SUBROUTINE CPD
      PARAMETER(LDM=15,LDPAR=250)
      IMPLICIT REAL*8 (A-H,O-Z)
      COMMON/AMCMC/WT,PRI,PDAT,PAR(LDPAR),ITER,NPAR,LRECRD,LWRITE,LERR
      LOGICAL LRECRD,LWRITE,LERR
      DIMENSION AN(LDM,LDM),AT(LDM,LDM),A1T(LDM,LDM),
     1 P(LDM,LDM),PT(LDM,LDM),
     2 S(LDM,LDM),ST(LDM,LDM)
C  Draw P matrix
      DO 10 I=1,NM
   10 CALL BET(NM,AT(1,I),PT(1,I))
      CALL DTRNRR(NM,NM,PT,LDM,NM,NM,P,LDM)
C  Store section
      IF(.NOT.(LRECRD.AND.LWRITE))RETURN
      CALL VEC(NM,NM,P,LDM,PAR)
      WT=0.0D0
      PRI=PRI0
      PDAT=0.0D0
      DO 20 I=1,NM
      PRI=PRI+BETK(NM,ST(1,I),PT(1,I))
   20 PDAT=PDAT+BETK(NM,A1T(1,I),PT(1,I))
      RETURN
      ENTRY PD0
C  Obtain Markov chain model specification
      CALL FOPEN('Model specification',1,.FALSE.)
      READ(1,*,END=210,ERR=220)NM
      IF((NM.LE.0).OR.(NM.GT.LDM))GO TO 230
C  Obtain data
      CALL DATFMC(2,1,NM,AN,LDM)
C  Obtain prior
      CALL BET0(1,NM,'first order Markov chain model',
     1  NM,S,LDM,PRI0,P,LDN)
C  Obtain sufficient statistics and log Bayes factor
      AA=PRI0
      DO 120 I=1,NM
      DO 110 J=1,NM
      AT(J,I)=S(I,J)+AN(I,J)
      A1T(J,I)=AT(J,I)+1.0D0
  110 ST(J,I)=ST(I,J)
  120 AA=AA-BETN(NM,A1T(1,I))
      WRITE(6,130)AA
  130 FORMAT(/,'  Log marginalized likelihood:',1PE15.7)
      IF(.NOT.LWRITE)RETURN
      NPAR=NM**2
      IF(NPAR.GT.LDPAR)GO TO 240
      RETURN
      ENTRY PD1
      RETURN
C  Error processing section
  210 CALL PEND('UNANTICIPATED END-OF-FILE READING PRIOR  LINE 1')
  220 CALL PEND('ERROR READING PRIOR LINE 1')
  230 CALL PINTI('NUMBER OF CATEGORIES M',NM,2,LDM)
  240 CALL PINTI('PARAMETER VECTOR LENGTH',NPAR,1,LDPAR)
      END