I noticed this paper by Anne Buckner & Dirk Froebrich on the scale heights of star clusters, in which they aim to estimate the parameters for a non-standard (i.e., not Normal, Student’s t, skew Normal, Gamma, etc.) probability model given univariate observed data assumed generated iid (from this model under some unknown parameterisation). Their solution is to attempt to set confidence intervals on the unknown parameters using the p-values of K-S tests between the observed data and the model’s cumulative density function. What puzzles me is why they didn’t just do a Bayesian or maximum likelihood analysis, since the likelihood is available from the product of the derivative of the CDF at each observed datapoint? Strange …

### Like this:

Like Loading...

*Related*

## One Response to

K-S test for parameter estimation … ?