Hi,
I would like to know if it is possible to specify a regression in brms that has a submodel for one (or more) of the predictors. In my example, I want to predict PM2.5 (air pollution), and for this example just with wind speed as a predictor variable. PM2.5 is a mean daily (24 hr) value. For wind speed, I have hourly values. So there are 24 hourly values associated with the mean daily PM2.5 value.
Instead of just summarizing the wind values, I would like to be able to have main level of the model that relates wind to PM2.5, and a sub-model in which a representative daily value for wind speed is drawn from the distribution of hourly values for a day.
I have done a similar things using JAGS and runjags in R before, but wondering if this is possible in brms and what the syntax would be. Below is an example of the the JAGS code that I would use for this. Thanks for any assistance with this.
Michael
JAGS.code ← "
model{
#Submodel for wind
for (i in 1:length(wind)) {
wind[i] ~ dnorm(mu.wind[dayid[i]], phi.wind[dayid[i]])
}
#wind priors for each day
for (i in 1:max(dayid)) {
mu.wind[i] ~ dexp(1/mu0)
phi.wind[i] ~ dexp(1/phi0)
}
#regression
for (i in 1:max(dayid)) {
pm[i]~dnorm(mu.pm[i], z.prec)
mu.pm[i]<-b0+b1*mu.wind[i]
}
#priors for regression parameters
b0~dnorm(0, 1)
b1~ dnorm(0, 1)
z.sd ~ dexp(1)
z.prec ← pow(z.sd, -2)
#Top level priors
phi0~dexp(1)
mu0~dexp(1)
}"