dnLogNormal help inaccurate

[ssolo]

mu <- 1.
sigma <- 1.
for (j in 1:10000) {x[j]~dnLnorm(mu,sigma)}
mean(x)
4.507144
var(x)
36.43327

The mean is exp(mu+sigma^2/2) and the variance is (exp(sigma^2)-1)exp(2mu+sigma^2).

The help on the other hand states the following:

Arguments

   mean : The mean in log-space (observed mean is exp(m)).
          Type:       Real, <any>, const reference
     sd : The standard deviation in log-space.
          Type:       RealPos, <any>, const reference

[ssolo]

Yes mean( ln(X) ) = mu, but the help states that "observed mean is exp(m)”, where as the observed mean is in fact exp(mu+sigma^2/2), this certainly seems inaccurate. I would also call “mean" “mu" and “sd" “sigma", as per the usual parametrisation. cf. Arguments mean : The mean in log-space (observed mean is exp(m)). Type: Real, <any>, const reference sd : The standard deviation in log-space. Type: RealPos, <any>, const reference

mu <- 1. sigma <- 1. for (j in 1:10000)

+ { + x[j]~dnLnorm(mu,sigma) + lnx[j] = ln(x[j]) + }

mean(x)

4.465145

mean(lnx)

0.9967692

mu <- 1. sigma <- 1. for (j in 1:10000)

+ { + x[j]~dnLnorm(mu,sigma) + lnx[j] = ln(x[j]) + }

exp(mu)

2.718282

exp(mu+sigma^2/2)

4.481689

mean(x)

4.400285

mean(lnx)

0.9854552

sqrt((exp(sigma^2)-1)*exp(2*mu+sigma^2))

5.874744

stdev(x)

6.030914

stdev(lnx)

0.9893643

…

On 12 Oct 2018, at 22:43, Fernando Roa ***@***.***> wrote: I think the help is accurate, try: for (j in 1:10000) {thelog[j]=ln(x[j])} mean(thelog) stdev(thelog) — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.