Testing Difference in Mean Between Two Specific Periods - 'MULTIPLE TIMES'.

[arii]

Hi, 

I need to test the difference in mean between two time periods in my data set. However, I need to carry out this test multiple times.

Precisely, for each 'date' in my data; I need to test the difference between the mean of 'X1' for '30 days before' and '30 days after'.

The code below works perfectly, but I need to manually specify each date (which is obviously too cumbersome). I would sincerely appreciate some suggestions!

Present Code:

/*date of interest: 16May2001*/

data xx;
set x;

Startdate1 = '16May2001'd - 30;
Enddate1 = '16May2001'd;

Startdate2 = '17May2001'd;
Enddate2 = '17May2001'd + 30;

If Startdate1 < date < Enddate1 then flag=0;
If Startdate2 < date < Enddate2 then flag=1;

run;

proc ttest data=xx;

class flag;

var X1;
run;

Sample Data:

(Date Format: mmddyear)

Date	X1
1/2/2001	22
5/16/2001	34
5/26/2001	65
5/27/2001	11
6/8/2001	56
10/11/2001	31
12/12/2001	91
1/13/2004	49
7/18/2005	56
9/19/2006	21
12/2/2007	67
12/21/2007	34
12/29/2007	67
1/23/2008	90
7/14/2008	78
9/25/2008	45
6/28/2009	45
1/29/2010	54

Thank you in anticipation!

[arii]

Hi, 

I need to test the difference in mean between two time periods in my data set. However, I need to carry out this test multiple times.

Precisely, for each 'date' in my data; I need to test the difference between the mean of 'X1' for '30 days before' and '30 days after'.

The code below works perfectly, but I need to manually specify each date (which is obviously too cumbersome). I would sincerely appreciate some suggestions!

Present Code:

/*date of interest: 16May2001*/

data xx;
set x;

Startdate1 = '16May2001'd - 30;
Enddate1 = '16May2001'd;

Startdate2 = '17May2001'd;
Enddate2 = '17May2001'd + 30;

If Startdate1 < date < Enddate1 then flag=0;
If Startdate2 < date < Enddate2 then flag=1;

run;

proc ttest data=xx;

class flag;

var X1;
run;

Sample Data:

(Date Format: mmddyear)

Date	X1
1/2/2001	22
5/16/2001	34
5/26/2001	65
5/27/2001	11
6/8/2001	56
10/11/2001	31
12/12/2001	91
1/13/2004	49
7/18/2005	56
9/19/2006	21
12/2/2007	67
12/21/2007	34
12/29/2007	67
1/23/2008	90
7/14/2008	78
9/25/2008	45
6/28/2009	45
1/29/2010	54

Thank you in anticipation!

[PGStats]

Here is one approach using a self join:

data have;
input Date :mmddyy. X1; 
format date yymmdd10.;
datalines;
1/2/2001 22 
5/16/2001 34 
5/26/2001 65 
5/27/2001 11 
6/8/2001 56 
10/11/2001 31 
12/12/2001 91 
1/13/2004 49 
7/18/2005 56 
9/19/2006 21 
12/2/2007 67 
12/21/2007 34 
12/29/2007 67 
1/23/2008 90 
7/14/2008 78 
9/25/2008 45 
6/28/2009 45 
1/29/2010 54 
;

proc sql;
create table test as
select
    a.date as middate,
    b.date > a.date as period,
    b.date,
    b.x1
from 
    have as a inner join
    have as b 
      on b.date between intnx('day',a.date,-30) and intnx('day',a.date,30) and
      a.date ne b.date
group by middate
having count(distinct period) = 2
order by middate, date;
quit;

ods select none;

proc ttest data=test plots=none;
ods output ttests=tt;
by middate;
class period;
var x1;
run;

ods select all;

proc print data=tt; run;

[arii]

Thank you for your response, PG Stats.

Is there a way to do this without the 'datalines' in the code? My actual data has close to 500 observations and the 'datalines approach' would be kinda clumsy for me.

[PGStats]

Of course, there are many ways. The first step in the code above is just there for testing. Create your have dataset with a datastep using infile or a SQL query or any other means. As long as it contains the date and x1 variables.

[arii]

Hi PGStats,

Unfortunately, I am still struggling with the codes you supplied (please pardon my basic SAS knowledge).

I am using the data-set approach because I have already created a data-set from my main data file. I have over a thousand observations, I don't think the 'datalines approach' would work well for me.

However, I still have lots of errors in the log file. I am not too sure what is wrong.

See the modified code I used below:

data xxx;
set xx;

input Date :mmddyy. X1;
format date yymmdd10.;

proc sql;
create table test as
select
a.date as middate,
b.date > a.date as period,
b.date,
b.x1
from
have as a inner join
have as b
on b.date between intnx('day',a.date,-30) and intnx('day',a.date,30) and
a.date ne b.date
group by middate
having count(distinct period) = 2
order by middate, date;
quit;

ods select none;

proc ttest data=test plots=none;
ods output ttests=tt;
by middate;
class period;
var x1;
run;

ods select all;

proc print data=tt; run;

See some errors from log file:

ERROR: No DATALINES or INFILE statement.
NOTE: The SAS System stopped processing this step because of errors.

ERROR: Variable MIDDATE not found.

ERROR: File WORK.TT.DATA does not exist

[PGStats]

The first thing you must do is to bring your data into a SAS dataset. That is often the first difficulty that new users encounter. Where is your main data file?

[arii]

I have attached the main data file herein.

[arii]

I doubt I had issues with the data import. My main challenge was with conducting the 'proc t test' multiple times - using the dates in my data file as a reference.

I have attached my full codes herein.

Thank you for your help, PGStats.