Can someone please explain how the perterbations relate to identifying the 95% CIs for a parameter estimate? I’m not sure I understand how you can know how much to vary the estimate a priori to identify what parameter values result in a change in the chi-squared value of 3.84
Hi Elliot, The approach is empirical and iterative. You specify some percentages perturbations and after the run finishes you check how much increase you have and adjust accordingly. The goal is find what value cause the objective function value to increase by 3.84. Then you can interpolate between the values if you did not hit exactly 3.84. Samer
I’m assuming we want it to increase and decrease by 3.84 on either side and it is the estimates which correspond to +/- 3.84 on either side of the starting estimatae which then become your 95% CI? if so, makes sense. thank you for explaining that. Elliot
Yes that is correct We want to get to a 3.84 difference on both sides of the estimate This way we don’t assume a symmetric CI. If you already have an asymptotic SE you can try to explore in the region of the asymptotic CI . Samer