I need help to better understand an aspect of marginal structural models (introduced by Robins). I am not clear on why this method requires artificial censoring at the first missing value. For instance, if there were 5 time points in a data set and a particular person had data for time points 1, 2, 4, and 5, but was missing data for time point 3 - to use this method you would have to artificially censor this individual at time point 3. It seems a shame to ignore the data for time points 4 and 5 just because you are missing time point 3. I know this has something to do with the calculation of the weights for treatment and censoring, but the discussion on the theory behind this is above my level of theoretical knowledge. Thank you in advance for your help!