What background survival formulation?

[n8thangreen]

• Simple approach:
  Corresponding to the frequentist model, in the first instance we assume we know with essentially no uncertainty the general population background mortality, but using the points values from life tables.

  • Life tables are over large populations so could consider the point values to be accurate

• Better approach:
  (Bayesian) Model the survival by ensuring that the model parameters are such that you "anchor" the survival to known values (while allowing for uncertainty).

  • Stratified by country and sex
  • How to include uncertainty:
    • Could fit a survival curve to the life table data (life table survival is log linear for adults) and then add uncertainty to the parameters
    • Or directly on each age. Separate/exchangeable uncertainty at each age or the same?