Creating episodes

[Sumedha Gupta]

Dear All,
I have an individual level panel data with 108 monthly waves.

	
	
* Example generated by -dataex-. To install: ssc install dataex
clear
input double(ID Month_count) float y
25316   1         0
25316   2         0
25316   3         0
25316   4         0
25316   5         0
25316   6         0
25316   7         0
25316   8         0
25316   9         0
25316  10         0
25316  11         0
25316  12         0
25316  13        10
25316  14 4.3396225
25316  15         5
25316  16         0
25316  17         5
25316  18         5
25316  19         0
25316  20         0
25316  21       200
25316  22         0
25316  23       800
25316  24      1270
25316  25  846.6666
25316  26  423.3333
25316  27       800
25316  28      1620
25316  29 1334.4578
25316  30 1652.1237
25316  31  1935.835
25316  32  1455.876
25316  33  975.9167
25316  34 1295.9578
25316  35  770.5695
25316  36 1045.1812
25316  37  495.0858
25316  38         0
25316  39       800
25316  40  228.5714
25316  41         0
25316  42       950
25316  43  316.6667
25316  44         0
25316  45       950
25316  46  1597.222
25316  47 2273.0903
25316  48   2953.94
25316  49  3411.617
25316  50 2878.5515
25316  51 2345.4863
25316  52 2912.4214
25316  53  3405.667
25316  54  3350.731
25316  55 3806.3596
25316  56  4230.675
25316  57 4133.3105
25316  58 4551.9956
25316  59 3848.0786
25316  60  3144.162
25316  61  2440.245
25316  62 1736.3282
25316  63 1032.4114
25316  64  328.4945
25316  65         0
25316  66         0
25316  67         0
25316  68       225
25316  69       645
25316  70 528.29785
25316  71 191.08646
25316  72         0
25316  73         0
25316  74         0
25316  75         0
25316  76         0
25316  77         0
25316  78         0
25316  79         0
25316  80         0
25316  81         0
25316  82         0
25316  83         0
25316  84         0
25316  85         0
25316  86         0
25316  87         0
25316  88         0
25316  89         0
25316  90         0
25316  91         0
25316  92         0
25316  93         0
25316  94         0
25316  95         0
25316  96         0
25316  97         0
25316  98         0
25316  99         0
25316 100         0
end

For each ID I want to create episodes that includes periods of 3+ months where y>0. For instance, in the example above, in Months 13-15 is the first episode as y>0 in 3 consecutive months. In months 17-18 y>0 but this spell only last 2 months (<3 months) and therefore does not count as an episode. Similarly, in month 21 y>0 but its only 1 month and should not be considered an episode. Month 23-37 is the second episode, 24-64 is third episode, and 68-71 is the fourth. It seems I should use some combination of egen or rangestat but I am struggling to specify a rule that will take care of this. I will be grateful for your guidance.
Sincerely,
Sumedha.

[Joro Kolev]

I am not on very clear about the rule that defines an episode, but is this doing it?

. gen episode = ((y[_n-2]>0)+ (y[_n-1]>0)+ (y[_n]>0) + (y[_n+1]>0) + (y[_n+2]>0))>2

[Rich Goldstein]

I also am not very clear about the definition, but suggest you look at -tsspell- a user-written command; use -search- or -findit- to locate and download

[Nick Cox]

Here I use tsspell from SSC. in two passes, to identify episodes (runs, spell) of positive y and then select only those at least length 3.

tsset ID Month_count

tsspell , pcond(y) spell(spell1) seq(seq1) end(end1)

bysort ID spell1 (Month): egen max = max(seq1)

sort ID Month

tsspell, cond(y > 0 & max >= 3) spell(spell2) seq(seq2) end(end2)

list if spell1, sepby(spell1)

     +---------------------------------------------------------------------------------+
     |    ID   Month_~t          y   seq1   spell1   end1   max   seq2   spell2   end2 |
     |---------------------------------------------------------------------------------|
 13. | 25316         13         10      1        1      0     3      1        1      0 |
 14. | 25316         14   4.339622      2        1      0     3      2        1      0 |
 15. | 25316         15          5      3        1      1     3      3        1      1 |
     |---------------------------------------------------------------------------------|
 17. | 25316         17          5      1        2      0     2      0        0      0 |
 18. | 25316         18          5      2        2      1     2      0        0      0 |
     |---------------------------------------------------------------------------------|
 21. | 25316         21        200      1        3      1     1      0        0      0 |
     |---------------------------------------------------------------------------------|
 23. | 25316         23        800      1        4      0    15      1        2      0 |
 24. | 25316         24       1270      2        4      0    15      2        2      0 |
 25. | 25316         25   846.6666      3        4      0    15      3        2      0 |
 26. | 25316         26   423.3333      4        4      0    15      4        2      0 |
 27. | 25316         27        800      5        4      0    15      5        2      0 |
 28. | 25316         28       1620      6        4      0    15      6        2      0 |
 29. | 25316         29   1334.458      7        4      0    15      7        2      0 |
 30. | 25316         30   1652.124      8        4      0    15      8        2      0 |
 31. | 25316         31   1935.835      9        4      0    15      9        2      0 |
 32. | 25316         32   1455.876     10        4      0    15     10        2      0 |
 33. | 25316         33   975.9167     11        4      0    15     11        2      0 |
 34. | 25316         34   1295.958     12        4      0    15     12        2      0 |
 35. | 25316         35   770.5695     13        4      0    15     13        2      0 |
 36. | 25316         36   1045.181     14        4      0    15     14        2      0 |
 37. | 25316         37   495.0858     15        4      1    15     15        2      1 |
     |---------------------------------------------------------------------------------|
 39. | 25316         39        800      1        5      0     2      0        0      0 |
 40. | 25316         40   228.5714      2        5      1     2      0        0      0 |
     |---------------------------------------------------------------------------------|
 42. | 25316         42        950      1        6      0     2      0        0      0 |
 43. | 25316         43   316.6667      2        6      1     2      0        0      0 |
     |---------------------------------------------------------------------------------|
 45. | 25316         45        950      1        7      0    20      1        3      0 |
 46. | 25316         46   1597.222      2        7      0    20      2        3      0 |
 47. | 25316         47    2273.09      3        7      0    20      3        3      0 |
 48. | 25316         48    2953.94      4        7      0    20      4        3      0 |
 49. | 25316         49   3411.617      5        7      0    20      5        3      0 |
 50. | 25316         50   2878.552      6        7      0    20      6        3      0 |
 51. | 25316         51   2345.486      7        7      0    20      7        3      0 |
 52. | 25316         52   2912.421      8        7      0    20      8        3      0 |
 53. | 25316         53   3405.667      9        7      0    20      9        3      0 |
 54. | 25316         54   3350.731     10        7      0    20     10        3      0 |
 55. | 25316         55    3806.36     11        7      0    20     11        3      0 |
 56. | 25316         56   4230.675     12        7      0    20     12        3      0 |
 57. | 25316         57   4133.311     13        7      0    20     13        3      0 |
 58. | 25316         58   4551.996     14        7      0    20     14        3      0 |
 59. | 25316         59   3848.079     15        7      0    20     15        3      0 |
 60. | 25316         60   3144.162     16        7      0    20     16        3      0 |
 61. | 25316         61   2440.245     17        7      0    20     17        3      0 |
 62. | 25316         62   1736.328     18        7      0    20     18        3      0 |
 63. | 25316         63   1032.411     19        7      0    20     19        3      0 |
 64. | 25316         64   328.4945     20        7      1    20     20        3      1 |
     |---------------------------------------------------------------------------------|
 68. | 25316         68        225      1        8      0     4      1        4      0 |
 69. | 25316         69        645      2        8      0     4      2        4      0 |
 70. | 25316         70   528.2979      3        8      0     4      3        4      0 |
 71. | 25316         71   191.0865      4        8      1     4      4        4      1 |
     +---------------------------------------------------------------------------------+

[Sumedha Gupta]

Thank you Prof. Goldstein and Prof. Cox. Following Prof. Goldstein's message I explored -tsspell- and wrote the following, which seems to work.

tsspell y, fcond((y>0 & L1.y==0)| (y==0 & L1.y>0)) 
replace _seq=0 if y==0
bysort ID _spell: egen ep_length=max(_seq)
recode ep_length (0/2=0)
rename (_spell _seq _end) (spell1 seq1 end1)
tsspell ep_length, fcond(ep_length>0 & spell1~=L1.spell1)  
gen episode=_spell
replace episode=0 if ep_length==0
drop _end _seq end1 _spell

But, Prof. Cox, your version is so much more elegant. Thank you so much! -tsspell- is a work of art!