IPTW for continuous treatments

[pzivich]

Might want to consider this. The difficulty will be that there is various approaches (not as simple as binary variable). This paper has potential options. Quite a few options, but quantile binning generally has the best performance (personal communications). Might preferentially add this alone (with option of adding other methods if requested)

Alternative is van der Laan's proposal for stochastic treatments (same for IPTW and TMLE). Likely will go this route, since it coordinates well with my plans for TMLE #52

[pzivich]

Update to this thread.

• I won't do binning of the continuous. The user will have to do that themselves because it gets cumbersome on my end

• Continuous is fairly easy to code on my end and the diagnostics for weights really narrow down to the weight distribution summary statistics

• This process is distinct from stochastic treatments. Stochastic treatments with continuous variables will have to be added to the StochasticIPTW class instead