# Tweedie location scale family Tweedie family in which the mean, scale and power parameters can all depend on smooth linear predictors. Restricted to estimation via the extended Fellner Schall method of Wood and Fasiolo (2017). Only usable with `gam`. Tweedie distributions are exponential family with variance given by $s*m^p$ where $s$ is a scale parameter, $p$ a parameter (here between 1 and 2) and $m$ is the mean. ```r twlss(link=list("log","identity","identity"),a=1.01,b=1.99) ``` ## Arguments - `link`: The link function list: currently no choise. - `a`: lower limit on the power parameter relating variance to mean. - `b`: upper limit on power parameter. ## Returns An object inheriting from class `general.family`. ## Details The first linear predictor defines the mean parameter $m$ (default link `log`). The Tweedie variance is given by $sm^p$. The second linear predictor defines the log scale parameter $r=log(s)$ (no link choice). The third linear predictor, $t$, defines $p=(a+b exp(t))/(1+exp(t))$ (again with no choice of link). A Tweedie random variable with 1

twlss function