ld-archer
commented
1 year ago NO NEED FOR CUMULATIVE PREDICTORS!!!!!!! IT WORKS!
Made lots of little changes to some of the transition models in the last week and it seems to be paying off.
- Changed the reference factor for loneliness and social isolation, specifically when predicting mortality.
- Now lnly3 and sociso6 are the reference factors, which are the highest levels of loneliness and social isolation. See below for the mortality model and a quick explanation of the coefficients.
- Massively improved the loneliness transition model
- Added a lot of new predictors, all based on literature and explained in detail in #116
- NOTE: Some of these new predictors are just being held static for now, some make sense to do this (childless) and some don't (hhres, gcareinhh1w, socyr). However I will not look a gift horse in the mouth, these can be fixed later.
- Changed atotb and itot to logatotb and logitot. Better to use the transformed values as predictors, and atotb is included in the prediction of loneliness
Going to close this issue as no longer necessary and open new ones to make the variables dynamic, as well as improve the social isolation model. I also need to add the chronic diseases back to the mortality model and add loneliness and social isolation as predictors of chronic disease.
Reference information:
time_scaled_probit
died
2
male .2518757
white .218156
hsless .0885608
college -.0248863
l2age65l .0382895
l2age6574 .0430296
l2age75p .054453
l2lnly1 -.7919944
l2lnly2 -.6713157
l2sociso1 -.2675508
l2sociso2 -.2196578
l2sociso3 -.1661182
l2sociso4 -.1401404
l2sociso5 -.1066686
_cons -4.23072Outcomes:
#### Life Years
[1] "No Loneliness"
[1] "Cohort Average Lifeyears: 41.59"
[1] "Intervention Average Lifeyears: 43.68"
[1] "Increase: 2.085"
[1] "No Social Isolation"
[1] "Cohort Average Lifeyears: 41.59"
[1] "Intervention Average Lifeyears: 42.87"
[1] "Increase: 1.279"
[1] "No Loneliness OR Social Isolation"
[1] "Cohort Average Lifeyears: 41.59"
[1] "Intervention Average Lifeyears: 44.83"
[1] "Increase: 3.238"
#### Disability Free Life Years
[1] "No Loneliness"
[1] "Cohort Average Disability Free Lifeyears: 14.79"
[1] "Intervention Average Disability Free Lifeyears: 15.99"
[1] "Increase: 1.206"
[1] "No Social Isolation"
[1] "Cohort Average Disability Free Lifeyears: 14.79"
[1] "Intervention Average Disability Free Lifeyears: 15.86"
[1] "Increase: 1.071"
[1] "No Loneliness OR Social Isolation"
[1] "Cohort Average Disability Free Lifeyears: 14.79"
[1] "Intervention Average Disability Free Lifeyears: 17.06"
[1] "Increase: 2.275"
#### Disease Free Life Years
[1] "No Loneliness"
[1] "Cohort Average Disease Free Lifeyears: 5.151"
[1] "Intervention Average Disease Free Lifeyears: 5.392"
[1] "Increase: 0.2411"
[1] "No Social Isolation"
[1] "Cohort Average Disease Free Lifeyears: 5.151"
[1] "Intervention Average Disease Free Lifeyears: 5.312"
[1] "Increase: 0.1613"
[1] "No Loneliness OR Social Isolation"
[1] "Cohort Average Disease Free Lifeyears: 5.151"
[1] "Intervention Average Disease Free Lifeyears: 5.552"
[1] "Increase: 0.4019"IMPORTANT NOTE: Loneliness and Social isolation are not yet included in the chronic disease or functional limitation models. I think the effect we are seeing here is purely the reduction in the probability of mortality directly from loneliness and social isolation, which is brilliant! Adding them to lots of other models could improve or worsen these results, so need to come back here and compare when I start updating things further.
Maybe it makes more sense to think about the length of time a simulant spends in each category as predictors here instead of just the 1 year lagged models. If someone is chronically lonely or social isolated they must have a different probability of certain outcomes compared with someone who has a particularly lonely year.
Reason for this train of thought is the life year and DFLY outcomes show an reduction in lifespan and healthy lifespan which is not what we hypothesised. This seems to go against the expectation when looking at the transition models and the cross tabulations, so maybe this is the other explanation.
Looking at the transition model, loneliness is a bit odd. It suggests that probability of mortality is highest in the low loneliness group, and smallest in the medium group. Social isolation is also very odd in this model, as compared with the reference factor of 1 (no isolation), all groups apart from 6 (highest) show a reduced probability of mortality. Then the highest isolation group have a higher probability of mortality. This is in contrast to probit models I have fit in R:
The relationships here are what we would expect from looking at the raw data: Highest loneliness group has the highest probability of mortality, as does the highest social isolation. We can also see that the magnitude of each level is roughly where we would expect, where the probability of mortality increases in order as loneliness and social isolation increase.
Perhaps this shows that the problem in the life years and dflys is not due to the outcomes chosen, but some quirk of the transition model estimation? I need to make sure here that the sample we are estimating the stata probits (top) on is correct, and try a version in R with this exact sample (R probits are fitted to entire transition sample NOT using the survey weights).