Hello, I am getting two completely different result when using nls (where I get the correct answer) and brms (where I cannot). The data is this
> r3
aux pin1 aux1 pin
1 1.388655 6.591023 1.385943 6.503267
2 1.370443 6.503267 1.388655 6.469410
3 1.382679 6.469410 1.370443 6.405328
4 1.347952 6.405328 1.382679 6.420060
5 1.357052 6.420060 1.347952 6.422814
6 1.275295 6.422814 1.357052 6.383797
7 1.186480 6.383797 1.275295 6.375547
8 1.212150 6.375547 1.186480 6.442755
9 1.257206 6.442755 1.212150 6.459890
10 1.234539 6.459890 1.257206 6.362459
11 1.288101 6.362459 1.234539 6.407614
12 1.347652 6.407614 1.288101 6.438193
13 1.367419 6.438193 1.347652 6.462612
14 1.325210 6.462612 1.367419 6.315244
15 1.403423 6.315244 1.325210 6.291391
16 1.276356 6.291391 1.403423 6.398825
17 1.343016 6.398825 1.276356 6.290331
18 1.290664 6.290331 1.343016 6.257015
19 1.201854 6.257015 1.290664 6.305973
this is running nls:
> nls(aux ~ (k * pin1 * aux1) / (k*pin + 0.0045), data = r3,start = list(k=1))
Nonlinear regression model
model: aux ~ (k * pin1 * aux1)/(k * pin + 0.0045)
data: data.frame(r3)
k
0.06409
residual sum-of-squares: 0.06201
Number of iterations to convergence: 7
Achieved convergence tolerance: 7.261e-08
the value of k is the expected one.
While with brms:
prior1 <- prior(normal(0, 1), nlpar = "k", lb=0) fit1 <- brm(bf(aux ~ (k * pin1 * aux1) / (k*pin + 0.0045), k~ 1, nl = TRUE), data = r3, prior = prior1)
> fit1
Population-Level Effects:
Estimate Est.Error l-95% CI u-95% CI Eff.Sample Rhat
k_Intercept 0.75 0.59 0.05 2.20 2138 1.00
Family Specific Parameters:
Estimate Est.Error l-95% CI u-95% CI Eff.Sample Rhat
sigma 0.06 0.01 0.05 0.09 1655 1.00
any ideas?
thans