Keep running a replace command until (0 real changes made)

[Helder Costa]

Hello!
I have a paneldata with the ages of individuals with around 50 million observations. The age of each individual should increase at the rate of 1 year per year, but it doesn't. Thus, I'm correcting this issue by checking if the increase (variable "delta") is different than 1. I'm also using a mode (egen tmp = mode(delta), by(ntrab)) of the delta to check if for at least some individuals, I have multiple years where the age increases at the correct rate. I then generate a new variable tmp2, which is simply to check if for two consecutive years the delta is equal to 1 (gen tmp2 = (delta == 1 & l.delta == 1)) and I replace the age variable (idade) to missing if tmp2 == 0. I then start correcting idade with the following code:

replace idade = f1.idade-1 if f1.idade ~= . & f1.tmp2 == 1

which basically goes to the first correct value and uses it subtracting one. This is an example of the data after running the code once. Stata has corrected the age for the year (variable ano) 2011, but I would need to run it again 3 more times until it corrects all years. Do note that this is an example and for some individuals I may have around 30 years, so I would actually like Stata to keep running the code until I get (0 real changes made). Is there anyway to do this?

Thanks for your time!
Hélder

ano | workerid | idade | delta | tmp | tmp2
2008 | 1 | . | -55 | 1 | 0
2009 | 1 | . | 57 | 1 | 0
2010 | 1 | . | -55 | 1 | 0
2011 | 1 | 21 | 1 | 1 | 0
2012 | 1 | 22 | 1 | 1 | 1
2013 | 1 | 23 | 1 | 1 | 1
2014 | 1 | 24 | 1 | 1 | 1

[Fei Wang]

Hélder, if I understand correctly, you'd like to correct ages for all individuals in all years. I wonder if you have a variable indicating the birth year of each individual. If yes, then "replace age = year - birthyear" will do. If there is no birth year, and the age for an individual is inconsistent over time, then theoretically we don't know which value is right and which is wrong if no other supporting information is available.

If we assume your mode hypothesis holds, then you may generate birth year and pick the mode of the birth year as the correct value. Codes are:

gen birthyear = year - age
bys personid: egen bthyr_mode = mode(birthyear)
replace age = year - bthyr_mode
drop birthyear bthyr_mode

[Helder Costa]

Thanks Fei Wang
I don't have the birth year, I only have the reported age in each year. Generating a birthyear variable for the "correct" age from the mode is a good idea!

[daniel klein]

The age of each individual should increase at the rate of 1 year per year

If a person is born in July, and interviewed first in August and again in June, then that person's age in years should not have increased by 1 year. In my experience, people tend to get their age right when asked; so if this is a frequent issue, I would think that it is due to varying interview dates rather than incorrectly reported ages.

[Helder Costa]

Thanks daniel klein, but that isn't an issue with the data. The collection date of the data is always the same.

[daniel klein]

I am not aware of a single survey study that has been able to conduct interviews of all respondents on the same day (week, or often even month). If those ages are not off by more than one year, I would not be suspicious.