Creating "previous history" variable based on a current year observation - multiple observations per person

[Andy Lalka]

Hello,
I have a dataset with multiple observations per person. There can be multiple incidences in a given year for up to 5 years. I want to create a binary variable that will update in a current year based on a previous year (different row) by unique identifier. The data is in long format. So far I've been able create dummy variables within each year for the injury of interest and a cumulative variable across all years but these don't address my issue.

Essentially I need a binary medical history variable.

* Example generated by -dataex-. For more info, type help dataex
clear
input float newid int school_year_key float(concussion17 concussion18 concussion19 concussion20 any_concussion)
 4 2018 0 0 0 0 0
 5 2018 0 0 0 0 0
 6 2019 0 0 0 0 0
 6 2020 0 0 0 0 0
 7 2019 0 0 0 0 0
 8 2018 0 1 0 0 2
 8 2017 0 0 0 0 2
 8 2019 0 0 1 0 2
 9 2017 0 0 0 0 1
 9 2020 0 0 0 0 1
 9 2019 0 0 0 0 1
 9 2017 0 0 0 0 1
 9 2018 0 0 0 0 1
 9 2019 0 0 0 0 1
 9 2017 1 0 0 0 1
 9 2020 0 0 0 0 1
10 2018 0 0 0 0 0
10 2017 0 0 0 0 0
10 2018 0 0 0 0 0
10 2019 0 0 0 0 0
end

[Andrew Musau]

Your question is unclear to me.

that will update in a current year based on a previous year (different row) by unique identifier.

Update in what way? You should manually create the variable for a few identifiers to illustrate what you want.

[Andy Lalka]

When I said current year I meant row/observation.

If we look at newid #8 for year 2019, we see that in year 2018 they had a concussion. I want to create a binary variable that indicates such for row/year 2019 prev_concussion18 = 1

I tried:
bysort newid: gen prev_concussion17 = any_concussion - (concussion18 + concussion19 + concussion20)

The above code fails and simply recreates any_concussion because it is not recognizing the differing rows despite sharing the newid variable.

[Jeph Herrin]

Try this:

bysort newid (school_year_key) : gen prev_concussion=any_concussion[_n-1]==1

hth,
Jeph

[Andrew Musau]

With the multiple observations per person per year, I would do it this way:

input float newid int school_year_key float(concussion17 concussion18 concussion19 concussion20 any_concussion)
 4 2018 0 0 0 0 0
 5 2018 0 0 0 0 0
 6 2019 0 0 0 0 0
 6 2020 0 0 0 0 0
 7 2019 0 0 0 0 0
 8 2018 0 1 0 0 2
 8 2017 0 0 0 0 2
 8 2019 0 0 1 0 2
 9 2017 0 0 0 0 1
 9 2020 0 0 0 0 1
 9 2019 0 0 0 0 1
 9 2017 0 0 0 0 1
 9 2018 0 0 0 0 1
 9 2019 0 0 0 0 1
 9 2017 1 0 0 0 1
 9 2020 0 0 0 0 1
10 2018 0 0 0 0 0
10 2017 0 0 0 0 0
10 2018 0 0 0 0 0
10 2019 0 0 0 0 0
end

egen concussion= rowtotal(concussion*)
frame put newid school_year_key concussion, into(concussions)
frame concussions{
    collapse (max) concussion, by(newid school_year_key)
    xtset newid school_year_key
    gen wanted=l.concussion
}
frlink m:1 newid school_year_key, frame(concussions)
frget wanted, from(concussions)
frame drop concussions

Res.:

. sort newid school_year_key

. l newid school_year_key concussion wanted, sepby(newid school_year)

     +--------------------------------------+
     | newid   school~y   concus~n   wanted |
     |--------------------------------------|
  1. |     4       2018          0        . |
     |--------------------------------------|
  2. |     5       2018          0        . |
     |--------------------------------------|
  3. |     6       2019          0        . |
     |--------------------------------------|
  4. |     6       2020          0        0 |
     |--------------------------------------|
  5. |     7       2019          0        . |
     |--------------------------------------|
  6. |     8       2017          0        . |
     |--------------------------------------|
  7. |     8       2018          1        0 |
     |--------------------------------------|
  8. |     8       2019          1        1 |
     |--------------------------------------|
  9. |     9       2017          1        . |
 10. |     9       2017          0        . |
 11. |     9       2017          0        . |
     |--------------------------------------|
 12. |     9       2018          0        1 |
     |--------------------------------------|
 13. |     9       2019          0        0 |
 14. |     9       2019          0        0 |
     |--------------------------------------|
 15. |     9       2020          0        0 |
 16. |     9       2020          0        0 |
     |--------------------------------------|
 17. |    10       2017          0        . |
     |--------------------------------------|
 18. |    10       2018          0        0 |
 19. |    10       2018          0        0 |
     |--------------------------------------|
 20. |    10       2019          0        0 |
     +--------------------------------------+

.