The log-KDE method: Density estimates for positive data

A previous article discusses the problem of fitting a kernel density estimate (KDE) to data that are strictly positive. If you use a standard KDE, the resulting density curve might estimate non-zero probability for negative values. This is unsatisfactory because quantities like lengths and mass cannot be negative.