What are some of the ways to forecast demographic census with some validation and calibration techniques?

Some of the concerns:

  • Census blocks vary in sizes as rural areas are a lot larger than condensed urban areas. Is there a need to account for the area size difference?
  • if let's say I have census data dating back to 4 - 5 census periods, how far can i forecast it into the future?
  • if some of the census zone change lightly in boundaries, how can i account for that change?
  • What are the methods to validate census forecasts? for example, if i have data for existing 5 census periods, should I model the first 3 and test it on the latter two? or is there another way?
  • what's the state of practice in forecasting census data, and what are some of the state of the art methods?
  • $\begingroup$ What kind of demographic data are you trying to forecast? Every data could (and should) be treated differently. $\endgroup$ – Wilhelm Jul 30 '10 at 13:43

I don't know about the first point. But for the second one, autoregressive (AR) functions could be simple. I would really chose a parametric method against a non-parametric one. The forecasting in AR is straight forward. And consensus data has lots of samples for each period so you can get robust parameter estimates at each time. And for the association x_n-1 to x_n function is simply a smoothed interpolation of any choice.

Changes in zones, well based on empirical data or prior belief?

State of the art in consenus? Those methods are arcane. They iterate over generations! Aeons. You could use Gaussian processes which would be quite advanced methodology for these problems. But most in the field stick to older methods given more 'tuning'. Best.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.