Creating a variable to show response patterns

[Sara Khan]

Dear all

I have 3 variables in long format.
1. pid
2. qnum - question number (1-16)
3. asthq - binary response to qunum

I would like to create a new variable to indicate a particular response pattern within id over the variable asthq (e.g., 0=all zeroes, and so forth).

I would be grateful for any advice on how to do this.

     +--------------------+
     | pid   qnum   asthq |
     |--------------------|
  1. |   1      1     Yes |
  2. |   1      2     Yes |
  3. |   1      3      No |
  4. |   1      4      No |
  5. |   1      5      No |
     |--------------------|
  6. |   1      6      No |
  7. |   1      7      No |
  8. |   1      8      No |
  9. |   1      9     Yes |
 10. |   1     10     Yes |
     |--------------------|
 11. |   1     11     Yes |
 12. |   1     12      No |
 13. |   1     13     Yes |
 14. |   1     14     Yes |
 15. |   1     15      No |
     |--------------------|
 16. |   1     16     Yes |
 17. |   2      1     Yes |
 18. |   2      2     Yes |
 19. |   2      3     Yes |
 20. |   2      4     Yes |
     |--------------------|
 21. |   2      5     Yes |
 22. |   2      6     Yes |
 23. |   2      7     Yes |
 24. |   2      8     Yes |
 25. |   2      9     Yes |
     |--------------------|
 26. |   2     10     Yes |
 27. |   2     11     Yes |
 28. |   2     12     Yes |
 29. |   2     13     Yes |
 30. |   2     14     Yes |
     +--------------------+

[Marcos Almeida]

The "particular response pattern" was not fully clarified in #1.

That said, you may wish to check - egen - command and options.

[Sara Khan]

Sorry - I want to identify all different types of response patterns across 16 variables (all 0s is just one of them).

[Nick Cox]

You seem to have 16 observations for each identifier, not 16 variables.

Perhaps what you want is

reshape wide asthq, i(pid) j(qnum) 

egen pattern = concat(asthq*) 

tab pattern 

[Sara Khan]

Thank you Nick! Now that I have my patterns, are you able to advise on how to assign a code to each pattern type which is equal to the frequency as I need to use this as a weight in my model? For example, from the patterns below, 99 individuals would have a weight of 99?

         pattern |      Freq.     Percent        Cum.
-----------------+-----------------------------------
0000000000000000 |         99       19.15       19.15
0000000000000100 |          5        0.97       20.12
0000000001000110 |          1        0.19       20.31
0000000001001010 |          3        0.58       20.89
0000000001001110 |          1        0.19       21.08
0000000001001111 |          1        0.19       21.28
0000000001101010 |          5        0.97       22.24
0000000001101011 |          3        0.58       22.82
0000000011000011 |          1        0.19       23.02
0000000011000110 |          1        0.19       23.21
0000000100000000 |          4        0.77       23.98
0000000111100011 |          1        0.19       24.18
0000001000000000 |         18        3.48       27.66
0000001001001010 |          1        0.19       27.85
0000001001001110 |          1        0.19       28.05
0000001001101010 |          2        0.39       28.43
0000001001101011 |          3        0.58       29.01
0000001111100111 |          1        0.19       29.21
0000100000000000 |          2        0.39       29.59
0000100001101000 |          1        0.19       29.79
0000100001101011 |          1        0.19       29.98
0001000000000000 |         80       15.47       45.45

[Nick Cox]

That sounds completely unnecessary: if a pattern is repeated so many times, that will be its weight in fitting a model automatically.