Hi.
I am currently trying to make a health index. you see my data below, being panel data sort by the id and then I have one ordered variable taking values from 1-5 where 1 is "excellent health" and 5 is "poor" health.
the rest of the variables Is binary variables taking the value 1 if the respondent has the condition and taking the value 0 if the respondent don't have the condition. I am trying to do the following
1. Estimate ten probit models each one with a different health condition on the lefthand-side and the remaining nine health conditions on the righthand-side.
2. For each probit model and for each observation, generate a prediction of the dependent variable
3. for each probit model, normalize the prediction to lie between zero and one, where the outcomes closer to one indicate better health.
4. For each observation, average across all ten prediction to calculate the observations final health index.
what I do is I use
where my dependent variable is 1-5 indicator and
on the binary variables and then predict using the predict command afterwards to obtain the predicted probabilities.
I then normalize the predictions using
i do this for all the probit models. I then take the average for all 10 normalizations. The problem and question is firstly if the approach is correct.
Then my problem is that the normalization should be done in a way where values closer to one means better health. So right now one variable is 1-5 were 1 is excellent health and 5 is poor health and the binary variables is 0 if you don't have the bad health condition and 1 is if you have the bad conditions, so how do I make sure that the values closest to one is the healthiest respondents?
I hope you can help me 
Kind regards
Mads
I am currently trying to make a health index. you see my data below, being panel data sort by the id and then I have one ordered variable taking values from 1-5 where 1 is "excellent health" and 5 is "poor" health.
the rest of the variables Is binary variables taking the value 1 if the respondent has the condition and taking the value 0 if the respondent don't have the condition. I am trying to do the following
1. Estimate ten probit models each one with a different health condition on the lefthand-side and the remaining nine health conditions on the righthand-side.
2. For each probit model and for each observation, generate a prediction of the dependent variable
3. for each probit model, normalize the prediction to lie between zero and one, where the outcomes closer to one indicate better health.
4. For each observation, average across all ten prediction to calculate the observations final health index.
what I do is I use
Code:
oprobit
Code:
probit
I then normalize the predictions using
Code:
su p_Self_reported_health, meanonly gen normal_p_Self_reported_health = (p_Self_reported_health - r(min)) / (r(max) - r(min))
Then my problem is that the normalization should be done in a way where values closer to one means better health. So right now one variable is 1-5 were 1 is excellent health and 5 is poor health and the binary variables is 0 if you don't have the bad health condition and 1 is if you have the bad conditions, so how do I make sure that the values closest to one is the healthiest respondents?
Code:
* Example generated by -dataex-. To install: ssc install dataex
clear
input double(HHIDPN Self_reported_health High_blood_pressure Diabetes Cancer Lung_problems Heart_problems Stroke Psychological_problems Arthritis) float obese
3010 4 0 0 0 0 1 0 0 0 0
3010 4 0 0 0 0 0 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
3010 4 0 0 0 0 1 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
3010 3 0 0 0 0 1 0 0 0 0
10013010 1 0 0 0 0 0 0 0 0 0
10013010 3 1 0 0 0 1 0 0 0 0
10013010 3 1 0 0 0 1 0 1 0 0
10013010 5 1 0 0 0 1 0 1 0 1
10013010 4 1 0 0 0 1 0 1 0 1
10013010 4 1 1 0 0 1 0 1 0 0
10013010 3 1 1 0 0 1 0 1 0 1
10013010 4 1 1 0 0 1 0 1 0 1
10013010 4 1 1 0 0 1 0 1 0 1
10038010 1 1 0 0 0 0 0 0 0 0
10038010 2 1 0 0 0 0 0 0 0 0
10038010 2 1 0 0 0 0 0 0 1 0
10038010 2 1 0 0 0 0 0 0 1 0
10038010 2 1 0 0 0 0 0 0 1 0
10038010 2 1 0 0 0 0 0 0 1 0
10038010 2 1 0 0 0 1 0 0 1 0
10038010 2 1 0 0 0 1 0 0 1 0
10038010 2 1 0 0 0 1 0 0 1 0
10038010 2 1 0 0 0 1 0 0 1 0
10063010 4 1 1 0 0 0 0 1 1 0
10063010 5 1 1 0 0 0 0 1 1 0
10063010 4 1 1 0 0 0 0 0 0 0
10063010 4 1 1 0 0 0 0 0 0 0
10063010 4 1 1 0 0 0 0 0 0 0
10097010 3 1 0 0 0 0 0 0 0 0
10097010 3 1 0 0 0 0 0 0 0 0
10114010 4 1 0 0 0 0 0 0 1 0
10114010 2 1 0 0 0 0 0 0 1 0
10114010 3 1 0 0 0 0 0 0 0 0
10210010 5 1 1 0 0 0 0 0 0 0
10210010 5 1 1 0 0 0 1 0 0 0
10210010 5 1 1 0 0 0 1 0 1 0
10210010 5 1 1 0 0 0 1 1 1 0
10210010 5 1 1 0 0 0 1 1 1 0
10210010 5 1 1 0 0 1 1 1 1 0
10225010 2 0 0 0 0 0 0 0 0 1
10225010 5 1 0 0 0 1 0 0 0 1
10225010 4 1 0 0 0 1 0 0 0 1
10225010 3 1 0 0 0 1 0 1 0 1
10225010 3 1 0 0 0 1 0 0 1 1
end
Kind regards
Mads
Comment