Count consecutive observation as cumulative sum

[Francois Durant]

Hi,

I have the following dataset:

 
USREC
myear

1
2001m4

1
2001m5

1
2001m6

1
2001m7

1
2001m8

1
2001m9

1
2001m10

1
2001m11

0
2001m12

0
2002m1

0
2002m2

0
2002m3

0
2002m4

0
2002m5

1
2002m6

1
2002m7

1
2002m8

1
2002m9

0
2002m10

0
2002m11

1
2002m12

1
2003m1

1
2003m2

0
2003m3

0
2003m4

I would like to count the consecutive "USREC"=="1", reinitialize the counter again when "0", not count the "0"s and start the counter over again when a new series a "1" starts. Manually, it looks as follow:

 
USREC
myear
rec_month

1
2001m4
1

1
2001m5
2

1
2001m6
3

1
2001m7
4

1
2001m8
5

1
2001m9
6

1
2001m10
7

1
2001m11
8

0
2001m12
0

0
2002m1
0

0
2002m2
0

0
2002m3
0

0
2002m4
0

0
2002m5
0

1
2002m6
1

1
2002m7
2

1
2002m8
3

1
2002m9
4

0
2002m10
0

0
2002m11
0

1
2002m12
1

1
2003m1
2

1
2003m2
3

0
2003m3
0

0
2003m4
0

I'm not sure how to program that. I have tried using "_n", unsuccessfully. Can someone help? Thanks!

[Stephen Jenkins]

Have a look at Nick Cox's tsspell on SSC. I think it has the functionality you seek.

[Nick Cox]

With tsspell (SSC) that variable comes for free as _seq when you go (e.g.)

tsset myear
tsspell, pcond(USREC)

If you are interested in spells, however called, that program may have other uses.

There are also various ways to get it directly. Here's one:

gen wanted = USREC in 1
replace wanted = cond(USREC, wanted[_n-1] + 1, 0) in 2/L

See also

http://www.stata-journal.com/sjpdf.h...iclenum=dm0029 (free to all)

for a detailed discussion of principles here.

In the cond() code, a moment's thought will show that USREC will do instead of USREC == 1 as evaluating USREC == 1 returns 1 or 0 precisely when evaluating USREC does. (If there are missings on USREC, you need something different, but it has the flavour of something always known.)

[Francois Durant]

Thanks a lot Nick &Stephen!

[Thong Nguyen]

I have a problem with my data
variables:
tmax:Maximum temperature
date: date of event

And I wanted to identify number of events and duration of each event.
Each event is identified when 2 following conditions are satisfied:
(1) tmax is equal 36.4 or more
(2) number of consecutive days of each event is 3 days or more
start date of the research is 1Jan2005


I used codes as below:
gen time=td(1jan2005)+date-1
tsspell time, cond(tmax>=36.4)
list _spell _seq date if tmax>=36.4


And I got results like this:

+---------------------------+
| _spell _seq date |
|---------------------------|
75. | 1 1 16mar2005 |
78. | 2 1 19mar2005 |
79. | 2 2 20mar2005 |
83. | 3 1 24mar2005 |
93. | 4 1 03apr2005 |
|---------------------------|
94. | 4 2 04apr2005 |
95. | 4 3 05apr2005 |
102. | 5 1 12apr2005 |
104. | 6 1 14apr2005 |
109. | 7 1 19apr2005 |
|---------------------------|
110. | 7 2 20apr2005 |
111. | 7 3 21apr2005 |


The problem here is Stata only determines each spell based on threshold of tmax, and not include number of consecutive days at least 3 days.

How can I make it right?
How can I generate a variable 'dur' as a number of consecutive days for each event.
How can I generate a variable as number of events weekly, monthly?

Thank you in advance!

[Nick Cox]

Stata just does what you tell it. The help for tsspell, which you use here but do not cite (SSC), does discuss two-stage identification of spells and gives a worked example. In your case the code might be

tsspell time, cond(tmax>=36.4)
egen dur = max(_seq), by(_spell) 
drop _seq _spell _end 
tsspell time, cond(tmax >= 36.4 & dur >= 3)

expect that any structure of different places calls for modifications.

The rest is counting using egen if I understand correctly. My own bias is that weeks and months are usually artificial time units in climatology.

[Thong Nguyen]

Thank you very much Nick Cox.
But when i use command egen nohw = count(_spell) if _spell!=0
It makes wrong because it count each _spell many times.
for example: for _spell #1 and duration of _spell #1 is 3 days, it determines _spell #1 as 3 different spells.
Can you help me!

[Nick Cox]

Don't do that then. The number of spells is the same as the number of observations with _seq == 1 or _end == 1.

Or it is the maximum value of _spell.

[Thong Nguyen]

Ok. I figure it out. Thank you so much Nick Cox! Have a nice day ^^

[Thong Nguyen]

How do I create a variable as maximum value of temperature for each spell. Thank you in advance Nick!

[Thong Nguyen]

Finally, I got it, thanks!

[Thong Nguyen]

I have a data of daily maximum temperature (tmax), and cases by month. How do i create a variable (mean_tmax_m) as "Mean Maximum Temperature (by month)" in order to link it with cases. Thanks for your help!

[Nick Cox]

These new questions are unrelated to the present thread. Please start a new thread for your next different question. Otherwise

help egen