topic Re: Booting in SAS Data Management

Booting

Rasheed — Mon, 06 Jul 2015 00:29:38 GMT

I have following data set from which I am running bootstrap samples

data bio;

infile datalines delimiter=',';

input pat seq period trt Y C;

logY=log(Y);

logC=log(C);

cards;

1,1,1,1,950.59,96.3

1,1,2,2,1637.71,194

1,1,3,2,2076.75,341

1,1,4,1,1485.93,316

2,1,1,1,774.44,87.6

2,1,2,2,585.89,56.2

2,1,3,2,801.26,89.1

2,1,4,1,773.51,84.6

38,2,1,2,2252.76,304

38,2,2,1,2262.88,255

38,2,3,1,1957.66,301

38,2,4,2,3084.05,685

;

Now I am using following cods for taking bootstrap sample from above data set and it works fine for a single bootstrap sample

%let anz=19;

proc datasets nolist;

delete out1;

run;

data _boot_;

do seq=1 to 2;

do pat=(seq-1)*&anz+1 to seq*&anz;

pat_old=(seq-1)*&anz+int(&anz*ranuni(0)+1);

output;

end;

run;

proc sql noprint;

create table out1 as

select _boot_.pat, bio.* from _boot_ left join bio(rename=(pat=pat_old))

on _boot_.pat_old=bio.pat_old

order by _boot_.pat, seq, pat_old, period;

quit;

Proc print data=out1;

run;

But when I want to generate multiple bootstrap sample so I have to run this code multiple times using do loop. The problem is that I don't understand where should I put do loop for multiple samples

pls guide me in this regard

Re: Booting

PGStats — Mon, 06 Jul 2015 02:18:54 GMT

Wouldn't you get the same result with SURVEYSELECT and running your subsequent analysis with BY processing? For example, three replicates of an ANCOVA:

proc surveyselect data=bio method=urs samprate=100 out=out1 outhits reps=3;

strata seq;

run;

proc sort data=out1; by Replicate seq; run;

proc glm data=out1;

by Replicate;

class seq;

model Y = seq C;

run;

Re: Booting

Reeza — Mon, 06 Jul 2015 03:37:03 GMT

You might find this paper useful:

http://www2.sas.com/proceedings/forum2007/183-2007.pdf

Re: Booting

Rasheed — Wed, 08 Jul 2015 06:58:56 GMT

Thank you very much for your reply this article is quite informative simulation through surveyselect.

Could you please mention the error in following code for booting

Data final;

do i = 1 to &rep ;

data _boot_&rep;

do seq=1 to 2;

do pat=(seq-1)*&anz+1 to seq*&anz;

pat_old=(seq-1)*&anz+int(&anz*ranuni(0)+1);

output;

end;

run;

output;

end;

stop;

I am getting error of un closed do block.

Please help

Re: Booting

Rasheed — Wed, 08 Jul 2015 07:00:17 GMT

%Let rep=2;

Data final;

do i = 1 to &rep ;

data _boot_&rep;

do seq=1 to 2;

do pat=(seq-1)*&anz+1 to seq*&anz;

pat_old=(seq-1)*&anz+int(&anz*ranuni(0)+1);

output;

end;

run;

output;

end;

stop;

I am getting error of un closed do block.

Please help

Re: Booting

Patrick — Wed, 08 Jul 2015 09:08:08 GMT

You can't use a "DATA" statement within another "DATA" statement. Does below amended code work for you? Make sure macro variables &rep and &anz get defined and populated before you try to execute the code.

Data final;

do i = 1 to &rep;

do seq=1 to 2;

do pat=(seq-1)*&anz+1 to seq*&anz;

pat_old=(seq-1)*&anz+int(&anz*ranuni(0)+1);

output;

end;

stop;

run;

Re: Booting

Rasheed — Thu, 09 Jul 2015 05:10:58 GMT

Thank you very much for your reply I run this code and its work fine but I did not get desired output yet

Actually initially I am running following program

data _boot_;

do seq=1 to 2;

do pat=(seq-1)*&anz+1 to seq*&anz;

pat_old=(seq-1)*&anz+int(&anz*ranuni(0)+1);

output;

end; end; run;

proc sql noprint;

create table out1 as

select _boot_.pat, bio.* from _boot_ left join bio(rename=(pat=pat_old))

on _boot_.pat_old=bio.pat_old

order by _boot_.pat, seq, pat_old, period;

quit;

Proc print data=out1;

run;

----------------------------------------

And I am getting following output

Obs pat pat_old seq period trt Y C logY logC

1	1	19	1	1	1	1845.61	235.0	7.52057	5.45959
2	1	19	1	2	2	2560.92	356.0	7.84812	5.87493
3	1	19	1	3	2	3088.04	248.0	8.03529	5.51343
4	1	19	1	4	1	2180.39	145.0	7.68726	4.97673
5	2	12	1	1	1	736.50	87.4	6.60191	4.47050
6	2	12	1	2	2	947.58	124.0	6.85391	4.82028
7	2	12	1	3	2	1426.96	151.0	7.26330	5.01728
8	2	12	1	4	1	681.66	75.5	6.52453	4.32413
9	3	5	1	1	1	2291.93	204.0	7.73715	5.31812
10	3	5	1	2	2	1223.74	126.0	7.10967	4.83628
11	3	5	1	3	2	1949.10	202.0	7.57512	5.30827
12	3	5	1	4	1	3184.15	365.0	8.06594	5.89990
13	4	6	1	1	1	1044.18	91.1	6.95099	4.51196
14	4	6	1	2	2	1023.00	111.0	6.93049	4.70953
15	4	6	1	3	2	1178.20	196.0	7.07174	5.27811
16	4	6	1	4	1	1155.25	104.0	7.05207	4.64439

.......................................................................

........................................................................

149	38	37	2	1	2	1061.92	97.4	6.96783	4.57883
150	38	37	2	2	1	987.86	94.1	6.89554	4.54436
151	38	37	2	3	1	1422.71	186.0	7.26032	5.22575
152	38	37	2	4	2	1220.58	103.0	7.10708	4.63473

Now I want this table multiple times say 3 times so what should I add in above code pls guide

thanks in advance

Regards

Re: Booting

Reeza — Thu, 09 Jul 2015 15:48:28 GMT

How do you want multiple tables? If the exact same table repeated:

data want;

set have have have;

run;

3 identical copies.

Or if you need to identify records:

data want;

set have(in=a) have(in=b) have(in=c);

record=a*1+b*2+3*c;

run;

Re: Booting

Rasheed — Thu, 09 Jul 2015 20:59:43 GMT

thank for reply actually i do not want identical copies

the above generated table is first bootstrap sample so I want multiple boot strap samples I think I have to include do loop in my code without macro

Could you please guide me

Re: Booting

Reeza — Thu, 09 Jul 2015 21:04:12 GMT

I honestly don't understand what you want. If you want more samples, why not just change the loops in your data step?

data _boot_;

do sample=1 to 10;

do seq=1 to 2;

do pat=(seq-1)*&anz+1 to seq*&anz;

pat_old=(seq-1)*&anz+int(&anz*ranuni(0)+1);

output;

end;

run;

Here's a tutorial on macro's from UCLA that might help you get started.

Statistical Computing Seminar: Introduction to SAS Macro Language

Re: Booting

PGStats — Thu, 09 Jul 2015 22:21:20 GMT

Hi Rasheed,

Have you tried the solution above based on SURVEYSELECT. I have used it for my own work and found it simpler and faster than alternatives based on datasteps and macro processing.