# Three-Level-Meta-Analyis in brms: "Error evaluating the log probability at the initial value."

Hi everybody,
I’m really new in R-/Stan-coding and want to conduct a Three-Level-Meta-Analysis in brms. But there is the following error in:

“Chain 1: Rejecting initial value:
Chain 1: Error evaluating the log probability at the initial value.
Chain 1: Exception: normal_lpdf: Scale parameter is 0, but must be > 0! (in ‘model2f243c61550a_bcf717c53c8e740f935e13413bc27e34’ at line 48)”

``````
```# set priors

priors <- c(prior(normal(0, .5), class = Intercept),
prior(cauchy(0, .5), class = sd))

# fit three-level-model

fit_sbi <- brm(r |se(SE) ~ 1 + (1|Study_ID/ES_ID),
data = dataset,
prior = priors,
iter = 4000,
warmup = 2000)

Data are structured as follows:

* Operating System: Windows 10, 64 bit
* brms Version: 2. 16.2

Thank you so much for your help!
Kind regards,
Hanna``````

I might help if you tell us a little more about your `r` data. How many cases do you have and what’s the mean and standard deviation of the `r` distribution?

Hi,
the data contain 40 correlation coefficients nested in 13 studies. Mean r = .35 (SD = .25). Some effect sizes are r = 0 and therefore SE = 0. I think this could be the problem. Therefore I chose sample sizes for weights, the new code looks like this:

priors ← c(prior(normal(0, .5), class = Intercept),
prior(cauchy(0, .5), class = sd))

fit_sbi ← brm(r |weights (N) ~ 1 + (1|Study_ID/ES_ID),
data = dataset,
prior = priors,
iter = 4000,
warmup = 2000,
control = list(max_treedepth = 12,
That info is helpful. It’s not clear, to me, why an r = 0 should also have an se = 0. I suspect that’s the origins of your problem. That aside, you might tighten up your second prior to `prior(normal(0, .5), class = sd)`.