Prior selection Sample Clauses

Prior selection. As for priors, we used weakly informative priors in our models similar to the ones provided in Xxxx Wiki on Github (Xxxxxxxxxx et al., 2020). In their blogpost, Xxxxxxxxxx et al. (2020) explains five different levels of priors: (i) flat priors, (ii) super-vague priors, (iii) weakly informative priors, (iv) generic weakly informative priors, (v) specific informative priors. In this context, a prior is considered informative or weak depending on its effect on the likelihood. Suppose the likelihood dominates the results, and the effect of a prior is either zero or unnoticeable. In that case, the prior is not informative. In our case, we chose priors that diminish the probability space fairly. We used a Normal(0,1) prior for the intercept and a Normal(0,1) prior for most of the slopes except for ungrammaticality and the interaction between ungrammaticality and the plural attractor. We set a Normal(-4,1) prior for the ungrammaticality and a Normal(1,0.5) prior for the interaction between the ungrammaticality and the plural attractor. These priors were set following previous findings. Lastly, Cauchy+(0,1) prior that is truncated at 0 for the standard deviations of random effects, and a LKJ(2) prior for correlation matrix for the random effects are used.