🕷️ Crawler Inspector

[Groups](https://groups.google.com/g/r-inla-discussion-group/c/my-groups) Groups Send feedback to Google Help Training [Sign in](https://accounts.google.com/ServiceLogin?hl=en&passive=true&continue=https://groups.google.com&ec=GAZA0AM) [Groups](https://groups.google.com/g/r-inla-discussion-group/c/my-groups) Groups ## R-inla discussion group [Conversations](https://groups.google.com/g/r-inla-discussion-group/c/g/r-inla-discussion-group) [About](https://groups.google.com/g/r-inla-discussion-group/c/g/r-inla-discussion-group/about) [Privacy](https://policies.google.com/privacy?hl=en_US) • [Terms](https://policies.google.com/terms?hl=en_US)     # Zero-inflated models 1,977 views Skip to first unread message   ### Laura Serra Saurina unread, Mar 21, 2012, 11:08:35 AM3/21/12    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to r-inla-disc...@googlegroups.com Hi, I would like to know the difference between the two types of zero-inflated models (type 0 and type 1). Looking at the formula I noticed that type 0 only includes positives values and type 1 includes all kind of values. Is there any other difference? Type 0: *Prob(y\|...) = p x 1\[y=0\] + (1 - p) x Poisson(y \| y \> 0)* Type 1: *Prob(y\|...) = p x 1\[y=0\] + (1 - p) x Poisson(y)* Moreover, could you give me the Type2's formula? I haven't found it. Thank you very much. Laura.  ### Finn Lindgren unread, Mar 21, 2012, 12:22:03 PM3/21/12    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to r-inla-disc...@googlegroups.com On Wednesday, 21 March 2012 11:08:35 UTC+1, Laura Serra Saurina wrote: > I would like to know the difference between the two types of zero-inflated models (type 0 and type 1). Looking at the formula I noticed that type 0 only includes positives values and type 1 includes all kind of values. Is there any other difference? > > Type 0: *Prob(y\|...) = p x 1\[y=0\] + (1 - p) x Poisson(y \| y \> 0)* > > Type 1: *Prob(y\|...) = p x 1\[y=0\] + (1 - p) x Poisson(y)* Type 1 is a mixture of a point mass at 0 and a regular Poisson distribution, whereas Type 0 is a mixture of a truncated Poisson (the y\>0 bit) and a point mass at 0, so that the probability at zero is governed directly by p. This means e.g. that Type 0 can have a \_lower\_ probability at 0 than a pure Poisson, relative to the probability at 1, whereas Type 1 can only increase the relative probability for 0. > Moreover, could you give me the Type2's formula? I haven't found it. The description of Type 2 is at the end of page 6 of <http://www.math.ntnu.no/~hrue/r-inla.org/doc/likelihood/zeroinflated.pdf> It's like Type 1 but with a different definition of p. /Finn  ### Victor unread, May 2, 2012, 5:30:19 PM5/2/12    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to r-inla-disc...@googlegroups.com  I am new to INLA and i was trying to fit a zero inflated poisson model (0) but i required the model to include a spatial component for example CAR model i seem to be fitting the model well without fail when i do not include the spatial component, but perhaps am missing some part of model specification or parameters my data distribution has a classical zero-inflated like histogram with very few large values my question is 1) is a specification of spatial component the right thing to do e.g using besag 2) how best do i initiate the parameters when i call INLA  ### Daniel Simpson unread, May 2, 2012, 5:31:34 PM5/2/12    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to Victor, r-inla-disc...@googlegroups.com Hi Victor, Could you send some example code that doesn't work? Thanks, Dan   ### Victor unread, May 2, 2012, 5:49:30 PM5/2/12    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to r-inla-disc...@googlegroups.com, Victor am still trying to figure out how best to include a sample data set, but basically (y) comprise of values with range from zero (0) (highest frequency) to 800 (just one value) hist(y, breaks = 400) f.mod1 \<- y ~ cov1 + f(ID, model="besag", graph.file = "dist.adj") \# where cov1 is a linear covariate mod1 \<- inla(f.mod1, family = "zeroinflatedpoisson0", data = mydata, E=pop, control.compute=list(dic=TRUE,cpo=TRUE)) basically the PIT values fail in above Thanks  ### Daniel Simpson unread, May 2, 2012, 8:14:57 PM5/2/12    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to Victor, r-inla-disc...@googlegroups.com Hi Victor, Without being able to try it on your dataset, it's hard to know exactly what happened. I ran the following code, where the graph and data were taken from the bym example. library(INLA) \#taken from BYM example. Read data and graph and organise dataframe data(Germany) g = system.file("demodata/germany.graph", package="INLA") source(system.file("demodata/Bym-map.R", package="INLA")) Germany = cbind(Germany,region.struct=Germany\$region) \#this is similar to your formula formula1 = Y ~ f(region.struct,model="besag",graph.file=g) +x \#this is similar to your inla call result = inla(formula1, family="zeroinflatedpoisson0",data=Germany, E=E, control.compute = list(dic=TRUE, cpo=TRUE)) summary(result) I get sensible results, but the following warning message is produced: "Warning message: In INLA::f(region.struct, model = "besag", graph.file = g) : Argument 'graph.file' in 'f()' is obsolete; please use the more general argument 'graph' instead." You can get rid of this by changing the formula to formula1 = Y ~ f(region.struct,model="besag",graph=g) +x So, can you please send me the exact error message that you got when you tried to run the problem. There are also some things that you can check: \-- Is INLA updated? Use inla.update(testing=TRUE) \-- Is your graph file valid? Use inla.debug.graph() \-- Is the data frame in order? Best wishes, Dan  Message has been deleted  ### Belay Birlie unread, Aug 12, 2016, 10:11:19 AM8/12/16    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to R-inla discussion group Dear All I am trying to fit a zero inflated model in INLA using *Prob(y\|...) = p x 1\[y=0\] + (1 - p) x Poisson(y \| y \> 0).* Is there a possibility to model p as a function of covariates. In JAGS I can specify this in this way. model { for( j in 1 : N ) { y\[j\] ~dpois(mu\[j\]) mu\[j\] \<- (1 - u\[j\])\*lambda\[j\] u\[j\]~dbern(p\[j\]) logit(p\[j\]) \<- a\[1\]+ a\[2\]\*time\[j\]+a\[3\]\*Trt\[j\]+a\[4\]\*Season\[j\]+b1\[id2\[j\]\] log(lambda\[j\]) \<-beta\[1\]+ beta\[2\]\*time\[j\]+beta\[3\]\*Trt\[j\]+ beta\[4\]\*Season\[j\]+beta\[5\]\*Trt\[j\]\*time\[j\]+b2\[id2\[j\]\] } Thus, I am looking for help to do the same thing in INLA Belay   ### Håvard Rue unread, Aug 12, 2016, 10:13:39 AM8/12/16    Reply to author Sign in to reply to author Forward Sign in to forward Delete You do not have permission to delete messages in this group Copy link Report message Show original message Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message to Belay Birlie, R-inla discussion group On Fri, 2016-08-12 at 01:11 -0700, Belay Birlie wrote: \> Dear All \> \> I am trying to fit a zero inflated model in INLA using Prob(y\|...) = \> p x 1\[y=0\] + (1 - p) x Poisson(y \| y \> 0). Is there a possibility to \> model p as a function of covariates. yes. since the poisson part is y\|y\>0, you can. see inla.doc("zeroinflated") and the spde-tutorial for a worked out example for zero-inflated rain data, which is the same as yours. you have to use two likelihoods, one for the 'p' and one for the 'y'. \-- Håvard Rue Department of Mathematical Sciences Norwegian University of Science and Technology N-7491 Trondheim, Norway Voice: [\+47-7359-3533](tel:+47%2073%2059%2035%2033) URL : <http://www.math.ntnu.no/~hrue> Mobile: [\+47-9260-0021](tel:+47%2092%2060%2000%2021) Email: [havar...@math.ntnu.no](https://groups.google.com/g/r-inla-discussion-group/c/bstKUeRrIBg) R-INLA: [www.r-inla.org](http://www.r-inla.org/) Reply all Reply to author Forward  0 new messages  Search Clear search Close search Google apps Main menu