Help with keep/drop

[Brendan Churchill]

Dear Statalisers

My dataset is longitudinal following people from 2008 to 2019. I would like to specifically follow those who were aged between 25 and 34 in 2008 over time as they age, but I'm having trouble with the keep/drop commands.

I have tried things like:

drop if (hgage <35 & year==2008) | (hgage >20 & year==2008)

I can't quite get it to work as whatever I try either includes people outside of this age range or I lose observations from 2009 onwards.

Any help is appreciated

best
Brendan

* Example generated by -dataex-. For more info, type help dataex
clear
input int year long pid int hgage
2008 100003 55
2009 100003 56
2010 100003 57
2011 100003 58
2012 100003 59
2013 100003 60
2014 100003 61
2015 100003 62
2016 100003 63
2017 100003 64
2018 100003 65
2019 100003 66
2008 100005 23
2009 100005 24
2010 100005 25
2011 100005 26
2012 100005 27
2013 100005 28
2014 100005 29
2015 100005 30
2016 100005 31
2017 100005 32
2018 100005 33
2019 100005 34
2008 100006 21
2009 100006 22
2010 100006 23
2011 100006 24
2012 100006 25
2013 100006 26
2014 100006 27
2015 100006 28
2016 100006 29
2017 100006 30
2018 100006 31
2019 100006 32
2008 100007 16
2009 100007 17
2010 100007 18
2011 100007 19
2012 100007 20
2008 100010 43
2009 100010 44
2010 100010 45
2011 100010 46
2012 100010 47
2013 100010 48
2014 100010 49
2015 100010 50
2016 100010 51
2017 100010 52
2018 100010 53
2019 100010 54
2008 100011 38
2009 100011 39
2008 100014 54
2009 100014 55
2010 100014 56
2011 100014 57
2012 100014 58
2013 100014 59
2014 100014 60
2015 100014 61
2016 100014 62
2018 100014 64
2019 100014 65
2008 100015 51
2009 100015 52
2010 100015 53
2011 100015 54
2012 100015 55
2013 100015 56
2014 100015 57
2015 100015 58
2016 100015 59
2017 100015 60
2018 100015 61
2019 100015 62
2008 100016 26
2009 100016 27
2010 100016 28
2011 100016 29
2012 100016 30
2013 100016 31
2014 100016 32
2015 100016 33
2016 100016 34
2017 100016 35
2018 100016 36
2019 100016 37
2008 100018 47
2009 100018 48
2010 100018 49
2011 100018 50
2012 100018 51
2013 100018 52
2014 100018 53
2015 100018 54
2016 100018 55
2017 100018 56
end
label values hgage VHGAGE

[Carlo Lazzaro]

Brendan:
the best way (for me, at least) to handle this issue is to get a precise idea of how many observations satisfy the -if- condition:

. tab hgage if year==2008

      hgage |      Freq.     Percent        Cum.
------------+-----------------------------------
         16 |          1       10.00       10.00
         21 |          1       10.00       20.00
         23 |          1       10.00       30.00
         26 |          1       10.00       40.00
         38 |          1       10.00       50.00
         43 |          1       10.00       60.00
         47 |          1       10.00       70.00
         51 |          1       10.00       80.00
         54 |          1       10.00       90.00
         55 |          1       10.00      100.00
------------+-----------------------------------
      Total |         10      100.00

In 2008, only one observation in your data excerpt fits the -if- condition.
Therefore, instead of -drop-ping variables, you may want to build an indicator to be used as an -if- condition in subsequent analyses:

. gen wanted=1 if hgage<35 & hgage>24 & year==2008

. list if wanted==1

     +--------------------------------+
     | year      pid   hgage   wanted |
     |--------------------------------|
 79. | 2008   100016      26        1 |
     +--------------------------------+

.

[Brendan Churchill]

Thanks Carlo for the solution. Unfortunately when I do this and then i do keep if wanted==1 then I only have observations for 2008. I want 25 to 34 year olds in 2008 and then to follow that cohort over ten years.

[Nick Cox]

The crunch here is that your syntax is about selecting individual observations, but you want to select people. That makes the problem a two-step problem. One way to see that you're missing part of the problem is that the syntax in #1 makes no use of the identifier pid.

Here are two ways to do it. I have to save the example data so that I can show that they give the same result, but you only need one method.

* Example generated by -dataex-. For more info, type help dataex
clear
input int year long pid int hgage
2008 100003 55
2009 100003 56
2010 100003 57
2011 100003 58
2012 100003 59
2013 100003 60
2014 100003 61
2015 100003 62
2016 100003 63
2017 100003 64
2018 100003 65
2019 100003 66
2008 100005 23
2009 100005 24
2010 100005 25
2011 100005 26
2012 100005 27
2013 100005 28
2014 100005 29
2015 100005 30
2016 100005 31
2017 100005 32
2018 100005 33
2019 100005 34
2008 100006 21
2009 100006 22
2010 100006 23
2011 100006 24
2012 100006 25
2013 100006 26
2014 100006 27
2015 100006 28
2016 100006 29
2017 100006 30
2018 100006 31
2019 100006 32
2008 100007 16
2009 100007 17
2010 100007 18
2011 100007 19
2012 100007 20
2008 100010 43
2009 100010 44
2010 100010 45
2011 100010 46
2012 100010 47
2013 100010 48
2014 100010 49
2015 100010 50
2016 100010 51
2017 100010 52
2018 100010 53
2019 100010 54
2008 100011 38
2009 100011 39
2008 100014 54
2009 100014 55
2010 100014 56
2011 100014 57
2012 100014 58
2013 100014 59
2014 100014 60
2015 100014 61
2016 100014 62
2018 100014 64
2019 100014 65
2008 100015 51
2009 100015 52
2010 100015 53
2011 100015 54
2012 100015 55
2013 100015 56
2014 100015 57
2015 100015 58
2016 100015 59
2017 100015 60
2018 100015 61
2019 100015 62
2008 100016 26
2009 100016 27
2010 100016 28
2011 100016 29
2012 100016 30
2013 100016 31
2014 100016 32
2015 100016 33
2016 100016 34
2017 100016 35
2018 100016 36
2019 100016 37
2008 100018 47
2009 100018 48
2010 100018 49
2011 100018 50
2012 100018 51
2013 100018 52
2014 100018 53
2015 100018 54
2016 100018 55
2017 100018 56
end
label values hgage VHGAGE

save safecopy

gen tag = year == 2008 & inrange(hgage, 25, 34)

bysort pid (tag) : keep if tag[_N]

list, sepby(pid)

use safecopy, clear

egen tag = max(year == 2008 & inrange(hgage, 25, 34)), by(pid)

keep if tag

list, sepby(pid)

. * Example generated by -dataex-. For more info, type help dataex
. clear

. input int year long pid int hgage

         year           pid     hgage
  1. 2008 100003 55
  2. 2009 100003 56
  3. 2010 100003 57
  4. 2011 100003 58
  5. 2012 100003 59
  6. 2013 100003 60
  7. 2014 100003 61
  8. 2015 100003 62
  9. 2016 100003 63
 10. 2017 100003 64
 11. 2018 100003 65
 12. 2019 100003 66
 13. 2008 100005 23
 14. 2009 100005 24
 15. 2010 100005 25
 16. 2011 100005 26
 17. 2012 100005 27
 18. 2013 100005 28
 19. 2014 100005 29
 20. 2015 100005 30
 21. 2016 100005 31
 22. 2017 100005 32
 23. 2018 100005 33
 24. 2019 100005 34
 25. 2008 100006 21
 26. 2009 100006 22
 27. 2010 100006 23
 28. 2011 100006 24
 29. 2012 100006 25
 30. 2013 100006 26
 31. 2014 100006 27
 32. 2015 100006 28
 33. 2016 100006 29
 34. 2017 100006 30
 35. 2018 100006 31
 36. 2019 100006 32
 37. 2008 100007 16
 38. 2009 100007 17
 39. 2010 100007 18
 40. 2011 100007 19
 41. 2012 100007 20
 42. 2008 100010 43
 43. 2009 100010 44
 44. 2010 100010 45
 45. 2011 100010 46
 46. 2012 100010 47
 47. 2013 100010 48
 48. 2014 100010 49
 49. 2015 100010 50
 50. 2016 100010 51
 51. 2017 100010 52
 52. 2018 100010 53
 53. 2019 100010 54
 54. 2008 100011 38
 55. 2009 100011 39
 56. 2008 100014 54
 57. 2009 100014 55
 58. 2010 100014 56
 59. 2011 100014 57
 60. 2012 100014 58
 61. 2013 100014 59
 62. 2014 100014 60
 63. 2015 100014 61
 64. 2016 100014 62
 65. 2018 100014 64
 66. 2019 100014 65
 67. 2008 100015 51
 68. 2009 100015 52
 69. 2010 100015 53
 70. 2011 100015 54
 71. 2012 100015 55
 72. 2013 100015 56
 73. 2014 100015 57
 74. 2015 100015 58
 75. 2016 100015 59
 76. 2017 100015 60
 77. 2018 100015 61
 78. 2019 100015 62
 79. 2008 100016 26
 80. 2009 100016 27
 81. 2010 100016 28
 82. 2011 100016 29
 83. 2012 100016 30
 84. 2013 100016 31
 85. 2014 100016 32
 86. 2015 100016 33
 87. 2016 100016 34
 88. 2017 100016 35
 89. 2018 100016 36
 90. 2019 100016 37
 91. 2008 100018 47
 92. 2009 100018 48
 93. 2010 100018 49
 94. 2011 100018 50
 95. 2012 100018 51
 96. 2013 100018 52
 97. 2014 100018 53
 98. 2015 100018 54
 99. 2016 100018 55
100. 2017 100018 56
101. end

. label values hgage VHGAGE

.
. save safecopy
.
. gen tag = year == 2008 & inrange(hgage, 25, 34)

.
. bysort pid (tag) : keep if tag[_N]
(88 observations deleted)

.
. list, sepby(pid)

     +-----------------------------+
     | year      pid   hgage   tag |
     |-----------------------------|
  1. | 2013   100016      31     0 |
  2. | 2009   100016      27     0 |
  3. | 2010   100016      28     0 |
  4. | 2015   100016      33     0 |
  5. | 2014   100016      32     0 |
  6. | 2019   100016      37     0 |
  7. | 2018   100016      36     0 |
  8. | 2012   100016      30     0 |
  9. | 2016   100016      34     0 |
 10. | 2017   100016      35     0 |
 11. | 2011   100016      29     0 |
 12. | 2008   100016      26     1 |
     +-----------------------------+

.
. use safecopy, clear

.
. egen tag = max(year == 2008 & inrange(hgage, 25, 34)), by(pid)

.
. keep if tag
(88 observations deleted)

.
. list, sepby(pid)

     +-----------------------------+
     | year      pid   hgage   tag |
     |-----------------------------|
  1. | 2008   100016      26     1 |
  2. | 2009   100016      27     1 |
  3. | 2010   100016      28     1 |
  4. | 2011   100016      29     1 |
  5. | 2012   100016      30     1 |
  6. | 2013   100016      31     1 |
  7. | 2014   100016      32     1 |
  8. | 2015   100016      33     1 |
  9. | 2016   100016      34     1 |
 10. | 2017   100016      35     1 |
 11. | 2018   100016      36     1 |
 12. | 2019   100016      37     1 |
     +-----------------------------+

For more on the technique see e.g. https://www.stata.com/support/faqs/d...ble-recording/

https://journals.sagepub.com/doi/pdf...867X1101100210