lags and leads for panel data, where an individual does an action multiple times during observational period

[James Glass]

I have an issue of not accurately reflecting the lag/lead variables if an individual migrates multiple times during the observational period.
I am using stata 15.0

pid = ID number of the individual, wave = time, M0 = If an individual migrated in that wave

I have made dummy variables for 3 lags and 3 leads for the act of migration, this is the code that I used.

tsset pid wave
gen p = wave if M0 ==1
bys pid: egen time = max(p)
gen timelength = time - wave

gen lead1=1 if timelength ==1
replace lead1=0 if lead1 != 1
gen lead2=1 if timelength == 2
replace lead2=0 if lead2 != 1
gen lead3=1 if timelength == 3
replace lead3=0 if lead3 != 1
gen lead4=1 if timelength == 4
replace lead4=0 if lead4 != 1
gen lead5=1 if timelength == 5
replace lead5=0 if lead5 != 1
gen lead6=1 if timelength == 2
replace lead6=0 if lead6 != 1
gen lead7=1 if timelength == 2
replace lead7=0 if lead7 != 1


gen lag1= 1 if timelength == -1
replace lag1 = 0 if lag1 != 1
gen lag2= 1 if timelength == -2
replace lag2 = 0 if lag2 != 1
gen lag3= 1 if timelength == -3
replace lag3 = 0 if lag3 != 1
gen lag4= 1 if timelength == -4
replace lag4 = 0 if lag4 != 1
gen lag5= 1 if timelength == -5
replace lag5 = 0 if lag5 != 1

This is it working as intended:

     +-----------------------------------------------------------------------------------------+
     |      pid   wave   M0   p   time   timele~h   lead1   lead2   lead3   lag1   lag2   lag3 |
     |-----------------------------------------------------------------------------------------|
131. | 10029133     12    0   .      .          .       0       0       0      0      0      0 |
132. | 10040331      6    0   .      7          1       1       0       0      0      0      0 |
133. | 10040331      7    1   7      7          0       0       0       0      0      0      0 |
134. | 10040331      8    0   .      7         -1       0       0       0      1      0      0 |
135. | 10040331      9    0   .      7         -2       0       0       0      0      1      0 |
     |-----------------------------------------------------------------------------------------|
136. | 10040331     10    0   .      7         -3       0       0       0      0      0      1 |
137. | 10040331     11    0   .      7         -4       0       0       0      0      0      0 |
138. | 10040331     12    0   .      7         -5       0       0       0      0      0      0 |
139. | 10040331     13    0   .      7         -6       0       0       0      0      0      0 |
140. | 10040331     14    0   .      7         -7       0       0       0      0      0      0 |
     |-----------------------------------------------------------------------------------------|
141. | 10040331     15    0   .      7         -8       0       0       0      0      0      0 |
142. | 10040331     16    0   .      7         -9       0       0       0      0      0      0 |
143. | 10040331     17    0   .      7        -10       0       0       0      0      0      0 |
144. | 10040331     18    0   .      7        -11       0       0       0      0      0      0 |
145. | 10040366      6    0   .      .          .       0       0       0      0      0      0 |
     |-----------------------------------------------------------------------------------------|
146. | 10040366      7    0   .      .          .       0       0       0      0      0      0 |
147. | 10040366      8    0   .      .          .       0       0       0      0      0      0 |
148. | 10040366     10    0   .      .          .       0       0       0      0      0      0 |
149. | 10040366     11    0   .      .          .       0       0       0      0      0      0 |
   

But when there are multiple acts of migration, it only takes account of the last migration of the individual

     +------------------------------------------------------------------------------------------+
     |      pid   wave   M0    p   time   timele~h   lead1   lead2   lead3   lag1   lag2   lag3 |
     |------------------------------------------------------------------------------------------|
210. | 10048243      6    0    .     15          9       0       0       0      0      0      0 |
211. | 10048243      7    0    .     15          8       0       0       0      0      0      0 |
212. | 10048243      8    0    .     15          7       0       0       0      0      0      0 |
213. | 10048243     10    0    .     15          5       0       0       0      0      0      0 |
214. | 10048243     11    0    .     15          4       0       0       0      0      0      0 |
     |------------------------------------------------------------------------------------------|
215. | 10048243     12    0    .     15          3       0       0       1      0      0      0 |
216. | 10048243     13    1   13     15          2       0       1       0      0      0      0 |
217. | 10048243     14    0    .     15          1       1       0       0      0      0      0 |
218. | 10048243     15    1   15     15          0       0       0       0      0      0      0 |
219. | 10048243     16    0    .     15         -1       0       0       0      1      0      0 |
     |------------------------------------------------------------------------------------------|
220. | 10048243     17    0    .     15         -2       0       0       0      0      1      0 |
221. | 10048243     18    0    .     15         -3       0       0       0      0      0      1 |

Is there a way to change the priority or should my code change in general?

[Nick Cox]

This is a sequel to https://www.statalist.org/forums/for...ced-panel-data which immediately raises some suggestions.

1. You should cross-reference relevant previous threads, in fact continue them if there is a progression.

2. Clyde advised you to look at

help tsvarlist

to understand lag operators and that really was good advice.

3. The request to provide readable data examples using dataex still stands. Permanently. No-one can use your data example here without editing hackery which is easily avoided if you just use dataex as requested.

That said, to understand your code here, I started to rewrite it. This block can be cut in two, as true or false conditions are automatically 1 if true and 0 if false.

gen lead1=1 if timelength ==1
replace lead1=0 if lead1 != 1
gen lead2=1 if timelength == 2
replace lead2=0 if lead2 != 1
gen lead3=1 if timelength == 3
replace lead3=0 if lead3 != 1
gen lead4=1 if timelength == 4
replace lead4=0 if lead4 != 1
gen lead5=1 if timelength == 5
replace lead5=0 if lead5 != 1
gen lead6=1 if timelength == 2
replace lead6=0 if lead6 != 1
gen lead7=1 if timelength == 2
replace lead7=0 if lead7 != 1

becomes

gen lead1= timelength == 1
gen lead2= timelength == 2
gen lead3= timelength == 3
gen lead4= timelength == 4
gen lead5= timelength == 5
gen lead6= timelength == 2
gen lead7= timelength == 2

Now I don't understand what's going on. Shouldn't that be 1 2 3 4 5 6 7? Either it's a bug you should fix, or there's some special logic you need to explain.

[James Glass]

I think this is in the right format:

The code working as intended version:

* Example generated by -dataex-. To install: ssc install dataex
clear
input long pid float(wave M0 p time timelength lead1 lead2 lead3 lag1 lag2 lag3)
10029133 12 0 . .   . 0 0 0 0 0 0
10040331  6 0 . 7   1 1 0 0 0 0 0
10040331  7 1 7 7   0 0 0 0 0 0 0
10040331  8 0 . 7  -1 0 0 0 1 0 0
10040331  9 0 . 7  -2 0 0 0 0 1 0
10040331 10 0 . 7  -3 0 0 0 0 0 1
10040331 11 0 . 7  -4 0 0 0 0 0 0
10040331 12 0 . 7  -5 0 0 0 0 0 0
10040331 13 0 . 7  -6 0 0 0 0 0 0
10040331 14 0 . 7  -7 0 0 0 0 0 0
10040331 15 0 . 7  -8 0 0 0 0 0 0
10040331 16 0 . 7  -9 0 0 0 0 0 0
10040331 17 0 . 7 -10 0 0 0 0 0 0
10040331 18 0 . 7 -11 0 0 0 0 0 0
10040366  6 0 . .   . 0 0 0 0 0 0
10040366  7 0 . .   . 0 0 0 0 0 0
10040366  8 0 . .   . 0 0 0 0 0 0
10040366 10 0 . .   . 0 0 0 0 0 0
10040366 11 0 . .   . 0 0 0 0 0 0
end

And the code showing the multiple observation for an individual:

* Example generated by -dataex-. To install: ssc install dataex
clear
input long pid float(wave M0 p time timelength lead1 lead2 lead3 lag1 lag2 lag3)
10048243  6 0  . 15  9 0 0 0 0 0 0
10048243  7 0  . 15  8 0 0 0 0 0 0
10048243  8 0  . 15  7 0 0 0 0 0 0
10048243 10 0  . 15  5 0 0 0 0 0 0
10048243 11 0  . 15  4 0 0 0 0 0 0
10048243 12 0  . 15  3 0 0 1 0 0 0
10048243 13 1 13 15  2 0 1 0 0 0 0
10048243 14 0  . 15  1 1 0 0 0 0 0
10048243 15 1 15 15  0 0 0 0 0 0 0
10048243 16 0  . 15 -1 0 0 0 1 0 0
10048243 17 0  . 15 -2 0 0 0 0 1 0
10048243 18 0  . 15 -3 0 0 0 0 0 1
end

[Nick Cox]

Sorry, but I remain confused on what you are asking.

The previous thread cited in #2 made standard suggestions about creating leading and lagged variables using lead and lag operators after a tsset

Either you want something extra, it seems, or you want to approach it differently, perhaps. I still don't know.

You have provided a brief data example in requested form, but otherwise you haven't answered any of the questions in #2.

Ignoring the singleton panel 10029133 this takes your example panels and shows you can create shifted versions of M0 for p(revious) and n(ext) waves.

* Example generated by -dataex-. To install: ssc install dataex
clear
input long pid float(wave M0 p)
10040331  6 0  .
10040331  7 1  7
10040331  8 0  .
10040331  9 0  .
10040331 10 0  .
10040331 11 0  .
10040331 12 0  .
10040331 13 0  .
10040331 14 0  .
10040331 15 0  .
10040331 16 0  .
10040331 17 0  .
10040331 18 0  .
10040366  6 0  .
10040366  7 0  .
10040366  8 0  .
10040366 10 0  .
10040366 11 0  .
10048243  6 0  .
10048243  7 0  .
10048243  8 0  .
10048243 10 0  .
10048243 11 0  .
10048243 12 0  .
10048243 13 1 13
10048243 14 0  .
10048243 15 1 15
10048243 16 0  .
10048243 17 0  .
10048243 18 0  .
end

 tsset pid wave 
 
forval j = 3(-1)1 { 
      gen pM0`j' = L`j'.M0 
 }
 
forval j = 1/3 { 
     gen nM0`j' = F`j'.M0 
} 

 list , sepby(pid) 

     +---------------------------------------------------------------------+
     |      pid   wave   M0    p   pM03   pM02   pM01   nM01   nM02   nM03 |
     |---------------------------------------------------------------------|
  1. | 10040331      6    0    .      .      .      .      1      0      0 |
  2. | 10040331      7    1    7      .      .      0      0      0      0 |
  3. | 10040331      8    0    .      .      0      1      0      0      0 |
  4. | 10040331      9    0    .      0      1      0      0      0      0 |
  5. | 10040331     10    0    .      1      0      0      0      0      0 |
  6. | 10040331     11    0    .      0      0      0      0      0      0 |
  7. | 10040331     12    0    .      0      0      0      0      0      0 |
  8. | 10040331     13    0    .      0      0      0      0      0      0 |
  9. | 10040331     14    0    .      0      0      0      0      0      0 |
 10. | 10040331     15    0    .      0      0      0      0      0      0 |
 11. | 10040331     16    0    .      0      0      0      0      0      . |
 12. | 10040331     17    0    .      0      0      0      0      .      . |
 13. | 10040331     18    0    .      0      0      0      .      .      . |
     |---------------------------------------------------------------------|
 14. | 10040366      6    0    .      .      .      .      0      0      . |
 15. | 10040366      7    0    .      .      .      0      0      .      0 |
 16. | 10040366      8    0    .      .      0      0      .      0      0 |
 17. | 10040366     10    0    .      0      0      .      0      .      . |
 18. | 10040366     11    0    .      0      .      0      .      .      . |
     |---------------------------------------------------------------------|
 19. | 10048243      6    0    .      .      .      .      0      0      . |
 20. | 10048243      7    0    .      .      .      0      0      .      0 |
 21. | 10048243      8    0    .      .      0      0      .      0      0 |
 22. | 10048243     10    0    .      0      0      .      0      0      1 |
 23. | 10048243     11    0    .      0      .      0      0      1      0 |
 24. | 10048243     12    0    .      .      0      0      1      0      1 |
 25. | 10048243     13    1   13      0      0      0      0      1      0 |
 26. | 10048243     14    0    .      0      0      1      1      0      0 |
 27. | 10048243     15    1   15      0      1      0      0      0      0 |
 28. | 10048243     16    0    .      1      0      1      0      0      . |
 29. | 10048243     17    0    .      0      1      0      0      .      . |
 30. | 10048243     18    0    .      1      0      0      .      .      . |
     +---------------------------------------------------------------------+

If that isn't what you want, I think you'll need to explain much more, but I will just back off if I can't follow you. Naturally, others may be able to help.

[James Glass]

Thank you, that has solved the issue