How to Create an Alternative Health Index?

[Jose Williams]

Hi,

I have 14 dummy health indicators. Each indicator has a value of 0 if the person is free of that disease and 1 if the person has it. I want to construct a simple index using the fraction of the conditions that the individual does that have so that a higher index value will indicate better health. I want to create it for each observation in the dataset. IDHH is the id variable and SURVEY is the time dummy.

* Example generated by -dataex-. To install: ssc install dataex
clear
input int SURVEY double IDHH int(MB3 MB4 MB5 MB6 MB7 MB8 MB9 MB10 MB11 MB12 MB13 MB14 MB15 MB17)
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 1 1 0 0 0 0 0 0 0 0 0 1
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 1 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201020 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010201 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 1 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 1 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201030 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010301 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201040 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010401 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201040 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010401 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201040 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010401 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201040 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010401 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201040 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010401 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201050 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010501 0 0 0 1 0 0 0 0 0 0 0 0 0 0
1  10201050 0 0 0 1 0 0 0 0 0 0 0 0 0 0
2 102010501 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201050 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010501 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201050 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010501 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201050 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010501 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201050 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010501 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201060 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010601 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010701 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 1
2 102010701 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010701 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010701 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010701 0 0 0 1 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010701 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201070 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010701 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201080 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010801 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201080 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010801 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201080 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010801 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201080 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010801 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201080 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010801 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201080 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010801 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201090 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010901 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201090 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010901 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201090 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010901 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201090 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010901 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201090 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 102010901 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1  10201090 0 0 0 0 0 0 0 0 0 0 0 0 0 1
2 102010901 0 0 0 0 0 0 0 0 0 0 0 0 0 0
end
label values SURVEY SURVEY
label def SURVEY 1 "IHDS1 1", modify
label def SURVEY 2 "IHDS2 2", modify
label values MB3 health
label values MB4 health
label values MB5 health
label values MB6 health
label values MB7 health
label values MB8 health
label values MB9 health
label values MB10 health
label values MB11 health
label values MB12 health
label values MB13 health
label values MB14 health
label values MB15 health
label values MB17 health
label def health 0 "No", modify
label def health 1 "Yes", modify

[Maarten Buis]

// first reverse the coding of the health variables: 1 no disease, 0 has disease
foreach var of varlist MB? MB?? {
    replace `var' = 1 - `var'
}
label def health 0 "yes" 1 "no", modify

// make index
egen health = rowmean(MB? MB??)

[Jose Williams]

Maarten Buis Thank you for the assistance. I had a quick follow-up. Can we adjust this measure based on age? I am trying to use the log of the health index. The majority of the observations have health scores equal to 1. Suppose the gap between the survey dummy is 7 years.

Will it be better to just divide the health variable by 7 or should I do inverse hyperbolic transformation?

[Romalpa Akzo]

I guess it is preferable to keep the original coding of health variables. Even if such reserving is needed, generating new (temporary) ones is still advisable.

One more thing: although you do mention a "fraction", you may consider the direct index of counting, which serves well for your desire.

egen wanted= anycount(MB*), v(0)

[Maarten Buis]

I guess it is preferable to keep the original coding of health variables.

In the original question Jose stated that he wanted an index "so that a higher index value will indicate better health", so that is why you would need to reverse the coding.

Even if such reserving is needed, generating new (temporary) ones is still advisable.

I used to think that to. However, one should always keep the original data unchanged, and only create analysis datasets using .do files. If you follow those rules, than there is not much value in creating new (temporary) variables.

One more thing: although you do mention a "fraction", you may consider the direct index of counting, which serves well for your desire.

If there are no missing values, then one is just a linear transformation of the other.

[Maarten Buis]

Can we adjust this measure based on age?

In what way?

[Jose Williams]

Maarten Buis Since the time gap between the two data is 7 years. I was thinking whether it would be appropriate to adjust for age to account for the time gap. However, I think so the problem will get solved by including age and age squared as regressors.