- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
From the data mentioned below, I want to generate id’s that had a flag=1 for two or more consecutive months. As described below, some id's had a flag=1 for 12 consecutive months, only 6 consecutive months, 4 consecutive months, and finally only a month. Also some id's had a flag=1 but not consecutively, for instance at month 12 of 2013 and month 2 of 2014. The output I want is only those ID’s who had a flag=1 for two or more consecutive months.
I really appreciate your help.
Id year month flag
1 2013 7 1
1 2013 8 1
1 2013 9 1
1 2013 10 1
1 2013 11 1
1 2013 12 1
1 2014 1 1
1 2014 2 1
1 2014 3 1
1 2014 4 1
1 2014 5 1
1 2014 6 1
2 2013 8 1
2 2013 9 1
2 2013 10 1
2 2013 11 1
2 2013 12 1
2 2014 1 1
3 2014 2 1
3 2014 3 1
3 2014 4 1
3 2014 5 1
4 2013 12 1
4 2014 1 1
5 2013 9 1
5 2014 3 1
6 2013 12 1
6 2014 2 1
7 2014 2 1
8 2013 7 1
8 2013 8 1
Output
Id
1
2
3
4
8
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Make use of the fine SAS date functions by converting your time data to SAS date values, and use the lag() function:
data have;
input id year month flag;
cards;
1 2013 7 1
1 2013 8 1
1 2013 9 1
1 2013 10 1
1 2013 11 1
1 2013 12 1
1 2014 1 1
1 2014 2 1
1 2014 3 1
1 2014 4 1
1 2014 5 1
1 2014 6 1
2 2013 8 1
2 2013 9 1
2 2013 10 1
2 2013 11 1
2 2013 12 1
2 2014 1 1
3 2014 2 1
3 2014 3 1
3 2014 4 1
3 2014 5 1
4 2013 12 1
4 2014 1 1
5 2013 9 1
5 2014 3 1
6 2013 12 1
6 2014 2 1
7 2014 2 1
8 2013 7 1
8 2013 8 1
;
run;
/* first, let's create SAS date values from year and month */
data int;
set have;
if flag = 1; * just to be sure;
period = mdy(month,1,year);
keep id period;
run;
/* now check if two observations within a group are only 1 month apart */
data want (keep=id);
set int;
by id period; * period to make sure that dataset is correctly ordered;
if not first.id and intck('month',lag(period),period) = 1; * this is why we created SAS date values;
run;
/* remove duplicates */
proc sort data=want nodupkey;
by id;
run;
Note how I presented your data in a data step; this enables everyone reading the post to recreate the dataset with a simple copy/paste and submit, no tedious typing from screen or reformatting needed. And it makes sure that the dataset hase the intended structure.
Also note how code looks when using the correct posting sub-window ("little running man" or {i} buttons).
Edit: fixed typo ("data functions" -> "date functions" in first line)
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Something like this
data have;
input Id$ year month flag;
datalines;
1 2013 7 1
1 2013 8 1
1 2013 9 1
1 2013 10 1
1 2013 11 1
1 2013 12 1
1 2014 1 1
1 2014 2 1
1 2014 3 1
1 2014 4 1
1 2014 5 1
1 2014 6 1
2 2013 8 1
2 2013 9 1
2 2013 10 1
2 2013 11 1
2 2013 12 1
2 2014 1 1
3 2014 2 1
3 2014 3 1
3 2014 4 1
3 2014 5 1
4 2013 12 1
4 2014 1 1
5 2013 9 1
5 2014 3 1
6 2013 12 1
6 2014 2 1
7 2014 2 1
8 2013 7 1
8 2013 8 1
;
proc sort data=have;
by ID year month;
run;
data want(keep=ID);
set have;
by ID year month;
dif=dif(month);
if first.ID then outflag=0;
else if flag=1 and dif in(1,-11) and outflag=0 then outflag=1;
retain outflag 0;
if last.ID and outflag=1;
run;
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I think you also need to check for the difference in years when dif = -11, or you might stumble over a sequence
year month 2012 12 2014 1
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
That is true, the main challenge to me was to make sure those who had the flag between month 12 and 1 are included. I will take care of this note. Thank you so much
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
@tesfuc wrote:
That is true, the main challenge to me was to make sure those who had the flag between month 12 and 1 are included. I will take care of this note. Thank you so much
That is why I recommend to use SAS date/datetime/time values wherever possible. As soon as you have them, usage of intck/intnx takes care of all the quirks of the date and time measuring systems (irregular days/month, non-decimal fractional notations like 60sec/min or 12months/year)
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
@Kurt_Bremser ah yes, did not think of that 🙂 Right as usual.
@Abimal_Zippi thank you, but I recommend that you go with Kurts solution
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Make use of the fine SAS date functions by converting your time data to SAS date values, and use the lag() function:
data have;
input id year month flag;
cards;
1 2013 7 1
1 2013 8 1
1 2013 9 1
1 2013 10 1
1 2013 11 1
1 2013 12 1
1 2014 1 1
1 2014 2 1
1 2014 3 1
1 2014 4 1
1 2014 5 1
1 2014 6 1
2 2013 8 1
2 2013 9 1
2 2013 10 1
2 2013 11 1
2 2013 12 1
2 2014 1 1
3 2014 2 1
3 2014 3 1
3 2014 4 1
3 2014 5 1
4 2013 12 1
4 2014 1 1
5 2013 9 1
5 2014 3 1
6 2013 12 1
6 2014 2 1
7 2014 2 1
8 2013 7 1
8 2013 8 1
;
run;
/* first, let's create SAS date values from year and month */
data int;
set have;
if flag = 1; * just to be sure;
period = mdy(month,1,year);
keep id period;
run;
/* now check if two observations within a group are only 1 month apart */
data want (keep=id);
set int;
by id period; * period to make sure that dataset is correctly ordered;
if not first.id and intck('month',lag(period),period) = 1; * this is why we created SAS date values;
run;
/* remove duplicates */
proc sort data=want nodupkey;
by id;
run;
Note how I presented your data in a data step; this enables everyone reading the post to recreate the dataset with a simple copy/paste and submit, no tedious typing from screen or reformatting needed. And it makes sure that the dataset hase the intended structure.
Also note how code looks when using the correct posting sub-window ("little running man" or {i} buttons).
Edit: fixed typo ("data functions" -> "date functions" in first line)
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Dear KurtBremser
I would like to try a different scenario, i.e. if there is a flag for three or more consecutive months, what changes should I made in the code.
From the above dataset, the desired output will be -
Id
1
2
3
Because all have flag=1 for three or more consecutive months.
I really appreciate your help again.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
While the lag() functions keeps a 1-stage FIFO chain, the lag2() function keeps 2 stages. With it you can "look back" for 2 observations.
When you now retain a counter variable that is set to 1 at first.id and incremented otherwise, you can expand my condition to check for 2 observations back when counter > 2.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Thank you so much,
I just used this code but couldn't get with lag2() function...
data want_2 (keep=id);
set int;
by id period;
if first.id then cnt=0;
if not first.id and intck('month',lag(period),period)= 1 then cnt+1;
retain cnt 0;
if cnt ge 2;
run;
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Don't think so complicated:
data want (keep=id);
set int;
by id period;
retain counter;
if first.id
then counter = 1;
else counter + 1;
if counter > 2 and intck('month',lag(period),period) = 1 and intck('month',lag2(period),period) = 2;
run;
Call this
retain counter;
if first.id
then counter = 1;
else counter + 1;
a "counter block"; you will use if often when doing by-group processing in a data step.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I really appreciate your help. That is perfect.