Friday, 12 June 2015

Some concepts that merit attention following conversation with Pete.

In non-linear systems, parameters may not be independent. Therefore, great care has to be given to the concept of parameter confidence intervals. It may be better to consider a confidence interval.

These can be calculated using bootstrapping, perhaps.

The nls tools package may be useful to understand and appreciate this.


Robustness test...
   Run your data with 90% of the data, randomly selected.
   If your values change dram

Concept 3
Testing down.
Put all the variables in and take them out one at a time.
Inspired by David F Hendry.



No comments:

Post a Comment