Add financial distress

[igelstorm]

What

• Add financial distress variable to the initial population data (dichotomous variable defined as 4 or 5 for UKHLS variable finnow)
• Add process to update financial distress (see below for model specification)
• Add code to include ethnicity in simulation
  • This is needed as a predictor for financial distress, and was already in initial population data but not yet used

Why

Resolves #181.

To do before merging

• Update training data to include financial distress
• Update expected output for integration test

Model specification for financial distress process

Logit model estimated in UKHLS. Predictors:

• Economic variables:
  • Employment transition
  • Working hours
  • Change in log income
  • Income decrease (binary)
  • Poverty transition
  • Gross personal non-employment capital income (L1)
  • Gross personal private pension income (L1)
• Lagged/fixed confounders
  • Receives benefits (L1)
  • Owns home (L1)
  • Partnership status (L1)
  • Number of children (L1)
  • SF-12 PCS (L1)
  • SF-12 MCS (L1)
  • Region (L1)
  • HH income quintile (L1)
  • Long-term sick/disabled (L1)
  • Financial distress (L1)
  • Gender
  • Age (L1)
  • Age^2 (L1)
  • Education
  • Ethnicity
  • Year

[andrewbaxter439]

This code seems good! Failing validation tests - is that likely caused by different initial population?

[igelstorm]

This code seems good! Failing validation tests - is that likely caused by different initial population?

I thought so at first, but I tried to keep the initial population unchanged except for the financial distress variable, and there are still small differences. (In theory, there shouldn't be, because the new process doesn't interact with any existing processes)

I think because I've added a process that involves additional random draws, all the subsequent random draws will be different, leading to slightly different random noise in the output. I have to say that diminishes the value of that test a bit. But maybe we can be content with verifying that the output statistics remain approximately the same?

[andrewbaxter439]

I had wondered if the additional draws were partly to blame, though these don't seem to affect the previous draws in the same Innovations object:

class InnovationsTest {

    @Test
    public void checkSame() {

        Innovations innovation1 = new Innovations(2, 100);
        Innovations innovation2 = new Innovations(3, 100);


        assertTrue(innovation1.getDoubleDraw(0) == innovation2.getDoubleDraw(0) );
        assertFalse(innovation1.getDoubleDraw(1) == innovation2.getDoubleDraw(2) );

    }

}

... passes fine?

[matteorichiardi]

...something to discuss with @van De Ven, Justin W ***@***.***> perhaps?

…

On Fri, 13 Jun 2025 at 09:47, Andrew Baxter ***@***.***> wrote: *andrewbaxter439* left a comment (simpaths/SimPaths#198) <#198 (comment)> I had wondered if the additional draws were partly to blame, though these don't seem to affect the previous draws in the same Innovations object: class InnovationsTest { @test public void checkSame() { Innovations innovation1 = new Innovations(2, 100); Innovations innovation2 = new Innovations(3, 100); assertTrue(innovation1.getDoubleDraw(0) == innovation2.getDoubleDraw(0) ); assertFalse(innovation1.getDoubleDraw(1) == innovation2.getDoubleDraw(2) ); } } ... passes fine? — Reply to this email directly, view it on GitHub <#198 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACOK4OAFPD7UVKW6JSKQDHL3DKFZ3AVCNFSM6AAAAAB65GG3VWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSNRZGU3TKMRVGI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[andrewbaxter439]

Awesome that everything's working now! Shouldn't add any breaking changes so if this were able to be merged (@pbronka /@justin-ven) I'd be keen to bring this new parameter into re-estimating mental health effects?

[igelstorm]

@andrewbaxter439 I've updated this based on discussions:

• Still using 5 ethnicity categories
• Including non-employment capital income and private pension income as covariates
• Using cross-sectional weights (dimxwt)
• Updated description of covariates to reflect lags correctly

[andrewbaxter439]

Awesome that this is working - think ready to merge I'd say!