Open ld-archer opened 1 year ago
Frequency based consumption measure is now in the model, renamed alcfreq
. It has 8 levels:
Predictor variables for alcfreq currently include:
This will likely change and expand when I find some literature associating characteristics with consumption frequency in older adults
Note: When including the dummy vars in the prediction of other variables, I have chosen level 4 (Once or twice a week) as the reference variable, because this is the most common category in the input populations and represents (I think) a moderate consumer. Again this could change if I find good evidence for something else.
Leaving this issue open so I can collect some literature here and provide evidence for or change some decisions.
From T-tests, looks like our predictions on average are slightly high (as in predicting slightly higher frequency than we see in ELSA). Need to look into this.
Also I need to think about the way we want to predict consumption. I could continue with a 2 step prediction (i.e. predict drinks at all, then if yes predict frequency) or I could move onto a single prediction that handles it all. Can use zero inflated poisson models for this similar to how MINOS does this. Need to talk to Bryan about whether these are already possible within the FEM framework or whether I would need to write my own C++ object (hopefully not...).
Unit level consumption has been hard to measure and predict, so we will be better off for the social isolation work in changing to a frequency-based measure -
scako
. This question asked the frequency of consumption in the last 12 months, with 8 categories ranging from none at all to almost every day. Will massively simplify the work on alcohol consumption.To justify this in the paper, need to find literature that says that high frequency of alcohol consumption is the main problem for older people, as opposed to binge drinking in younger ages.
TODO: