Duration analysis with longitudinal data

[UNBRDC]

Hello,

I am a new user of SAS so that my question may be very easy but unfortunately I do not know the aswer. I will appreciate for any of your suggestions.

Thank you very much for your time and Merry Christmas!

The following is the description of the dataset and the results that I would like to find:

Data description: longitudinal data, edu is a dummy which 1 represents to have education in the corresponding tax year; 0 not.

I would like to get (1) the duration of the waiting period for a person to go to education, and the number of years s/he participated in education. (2) Also, the mean of the waiting time and the number of years in edu for all people.

For (1), I would like to have a variable to show the first person wait for 1 year to start education, and stay in school for 2 years; the 2nd person waits 0 year and stay in school for 3 years, 4th wait for 2 years and in school for 2 years. For person 3, he has multiple entry of education but I would like to count only the first time and treat the 2nd or any furhter edu as no edu in the period.

For (2), I would like to think it is good if the mean of waiting time and years of edu can be calculated directly.

Sample data:

id            tax_year              edu                       

1              1                              0                             

1              2                              1                             

1              3                              1                             

1              4                              0             

1              5                              0

2              1                              1             

2              2                              1             

2              3                              1             

2              4                              0             

2              5                              0

3              1                              0             

3              2                              1             

3              3                              0             

3              4                              1             

3              5                              0

4              1                              0             

4              2                              0             

4              3                              1             

4              4                              1

5              5                              0             

[Astounding]

Calculating means is easy.  Preparing the numbers is a bit trickier.  Assuming your data is sorted as indicated, BY ID TAX_YEAR:

data want;

set have;

by id;

if first.id then do;

   waiting_increment=1;

   education_increment=1;

   waiting_years=0;

   education_years=0;

end;

retain waiting_increment education_increment;

if edu=0 then do;

   if education_years > 0 then education_increment=0;

   waiting_years + waiting_increment;

end;

else do;

   waiting_increment=0;

   education_years + education_increment;

end;

if last.id;

run;

Then run PROC MEANS to get the means:

proc means data=want;

var waiting_years education_years;

run;

It's untested ... should be fine by write back if there are any difficulties.

[UNBRDC]

Thanks. Astounding.

I tested the code and it works. I am wondering whether I can ask you the logic behind the code. Please forgive me if my question is too basic.

I did a few tests and try to find the logic when SAS runs this code. I think SAS handles it as kind of loop. The reason that I think so is that: before running the "else do....end" part, the code generates the total of all non-edu period, or the counts for all edu = 0. However, when the else do...end added, the periods after an edu = 1 have been deducted from the previous totals. This generates the actual waiting time. The reason I think it is a loop is from here-- the Else do changed the condition of calculating total and recalculated it. The conclusion also comes from the education_years, since when SAS run the “If edu_years >0 then …” , all edu_years are equal to zeros.

I am wondering whether you can point out my understanding is correct or not.

Thank you very much for your time.

[ndp]

Try this: edu2 and edu3 datasets will have appropriate variables you need. You can then use any of the procs univariate/means to count means.

data edu2(keep=id edu1_yr) edu3(keep=id tot_edu);
	set edu;
	by id tax_year;
	where edu=1;
	retain tot_edu;
	new_yr=lag(tax_year);
	if first.id then new_yr=tax_year;
	if edu=1 and tax_year-new_yr>1 then new_edu=0;
	else new_edu=edu;
	if first.id then do;
                   edu1_yr=tax_year-1;
                   output edu2;
                end;
	if first.id then tot_edu=0;
	tot_edu=tot_edu+new_edu;
	if last.id then output edu3;
run;