Closed ld-archer closed 1 year ago
T-tests show we are over-predicting the inactive group compared to the input data, however I think this is probably because of other factors i.e. difference in mortality rate and resulting age distributions. The difference is very small by wave 9 however due to tiny sample sizes (0.08 simulated vs 0.04 in raw data), and a similar pattern is seen in the minimal models (0.07 simluated vs 0.04 raw data). I think this is evidence that another factor is at work.
This shows that prevalence of employment drops to very low over time (4% by end of sim). This is too low to be accounted for by people living longer and having more retired people, so I need to figure out why this is. Maybe this would be a good place to try out a 2 step model, where 50-65 year olds have one model which decides between employed and inactive, and another model for the 65+ which includes retired. Think I'm going to make this an issue in itself.
Currently workstat is modelled as employed - unemployed - retired/disabled. Retired and disabled are joined together because at the time I thought they represented a similar outcome (i.e. not working and will not return to the workforce). However, another paper using ELSA (Iparraguirre (2015)) separates them into 3 categories:
This might be a nice way of changing things around, so lets have a look at it.