yes On Fri, Mar 13, 2009 at 2:34 PM, charlotte cavaille < charlotte.cavaille at gmail.com> wrote: -- Patrick Lam Department of Government and Institute for Quantitative Social Science, Harvard University http://www.people.fas.harvard.edu/~plam

you should be plotting proportion of 1s correct versus proportion of zeros correct for various values of C. it may end up being a bunch of points on the graph, but you can connect them and smooth it out to a line. On Fri, Mar 13, 2009 at 3:14 PM, charlotte cavaille < charlotte.cavaille at gmail.com> wrote: -- Patrick Lam Department of Government and Institute for Quantitative Social Science, Harvard University http://www.people.fas.harvard.edu/~plam

Charlotte, If you are using a for-loop that may explain why it is taking so long to run. If you use apply() it takes under a second even for a long vector of cutpoints. Brandon On Sat, Mar 14, 2009 at 5:12 PM, charlotte cavaille < charlotte.cavaille at gmail.com> wrote:

One thing jumps out...you seem to be missing a fourth "0" in your optim() line. It should read: optim.out <- optim(par = c(0,0,0,0), scobit.ll, y = y, X = X, method="BFGS",control=list(fnscale = -1), hessian = T) From: gov2001-l-bounces at lists.fas.harvard.edu [mailto:gov2001-l-bounces at lists.fas.harvard.edu] On Behalf Of Lauren Peritz Sent: Tuesday, March 17, 2009 10:42 PM To: gov2001-l at lists.fas.harvard.edu Subject: Re: [gov2001-l] 1.2 in pset Hi All - Patrick suggested using 0's as starting values for the simulation in part 1.2. I did so but seem to be getting an error when I run optim: "...initial value in 'vmmin' is not finite" Has anyone else encountered this problem? My code is below - perhaps I am missing something entirely obvious? Thanks in advance, Lauren --------------- # "Scobit" log-likelihood scobit.ll <- function(par, X, y){ #par=parameters #X=matrix of covariates with intercept #y=dependent variable beta <- par[1:ncol(X)] mu <- X %*% beta a <- 0.01 outA <- sum(-y*log((1+exp(-mu))^a)) outB <- sum((1-y)*log(1-(1+exp(-mu))^a)) out<-outA+outB return(out) } #### Problem 1.2 use "turnout" data and run model using ll.scobit library(Zelig) data(turnout) # select indept variables from turnout data for X X <- matrix(NA,2000,4) X[,1] <- 1 X[,2] <- as.numeric(turnout[,"income"]) X[,3] <- as.numeric(turnout[,"educate"]) X[,4] <- as.numeric(turnout[,"age"]) colnames(X) <- c("intercept","income","educate","age") # select dept variable from turnout data y <- as.numeric(turnout[,"vote"]) # Maximize the log likelihood for "Scobit" optim.out <- optim(par = c(0,0,0), scobit.ll, y = y, X = X, method="BFGS",control=list(fnscale = -1), hessian = T) On Fri, Mar 13, 2009 at 2:13 PM, Patrick Lam <plam at fas.harvard.edu<mailto:plam at fas.harvard.edu>> wrote: Hi all, Just a slight tip. To avoid numerical issues, use 0s as your starting values for 1.2 in the problem set. -- Patrick Lam Department of Government and Institute for Quantitative Social Science, Harvard University http://www.people.fas.harvard.edu/~plam _______________________________________________ gov2001-l mailing list gov2001-l at lists.fas.harvard.edu<mailto:gov2001-l at lists.fas.harvard.edu> http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l