<div dir="ltr"><div><div><div><div><div><div><div>Dear group,<br><br></div>To estimate the parameters of a von Bertalanffy model (length <- (Linf-(Linf-Lo)*exp(k*age))), i log-transform the data and the prediction of L (L=length). There is however an offset in L_inf estimates and this can be due to the log-transformation. I try to correct the estimates using a bias correction method where you multiply the estimates by 10^(Mean_Square_Error/2) but it's seems that it doesn't work. (<a href="http://www.vims.edu/people/newman_mc/pubs/Newman1993.pdf">http://www.vims.edu/people/newman_mc/pubs/Newman1993.pdf</a>)<br>
<br></div>Anyway, i have done a simulation framework to generate data of length-at-age with a small noise (5%) and I boostrap this and estimate the parameters using ADMB, nls and STAN for comparison purposes. I then estimates the estimation errors for each bootstrap without correction (without_corr.pdf). ADMB and STAN are very similar (although STAN is way longer) while nls is clearly not good. with_corr.pdf is the same with the correction factor...<br>
<br></div>Does anyone have an idea how to correct the bias of parameter estimation?<br><br></div>Thanks!<br><br>Sylvain<br><br></div>PS: simulation_VB.R is the code to simulate the data and fit the VB<br></div>vb.stan is the STAN code<br>
</div>vb_simu.tpl is the ADMB code<br></div>