kralljr/sheds-dist

Files for project looking at exposure distributions using SHEDS data

sheds-dist's Issues

Separate out mean effect

If have difficulty with flat beta1, try modeling mean effect separate
Be sure to check correlation between mean and quantiles.

Right-skewed sigma2

When relax prior for sigma2, get large values of sigma2 that impact estimation of beta

Cluster distributions

Idea: cluster distribution days to characterize the shape of the exposure distribution on most toxic days (holding pm2.5 constant)

1. Use kolmogorov-smirnov test to separate normal vs. non-normal. Include main effects + interaction with mean PM2.5 in health model
2. Other ways of clustering distribution days?

Use true X's

Instead of randomly distributed X's, use quantiles that are highly correlated

Penalize quantiles

May be enough to have prior around 0, but think about this