Problems calculating Hessian for confidence intervals

CoVaRR-NET / duotang

Scripts and data for the CoVaRR-Net Pillar 6 notebook

https://covarr-net.github.io/duotang/duotang.html

MIT License

1 stars 2 forks source link

Problems calculating Hessian for confidence intervals #275

Closed ArtPoon closed 10 months ago

ArtPoon commented 11 months ago

to calculate confidence intervals on fitting logistic growth models to variant frequencies over time, we are using the mle2 optimization method
mle2 calculates the Hessian matrix to estimate the confidence interval associated with the maximum likelihood parameter estimates
this fails in some cases, e.g., where the estimated frequencies are close to zero

ArtPoon commented 11 months ago

In meeting I suggested looking into a logit transform $\log(p/(1-p))$ from frequencies (0,1) to (-inf, inf), which could make the Hessian (second derivatives) easier to compute

bfjia commented 10 months ago

This issue had been resolved by using a sliding window of date ranges, in increments of 10, until we find a point where we re able to estimate confidence intervals.