Work status

[ld-archer]

Think the best solution to all this would be to tie all the work and unemployment information together into a workstat variable. The labour force status variable r*lbrf_e seems a good place to start, here are the categories to that var:

Value----------------------|
.d:DK
.o:Other
.r:Refuse
1.employed
2.self-employed
3.unemployed
4.partly ret
5.retired
6.disabled
7.looking after home or fam

• Employed
  • 1 Employed
  • 2 Self-employed
  • 4 Partly retired
• Unemployed (but still in labour force)
  • 3 Unemployed
• Not working (retired, disabled, looking after home...) Will probably label this as retired.
  • 5 Retired
  • 6 Disabled
  • 7 Looking after home or family

Going to open this as another issue, then link both of these issues into a pull request for the work and retirement status variables.

_Originally posted by @ld-archer in https://github.com/ld-archer/E_FEM/issues/43#issuecomment-739250946_

Todo

• [x] Add labour force status variable into model
  • [x] reshape_long.do
  • [x] ELSA_covariate_definitionsELSA.do
    • and minimal
  • [x] sample_selections
  • [x] measures_subpop
  • [x] Vars.cpp
    • Current
    • Lag
    • Probability
    • Dummy vars
  • [x] Vars.h
  • [x] settings spreadsheet (.csv)
  • [x] settings.txt
  • [x] Cross Validation script
• [x] Test and Impute (if necessary)
• [x] Derive and include dummy vars for:
  • [x] working
  • [x] unemployed
  • [x] retired
  • [x] Make sure all these variables are linked to workstat in Vars.cpp
• [x] Do accounting in HealthModule for any changes between these states
• [x] Check against current work, unemp and retired variables
  • [x] Remove old vars
    • [x] work
    • [x] retemp
    • [x] retage
    • [x] unemp

DON'T TRANSITION THE WORKSTAT VAR. Instead estimate transitions for each of the states (careful with the sample selections) and account. The workstat var is only so these variables are linked and therefore cannot occur at the same time.

[ld-archer]

Update This is mostly complete, and just waiting on a reply from Bryan, after which I will close this issue. I asked Bryan whether it would be better to estimate a transition model for the ordered variable (workstat) and relate this back to the dummy variables, or to transition the dummys directly and relate back to the status var. The current implementation is the latter, with a model estimated for each of employed, unemployed, retired, and relayed back to the ordered variable.

[ld-archer]

Bryan suggested that I rearrange the transition estimation for this group of variables. Instead of transitioning the employed, unemployed, retired vars separately, I am now transitioning the workstat var and relating this back to the indicator variables in HealthModule. Testing now, will remove old variables and close this issue if tests look good.