Importance weights

[ghost]

• It may be worth introducing some built-in checking of the importance weights in the controls.csv file. Having high importance of total households relative to all other controls is a theoretical requirement for our specific use case of population synthesis, but it is not a strict requirement for the code itself. There’s nothing stopping someone from using 1000 importance for all controls, but the output from this would not be feasible because you’re going to have non-zero differences between total households compared to controls.

• The current setting and importance guidelines allow the user to assign different importance weights for different categories of the same attribute type. For example, we can assign 5000 to HHSIZE4 and assign 100 to HHSIZE1. I believe such settings can be hard to defend from a theoretical standpoint. I would recommend adding input to group those categories based on the attribute type (for example, HHSIZE) and allowing the user to do some checking or weighting at the attribute type level.

• Users would benefit from more refined guidelines for setting importance and providing some idea on the impact changing importance from 1000 to 10000. Setting importance provides a significant amount of influence over the result to users, which isn’t inherently a bad thing, but also assumes significant understanding from users.