Skip to content

A mess with which I am comfortable

Having established that survey weighting is a mess, I should also acknowledge that, by this standard, regression modeling is also a mess, involving many arbitrary choices of variable selection, transformations and modeling of interaction. Nonetheless, regression modeling is a mess with which I am comfortable and, perhaps more relevant to the discussion, can be extended using multilevel models to get inference for small cross-classifications or small areas.

We’re working on it.


  1. LJ says:

    Looking at your recent postings regarding to statistical methods, it is not surprising that you have thought about this “mess”. My impression is that you are trying to find the methods to include higher order interactions in multilevel modeling. With my short experiences for this type of data, I am just wondering whether the higher order interactions are really key to entangle existing complexity. It will be very interesting to see why you decided to go for interactions, not other aspects of modeling through your work. However, though I have worked on this problem little, I would rather keep my hierarchical model simple by pushing all uncertainties in data into the random effects for space and time. Well, I cannot explain yet why or how I come up with this kind of thoughts, I believe keeping a model simple other than random effects and the hierarchical structures due to them would be the best way to benefit this method the most. Will see how your thoughts develop and how I can define my thoughts after seeing your thoughts.

  2. Shira says:

    Hi Andrew,

    Very interesting paper. I have a minor question about Table 3. You say that “ignoring the weighting or treating the weights as constant underestimates uncertainty, whereas uncertainty is overestimated by treating the weights as inverse probabilities”, and maybe I’m backwards here but the column for “conditioning on weights” looks like higher numbers than the column for “assuming inv-prob”. My apologies if I’m totally confused!