# Changes related to the `mh_jscale`-option of `estimation`
The default value of `mh_jscale` is now `2.38^2/npar`, where `npar` denotes the number of estimated parameters, instead of a fixed `0.2`. The new value corresponds to the optimal value for a Gaussian symmetric random walk Metropolis-Hastings (Gelman et al. 1995). Particularly for big models, the new default value should perform better.
At the same time, the `mh_init_scale`-option has become deprecated. Use the new `mh_init_scale_factor`-option to govern the overdispersion of the starting values for multiple MCMC chains. This option specifies the multiple of `mh_jscale` to be used.
