Generating dummy variables!

[Vijay Kumar]

Hi all! I am working on a student school admission data set which has admission status of the child, the order of preference of the schools (applied to), and a bunch of SES variables for each child. I want to create a dummy variable for each school. To be clear, this is how my data looks like-
A,B,C,D,E are school names;a,b,c,d,e are student id's; P1-P5 are the school preferences.

id	P1	P2	P3	P4	P5
a	A	B	C
b	C	B	A	D
c	A	B	C	D	E
d	D	E
e	C	A	E

I now want to create dummy variables for A, B,C,D,E and make my data set look like this

id	P1	P2	P3	P4	P5	A	B	C	D	E
a	A	B	C			1	1	1	0	0
b	C	B	A	D		1	1	1	1	0
c	A	B	C	D	E	1	1	1	1	1
d	D	E				0	0	0	1	1
e	C	A	E			1	0	1	0	1

Obviously tab P1, generate (s) doesn't word here.
Thank you!

[Nick Cox]

Using dataex (SSC) is preferred please (FAQ Advice #12).

* Example generated by -dataex-. To install: ssc install dataex
clear
input str1(id p1 p2 p3 p4 p5)
"a" "A" "B" "C" ""  "" 
"b" "C" "B" "A" "D" "" 
"c" "A" "B" "C" "D" "E"
"d" "D" "E" ""  ""  "" 
"e" "C" "A" "E" ""  "" 
end

foreach v in A B C D E {
     gen `v' = 0
     quietly forval j = 1/5 {
           replace `v' = 1 if p`j' == "`v'"
     }
}


 list 

     +-------------------------------------------------+
     | id   p1   p2   p3   p4   p5   A   B   C   D   E |
     |-------------------------------------------------|
  1. |  a    A    B    C             1   1   1   0   0 |
  2. |  b    C    B    A    D        1   1   1   1   0 |
  3. |  c    A    B    C    D    E   1   1   1   1   1 |
  4. |  d    D    E                  0   0   0   1   1 |
  5. |  e    C    A    E             1   0   1   0   1 |
     +-------------------------------------------------+

[Mathias Pedersen Heinze]

Dear Vijay,

This should do the trick:

foreach x in A B C D E {
    gen `x' = 0
    replace `x' = 1 if inlist("`x'", p1, p2, p3, p4, p5)
}

[Nick Cox]

Mathias' trick is better. See also (e.g.) http://www.stata-journal.com/sjpdf.h...iclenum=dm0058 for more discussion.

In fact, this will work too:

foreach x in A B C D E {
    gen `x' = inlist("`x'", p1, p2, p3, p4, p5)
}

[Mathias Pedersen Heinze]

See also (e.g.) http://www.stata-journal.com/sjpdf.h...iclenum=dm0058 for more discussion.

That is indeed a great column, Nick. I will make sure to distribute it next time I teach Stata.

[Vijay Kumar]

Dear Nick and Mathias.. Thank you so much for the code. I am however not able to use it as I have a large number of schools (503) and preferences from P1-P178. So, how do I get all my school id/names in the first line of the code?

foreach x in A B C D E.

I am hoping this will work for the second line
gen `x' = inlist("`x'", p1-p178) Thanks again!

[Vijay Kumar]

* Example generated by -dataex-. To install: ssc install dataex
clear
input double id str7(p1 p2 p3 p4 p5 p6 p7)
20160000015 "1412148" "1413329" "1413240" "1413245" "1413215" "1411253" "1411219"
20160000041 "1821202" "1821210" "1821232" "."       "."       "."       "."      
20160000053 "1617224" "1617166" "1617186" "1617204" "1617220" "1411253" "1617192"
20160000058 "1821202" "1821146" "1821210" "1618189" "."       "."       "."      
20160000088 "1618249" "1821143" "1618232" "1514087" "1720147" "1514073" "1617192"
20160000090 "1720158" "1720136" "1514085" "1514086" "1515105" "1516110" "1516113"
20160000097 "1003262" "1002318" "1002294" "1001197" "1002306" "1003217" "1003225"
20160000106 "1617204" "1617186" "1617192" "1617181" "1617229" "1618253" "1515114"
20160000107 "1617192" "1617204" "."       "."       "."       "."       "."      
20160000115 "1412257" "1412152" "1413330" "1413225" "1413227" "1413329" "1412148"
20160000122 "1617229" "1411253" "1617182" "1617186" "1413283" "1413183" "1413215"
20160000149 "1411219" "."       "."       "."       "."       "."       "."      
20160000150 "1923255" "1720142" "1720145" "."       "."       "."       "."      
20160000157 "1413329" "1411253" "1617229" "."       "."       "."       "."      
20160000228 "1411183" "1411214" "1411201" "1207188" "1411199" "1207186" "1207181"
20160000264 "1617192" "1617229" "1617186" "1617182" "1617181" "1411253" "1514086"
20160000270 "1516116" "."       "."       "."       "."       "."       "."      
20160000276 "1105208" "1106190" "1104301" "1105233" "1106218" "1104310" "1104275"
20160000283 "1411253" "1412135" "1412141" "1412151" "1412156" "1413205" "1413215"
20160000288 "1821141" "1821232" "1821168" "1821145" "1821151" "1821152" "1821143"
end

This is how my data looks like- though I have 178 preference variables from p1-p178. Thanks in advance for your help.

[Nick Cox]

If you change the question radically the answer may change! You want how many dummies? Is it one for every school mentioned?

[Vijay Kumar]

Hi Nick. I want to create a dummy for each of my 503 schools. The dummy should take a value 1 if the school is in the preference list of the student and 0 otherwise. Thank you.

[Nick Cox]

Are you sure you know how you want to use those variables? You can do it but it may not be a good idea. Do you have the school identifiers anywhere in a single variable or do they appear only in the preference variables?

[Vijay Kumar]

Hi Nick.. I want to run a linear probability/ logit model to estimate the predicted probabilities of admission of each of the 15761 children in my data with dummies for neighborhoods ( 1283 neighborhoods) and schools (503) as the independent variables. I do have 503 school identifiers in the variable schoolid which takes the value of the school identifier if the child got admitted and "." if not admitted. My data with 7 preferences, schoolid, neighborhood variables looks like this.

* Example generated by -dataex-. To install: ssc install dataex
clear
input double id str50 N1 str7(p1 p2 p3 p4 p5 p6 p7 schoolid) float treatment
20160000015 "Rohini Extension" "1412148" "1413329" "1413240" "1413245" "1413215" "1411253" "1411219" "1412148" 1
20160000041 "Kakraula Village" "1821202" "1821210" "1821232" "."       "."       "."       "."       "."       0
20160000053 "JJ Colony 1"      "1617224" "1617166" "1617186" "1617204" "1617220" "1411253" "1617192" "."       0
20160000058 "Sadh Nagar I"     "1821202" "1821146" "1821210" "1618189" "."       "."       "."       "."       0
20160000088 "Uttam Nagar East" "1618249" "1821143" "1618232" "1514087" "1720147" "1514073" "1617192" "."       0
20160000090 "Budh Nagar"       "1720158" "1720136" "1514085" "1514086" "1515105" "1516110" "1516113" "1720158" 1
20160000097 "BlockWb"          "1003262" "1002318" "1002294" "1001197" "1002306" "1003217" "1003225" "1105191" 1
20160000106 "Nihal Vihar"      "1617204" "1617186" "1617192" "1617181" "1617229" "1618253" "1515114" "1617204" 1
20160000107 "GH 6"             "1617192" "1617204" "."       "."       "."       "."       "."       "."       0
20160000115 "Karala"           "1412257" "1412152" "1413330" "1413225" "1413227" "1413329" "1412148" "1413330" 1
20160000122 "Block O"          "1617229" "1411253" "1617182" "1617186" "1413283" "1413183" "1413215" "."       0
20160000149 "Pitampura"        "1411219" "."       "."       "."       "."       "."       "."       "1411219" 1
20160000150 "Block A"          "1923255" "1720142" "1720145" "."       "."       "."       "."       "."       0
20160000157 "Sector 20"        "1413329" "1411253" "1617229" "."       "."       "."       "."       "."       0
20160000228 "Block A"          "1411183" "1411214" "1411201" "1207188" "1411199" "1207186" "1207181" "."       0
20160000264 "GH 6"             "1617192" "1617229" "1617186" "1617182" "1617181" "1411253" "1514086" "."       0
20160000270 "BlockB"           "1516116" "."       "."       "."       "."       "."       "."       "."       0
20160000276 "Taharpur Village" "1105208" "1106190" "1104301" "1105233" "1106218" "1104310" "1104275" "1104284" 1
20160000283 "Block F"          "1411253" "1412135" "1412141" "1412151" "1412156" "1413205" "1413215" "."       0
20160000288 "Palam Village"    "1821141" "1821232" "1821168" "1821145" "1821151" "1821152" "1821143" "1821173" 1
end

Thank you co much!

[Nick Cox]

I am not convinced in your own best interests that fitting a model with so many predictors is a good idea. Even with plain regression observations/parameters should be much more than about 10; with logit-type models you should want many, many more.

It's a personal rule not to give code willingly for what looks like a bad idea.

But

1. What's clear with so many schools is that the inlist() trick is ruled out.

2. The basic coding principle is just like the code in #2. You clearly don't want to type all the values which is trivial with A B C D E but you can try automating their collection with levelsof.

Nevertheless there are many people here who do this kind of analysis routinely and they should have better and more detailed advice, but I think you would need to start a new thread to get that.

[Vijay Kumar]

Thanks a lot Nick. Really appreciate your time and advice. I will rethink my strategy and start a new thread, if required.