DATA Step, Macro, Functions and more

importing macro for analysis

Accepted Solution Solved
Reply
Super Contributor
Posts: 297
Accepted Solution

importing macro for analysis

Hello I wish to call the following macro and run the analysis below. The macro runs well but the last part to simulate a data and runs does not work. Any help?

The problem is from the data test (marked bold below) , no problem with the macro. The error is at the tail end

%macro indirect(data=,y=,x=,m=,c=0,boot=1000,conf=95,percent=0,bc=1,bca=0, normal=0,contrast=0);

proc iml;

use &data;

read all var{&y &x &m} into dd;

nm={&y &x &m};

xx=(dd = .);xx=xx[,+];

j=1;do i = 1 to nrow(dd);if xx[i,1]=0 then;do;dd[j,]=dd[i,];j=j+1;end;end;

dd=dd[1:j-1,];

nm = nm`;

n = nrow(dd);

nv = ncol(dd);

nc = &c;

con=j(n,1,1);

dt2 = dd;

dt = dd;

resid = j(n,(nv-nc),0);

info = j((2*(nv-nc-2)+1),(2*(nv-nc-2)+1),0);

imat = j(ncol(info),4,1);

imat[1Smiley Sadnv-nc-2),1]=(2Smiley Sadnv-nc-1))`;

imat[1Smiley Sadnv-nc-2),3]=(2Smiley Sadnv-nc-1))`;

imat[(nv-nc-1)Smiley Sadncol(info)-1),2]=(2Smiley Sadnv-nc-1))`;

imat[(nv-nc-1)Smiley Sadncol(info)-1),4]=(2Smiley Sadnv-nc-1))`;

imat[(nv-nc-1)Smiley Sadncol(info)-1),1]=j((nv-nc-2),1,(nv-nc));

imat[(nv-nc-1)Smiley Sadncol(info)-1),3]=j((nv-nc-2),1,(nv-nc));

imat[ncol(info),1:4]={1 1 1 1};

imat[ncol(info),1]=nv-nc;

imat[ncol(info),3]=nv-nc;

bzx = j(nv-2-nc,1,0);

bzxse = j(nv-2-nc,1,0);

b=j((nv-1-nc),(nv-1-nc),0);

cname={"C1", "C2", "C3", "C4", "C5", "C6", "C7", "C8", "C9", "C10", "C11", "C12", "C13", "C14", "C15", "C16"};

cname=cname//{"C17", "C18", "C19", "C20", "C21", "C22", "C23", "C24", "C25", "C26", "C27", "C28", "C29"};

cname=cname//{"C30", "C31", "C32", "C33", "C34", "C35", "C36", "C37", "C38", "C39", "C40", "C41", "C42"};

cname=cname//{"C43", "C44", "C45"};

p0 = -0.322232431088;

p1 = -1;

p2 = -0.342242088547;

p3 = -0.0204231210245;

p4 = -.0000453642210148;

q0 = 0.0993484626060;

q1 = 0.588581570495;

q2 = 0.531103462366;

q3 = 0.103537752850;

q4 = 0.0038560700634;

conf=round(&conf);

lowalp = 0.5*(1-(conf/100));

upalp = 0.5*(1+(conf/100));

zbca = lowalp//upalp;

btn = 1;

if (&boot > 999) then;

  do;

  btn = floor(&boot/1000)*1000;

  end;

blowp = floor(lowalp*btn);

if (blowp < 1) then;

  do;

  blowp = 1;

  end;

bhighp = floor((upalp*btn)+1);

if (bhighp > btn) then;

  do;

  bhighp = btn;

  end;

indeff = j((n+1+btn),(nv-1-nc),0);

do d = 1 to (n+1+btn);

  if (d = (n+2)) then;

    do;

  dt = dt2;

  con = j(n,1,1);

  end;

  if (d > 1) then if (d < (n+2)) then;

    do;

    if (d = 2) then;

   do;

   con = j((n-1),1,1);

   dt = dt2[2:n,];

   end;

  if (d = (n+1)) then;

   do;

      dt = dt2[1Smiley Sadn-1),];

   end;

    if (d > 2) then if (d < (n+1)) then;

   do;

      dt = dt2[1Smiley Sadd-2),]//dt2[(d:n),];

      end;

  end;

  if (d > (n+1)) then;

    do;

   do nn = 1 to n;

      v = int(ranuni(0)*n)+1;

   dt[nn,1:nv]=dt2[v,1:nv];

   end;

  end;

x = dt[,2];

m = dt[,3Smiley Sadnv-nc)];

y = dt[,1];

xz = dt[,2:nv];

xo = con||x;

if (nc > 0) then;

   do;

   c = dt[,(nv-nc+1):nv];

   xo = xo||c;

   end;

do k = 3 to (nv-nc);

  ytmp = dt[,k];

  bzxt = inv(xo`*xo)*xo`*ytmp;

  bzx[(k-2),1]=bzxt[2,1];

  if (d = 1) then;

    do;

  resid[,(k-1)]=ytmp-(xo*bzxt);

  mse = sum((ytmp-(xo*bzxt))##2)/(n-2-nc);

  olscm = (mse*inv(xo`*xo));

  bzxse[(k-2),1]=sqrt(olscm[2,2]);

  end;

end;

if (d = 1) then;

  do;

  if (nc > 0) then;

    do;

    cnt = dd[,(nv-(nc-1)):nv];

    xo = con||x||cnt;

  end;

  if (nc = 0) then;

    do;

    xo = con||x;

  end;

  byx = inv(xo`*xo)*xo`*y;

  mse = sum((y-(xo*byx))##2)/(n-2-nc);

  olscm = (mse*inv(xo`*xo));

  byxse = sqrt(olscm[2,2]);

  byx = byx[2,1];

  end;

xzo = con||xz;

byzx = inv(xzo`*xzo)*xzo`*y;

byzx2 = byzx[3Smiley Sadnv-nc),1];

if (d = 1) then;

  do;

  resid[,ncol(resid)]=y-(xzo*byzx);

  mse = sum((y-(xzo*byzx))##2)/(n-nv);

  covmat = mse*inv(xzo`*xzo);

  olscm = vecdiag(covmat);

  sse = mse*(n-nv);

  sst = sum((y-(sum(y)/n))##2);

  r2 = 1-(sse/sst);

  ar2 = 1-(mse/(sst/(n-1)));

  fr = ((n-nv)*r2)/((1-r2)*ncol(xz));

  pfr = 1-probf(fr,ncol(xz),(n-nv));

  if (nc > 0) then;

    do;

    bcon = byzx[(nv-nc+1):nv,1];

  bconse = sqrt(olscm[(nv-nc+1):nv,1]);

    end;

  byzx2se = sqrt(olscm[3Smiley Sadnv-nc),1]);

  cprime = byzx[2,1];

  cprimese=sqrt(olscm[2,1]);

end;

indeff2 = (bzx#byzx2);

zs = (bzx/bzxse)#(byzx2/byzx2se);

temp = t(sum(indeff2)//indeff2);

indeff[d,]=temp;

if (d = 1) then;

  do;

  vs = nm[1Smiley Sadnv-nc),1];

  rn = {"DV = " "IV = " "MEDS = "};

  print "Dependent, Independent, and Proposed Mediator Variables";

  print vs [rowname = rn];

  if (nc > 0) then;

    do;

  vs = nm[(nv-nc+1):nv,1];

  print "Statistical Controls";

  rn = {"CONTROLS="};

  print vs [rowname = rn];

  end;

  print "Sample size";

  print n;

  nms = nm[3Smiley Sadnv-nc),1];

  te = bzx/bzxse;

  df = n-2-nc;

  p = 2*(1-probt(abs(te),df));

  bzxmat = bzx||bzxse||te||p;

  b[2Smiley Sadnv-1-nc),1]=bzx;

  se2 = bzxse#bzxse;

  cnm = {"Coeff" "se" "t" "p"};

  print "IV to Mediators (a paths)";

  print bzxmat [rowname = nms colname = cnm format = 9.4];

  te = byzx2/byzx2se;

  df = n-nv;

  p = 2*(1-probt(abs(te),df));

  byzx2mat=byzx2||byzx2se||te||p;

  print "Direct Effects of Mediators on DV (b paths)";

  print byzx2mat [rowname = nms colname = cnm format = 9.4];

  te=byx/byxse;

  df = n-2-nc;

  p=2*(1-probt(abs(te),df));

  byxmat = byx||byxse||te||p;

  xnm=nm[2,1];

  print "Total effect of IV on DV (c path)";

  print byxmat [rowname = xnm colname = cnm format = 9.4];

  te=cprime/cprimese;

  df = n-nv;

  p=2*(1-probt(abs(te),df));

  cprimmat = cprime||cprimese||te||p;

  print "Direct Effect of IV on DV (c' path)";

  print cprimmat [rowname = xnm colname = cnm format = 9.4];

  if (nc > 0) then;

    do;

  df = n-nv;

  nms = nm[(nv-nc+1):nv,1];

  te=bcon/bconse;

  p=2*(1-probt(abs(te),df));

  bconmat = bcon||bconse||te||p;

  print "Partial Effect of Control Variables on DV";

  print bconmat [rowname = nms colname = cnm format = 9.4];

  end;

  dvms=r2||ar2||fr||ncol(xz)||(n-nv)||pfr;

  print "Fit Statistics for DV Model";

  cnm = {"R-sq" "adj R-sq" "F" "df1" "df2" "p"};

  print dvms [colname = cnm format = 9.4];

  if (&normal =^ 0) then;if(&c = 0) then;do;

    do;

      bmat=j((nv-nc),(nv-nc),0);

      bmat[2Smiley Sadnv-nc-1),1]=bzx;

   bmat[(nv-nc),2Smiley Sadnv-nc-1)]=byzx2`;

      bmat[(nv-nc),1]=cprime;

      imbinv = inv(i(ncol(bmat))-bmat);

   imbtinv = inv(i(ncol(bmat))-bmat`);

      resid[,1]=x-(x[+]/n);

   psi = (resid`*resid)/(n-1);

   invpsi = inv(psi);

   ibpsiib = imbinv*psi*imbtinv;

   do ic = 1 to ncol(info);

     do ic2 = 1 to ncol(info);

   info[ic,ic2]=(n-1)*((imbinv[imat[ic2,4],imat[ic,1]]*imbinv[imat[ic,2],imat[ic2,3]])+(ibpsiib[imat[ic2,4],imat[ic,2]]*invpsi[imat[ic,1],imat[ic2,3]]));

        end;

   end;

   varcov=inv(info);

   varcov=varcov[1Smiley Sad2*(nv-nc-2)),1Smiley Sad2*(nv-nc-2))];

   ses = vecdiag(varcov);

   avar = ses[1:nrow(bzxse),1];

   bvar = ses[(nrow(bzxse)+1):nrow(ses),1];

   if ((nv-nc-2) > 1) then;do;if(&contrast = 1) then;do;

     prws=j(((nv-nc-2)*(nv-nc-3)/2),1,0);

  prwse=prws;

  kk=1;

  do ic = 1 to (nv-nc-3);

   do ic2 = (ic+1) to (nv-nc-2);

     vf2=((byzx2[ic,1]##2)*varcov[ic,ic])-(2*byzx2[ic,1]*byzx2[ic2,1]*(varcov[ic,ic2]));

  vf2=vf2+((byzx2[ic2,1]##2)*varcov[ic2,ic2])+((bzx[ic,1]##2)*(bvar[ic,1]));

  vf2=vf2-(2*bzx[ic,1]*bzx[ic2,1]*covmat[(2+ic),(2+ic2)])+((bzx[ic2,1]##2)*(bvar[ic2,1]));

  cnt=indeff2[ic,1]-indeff2[ic2,1];

  prws[kk,1]=cnt;

  prwse[kk,1]=sqrt(vf2);

  kk=kk+1;

   end;

     end;

   cnam2=cname[1Smiley Sadkk-1),1];

   end;end;

   dermat=byzx2//bzx;

   totse=sqrt((dermat)`*varcov*dermat);

   specse = sqrt((byzx2#byzx2)#(avar)+(bzx#bzx)#(bvar));

   specse = totse//specse;

   indsum=indeff2[+];

      specz = (indsum//indeff2)/specse;

   ind22 = indsum//indeff2;

   nms = {"TOTAL"}//nm[3Smiley Sadnv-nc),1];

   if ((nv-nc-2) > 1) then;do;if(&contrast = 1) then;do;

     ind22 = ind22//prws;

  specse = specse//prwse;

  specz2 = prws/prwse;

  specz = specz//specz2;

  nms = nms//cnam2;

  end;end;

      pspec = 2*(1-probnorm(abs(specz)));

      spec = ind22||specse||specz||pspec;

      cnm = {"Effect" "se" "Z" "p"};

      print "*****************************************************";

      print "NORMAL THEORY TESTS FOR INDIRECT EFFECTS";

      print "Indirect Effects of IV on DV through Mediators (ab paths)";

      print spec [rowname = nms colname = cnm format = 9.4];

    end;end;

  end;

end;

if (btn > 1) then;do;

  nms = {"TOTAL"}//nm[3Smiley Sadnv-nc),1];

  if (nv-nc-2) > 1 then do;if (&contrast = 1) then do;

     crst=j((n+1+btn),((nv-nc-2)*(nv-nc-3)/2),0);

  kk=1;

  do ic = 2 to (nv-nc-2);

    do ic2 = (ic+1) to (nv-nc-1);

      crst[,kk]=indeff[,ic]-indeff[,ic2];

  kk=kk+1;

    end;

  end;

     indeff = indeff||crst;

  cnam2=cname[1Smiley Sadkk-1),1];

  nms = nms//cnam2;

  end;

  end;

lvout = indeff[2Smiley Sadn+1),];

tdotm = lvout[+,]/n;

tm = j(n,ncol(lvout),1)*diag(tdotm);

topa=(((n-1)/n)*(tm-lvout))##3;

topa=topa[+,];

bota =((((n-1)/n)*(tm-lvout))##2);

bota=bota[+,];

bota=6*sqrt(bota##3);

ahat = topa/bota;

indsam = indeff[1,]`;

boot = indeff[(n+2):nrow(indeff),];

mnboot = (boot[+,]/btn)`;

xt=boot-j(btn,1)*boot[:,];

cv=(xt`*xt)/btn;

se=sqrt(vecdiag(cv));

  create bootstp from boot [colname='indirect'];

  append from boot;

  nnn = j(1,ncol(indeff),-999);

  boot = nnn//boot;

  do e = 1 to (ncol(indeff));

  do i = 2 to (btn+1);

    ix = boot[i,e];

  do k = i to 2 by -1;

   k2 = k;

   if (boot[(k-1),e] > ix) then;

     do;

  boot[k,e]=boot[(k-1),e];

  end;

   else;

   if (boot[(k-1),e] <= ix) then;

     do;

  goto stpit;

  end;

  end;

  stpit:

  boot[k2,e]=ix;

    end;

    end;

  boot = boot[2Smiley Sadbtn+1),];

xp=j((nrow(mnboot)+2),1,0);

do i = 1 to (nrow(mnboot)+2);

  if (i <= nrow(mnboot)) then;

    do;

    pv = (boot[,i] < indsam[i,1]);

  pv = pv[+,]/btn;

    end;

  else;

    pv = zbca[(i-nrow(mnboot)),1];

  p=pv;

  if (pv > 0.5) then;

    do;

  p = 1-pv;

  end;

  y5 = sqrt(-2*log(p));

  xp[i,1]=y5+((((y5*p4+p3)*y5+p2)*y5+p1)*y5+p0)/((((y5*q4+q3)*y5+q2)*y5+q1)*y5+q0);

  if (pv <= 0.5) then;

    do;

  xp[i,1]=-xp[i,1];

  end;

end;

bbb = nrow(mnboot);

zz = xp[1:bbb,1];

zlo = zz + ((zz+xp[(bbb+1),1])/(1-ahat`#(zz+xp[(bbb+1),1])));

zup = zz + ((zz+xp[(bbb+2),1])/(1-ahat`#(zz+xp[(bbb+2),1])));

ahat = 0;

zlobc = zz + ((zz+xp[(bbb+1),1])/(1-ahat`#(zz+xp[(bbb+1),1])));

zupbc = zz + ((zz+xp[(bbb+2),1])/(1-ahat`#(zz+xp[(bbb+2),1])));

zlo = probnorm(zlo);

zup = probnorm(zup);

zlobc = probnorm(zlobc);

zupbc = probnorm(zupbc);

blow = int(zlo*(btn+1));

bhigh = int(zup*(btn+1))+1;

blowbc = int(zlobc*(btn+1));

bhighbc = int(zupbc*(btn+1))+1;

lowbca = j(nrow(blow),1,0);

upbca = lowbca;

do i = 1 to nrow(blow);

  if (blow[i,1] < 1) then;

    do;

  blow[i,1]=1;

  end;

  lowbca[i,1]=boot[blow[i,1],i];

  if (bhigh[i,1] > btn) then;

    do;

  bhigh[i,1]=btn;

  end;

  upbca[i,1]=boot[bhigh[i,1],i];

end;

lowbc = j(nrow(blow),1,0);

upbc = lowbca;

do i = 1 to nrow(blowbc);

  if (blowbc[i,1] < 1) then;

    do;

  blowbc[i,1]=1;

  end;

  lowbc[i,1]=boot[blowbc[i,1],i];

  if (bhighbc[i,1] > btn) then;

    do;

  bhighbc[i,1]=btn;

  end;

  upbc[i,1]=boot[bhighbc[i,1],i];

end;

print "*****************************************************";

print "BOOTSTRAP RESULTS FOR INDIRECT EFFECTS";

res = indsam||mnboot||(mnboot-indsam)||se;

cn = {"Data" "Boot" "Bias" "SE"};

print "Indirect Effects of IV on DV through Mediators (ab paths)";

print res [rowname = nms colname = cn format = 9.4];

lowperc = boot[blowp,];

upperc = boot[bhighp,];

ci = lowbca||upbca;

cn = {"Lower" "Upper"};

if (&bca ^= 0) then;

  do;

  print "Bias Corrected and Accelerated Confidence Intervals";

  print ci [rowname = nms colname = cn format = 9.4];

  end;

if (&bc ^= 0) then;

  do;

  ci = lowbc||upbc;

  print "Bias Corrected Confidence Intervals";

  print ci [rowname = nms colname = cn format = 9.4];

  end;

if (&percent ^= 0) then;

  do;

  ci = lowperc`||upperc`;

  print "Percentile Confidence Intervals";

  print ci [rowname = nms colname = cn format = 9.4];

  end;

print "*****************************************************";

print "Level of Confidence for Confidence Intervals";

print conf;

print "Number of Bootstrap Resamples";

print btn;

end;

if (&normal = 1) then;do;if (&c = 0) then;do;prt =1;end;end;

if (btn > 999) then;do;prt = 1;end;

if ((nv-nc-2) > 1) then;do;if (&contrast = 1) then;do;

if (prt = 1) then;do;

print "*****************************************************";

print "Indirect Effect Contrast Definitions: IndEff_1 minus IndEff2";

kk=1;

prwsv = j(((nv-nc-2)*(nv-nc-3)/2),2,"XXXXXXXX");

do ic = 1 to (nv-nc-3);

  do ic2 = (ic+1) to (nv-nc-2);

    prwsv[kk,1]=nm[ic+2,1];

  prwsv[kk,2]=nm[ic2+2,1];

  kk=kk+1;

  end;

end;

prwsv = cnam2||prwsv;

cn = {"Contrast" "IndEff_1" "IndEff_2"};

print prwsv [colname = cn];

end;end;end;

quit;

%mend;

/* Macro ends*/;

/*Call macro into data set*/;

data test;

seed=-1;

do i = 1 to 100;

Y=10+(.5)*rannor(1);

  X=2+(0.3)*rannor(1);

  /*** These are the mediators *****/;

  M1=1+(0.3)*rannor(1);

  M2=3+(0.4)*rannor(1);

  M2=5+(0.8)*rannor(1);

  M4=0+(.9)*rannor(1);

  /***** These are the covariates *******/;

  gender=ranbin(0,1,.5);

  age=60+(5)*rannor(1);

  edu=12+(1)*rannor(1);

    output;

  end;

run;

/*So i want to call the macro indirect and fit this model, can any one help with the last step*/

%INDIRECT [DATA = test, Y = yvar, X = xvar, M = M1 M2 M3 M4{gender age edu}];    /* Note M=Mvlist are M1, M2, M3 and M4 and covlist are age gender and edu */

{,C = {cov}(0**)}

{,BOOT = {z}(1000**)}

{,CONF = {ci}(95**)}

{,NORMAL = {t}(0**)}

{,CONTRAST = {n}(0**)}

{,PERCENT = {p}(0**)}

{,BC = {b}(1**)}

{,BCA = {d}(0)};

ERROR:

ERROR

982

983  /*So i want to call the macro indirect and fit this model, can any one help with the last

983! step*/

984  %INDIRECT [DATA = test, Y = y, X = x, M = M1 M2 M3 M4{gender age edu}];    /* Note M=Mvlist

984! are M1, M2, M3 and M4 and covlist are age gender and edu */

NOTE: IML Ready

NOTE: Line generated by the invoked macro "INDIRECT".

1                 use &data;

                            -


Accepted Solutions
Solution
‎02-05-2015 01:21 PM
Super Contributor
Posts: 297

Re: importing macro for analysis

Works now

%INDIRECT (DATA = test, Y = Y, X = X, M = M1 M2 M3 M4 gender age edu, C=3, CONTRAST = 1, NORMAL

= 1, BOOT = 500);

View solution in original post


All Replies
Super User
Posts: 10,483

Re: importing macro for analysis

From your other thread on this topic:

%INDIRECT [DATA = test, Y = y, X = x, M = M1 M2 M3 M4{gender age edu}];

generates an error in the READ ALL because it resolves to

read all var{y x  M1 M2 M3 M4{gender age edu}} into dd;

and the extra {} inside the var{ } are not acceptable syntax.

Also, did you start with a program that actually produced the desired output BEFORE writing this macro?

Solution
‎02-05-2015 01:21 PM
Super Contributor
Posts: 297

Re: importing macro for analysis

Works now

%INDIRECT (DATA = test, Y = Y, X = X, M = M1 M2 M3 M4 gender age edu, C=3, CONTRAST = 1, NORMAL

= 1, BOOT = 500);

☑ This topic is SOLVED.

Need further help from the community? Please ask a new question.

Discussion stats
  • 2 replies
  • 266 views
  • 3 likes
  • 2 in conversation