How to set priors when computing ordinal model

[wenboyu0803]

I am currently running an ordinal regression model with brms, following the guideline of Paul-Buerkner and Matti Vuorre.

here is my code with default priors:

hd.model1 <- brm(Rating ~ 1 + GAP + (1 + GAP| sub) + (1 + GAP| item),
                 data = hd,
                 init = 0,
                 family = cumulative("probit"),
                 chains = 4, 
                 warmup = 1000,
                 iter = 5000,
                 thin = 1,
                 cores = **4)**

the results of MCMC is good, and the value of beta is consistent with my assumption:

Population-Level Effects: 
             Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept[1]    -2.33      0.19    -2.70    -1.98 1.00     2318     4621
Intercept[2]    -1.40      0.17    -1.74    -1.06 1.00     2099     4323
Intercept[3]    -0.64      0.17    -0.97    -0.31 1.00     2057     4224
Intercept[4]    -0.02      0.17    -0.35     0.31 1.00     2046     4130
Intercept[5]     0.85      0.17     0.52     1.18 1.00     2094     4309
Intercept[6]     1.86      0.18     1.51     2.21 1.00     2298     4752
GAPg350Mg200    -0.12      0.14    -0.39     0.15 1.00    12602    11173
GAPg500Mg350    -0.10      0.13    -0.35     0.14 1.00    12118    11375
GAPg650Mg500     0.05      0.12    -0.19     0.28 1.00    13454    11812
GAPg800Mg650    -0.34      0.12    -0.59    -0.09 1.00    13136    11935
GAPg950Mg800    -0.07      0.13    -0.32     0.19 1.00    13810    11633

Then i want to set some priors. Related research (based on the NHST) has found that all beta values are negative and significant between 650 and 800 level. So I want to set the prior to a normal distribution with an average of -0.2 and a standard deviation of 0.5.

my code is here:

prior.1 <- c(prior(normal(-0.2, 0.5), class = "b", coef = "GAPg350Mg200"),
             prior(normal(-0.2, 0.5), class = "b", coef = "GAPg500Mg350"),
             prior(normal(-0.2, 0.5), class = "b", coef = "GAPg650Mg500"),
             prior(normal(-0.2, 0.5), class = "b", coef = "GAPg800Mg650"),
             prior(normal(-0.2, 0.5), class = "b", coef = "GAPg950Mg800"))
hd.model1 <- brm(Rating ~ 1 + GAP + (1 + GAP| sub) + (1 + GAP| item),
                 data = hd,
                 init = 0,
                 family = cumulative("probit"),
                 chains = 4, 
                 warmup = 1000,
                 iter = 5000,
                 prior = prior.1,
                 sample_prior = "only",
                 thin = 1,
                 cores = 4)

but the result is interesting:

Population-Level Effects: 
             Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept[1]    -5.11      4.43   -20.21    -0.27 1.01      414      117
Intercept[2]    -2.07      1.97    -6.74     0.99 1.00     1236      899
Intercept[3]    -0.61      1.50    -3.84     2.16 1.00     5060     3504
Intercept[4]     0.58      1.49    -2.30     3.71 1.00     6105     3309
Intercept[5]     1.99      1.83    -1.02     6.27 1.00     7118     6627
Intercept[6]     4.79      4.76     0.15    15.19 1.00     4316     5657
GAPg350Mg200    -0.19      0.50    -1.19     0.80 1.00     9648     8571
GAPg500Mg350    -0.20      0.51    -1.17     0.81 1.00     6132     5808
GAPg650Mg500    -0.21      0.50    -1.18     0.79 1.00     5494     7874
GAPg800Mg650    -0.20      0.49    -1.15     0.76 1.00     9779    10689
GAPg950Mg800    -0.21      0.51    -1.21     0.77 1.00     5440     5966

So, My question is, why are all beta values basically the same after I set the prior, and why is there such a big difference from the default prior; Or is my prior setting incorrect? I hope to set a prior, after all, previous studies have reached such conclusions, and I also need to calculate the Bayesian factor.

Thanks