Panel data - dropping all observations after a switch

[Graham Wells]

Hi everyone,

I have panel data below (just one id shown), it is sorted by id and date
I want to be able to drop all observations once the type is different from above - i.e. I only want to keep the observations below nn<=45
How can I code this so that the order of 1 or 2 doesn't matter (some id have first type as 2, some have first type as 1 - I want to keep only the observations where the first type occurs but none after they switch, even if they switch back)

any help would be appreciated

* Example generated by -dataex-. To install: ssc install dataex
clear
input long id int date byte type float nn
165618 16819 1   1
165618 16845 1   2
165618 16867 1   3
165618 16887 1   4
165618 16897 1   5
165618 16918 1   6
165618 16940 1   7
165618 16965 1   8
165618 16987 1   9
165618 17010 1  10
165618 17034 1  11
165618 17058 1  12
165618 17080 1  13
165618 17106 1  14
165618 17129 1  15
165618 17153 1  16
165618 17177 1  17
165618 17199 1  18
165618 17221 1  19
165618 17245 1  20
165618 17274 1  21
165618 17303 1  22
165618 17331 1  23
165618 17359 1  24
165618 17386 1  25
165618 17421 1  26
165618 17456 1  27
165618 17485 1  28
165618 17513 1  29
165618 17539 1  30
165618 17567 1  31
165618 17594 1  32
165618 17619 1  33
165618 17645 1  34
165618 17671 1  35
165618 17696 1  36
165618 17723 1  37
165618 17748 1  38
165618 17772 1  39
165618 17797 1  40
165618 17806 1  41
165618 17811 1  42
165618 17834 1  43
165618 17868 1  44
165618 17902 1  45
165618 17918 2  46
165618 17930 2  47
165618 17930 1  48
165618 17958 2  49
165618 17958 1  50
165618 17986 1  51
165618 18014 2  52
165618 18014 1  53
165618 18028 2  54
165618 18031 2  55
165618 18039 2  56
165618 18056 2  57
165618 18057 1  58
165618 18070 2  59
165618 18084 2  60
165618 18088 1  61
165618 18088 2  62
165618 18088 1  63
165618 18100 2  64
165618 18100 1  65
165618 18129 2  66
165618 18129 1  67
165618 18161 1  68
165618 18161 2  69
165618 18189 1  70
165618 18189 2  71
165618 18213 1  72
165618 18213 2  73
165618 18241 2  74
165618 18241 1  75
165618 18274 2  76
165618 18274 1  77
165618 18274 1  78
165618 18274 2  79
165618 18302 1  80
165618 18302 2  81
165618 18329 1  82
165618 18329 2  83
165618 18351 1  84
165618 18351 2  85
165618 18381 2  86
165618 18381 1  87
165618 18415 1  88
165618 18415 2  89
165618 18441 2  90
165618 18441 1  91
165618 18465 2  92
165618 18465 1  93
165618 18492 1  94
165618 18519 1  95
165618 18553 1  96
165618 18561 1  97
165618 18577 1  98
165618 18605 1  99
165618 18631 1 100
end
format %tdnn/dd/CCYY date

[Graham Wells]

I think I found a way:

bysort id (date): gen switch_=1 if type!=type[_n-1]
bysort id (date): gen switch_sum= sum(switch_)
drop if switch_sum>1

[Nick Cox]

Thanks for posting a nice answer.

Now that you've seen that, note that it condenses to

bysort id (date): drop if sum(type!=type[_n-1]) > 1

Related discussion at https://www.stata.com/support/faqs/d...t-occurrences/