Add UC direct effects on mental health

[andrewbaxter439]

Adding to each outcome directly affected:

• Estimates of UC*work hours effects on health from @dkopasker estimates

Adding to outcomes:

• MCS
• PCS
• GHQ
• Life Satisfaction

Adding to initial populations (possibly) at ParametersL104:

• dot
• econ_benefits_uc
• econ_benefit_nonuc

[andrewbaxter439]

Checks for next steps

New Person enums to add

• labour supply weekly, six categories
  • Lhw_ZERO Lhw_TEN Lhw_TWENTY Lhw_THIRTY Lhw_FORTY
• partner labour supply weekly, six categories
  • Lhwsp_ZERO Lhwsp_TEN Lhwsp_TWENTY Lhwsp_THIRTY Lhwsp_FORTY?
• UC interaction with labour supply weekly, six categories
  • Lhw_ZERO_Econ_benefits_UC Lhw_TEN_Econ_benefits_UC Lhw_TWENTY_Econ_benefits_UC Lhw_THIRTY_Econ_benefits_UC Lhw_FORTY_Econ_benefits_UC ?
  • Ethnicity - two categories. Enum name Deth or similar?

Getting UC receipt status

• Setting D_Econ_benefits_NonUC and D_Econ_benefits_UC
• Detecting bsauc_s > 0 if D_Econ_benefits == 1?
• How best to do this in current model?

[dkopasker]

I think you also need a "no partner" option in partner's hours of labour supplied. If I remember correctly, this is the base category in the regression.

Not sure the interaction needs to be added since it can be defined by other characteristics and the health effect already combines the two elements.

Ethnicity has been called "dot" in the initial population files and has five categories.

[andrewbaxter439]

I think you also need a "no partner" option in partner's hours of labour supplied. If I remember correctly, this is the base category in the regression.

Ah thanks for pointing that out, that would correspond to ib999.lhwsp_c6 in the Stata regressions I presume?

Not sure the interaction needs to be added since it can be defined by other characteristics and the health effect already combines the two elements.

Following #124 (comment) I was aiming to add the five variables for UC effect as interacting with work hours, estimated by i.econ_benefits_uc##i.lhw_c5. Am I following correctly what suits best for the mental health2/causal model?

Ethnicity has been called "dot" in the initial population files and has five categories.

Ah that's helpful, though seems to be missing in training populations so will have to re-compile that I think!

[dkopasker]

Yes, I used code 999 for "no partner".

In the causal element of the health module effects will be applied for those with the UC marker equal to one which vary by own hours of labour supplied. As those two variables will already be present, the effects can be assigned by existing variables. Defining a new variable that combines the two would also work, but seems somewhat unnecessary given that it will not be used elsewhere. I would tend to favour the most parsimonious approach, but either would provide the same health outcomes.

[andrewbaxter439]

That makes sense, they would then not need the variable created and stored in the dataset. My understanding is that to process the coefficient of i.econ_benefits_uc##i.lhw_c5 = "Yes#10" the Java side would need a named enum in the regression table to look-up (e.g. Lhw_TEN_Econ_benefits_UC) which would then fetch the 'interaction' within Java (testing whether receives UC plus works TEN hours).

Revisiting and thinking clearly, we don't need the labour supply lookup enums added if only needing accessed for the UC recipients

[andrewbaxter439]

Closed in #202