: U[6,6] = 0"
Does this indicate that my variables are col-linear?, overfitting? I have
tried to remove all but one variable and rerun and still get an error code.
Can anyone help me get over this hump or help suggest a different modeling
approach?
I am mainly interested in the probability of choosing to react (fly or
dive) or not, and then once a reaction has been made, which one is chosen
and how these decisions relate to perpendicular distance to the ship's path.
Thanks, Tim
--
Tim Marcella
508.498.2989
timmarcella@gmail.com
[[alternative HTML version deleted]]