I´m tring to fit the Bivariate Poisson distribution to a set of sports results for my masters thesis.
But using the model given in Bayesian Modeling Using WinBUGS (Ch. 8, Section 3.4, pgs 292-294), I can't seem to get the model to work.
The model code is given below:
m1_string <- "model {
C <- 0
for(i in 1:n_games) {
minscore[i] <- min( HomeGoals[i],AwayGoals[i]) + 1
# Generate minscore latent variable
u[i] ~ dunif(0,minscore[i])
z3[i] <- trunc( u[i] )
HomeGoals[i] <- z1[i] + z3[i]
AwayGoals[i] <- z2[i] + z3[i]
Zeros[i] <- 0
Zeros[i] ~ dpois( zeros.mean[i] )
zeros.mean[i] <- -l[i] + C
l[i] <- -lambda_home[HomeTeam[i],AwayTeam[i]] + z1[i]*log( lambda_home[HomeTeam[i],AwayTeam[i]] ) - loggam( z1[i]+1 )
-lambda_away[HomeTeam[i],AwayTeam[i]] + z2[i]*log( lambda_away[HomeTeam[i],AwayTeam[i]] ) - loggam( z2[i]+1 )
-lambda_3[HomeTeam[i],AwayTeam[i]] + z3[i]*log( lambda_3[HomeTeam[i],AwayTeam[i]] ) - loggam( z3[i]+1 )
}
for(home_i in 1:n_teams) {
for(away_i in 1:n_teams) {
log( lambda_home[home_i, away_i] ) <- mu + skill[home_i] - skill[away_i]
log( lambda_away[home_i, away_i] ) <- mu + skill[away_i] - skill[home_i]
log( lambda[home_i, away_i] ) <- beta
}
}
skill[1] <- 0
for(j in 2:n_teams) {
skill[j] ~ dnorm(group_skill, group_tau)
}
group_skill ~ dnorm(0, 0.0625)
group_tau <- 1 / pow(group_sigma, 2)
group_sigma ~ dunif(0, 3)
mu ~ dnorm(0, 0.0625)
beta ~ dnorm(0, 0.0625)
}"
As data supplied to JAGS through the jags.model() function in R, I am providing the following list:
HomeTeam/AwayTeam -the home and away teams
HomeGoals/AwayGoals - the observed scores
n_games - the number of games in the dataset
n_teams - the number of teams in the dataset
When I try to compile:
# Compiling model 1
m1 <- jags.model(textConnection(m1_string), data = data_list, n.chains = 3,
n.adapt = 5000)
message ERROR: RUNTIME ERROR
Any help would be greatly appreciated.
