Generating a Unique within group Identifier

[George Kariuki]

I have data on clusters. Within each cluster (read "administrative rural village"), I have a number of observations. I now want to generate unique Identifier for every observation assigning a

==1

to the highest value for each cluster,

==2

for the second highest value and so on for all clusters. Thus I will be able to aptly use a qualification

..,if id==1

when I want to analyse just the highest scores for all clusters in the dataset. I am having problems figuring out how to go about it.

Thanks

[Nick Cox]

George: You should know Statalist well enough to know that we ask for data examples. I could guess at your data structure, variable names, and so forth, but it's your job to show us more (please). Highest value of what? What do you want to happen if there are ties?

Alternatively, isn't this just egen's rank() function?

[George Kariuki]

Now, here is what i mean and for a sample example of my dataset, I have a

district_id

variable that uniquely identify's my cluster. As in the example

239

refers to District ID while the variable turnout_total refers to the number in each village in the district that turned up for a relief ration. I have many districts with a similar way of presentation but different

district_id

. I have sorted the data such that for each district, I have the highest village turnout to the least. I now want to have a new unique ID with ==1 for the highest turnout in each district followed by ==2 for the second highest in that sequence. An example dataset for one district is as this;

* Example generated by -dataex-. To install: ssc install dataex
clear
input int(district_id turnout_total)
239 959
239 902
239 892
239 841
239 806
239 769
239 764
239 746
239 736
239 719
239 705
239 700
239 690
239 687
239 683
239 678
239 677
239 676
239 676
239 675
end

[Nick Cox]

Thanks. So you can use egen here.

egen rank = rank(-turnout), by(district_id)

Don't forget the minus sign.

(You didn't answer my question about ties.)

[George Kariuki]

Admittedly, I had not thought about such a situation with tallying turnouts but for the purposes of my analysis, I would wish to have the tallies assigned the same value. That is to mean, if there are two or more villages with equal highest turnout in a given district, assign all ==1. Running the egen command above works but seems to average the observations that have tied.

[George Kariuki]

The unique option seems to arbitrarily assign a successive rank for the ties or I did not grasp properly how to instrument the option.

[Nick Cox]

Correct about the unique option -- which is really intended for graphical purposes -- but is the field option then what you need?

(Note: when Rich Goldstein and I wrote the original code behind these options about 1999 I searched high and low for standard names: in the end I gave up and invented "field" and "track". Even as someone with minimal interest in sports I was aware that big numbers win in field events and small numbers win in track events. Rich may have alternative facts.)

Since then I discovered "schoolmaster's rank" as an alternative in print to "field", which is probably too obscure (hinging on certain British practices probably now obsolete) and too sexist to pass muster, and in any case what about track?)

. sysuse auto, clear
(1978 Automobile Data)

. egen rank = rank(mpg), field

. sort mpg rank

. sort rank

. list rank  mpg, sepby(rank)

     +------------+
     | rank   mpg |
     |------------|
  1. |    1    41 |
     |------------|
  2. |    2    35 |
  3. |    2    35 |
     |------------|
  4. |    4    34 |
     |------------|
  5. |    5    31 |
     |------------|
  6. |    6    30 |
  7. |    6    30 |
     |------------|
  8. |    8    29 |
     |------------|
  9. |    9    28 |
 10. |    9    28 |
 11. |    9    28 |
     |------------|
 12. |   12    26 |
 13. |   12    26 |
 14. |   12    26 |
     |------------|
 15. |   15    25 |
 16. |   15    25 |
 17. |   15    25 |
 18. |   15    25 |
 19. |   15    25 |
     |------------|
 20. |   20    24 |
 21. |   20    24 |
 22. |   20    24 |
 23. |   20    24 |
     |------------|
 24. |   24    23 |
 25. |   24    23 |
 26. |   24    23 |
     |------------|
 27. |   27    22 |
 28. |   27    22 |
 29. |   27    22 |
 30. |   27    22 |
 31. |   27    22 |
     |------------|
 32. |   32    21 |
 33. |   32    21 |
 34. |   32    21 |
 35. |   32    21 |
 36. |   32    21 |
     |------------|
 37. |   37    20 |
 38. |   37    20 |
 39. |   37    20 |
     |------------|
 40. |   40    19 |
 41. |   40    19 |
 42. |   40    19 |
 43. |   40    19 |
 44. |   40    19 |
 45. |   40    19 |
 46. |   40    19 |
 47. |   40    19 |
     |------------|
 48. |   48    18 |
 49. |   48    18 |
 50. |   48    18 |
 51. |   48    18 |
 52. |   48    18 |
 53. |   48    18 |
 54. |   48    18 |
 55. |   48    18 |
 56. |   48    18 |
     |------------|
 57. |   57    17 |
 58. |   57    17 |
 59. |   57    17 |
 60. |   57    17 |
     |------------|
 61. |   61    16 |
 62. |   61    16 |
 63. |   61    16 |
 64. |   61    16 |
     |------------|
 65. |   65    15 |
 66. |   65    15 |
     |------------|
 67. |   67    14 |
 68. |   67    14 |
 69. |   67    14 |
 70. |   67    14 |
 71. |   67    14 |
 72. |   67    14 |
     |------------|
 73. |   73    12 |
 74. |   73    12 |
     +------------+

[George Kariuki]

Thank you so much. The expound on field and track option is undoubtedly my today's big "learn". With the sequence of code, I think track works great for me and gives the answer am looking for. Field leads to numbers larger than the score by some +1 (obvious from its construct). Their arguments seem to be a mirror of the other and thus allows individuals to have both approaches - mine and the opposite of it. Appreciated!