Creating a variable with three condtions

[Arbnor Gashi]

I am using a data set of 50 countries which are separated into four regions.
I want to create a variable of regional contagion, where recgontagion for country j takes one if and only if there is a crisis in at least one country other than country j and this country is in the same region as country j.

I have been the below command but it doesn't work:

replace regcontagion=1 if (evnt3w==1&country!=country &region_id==region_id)

[Maarten Buis]

country!=country is never true

[Arbnor Gashi]

Which command should I use then?

[Nick Cox]

rangestat from SSC should help. Syntax would be something like

rangestat (max) evnt3w, by(region) int(country . .) excludeself

noting that

1. You must install rangestat first

2. For this to work country must be numeric. You don’t give a data example sufficient for us to be clear (please do read and act on FAQ Advice #12) but at worst the variable is string and you need to do something like

encode country, gen(Country)

and use the new variable instead.

[Arbnor Gashi]

clear
input long country float id int year float(evnt3w region_id regcontagion)
3 3 1980 . 3 0
3 3 1981 . 3 0
3 3 1982 1 3 0
3 3 1983 . 3 0
3 3 1984 . 3 0
3 3 1985 . 3 0
3 3 1986 0 3 0
3 3 1987 0 3 0
3 3 1988 0 3 0
3 3 1989 0 3 0
3 3 1990 0 3 0
8 8 1980 . 3 0
8 8 1981 . 3 0
8 8 1982 1 3 0
8 8 1983 . 3 0
8 8 1984 . 3 0
8 8 1985 . 3 0
8 8 1986 . 3 0
8 8 1987 . 3 0
8 8 1988 0 3 0
8 8 1989 0 3 0
8 8 1990 0 3 0
11 11 1980 . 3 0
11 11 1981 1 3 0
11 11 1982 . 3 0
11 11 1983 . 3 0
11 11 1984 . 3 0
11 11 1985 0 3 0
11 11 1986 0 3 0
11 11 1987 0 3 0
11 11 1988 0 3 0
11 11 1989 0 3 0
11 11 1990 0 3 0
21 21 1980 0 4 0
21 21 1981 0 4 0
21 21 1982 0 4 0
21 21 1983 . 4 0
21 21 1984 . 4 0
21 21 1985 . 4 0
21 21 1986 1 4 0
21 21 1987 . 4 0
21 21 1988 . 4 0
21 21 1989 . 4 0
21 21 1990 0 4 0
33 33 1980 0 4 0
33 33 1981 0 4 0
33 33 1982 0 4 0
33 33 1983 . 4 0
33 33 1984 . 4 0
33 33 1985 . 4 0
33 33 1986 1 4 0
33 33 1987 . 4 0
33 33 1988 . 4 0
33 33 1989 . 4 0
33 33 1990 . 4 0
end
format %ty year
label values country country
label def country 3 "Bolivia", modify
label def country 8 "Chile", modify
label def country 11 "Costa Rica", modify
label def country 21 "Guinea", modify
label def country 33 "Nigeria", modify
[/CODE]

[Nick Cox]

So, thanks for the extra detail. Evidently, country is numeric and the first line of code is still what is recommended.

[Arbnor Gashi]

Sorry Nick, but I still don't get how this command works. I already have the variable

regcontagion which has just zeros. So, how should I replace the values with 1 if the conditions are fullfilled. I tried your command once but I got all zero.

[Nick Cox]

Not so. Your example data came in #5 after my code in #4, but you could have applied it.

I now do so and find that all the results are all 1. See below for evnt3w_max.

. rangestat (max) evnt3w, by(region) int(country . .) excludeself  

. 
. 
end of do-file

. su

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
     country |         55        15.2    10.76414          3         33
          id |         55        15.2    10.76414          3         33
        year |         55        1985    3.191424       1980       1990
      evnt3w |         26    .1923077    .4019185          0          1
   region_id |         55         3.4    .4944132          3          4
-------------+---------------------------------------------------------
regcontagion |         55           0           0          0          0
  evnt3w_max |         55           1           0          1          1

.

This fits your example data, and I can work with no other data. For example, there are two countries in region 4, 21 and 33: it's true for each of them that there is a 1 on record for the other country.

Incidentally, your specification says nothing whatsoever about year. If being in the same year is part of the recipe that needs to be explicit.

Also, my code does nothing to your existing regcontagion variable. I wasn't trying to overwrite it or re-create it.

[Arbnor Gashi]

Actually I forgot to mention the year. It should take one if and only if there is a crisis in the region in the same year and not for all years.

Definition: The contagion variable for country j takes the value 1 if and only if there is a crisis in at least
one country other than country j and this country is in the same geographical region as
country j.

This can mathematical be expressed as:

[Nick Cox]

That was my guess, but it wasn’t what you said in #1.

Now that is cleared up, my code needs to be by(region year).

Your claim in #7 remains unsubstantiated. My code will produce 1s when the data so imply.

[Arbnor Gashi]

rangestat (max) evnt3w, by(region_id year) int(id . .) excludeself

This works perfectly. Thanks Nick.

[Nick Cox]

Thanks for the closure.