-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathClass13_script_2019_part2.R
123 lines (89 loc) · 9.55 KB
/
Class13_script_2019_part2.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
# Script Class 13 (second part)
# GLMMs for count data -- zero inflation
# PLS 298 -- F2019
# Andrew Latimer
library(lme4)
library(rjags)
library(pscl)
# If we think the processes generating some of the "excess zeros" in our data are also generating the counts, then we want to use a zero-inflated model. If we think the zeros process is independent of the count process, it makes sense use a hurdle model instead. In this exercise we will try both approaches.
#### Part 1 -- Zero-inflated Poisson model ####
# This is a model that makes sense when we believe the data are generated from a combination of two processes: one that's Poisson, and one that's a process that generates extra ("excess") zeros.
# Load example data -- from Ntzoufras (2009) "Bayesian Modeling Using WinBUGS: An Introduction"
# Chapter 8
# http://stat-athens.aueb.gr/~jbn/winbugs_book/
zipmodeldata = list(
n =550,
y = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 12, 12,
12, 12, 12, 12, 13, 13, 13, 15, 15, 15, 16, 16, 16, 20, 20, 20, 20, 20, 20, 20, 24, 25, 30, 30, 30, 50, 60, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 13, 13, 13, 14, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 16, 17, 18, 20, 20, 20, 20, 20, 20, 22, 23, 25, 25, 25, 27, 30),
gender = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
)
# "y" is a series of counts of events and "gender" is a dummy variable for male/female.
# How do these data (y) compare to a Poisson distribution? To a negative binomial distribution? Does it look like there are excess 0's compared to a Poisson distribution?
# Can any excess 0's or overdispersion be accounted for by differences between genders?
# First let's try fitting a zero-inflated Poisson in R
# Make a data frame
d <- data.frame(y=zipmodeldata$y, gender=zipmodeldata$gender)
# GLM fit via pscl library
zip.glm = zeroinfl(y~gender, data=d, dist="poisson", EM=TRUE)
summary(zip.glm)
# Q: What effect does "gender" have on the outcome for excess zeros (zero-inflation model)? On the counts of events (count model)?
# Now the JAGS version:
# ZIP_model_jags.txt
# For this one, we can supply initial values for the chains.
zip.inits <- list(list(beta=c(0,0), gamma=c(0,0)),
list(beta=c(3,1), gamma=c(3,-1)),
list(beta=c(1,3), gamma=c(-2, 2)))
zip.jags <- jags.model("ZIP_model_jags.txt", data=zipmodeldata, inits=zip.inits,
n.chains=3, n.adapt=3000)
zip.test <- coda.samples(zip.jags, c("beta", "gamma"), n.iter=500)
par(mar=rep(2, 4))
plot(zip.test)
gelman.diag(zip.test)
acfplot(zip.test)
zip.samp <- coda.samples(zip.jags, c("beta", "gamma"), n.iter=5000, thin=5)
gelman.diag(zip.samp)
summary(zip.samp)
# Q: how would you interpret the coefficients for "gender" in the "excess zeros" part of the model and the Poisson part of the model?
# A take-home problem would be to modify this ZIP model to
# 1) include random effects
# 2) and/or handle overdispersion via a Poisson-gamma likelihood (i.e. negative binomial), or a Poisson-lognormal likelihood.
# This then would allow you to handle all combinations of overdispersion and zero inflation (or even too few zeros), with whatever random effects structure you want. This includes and even goes beyond one of the applications Bolker et al. worked on at NCEAS in the working group described here:
# https://groups.nceas.ucsb.edu/non-linear-modeling/projects ("Owl nesting negotiation")
########################
# Part 2 -- Hurdle model
# Set up the data for the hurdle model.
# This is slightly more complicated than for the ZIP/mixture model above, because now we are splitting the model into 2 parts and only the positive values of y will be modeled via a truncated Poisson distribution. So we need an index that tells the model when the values of y are greater than 0.
positives <- which(zipmodeldata$y>0)
n.positive <- length(positives)
y.positive <- zipmodeldata$y[positives]
gender.positive <- zipmodeldata$gender[positives]
y.presabs <- as.integer(zipmodeldata$y>0)
# Assemble the data
hurdlemodeldata <- list(
n =550, n.positive = n.positive,
y.presabs = y.presabs,
y.positive = y.positive, gender.positive=gender.positive,
gender = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
)
# specify intitial values
hurdle.inits <- list(list(beta=c(0,0), g=c(0,0)),
list(beta=c(3,1), g=c(-1, 1)),
list(beta=c(1,3), g=c(2, -2)))
# compile and run
hurdle.jags <- jags.model("hurdle_model_jags.txt", data=hurdlemodeldata, inits=zip.inits,
n.chains=3, n.adapt=1000)
hurdle.test <- coda.samples(hurdle.jags, c("beta", "gamma"), n.iter=500)
plot(hurdle.test)
gelman.diag(hurdle.test)
acfplot(hurdle.test)
hurdle.samp <- coda.samples(hurdle.jags, c("beta", "gamma"), n.iter=5000, thin=5)
gelman.diag(hurdle.samp)
summary(hurdle.samp)
# Comparison to a hurdle glm in R (also in library pscl)
hurdle.glm <- hurdle(y~gender, data=d, dist="poisson", zero.dist="binomial")
summary(hurdle.glm)
# Because we used weak priors, the results should be very similar. The advantage of running the model in JAGS (or Stan) is that now we can add random effects, impute missing data, etc.