I am using PISA data and modelling 2-level (school+students) multinomial logistic regression using brms. PISA has complex survey design, and final sampling weights reflect different probabilities of school and student selection and non-response adjustment. Can I use sampling weights with the brms package? Can I normalize them?
Example of my code:
fit_brms_deu_imp1 ← brm(
formula = ladder30_5 | weights(w_fstuwt_norm) ~ s_pv1math + s_pv1read + escs + female + siblings + immig +
(1|cntschid),
data = com1,
family = categorical(link = “logit”),
prior = c(
set_prior(“normal(0, 2)”, class = “b”, dpar = “mu1”),
set_prior(“normal(0, 2)”, class = “b”, dpar = “mu2”),
set_prior(“normal(0, 2)”, class = “b”, dpar = “mu4”),
set_prior(“normal(0, 2)”, class = “b”, dpar = “mu5”)
),
iter = 4000,
warmup = 2000,
chains = 4,
thin = 10,
control = list(adapt_delta = 0.99)
)
I know it is a pseudo-Bayesian analysis.
I appreciate your help!