Implement intialisation strategy for paid_employment

[pausz]

Current implementation for paid_employment:

• for women aged < 15: assign probability 0
• for women aged >=50: assign probability 0 which seems unrealistic.

[pausz]

Update: @mzimmermann-IDM will look into extrapolating values in the input datasets

[mzimmermann-IDM]

I put the inflection point for this one at age 25. It does look like there's more of a peak from age 35-45, so maybe this is too simplified. But we don't necessarily need all the detail.

Parameterization:
0.6198487 at age 25
slope <25, 6.216042e-02 (SE 2.062729e-03)
slope >25, 0.0008010242 (SE 0.0592966648)

[pausz]

@mzimmermann-IDM Once I implemented the piecewise linear parameterisation, I noticed that for ages < 15 we would get negative proportions/probabilities. To overcome this I've implemented an asymmetric logistic function.
Data, linear and logistic fits are shown below:

The fitting process is done right after loading empowerment data , so in principle we can change the function that we fit to data. We wouldn't have to change the parameterisation if the data source changes either (eg, we use a dataset from a different year).

Let me know if you think of any potential issues we may have with this approach. For instance, I noticed that the logistic fit in the figure doesn't reach to 0 for ages < 5 years old -- it's very small approx 0.04. I'll look into that next.

If the logistic fit is not the best approximation, we can use a multi-piecewise linear approximation:

• women aged < 15: assign probability 0
• women aged >=15 <60: piecewise linear approximation of DHS data with the parameterisation you suggested.
• women aged >= 60 (retirement age in Kenya) linear approximation with a negative slope

[pausz]

Paid employment with more segments to the piecewiselinear parameterisation:

Choice of additional inflection points and slopes are in kenya.py