!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! gibbsit 1-10-95 !
! !
! Gibbsit -- Version 2.0, by Adrian E. Raftery and Steven M. Lewis !
! (with thanks to Jill Kirby and Jan de Leeuw) !
! !
! This program calculates the number of iterations required in a MCMC !
! run. The user has to specify the precision required. The program !
! returns the number of iterations required to estimate the posterior !
! cdf of the q-quantile of the quantity of interest (a function of !
! the parameters) to within +-r with probability s. It also gives !
! the number of "burn-in" iterations required for the conditional !
! distribution given any starting point (of the derived two-state !
! process) to be within epsilon of the actual equilibrium distri- !
! bution. !
! !
! !
! References: !
! !
! Raftery, A.E. and Lewis, S.M. (1992). How many iterations in the !
! Gibbs sampler? In Bayesian Statistics, Vol. 4 (J.M. Bernardo, J.O. !
! Berger, A.P. Dawid and A.F.M. Smith, eds.). Oxford, U.K.: Oxford !
! University Press, 763-773. !
! This paper is available via the World Wide Web by linking to URL !
! http://www.stat.washington.edu/tech.reports/pub/tech.reports !
! and then selecting the "How Many Iterations in the Gibbs Sampler" !
! link. !
! This paper is also available via regular ftp using the following !
! commands: !
! ftp ftp.stat.washington.edu (or 128.95.17.34) !
! login as anonymous !
! enter your email address as your password !
! ftp> cd /pub/tech.reports !
! ftp> get raftery-lewis.ps !
! ftp> quit !
! !
! Raftery, A.E. and Lewis, S.M. (1992). One long run with diagnos- !
! tics: Implementation strategies for Markov chain Monte Carlo. !
! Statistical Science, Vol. 7, 493-497. !
! !
! Raftery, A.E. and Lewis, S.M. (1995). The number of iterations, !
! convergence diagnostics and generic Metropolis algorithms. In !
! Practical Markov Chain Monte Carlo (W.R. Gilks, D.J. Spiegelhalter !
! and S. Richardson, eds.). London, U.K.: Chapman and Hall. !
! This paper is available via the World Wide Web by linking to URL !
! http://www.stat.washington.edu/tech.reports/pub/tech.reports !
! and then selecting the "The Number of Iterations, Convergence !
! Diagnostics and Generic Metropolis Algorithms" link. !
! This paper is also available via regular ftp using the following !
! commands: !
! ftp ftp.stat.washington.edu (or 128.95.17.34) !
! login as anonymous !
! enter your email address as your password !
! ftp> cd /pub/tech.reports !
! ftp> get raftery-lewis2.ps !
! ftp> quit !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Input: !
! !
! Example values of q, r, s: !
! 0.025, 0.005, 0.95 (for a long-tailed distribution) !
! 0.025, 0.0125, 0.95 (for a short-tailed distribution); !
! 0.5, 0.05, 0.95; 0.975, 0.005, 0.95; etc. !
! !
! The result is quite sensitive to r, being proportional to the !
! inverse of r^2. !
! !
! For epsilon, we have always used 0.001. It seems that the result !
! is fairly insensitive to a change of even an order of magnitude in !
! epsilon. !
! !
! One way to use the program is to run it for several specifications !
! of r, s and epsilon and vectors q on the same data set. When one !
! is sure that the distribution is fairly short-tailed, such as when !
! q=0.025, then r=0.0125 seems sufficient. However, if one is not !
! prepared to assume this, safety seems to require a smaller value of !
! r, such as 0.005. !
! !
! The program takes as input the name of a file containing an initial !
! run from a MCMC sampler. If the MCMC iterates are independent, !
! then the minimum number required to achieve the specified accuracy !
! is about $\Phi^{-1} (\frac{1}{2}(1+s))^2 q(1-q)/r^2$ and this would !
! be a reasonable number to run first. !
! When q=0.025, r=0.005 and s=0.95, this number is 3,748; !
! when q=0.025, r=0.0125 and s=0.95, it is 600. !
! !
! !
! Output: !
! !
! nmin = minimum number of iterations, assuming independence, !
! required to achieve the specified accuracy. !
! nburn = number of iterations to be discarded at the beginning. !
! nprec = number of subsequent iterations required. !
! kthin = skip parameter for a first-order Markov chain. The !
! desired accuracy will be achieved if at most every !
! kthin-th iterate is used. !
! kind = skip parameter sufficient to achieve an independence !
! chain. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The development of this computer program was supported by the !
! Office of Naval Research under contracts N-00014-88-K-0265 and !
! N-00014-91-J-1074 and by the National Institutes of Health under !
! grant 5R01HD26330. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Development history: !
! !
! Version 0.3 (May 1, 1991): Initial version of the program. !
! Version 0.4 (June 8, 1991): This incorporates Jill Kirby's !
! suggestion that lots of things be initialized to zero and her code !
! that does that. !
! !
! Version 1.0 (Sept. 14, 1994): A reorganization of the program by !
! Steven Lewis. This included multiplying nprec by an additional !
! missing factor, (2-alpha-beta), modifying the way in which ties are !
! handled in the calculation of the empirical quantile, and taking !
! the logarithm of abs(1-alpha-beta) in the calculation of nburn. !
! !
! Version 2.0 (Dec. 15, 1994): Add ability to read a matrix of input !
! values, add the I_rl statistic to the output, add calculation of !
! kind (the spacing needed to achieve an independence chain) to the !
! output, add ability to read in a vector of quantiles and add the !
! ability to estimate a probability such as described in the Raftery !
! and Lewis Bayesian Statistics paper. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The dimensions are currently set for input series of length at most !
! 50000, but this is easily changed by editing the line below where !
! maxiterates is set. !
! !
! The maximum number of variables which can be read from the input !
! file is given by the maxseries parameter. This is currently set to !
! 20, but is also easily changed. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
program gibbsit
integer maxiterates
parameter (maxiterates=50000)
integer maxseries
parameter (maxseries=20)
integer maxqcnt
parameter (maxqcnt=20)
integer wasize
parameter (wasize=maxiterates*2)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
! data - contains the original input MCMC series !
! qs,r,s - parameters used to specify the required precision: !
! the q-quantile is to be estimated to within +-r with !
! probability s !
! irl - the I statistic from the Raftery and Lewis Statistical !
! Science paper !
! !
! inpfile - name of file containing the initial MCMC input matrix !
! !
! workarea - vector used internally by the gibbsmain subroutine !
! iteracnt - the number of iterations (ie., rows) in the input data !
! varcnt - number of variables (ie., columns) in the input data !
! nburn - the number of iterations required for burn-in !
! nprec - the number of iterations required to achieve the !
! specified precision !
! kthin - skip parameter for a first-order Markov chain !
! kmind - minimum skip parameter to get an independence chain !
! nmin - minimum number of independent iterates required !
! kind - skip parameter sufficient to achieve an independence !
! chain (ie., the greater of ceiling(irl) and kmind) !
! ccnt - number of control parameters entered by the user !
! qcnt - number of quantiles entered by the user !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
double precision data(maxiterates,maxseries)
double precision qs(maxqcnt)
double precision irl
double precision controlparms(3)
double precision r
equivalence (controlparms(1),r)
double precision s
equivalence (controlparms(2),s)
double precision epsilon
equivalence (controlparms(3),epsilon)
character inpfile*24
integer workarea(wasize)
integer argcount
integer iteracnt
integer varcnt
integer nburn
integer nprec
integer kthin
integer kmind
integer iargc
integer nmin
integer kind
integer ccnt
integer qcnt
integer q1
integer v1
integer rc
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Process whatever command line arguments were passed to gibbsit. !
! Currently at most one argument is expected. If present, the first !
! argument contains the name of the input data file. If not present, !
! get the name of the input data file from the user's console. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
argcount = iargc()
if (argcount.ge.1) then
call getarg(1,inpfile)
else
write (0,*) 'Enter the name of the input file'
read (5,*) inpfile
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Open the input data file. If unable to do so, stop the program. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
open (unit=7,file=inpfile,status='old',iostat=rc)
if (rc.ne.0) go to 900
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Read the input data into the data matrix. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
call matinput(7,maxiterates,maxseries,data,iteracnt,varcnt,rc)
if (rc.ne.0) then
write (0,*) 'matinput exited with a nonzero error code of', rc
go to 900
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The main program loop follows. To begin each loop, get the input !
! control parameters from the user's console. If the user enters !
! an end-of-data or r=99, stop this program. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
300 write (0,*)
+ 'Enter r,s,epsilon (e.g. .0125 .95 .001). r=99 to stop'
call vecinput(5,3,controlparms,ccnt,rc)
if (rc.gt.0) then
write (0,*) 'vecinput exited with a nonzero error code of', rc
go to 900
end if
if (rc.lt.0 .or. r.eq.99) go to 900
if (ccnt.ne.3) then
write (0,*) 'r, s, and epsilon are all required'
go to 300
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Next get a vector of quantiles at which MCMC settings for each !
! variable input series are to be calculated. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
write (0,'(''Enter a vector of quantiles (e.g. .025 .975). '',
+ ''q=0 to estimate probability'')')
call vecinput(5,maxqcnt,qs,qcnt,rc)
if (rc.ne.0) then
write (0,*) 'vecinput exited with a nonzero error code of', rc
go to 900
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Loop through the vector of quantiles, calculating nmin, kthin, !
! nburn, nprec and kind for each variable input series at each given !
! quantile value, separating each set by an extra blank line. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 700 q1=1,qcnt
write (6,'(/''q = '',f5.3,'', r = '',f6.4,'', s = '',f4.2,
+ '', epsilon = '',f6.4,'':'')') qs(q1), r, s, epsilon
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Now execute the gibbmain subroutine once for each variable input !
! series for the current quantile value, to perform all of the real !
! work. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 500 v1=1,varcnt
call gibbmain(data(1,v1),iteracnt,qs(q1),r,s,epsilon,workarea,
+ nmin,kthin,nburn,nprec,kmind,rc)
if (rc.ne.0) then
if (rc.eq.12) then
write (0,'(''When q=0 the input series must consist of '',
+ ''only 0''''s and 1''''s'')')
else
write (0,'(''gibbmain exited with a nonzero error code '',
+ ''of '',i2)') rc
end if
go to 900
end if
irl = dble(nburn + nprec) / dble(nmin)
kind = max( int(irl + 1.0d0), kmind )
write (6,'('' ('',i2,'') kthin='',i3,'', nburn='',i5,
+ '', nprec='',i8,'', nmin='',i5,'', I='',f6.2,'', kind='',
+ i3)') v1, kthin, nburn, nprec, nmin, irl, kind
500 continue
700 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Repeat the main program loop with possibly new control parameters !
! and/or a new vector of quantiles, leaving an extra blank line at !
! the end of the current set of outputs. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
write (6,'()')
go to 300
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! At this point we are done with the gibbsit program. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
900 stop
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! matinput 11-04-94 !
! !
! This subroutine inputs a matrix of double precision numbers from a !
! designated file. The input file is assumed to have the same number !
! of blank delimited numbers on each record of the file. There is an !
! upper limit of twenty numbers per line. This subroutine reads the !
! numbers from the first line of the input file into the first row of !
! the matrix, from the second line of the input file into the second !
! row, and so on. It is important to check the return code from this !
! subroutine since there are a number of different error conditions !
! which may occur. Error codes less than 0 are just warnings, so in !
! most cases may be ignored. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! uid = an integer containing the unit identifier number of an !
! already opened external input file. The uid must be a !
! non-negative number. If not, matinput will return to !
! the caller with an error code of 4 without having read !
! anything into the output matrix, matout. !
! !
! rowmax = an integer containing the allocated number of rows in !
! the output matrix, matout. If the input file contains !
! more than rowmax records, the first rowmax records !
! will be read into the output matrix and matinput will !
! return to the caller with an error return code of -4. !
! The actual number of rows used is returned in the !
! rowused argument. !
! !
! colmax = an integer containing the allocated number of columns !
! in the output matrix, matout. The maximum number of !
! numbers per line which matinput can read into the !
! output matrix is the lesser of colmax and twenty. !
! The actual number of columns used is returned in the !
! colused argument. !
! !
! !
! Outputs: !
! !
! matout = a double precision matrix in which this subroutine is !
! to return the matrix of numbers read in from the input !
! file. This matrix consists of rowmax rows by colmax !
! columns. !
! !
! rowused = an integer containing the actual number of rows of the !
! output matrix, matout, into which matinput has read !
! data. The rest of the output matrix has not been !
! altered by this subroutine. !
! !
! colused = an integer containing the actual number of columns of !
! the output matrix, matout, into which matinput has !
! read data. The rest of the output matrix has not been !
! altered by this subroutine. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Outputs (continued): !
! !
! r15 = an integer valued error return code. This variable !
! is set to 0 if no errors were encountered. !
! Otherwise, r15 can assume the following values: !
! !
! -4 = the end of file from the input file had not yet !
! been reached before running out of the maximum !
! number of rows available in the output matrix. !
! The first rowmax records from the input file !
! will have been read into the output matrix. !
! 4 = the input file unit identifier was negative. !
! 8 = the rowmax argument was not a postitive number. !
! 12 = the oneparse subroutine returned with a nonzero !
! error return code. !
! 16 = an error occurred while converting one of the !
! input numbers into internal double precision !
! format. Any input numbers after the number !
! causing the error will not have been read into !
! output matrix. !
! No other possible values are currently in use. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine matinput(uid,rowmax,colmax,matout,rowused,colused,r15)
integer uid
integer rowmax
integer colmax
double precision matout(rowmax,colmax)
integer rowused
integer colused
integer r15
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine such as loop counters, indices and so forth. !
! !
! curterms - parsed version of the current input record !
! currecrd - the most recently read record from the input file !
! delimit - a single character to be used as the separator between !
! numbers in the input file !
! collimit - lesser of colmax and twenty !
! curcnt - number of tokens parsed from the current input record !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
character curterms(20)*24
character currecrd*512
character delimit!1 /' '/
integer collimit
integer curcnt
integer c1
integer rc
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Make sure that the external unit identifier of the input file is a !
! non-negative number. If it isn't, return to the caller with an !
! error code of 4. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (uid.lt.0) then
write (0,*) 'unit identifier passed to matinput is negative'
r15 = 4
return
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Make sure that the rowmax argument is greater than 0. If it isn't, !
! return to the caller with an error code of 8. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (rowmax.le.0) then
write (0,*) 'output matrix must have a positive number of rows'
r15 = 8
return
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The maximum number of columns of the output matrix which matinput !
! can read into is the lesser of the colmax argument and twenty, but !
! it must be at least 1. !
! !
! For the moment, initialize the number of columns used to this just !
! determined maximum number of columns. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (colmax.lt.20) then
collimit = max(colmax,1)
else
collimit = 20
end if
colused = collimit
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Initialize the number of rows used argument to 0. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
rowused = 0
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Read the next record from the input file as a character string. If !
! there are no more records in the input file, we have completed our !
! job successfully. In this latter case we can jump to the end of !
! the subroutine to set the error code to 0 before returning. !
! !
! Note that I chose to read at least one non-blank record from the !
! input file before checking the number of rows used in the output !
! matrix so that the error code will be set to 0 in the case where !
! the number of non-blank records in the input file is precisely !
! equal to the number of rows in the output matrix. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
200 read (uid,'(a)',end=600) currecrd
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Call the oneparse subroutine to parse the current record into up to !
! collimit blank separated tokens. If any blank lines are read from !
! the input file, these should be ignored. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
call oneparse(currecrd,delimit,collimit,curterms,curcnt,rc)
if (rc.ne.0) then
write (0,*) 'oneparse exited with a nonzero error code of', rc
r15 = 12
return
end if
if (curcnt.lt.1) go to 200
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Make sure that the number of rows used in the output matrix is !
! still less than the maximum number of rows in the matrix. If it !
! is not, then we have read as much as we can into the matrix. We !
! can return to the caller but set the error code to -4 to warn the !
! caller that not all the data from the input file could be read !
! into the output matrix. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (rowused.ge.rowmax) then
r15 = -4
return
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! To make sure that every significant element of the output matrix !
! will have been set by the matinput subroutine, this subroutine sets !
! the actual number of columns used equal to the fewest number of !
! tokens read in from any one record of the input file. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (curcnt.lt.colused) colused = curcnt
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Increment the number of rows used in the output matrix. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
rowused = rowused + 1
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! After parsing the input record into separate numbers, convert each !
! of the input quantities into internal double precision numbers. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 300 c1=1,colused
read (curterms(c1),'(f24.0)',err=400) matout(rowused,c1)
300 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Go read the next record from the input file. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
go to 200
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! An error occurred in trying to convert one of the input numbers !
! into a floating point number. The most prudent thing to do would !
! be to return to the caller without finishing reading from the input !
! file. The error return code is set to 16 before returning. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
400 r15 = 16
return
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Everything has gone as expected if we made it to this point in the !
! program, so return to the caller with the good news. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
600 r15 = 0
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! vecinput 12-07-94 !
! !
! This subroutine reads a vector of double precision numbers from a !
! designated file (which may be the standard input). Only a single !
! line is read from the input file. This line is assumed to consist !
! of up to twenty blank delimited numbers. !
! !
! Vecinput is a trimmed down version of the matinput subroutine. !
! Instead of reading in an entire matrix, vecinput only reads in a !
! single line of input into a vector. Since matrix input need not be !
! supported by this subroutine, the code may be greatly simplified. !
! First, there is no need in vecinput to keep track of row numbers in !
! a matrix, so any code referring to row number may be removed. The !
! other key simplification possible in this subroutine is a result of !
! the fact that only a single line of input is to be read by this !
! subroutine, whereas the matinput subroutine always attempts to read !
! more than one line of input. This difference in behavior is not !
! worth worrying about, except in the case where the input is being !
! read from the user's terminal. Any extra attempted reads to the !
! terminal would not ordinarily be expected by a user and hence need !
! to be prevented. It is particularly this last reason which led me !
! to implement a separate subroutine to read in a simple vector of !
! numbers (which is very likely to be entered interactively). !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! uid = an integer containing the unit identifier number of an !
! already opened external input file. The uid must be a !
! non-negative number. If not, vecinput will return to !
! the caller with an error code of 4 without having read !
! anything into the output vector, vecout. !
! !
! vecmax = an integer containing the allocated number of elements !
! in the output vector, vecout. The maximum number of !
! numbers which vecinput can read into the output vector !
! is the lesser of vecmax and twenty. The actual number !
! of elements used is returned in the vecused argument. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Outputs: !
! !
! vecout = a double precision vector in which this subroutine is !
! to return the vector of numbers read from the single !
! line of the input file. This vector must contain at !
! least vecmax elements. !
! !
! vecused = an integer containing the actual number of elements of !
! the output vector, vecout, into which vecinput has !
! read data. The rest of the output vector will not !
! have been modified by this subroutine. !
! !
! r15 = an integer valued error return code. This variable !
! is set to 0 if no errors were encountered. !
! Otherwise, r15 can assume the following values: !
! !
! -4 = the user entered an end-of-data when prompted !
! for the single line of input. !
! 4 = the input file unit identifier was negative. !
! 8 = the oneparse subroutine returned with a nonzero !
! error return code. !
! 12 = an error occurred while converting one of the !
! input numbers into internal double precision !
! format. Any input numbers after the number !
! causing the error will not have been read into !
! output vector. !
! No other possible values are currently in use. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine vecinput(uid,vecmax,vecout,vecused,r15)
integer uid
integer vecmax
double precision vecout(vecmax)
integer vecused
integer r15
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine such as loop counters, indices and so forth. !
! !
! septerms - parsed version of the line read from the input file !
! charinpt - character form of the line read from the input file !
! delimit - a single character to be used as the separator between !
! numbers in the input file !
! veclimit - lesser of vecmax and twenty !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
character septerms(20)!24
character charinpt*512
character delimit!1 /' '/
integer veclimit
integer v1
integer rc
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Make sure that the external unit identifier of the input file is a !
! non-negative number. If it isn't, return to the caller with an !
! error code of 4. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (uid.lt.0) then
write (0,*) 'unit identifier passed to vecinput is negative'
r15 = 4
return
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The maximum number of elements of the output vector which vecinput !
! can read into is the lesser of the vecmax argument and twenty, but !
! it must be at least 1. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (vecmax.lt.20) then
veclimit = max(vecmax,1)
else
veclimit = 20
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Read a single line from the input file as a character string. If !
! an end-of-data or an end-of-file occurs, we just need to set the !
! error return code to -4 before returning to the caller. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
read (uid,'(a)',end=400) charinpt
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Call the oneparse subroutine to parse the line read from the input !
! file into up to veclimit blank separated terms. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
call oneparse(charinpt,delimit,veclimit,septerms,vecused,rc)
if (rc.ne.0) then
write (0,*) 'oneparse exited with a nonzero error code of', rc
r15 = 8
return
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! After parsing the input record into separate numbers, convert each !
! of the input quantities into internal double precision numbers. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 300 v1=1,vecused
read (septerms(v1),'(f24.0)',err=500) vecout(v1)
300 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Everything has gone as expected if we made it to this point in the !
! program, so return to the caller with the good news. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
r15 = 0
return
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! An end-of-data or an end-of-file occurred when we attempted to read !
! the single line from the input file. Set the error return code to !
! -4 before returning to the caller. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
400 r15 = -4
return
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! An error occurred in trying to convert one of the input numbers !
! into a floating point number. Set the error return code to 12 !
! before returning to the caller. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
500 r15 = 12
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! oneparse 2-17-94 !
! !
! This subroutine is a very rudimentary parser. The input character !
! string passed as the first argument is parsed into 0 or more tokens !
! where the tokens are separated by instances of the single character !
! delimiter passed as the delimit argument. The separate tokens are !
! returned in the output character vector, tokens. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! instring = a character string to be parsed into separate tokens. !
! !
! delimit = a single character to be used to separate the input !
! into individual tokens. !
! !
! maxtok = an integer containing the maximum number of tokens !
! which may be returned. This is the number of elements !
! available in the tokens vector. !
! !
! !
! Outputs: !
! !
! tokens = a vector of character strings containing the parsed !
! version of the input character string, instring. Each !
! entry in tokens will be left justified (leading blanks !
! removed). The number of tokens found in the input !
! string is returned in tokcnt. !
! !
! tokcnt = an integer containing the number of tokens actually !
! returned in tokens. This will be a number between 0 !
! and maxtok, inclusive. !
! !
! r15 = an integer valued error return code. This variable !
! is set to 0 if no errors were encountered. !
! Otherwise, r15 can assume the following values: !
! !
! 4 = the input character string contained more than !
! maxtok tokens. The first maxtok tokens have !
! been returned in tokens and tokcnt has been set !
! equal to maxtok. !
! No other possible values are currently in use. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine oneparse(instring,delimit,maxtok,tokens,tokcnt,r15)
character instring*(*)
character delimit*1
integer maxtok
character tokens(maxtok)*(*)
integer tokcnt
integer r15
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine such as loop counters, indices and so forth. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
integer inlen
integer index
integer bpos
integer dpos
integer epos
integer len
integer tn
integer pn
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Parse the input character string into its separate tokens. The !
! input string is assumed to contain from 0 to maxtok tokens !
! separated by the token separator character, delimit. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
inlen = len(instring)
bpos = 1
tn = 0
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Locate the beginning and the end of the next token within the !
! input string. !
! !
! First, find the next nonblank character in the input string. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
400 do 450 pn=bpos,inlen
if (instring(pn:pn).ne.' ') go to 500
450 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! If no nonblank characters were found in the input string, there !
! are no more tokens in the input, so we have finished parsing this !
! input string. !
! !
! At this point set tokcnt to the number of tokens found and set the !
! error return code to indicate that all went as would be expected !
! before returning to the caller. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
tokcnt = tn
r15 = 0
return
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The beginning of the next token has been located. Increment the !
! subscript to use for this token. The element of the tokens vector !
! addressed by this subscript will be the one used to return the !
! current token. !
! !
! If this subscript is greater than maxtok, the passed input string !
! contained more tokens than can be returned in the tokens vector. !
! In this case, tokcnt is set equal to maxtok and the error return !
! code is set to 4 before returning to the caller. !
! !
! To find the end of this token, find the next token separator or the !
! end of the input string if there are no more separators. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
500 tn = tn + 1
if (tn.gt.maxtok) then
tokcnt = maxtok
r15 = 4
return
end if
bpos = pn
dpos = index(instring(bpos:),delimit)
if (dpos.eq.0) dpos = inlen
epos = bpos + dpos - 2
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! We now have the positions of both the beginning and the end of the !
! next token within the input string, so save this token in the !
! correct element of the tokens vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
tokens(tn) = instring(bpos:epos)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Start looking for the next token immediately following the most !
! recently located token separator (the one which terminated the !
! previous token). !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
bpos = epos + 2
go to 400
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! gibbmain 1-09-95 !
! !
! This program calculates the number of iterations required in a run !
! of MCMC. The user has to specify the precision required. This !
! subroutine returns the number of iterations required to estimate !
! the posterior cdf of the q-quantile of the quantity of interest (a !
! function of the parameters) to within +-r with probability s. It !
! also gives the number of "burn-in" iterations required for the !
! conditional distribution given any starting point (of the derived !
! two-state process) to be within epsilon of the actual equilibrium !
! distribution. !
! !
! If q<=0, then gibbmain is to treat the original input series as a !
! vector of 0-1 outcome variables. In this case no quantile needs to !
! be found. Instead, this subroutine just needs to calculate kthin, !
! nburn, nprec and kmind tuning parameters such that a MCMC run based !
! on these tuning parameters should be adequate for estimating the !
! probability of an outcome of 1 within the prescribed (by r, s, and !
! epsilon) probability. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! original = a double precision vector containing the original MCMC !
! generated series of parameter estimates. This vector !
! contains iteracnt elements. !
! !
! iteracnt = an integer containing the number of actual iterations !
! provided in the sample MCMC output series, original. !
! !
! q,r,s = double precision numbers in which the caller specifies !
! the required precision: the q-quantile is to be !
! estimated to within r with probability s. !
! !
! epsilon = a double precision number containing the half width of !
! the tolerance interval required for the q-quantile. !
! !
! work = an integer vector passed to various subroutines to !
! hold a number of internal vectors. There must be at !
! least (iteracnt ! 2) elements in this vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Outputs: !
! !
! nmin = an integer which will be set to the minimum number of !
! independent Gibbs iterates required to achieve the !
! specified accuracy for the q-quantile. !
! !
! kthin = an integer which will be set to the skip parameter !
! sufficient to produce a first-order Markov chain. !
! !
! nburn = an integer which will be set to the number of !
! iterations to be discarded at the beginning of the !
! simulation, i.e. the number of burn-in iterations. !
! !
! nprec = an integer which will be set to the number of !
! iterations not including the burn-in iterations which !
! need to be obtained in order to attain the precision !
! specified by the values of the q, r and s input !
! parameters. !
! !
! kmind = an integer which will be set to the minimum skip !
! parameter sufficient to produce an independence chain. !
! !
! r15 = an integer valued error return code. This variable !
! is set to 0 if no errors were encountered. !
! Otherwise, r15 can assume the following values: !
! !
! 12 = the original input vector contains something !
! other than a 0 or 1 even though q<=0. !
! No other possible values are currently in use. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine gibbmain(original,iteracnt,q,r,s,epsilon,work,nmin,
+ kthin,nburn,nprec,kmind,r15)
integer iteracnt
double precision original(iteracnt)
double precision q
double precision r
double precision s
double precision epsilon
integer work(iteracnt*2)
integer nmin
integer kthin
integer nburn
integer nprec
integer kmind
integer r15
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
! cutpt - the q-th empirical quantile !
! qhat - when q=0, proportion of 1's in the input data vector, !
! when q>0, qhat is set equal to the passed value of q !
! g2 - G2 for the test of first-order vs second-order Markov !
! bic - the corresponding BIC value !
! phi - \PHI^{-1} ((s+1)/2) !
! alpha - probability of moving from below the cutpt to above !
! beta - probability of moving from above the cutpt to below !
! probsum - sum of alpha + beta !
! !
! The first iteracnt elements of the work vector will be used to !
! store a binary 0-1 series indicating which elements are less than !
! or equal to the cutpt (set to 1) and which elements are less than !
! the cutpt (set to 0). !
! !
! The remaining iteracnt elements of the work vector are to be used !
! to hold thinned versions of the 0-1 series, where the amount of !
! thinning is determined by the current value of kthin (or kmind). !
! That is, for each proposed value of kthin (or kmind), only every !
! kthin-th (or kmind-th) element of the 0-1 series is copied to this !
! thinned copy of the series. !
! !
! ixkstart is the subscript of the first element of the thinned !
! series. That is, ixkstart = iteracnt + 1. !
! !
! thincnt is the current length of the thinned series. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
double precision empquant
double precision cutpt
double precision qhat
double precision g2
double precision bic
double precision phi
double precision alpha
double precision beta
double precision probsum
double precision tmp1
double precision tmp2
real ppnd7
integer ixkstart
integer thincnt
integer i1
integer rc
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! If the q argument is a postive number, interpret it as the quantile !
! which is to be ascertained using MCMC. It should be a positive !
! number less than 1. Set qhat to the passed value of q (we will use !
! qhat later when we calculate nmin). !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
if (q.gt.0.0d0) then
qhat = q
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Find the q-th quantile of the original MCMC series of parameter !
! estimates. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
cutpt = empquant(original,iteracnt,qhat,work)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Calculate a binary 0-1 series indicating which elements are less !
! than or equal to the cutpt (set to 1) and which elements are !
! greater than the cutpt (set to 0). The resulting series is stored !
! in the work vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
call dichot(original,iteracnt,cutpt,work)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Otherwise treat the original input series as a binary 0-1 series of !
! outcomes whose probability needs to be estimated using MCMC. This !
! is easily accomplished by copying the input series into the first !
! iteracnt elements of the work vector, converting the double preci- !
! sion input into an equivalent integer vector of 0's and 1's. For !
! this case we will also need to set qhat equal to the proportion of !
! 1's in the original input data vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
else
qhat = 0.0d0
do 300 i1=1,iteracnt
if (original(i1).eq.0.0d0 .or. original(i1).eq.1.0d0) then
work(i1) = int( original(i1) )
qhat = qhat + original(i1)
else
r15 = 12
return
end if
300 continue
qhat = qhat / dble( iteracnt )
end if
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Find kthin, the degree of thinning at which the indicator series is !
! first-order Markov. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
ixkstart = iteracnt + 1
kthin = 1
500 call thin(work,iteracnt,kthin,work(ixkstart),thincnt)
call mctest(work(ixkstart),thincnt,g2,bic)
if (bic.le.0.0d0) go to 600
kthin = kthin + 1
go to 500
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Calculate both the alpha and beta transition probabilities (in the !
! Cox & Miller parametrization) of the two state first-order Markov !
! chain determined above. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
600 call mcest(work(ixkstart),thincnt,alpha,beta)
kmind = kthin
go to 750
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Now compute just how big the spacing needs to be so that a thinned !
! chain would no longer be a Markov chain, but rather would be an !
! independence chain. This thinning parameter must be at least as !
! large as the thinning parameter required for a first-order Markov !
! chain. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
700 call thin(work,iteracnt,kmind,work(ixkstart),thincnt)
750 call indtest(work(ixkstart),thincnt,g2,bic)
if (bic.le.0.0d0) go to 800
kmind = kmind + 1
go to 700
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Estimate the first-order Markov chain parameters and find the !
! burn-in and precision number of required iterations. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
800 probsum = alpha + beta
tmp1 = dlog(probsum * epsilon / max(alpha,beta)) /
+ dlog( dabs(1.0d0 - probsum) )
nburn = int( tmp1 + 1.0d0 ) * kthin
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Note: ppnd7 is the routine that implements AS algorithm 241. !
! It calculates the specified percentile of the Normal distribution. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
phi = dble(ppnd7( ((real(s) + 1.0) / 2.0), rc ))
tmp2 = (2.0d0 - probsum) * alpha * beta * phi**2 / (probsum**3 *
+ r**2)
nprec = int( tmp2 + 1.0d0 ) * kthin
nmin = int( ((1.0d0-qhat) * qhat * phi**2 / r**2) + 1.0d0 )
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! At this point we have calculated nmin, kthin, nburn, nprec and !
! kmind, so we can return to the calling program. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
r15 = 0
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! empquant 9-13-94 !
! !
! This function finds the q-th empirical quantile of the input double !
! precsion series, data, of length iteracnt. !
! !
! The algorithm used by this subroutine is the one used in the SPLUS !
! quantile function. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! data = a double precision vector of numbers whose q-th !
! empirical quantile is to be calculated. !
! !
! iteracnt = an integer containing the number of elements in the !
! input data vector, data. There must also be this !
! many elements in the work vector. !
! !
! q = a double precision number between 0.0d0 and 1.0d0, !
! inclusive, specifying which empirical quantile is !
! wanted. !
! !
! work = a double precision vector to be used as a work area !
! for the sort subroutine called by empquant. This !
! vector must contain at least iteracnt elements. !
! !
! !
! Outputs: !
! !
! empquant = a double precision number corresponding to the q-th !
! level of the sorted vector of input values. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
function empquant(data,iteracnt,q,work)
double precision empquant
integer iteracnt
double precision data(iteracnt)
double precision q
double precision work(*)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
double precision order
double precision fract
integer low
integer high
integer i1
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Copy the input series of double precision numbers into the work !
! area provided by the caller. In this way the original input will !
! not be modified by this subroutine. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 300 i1=1,iteracnt
work(i1) = data(i1)
300 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Sort the input series into ascending order. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
call ssort(work,work,iteracnt,int(1))
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Now locate the q-th empirical quantile. This apparently longer !
! than necessary calculation is used so as to appropriately handle !
! the case where there are two or more identical values at the !
! requested quantile. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
order = dble(iteracnt-1) * q + 1.0d0
fract = mod(order, 1.0d0)
low = max(int(order), 1)
high = min(low+1, iteracnt)
empquant = (1.0d0 - fract) * work(low) + fract * work(high)
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! dichot 9-13-94 !
! !
! This subroutine takes a double precision vector, data, of length !
! iteracnt and converts it into a 0-1 series in zt, depending on !
! which elements of data are less than or greater than cutpt. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! data = a double precision vector containing a series of !
! numbers which are to be compared to cutpt in order to !
! determine which elements of zt are to be set to 1 and !
! which are to be set to 0. !
! !
! iteracnt = an integer containing the number of elements in the !
! input data vector. !
! !
! cutpt = a double precision number indicating the boundary !
! about which the input data vector is to be dichoto- !
! mized, i.e. set to 1 when less than or equal to the !
! cutpoint and to 0 when greater than the cutpoint. !
! !
! !
! Outputs: !
! !
! zt = an integer vector containing zeros and ones depending !
! on whether or not the corresponding elements of data !
! were less than the cutpoint or not. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine dichot(data,iteracnt,cutpt,zt)
integer iteracnt
double precision data(iteracnt)
double precision cutpt
integer zt(iteracnt)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
integer i1
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! If the entry in the input data vector is less than or equal to the !
! cutpoint, set the corresponding element of zt to 1, otherwise set !
! it to 0. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 500 i1=1,iteracnt
if (data(i1).le.cutpt) then
zt(i1) = 1
else
zt(i1) = 0
end if
500 continue
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! thin 9-13-94 !
! !
! This subroutine takes the integer-valued vector series of length !
! iteracnt and outputs elements 1,1+kthin,1+2kthin,1+3kthin,... in !
! the result vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! series = an integer vector containing the sequence of numbers !
! from which this subroutine is to select every kthin'th !
! number to be copied to the output vector, starting !
! with the first number. There are iteracnt elements in !
! this vector. !
! !
! iteracnt = an integer containing the number of elements in the !
! input vector of numbers to be thinned, series. If !
! kthin can be as little as 1, the output result vector !
! must also contain iteracnt elements. !
! !
! kthin = an integer specifying the interval between elements of !
! the input data vector, series, which are to be copied !
! to the output vector, result. If kthin is 1, then all !
! of series is to be copied to result. If kthin is 2, !
! then only every other element of series is to be !
! copied to result. If kthin is 3, then every third !
! element is copied and so forth. !
! !
! !
! Outputs: !
! !
! result = an integer vector containing the thinned subset of the !
! input data vector, series, starting with the first !
! element and copying every kthin'th from there on. The !
! number of meaningful elements in this vector will be !
! returned as thincnt. !
! !
! thincnt = an integer containing the number of elements actually !
! copied to the result vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine thin(series,iteracnt,kthin,result,thincnt)
integer iteracnt
integer series(iteracnt)
integer kthin
integer result(iteracnt)
integer thincnt
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
integer from
integer i1
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The specified subset of the input data vector, series, is copied to !
! sequential elements of the output vector, result. Stop copying !
! when the entries in the input data vector run out. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 300 i1=1,iteracnt
from = (i1-1) * kthin + 1
if (from.gt.iteracnt) go to 600
result(i1) = series(from)
300 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Calculate how many elements have been copied to the output vector, !
! result, and return this number as thincnt. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
600 thincnt = i1 - 1
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! mctest 12-05-94 !
! !
! This subroutine tests for a first-order Markov chain against a !
! second-order Markov chain using the log-linear modeling !
! formulation. Here the first-order model is the [12][23] model, !
! while the 2nd-order model is the saturated model. The [12][23] !
! model has closed form estimates - see Bishop, Feinberg and Holland. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! data = an integer vector containing the series of 0's and 1's !
! for which this subroutine is to determine whether a !
! first-order Markov chain is sufficient or whether a !
! second-order Markov chain is needed to model the data. !
! There must be at least datacnt elements in the data !
! vector. !
! !
! datacnt = an integer containing the number of elements in the !
! data argument. !
! !
! !
! Outputs: !
! !
! g2 = a double precision number in which this subroutine is !
! to return the log likelihood ratio statistic for !
! testing a second-order Markov chain against only a !
! first-order Markov chain. Bishop, Feinberg and !
! Holland denote this statistic as G2. !
! !
! bic = a double precision number in which this subroutine is !
! to return the BIC value corresponding to the log !
! likelihood ratio statistic, g2. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine mctest(data,datacnt,g2,bic)
integer datacnt
integer data(datacnt)
double precision g2
double precision bic
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
double precision fitted
double precision focus
integer tran(2,2,2)
integer i1
integer i2
integer i3
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Initialize the transition counts array to all zeroes. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 300 i1=1,2
do 200 i2=1,2
do 100 i3=1,2
tran(i1,i2,i3) = 0
100 continue
200 continue
300 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Count up the number of occurrences of each possible type of !
! transition. Keep these counts in the transition counts array. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 400 i1=3,datacnt
tran( data(i1-2)+1, data(i1-1)+1, data(i1)+1 ) =
+ tran( data(i1-2)+1, data(i1-1)+1, data(i1)+1 ) + 1
400 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Compute the log likelihood ratio statistic for testing a second- !
! order Markov chain against only a first-order Markov chain. This !
! is Bishop, Feinberg and Holland's G2 statistic. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
g2 = 0.0d0
do 700 i1=1,2
do 600 i2=1,2
do 500 i3=1,2
if (tran(i1,i2,i3).eq.0) go to 500
fitted = dble( (tran(i1,i2,1) + tran(i1,i2,2)) *
+ (tran(1,i2,i3) + tran(2,i2,i3)) ) / dble( tran(1,i2,1) +
+ tran(1,i2,2) + tran(2,i2,1) + tran(2,i2,2) )
focus = dble( tran(i1,i2,i3) )
g2 = g2 + dlog( focus / fitted ) * focus
500 continue
600 continue
700 continue
g2 = g2 ! 2.0d0
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Finally calculate the associated bic statistic and return to the !
! caller. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
bic = g2 - dlog( dble(datacnt-2) ) * 2.0d0
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! indtest 11-23-94 !
! !
! This subroutine tests for an independence chain against a first- !
! order Markov chain using the log-linear modeling formulation. In !
! our case the independence model is the [1][2][3] model, while the !
! first-order model is the [12][23] model. Both the [1][2][3] and !
! the [12][23] models have closed form estimates - see Bishop, !
! Feinberg and Holland (1975). !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! data = an integer vector containing the series of 0's and 1's !
! for which this subroutine is to determine whether an !
! independence chain is sufficient or whether a first- !
! order Markov chain is needed to model the data. There !
! must be at least datacnt elements in the data vector. !
! !
! datacnt = an integer containing the number of elements in the !
! data argument. !
! !
! !
! Outputs: !
! !
! g2 = a double precision number in which this subroutine is !
! to return the log likelihood ratio statistic for !
! testing a first-order Markov chain against simply an !
! independence chain. Bishop, Feinberg and Holland !
! denote this statistic as G2. !
! !
! bic = a double precision number in which this subroutine is !
! to return the BIC value corresponding to the log !
! likelihood ratio statistic, g2. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine indtest(data,datacnt,g2,bic)
integer datacnt
integer data(datacnt)
double precision g2
double precision bic
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
double precision fitted
double precision focus
double precision dcm1
integer tran(2,2)
integer i1
integer i2
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Initialize the transition counts array to all zeroes. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 300 i1=1,2
do 200 i2=1,2
tran(i1,i2) = 0
200 continue
300 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Count up the number of occurrences of each possible type of !
! transition. Keep these counts in the transition counts array. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 400 i1=2,datacnt
tran( data(i1-1)+1, data(i1)+1 ) = tran( data(i1-1)+1,
+ data(i1)+1 ) + 1
400 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Compute the log likelihood ratio statistic for testing a first- !
! order Markov chain against simply an independence chain. This is !
! Bishop, Feinberg and Holland's G2 statistic. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
dcm1 = dble( datacnt-1 )
g2 = 0.0d0
do 700 i1=1,2
do 600 i2=1,2
if (tran(i1,i2).eq.0) go to 600
fitted = dble( (tran(i1,1) + tran(i1,2)) * (tran(1,i2) +
+ tran(2,i2)) ) / dcm1
focus = dble( tran(i1,i2) )
g2 = g2 + dlog( focus / fitted ) * focus
600 continue
700 continue
g2 = g2 * 2.0d0
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Finally calculate the associated bic statistic and return to the !
! caller. Note that the first-order Markov chain model contains just !
! one more parameter than does the independence chain model, so p=1. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
bic = g2 - dlog( dcm1 )
return
end
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! mcest 12-05-94 !
! !
! Estimate the parameters of a first-order Markov chain (in the Cox !
! & Miller parametrization) from a series of binary, i.e. 0-1, data !
! passed in the data vector argument. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Inputs: !
! !
! data = an integer vector containing the series of 0's and 1's !
! from which this subroutine is to calculate empirical !
! probabilities of a transition from a 0 to a 1 or a !
! transition from a 1 to a 0. There must be at least !
! datacnt elements in this vector. !
! !
! datacnt = an integer containing the number of elements in the !
! data argument. !
! !
! !
! Outputs: !
! !
! alpha = a double precision number in which this subroutine is !
! to return the empirical probability of a 1 following !
! a 0 in the input data vector. !
! !
! beta = a double precision number in which this subroutine is !
! to return the empirical probability of a 0 following !
! a 1 in the input data vector. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
subroutine mcest(data,datacnt,alpha,beta)
integer datacnt
integer data(datacnt)
double precision alpha
double precision beta
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! The following variables hold various temporary values used in this !
! subroutine. This includes any do-loop counters and similar such !
! temporary subscripts and indices. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
integer tran(2,2)
integer i1
integer i2
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Initialize the transition counts array to all zeroes. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 200 i1=1,2
do 100 i2=1,2
tran(i1,i2) = 0
100 continue
200 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Count up the number of occurrences of each possible type of !
! transition. Keep these counts in the transition counts array. !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
do 400 i1=2,datacnt
tran( data(i1-1)+1, data(i1)+1 ) = tran( data(i1-1)+1,
+ data(i1)+1 ) + 1
400 continue
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! !
! Calculate the empirical transition probabilities between 0's and !
! 1's in the input (returned in alpha) and between 1's and 0's in the !
! input (returned in beta). !
! !
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
alpha = dble(tran(1,2)) / dble( (tran(1,1) + tran(1,2)) )
beta = dble(tran(2,1)) / dble( (tran(2,1) + tran(2,2)) )
return
end
REAL FUNCTION PPND7(P,IFAULT)
! ALGORITHM AS241 APPL. STATIST. (1988) VOL. 37, NO. 3, 477-
! 484.
! Produces the normal deviate Z corresponding to a given lower
! tail area of P; Z is accurate to about 1 part in 10!!7.
! The hash sums below are the sums of the mantissas of the
! coefficients. They are included for use in checking
! transcription.
REAL ZERO, ONE, HALF, SPLIT1, SPLIT2, CONST1, CONST2, A0, A1,
+ A2, A3, B1, B2, B3, C0, C1, C2, C3, D1, D2, E0, E1, E2,
+ E3, F1, F2, P, Q, R
PARAMETER (ZERO = 0.0, ONE = 1.0, HALF = 0.5,
+ SPLIT1 = 0.425, SPLIT2 = 5.0,
+ CONST1 = 0.180625, CONST2 = 1.6)
INTEGER IFAULT
! Coefficients for P close to 0.5
PARAMETER (A0 = 3.38713 27179E+00, A1 = 5.04342 71938E+01,
+ A2 = 1.59291 13202E+02, A3 = 5.91093 74720E+01,
+ B1 = 1.78951 69469E+01, B2 = 7.87577 57664E+01,
+ B3 = 6.71875 63600E+01)
! HASH SUM AB 32.31845 77772
! Coefficients for P not close to 0, 0.5 or 1.
PARAMETER (C0 = 1.42343 72777E+00, C1 = 2.75681 53900E+00,
+ C2 = 1.30672 84816E+00, C3 = 1.70238 21103E-01,
+ D1 = 7.37001 64250E-01, D2 = 1.20211 32975E-01)
! HASH SUM CD 15.76149 29821
! Coefficients for P near 0 or 1.
PARAMETER (E0 = 6.65790 51150E+00, E1 = 3.08122 63860E+00,
+ E2 = 4.28682 94337E-01, E3 = 1.73372 03997E-02,
+ F1 = 2.41978 94225E-01, F2 = 1.22582 02635E-02)
! HASH SUM EF 19.40529 10204
IFAULT = 0
Q = P - HALF
IF (ABS(Q) .LE. SPLIT1) THEN
R = CONST1 - Q * Q
PPND7 = Q * (((A3 * R + A2) * R + A1) * R + A0) /
+ (((B3 * R + B2) * R + B1) * R + ONE)
RETURN
ELSE
IF (Q .LT. ZERO) THEN
R = P
ELSE
R = ONE - P
END IF
IF (R .LE. ZERO) THEN
IFAULT = 1
PPND7 = ZERO
RETURN
END IF
R = SQRT(-LOG(R))
IF (R .LE. SPLIT2) THEN
R = R - CONST2
PPND7 = (((C3 * R + C2) * R + C1) * R + C0) /
+ ((D2 * R + D1) * R + ONE)
ELSE
R = R - SPLIT2
PPND7 = (((E3 * R + E2) * R + E1) * R + E0) /
+ ((F2 * R + F1) * R + ONE)
END IF
IF (Q .LT. ZERO) PPND7 = - PPND7
RETURN
END IF
END
SUBROUTINE SSORT(X,Y,N,KFLAG)
!!!!BEGIN PROLOGUE SSORT
!!!!REVISION OCTOBER 1,1980
!!!!CATEGORY NO. M1
!!!!KEYWORD(S) SORTING,SORT,SINGLETON QUICKSORT,QUICKSORT
!!!!DATE WRITTEN NOVEMBER,1976
!!!!AUTHOR JONES R.E., WISNIEWSKI J.A. (SLA)
!!!!PURPOSE
! SSORT SORTS ARRAY X AND OPTIONALLY MAKES THE SAME
! INTERCHANGES IN ARRAY Y. THE ARRAY X MAY BE SORTED IN
! INCREASING ORDER OR DECREASING ORDER. A SLIGHTLY MODIFIED
! QUICKSORT ALGORITHM IS USED.
!!!!DESCRIPTION
! SANDIA MATHEMATICAL PROGRAM LIBRARY
! APPLIED MATHEMATICS DIVISION 2646
! SANDIA LABORATORIES
! ALBUQUERQUE, NEW MEXICO 87185
! CONTROL DATA 6600/7600 VERSION 8.1 AUGUST 1980
!
! WRITTEN BY RONDALL E JONES
! MODIFIED BY JOHN A. WISNIEWSKI TO USE THE SINGLETON QUICKSORT
! ALGORITHM. DATE 18 NOVEMBER 1976.
!
! ABSTRACT
! SSORT SORTS ARRAY X AND OPTIONALLY MAKES THE SAME
! INTERCHANGES IN ARRAY Y. THE ARRAY X MAY BE SORTED IN
! INCREASING ORDER OR DECREASING ORDER. A SLIGHTLY MODIFIED
! QUICKSORT ALGORITHM IS USED.
!
! REFERENCE
! SINGLETON,R.C., ALGORITHM 347, AN EFFICIENT ALGORITHM FOR
! SORTING WITH MINIMAL STORAGE, CACM,12(3),1969,185-7.
!
! DESCRIPTION OF PARAMETERS
! X - ARRAY OF VALUES TO BE SORTED (USUALLY ABSCISSAS)
! Y - ARRAY TO BE (OPTIONALLY) CARRIED ALONG
! N - NUMBER OF VALUES IN ARRAY X TO BE SORTED
! KFLAG - CONTROL PARAMETER
! =2 MEANS SORT X IN INCREASING ORDER AND CARRY Y ALONG.
! =1 MEANS SORT X IN INCREASING ORDER (IGNORING Y)
! =-1 MEANS SORT X IN DECREASING ORDER (IGNORING Y)
! =-2 MEANS SORT X IN DECREASING ORDER AND CARRY Y ALONG.
!
!!!!REFERENCE(S)
! SINGLETON,R.C., ALGORITHM 347, AN EFFICIENT ALGORITHM FOR
! SORTING WITH MINIMAL STORAGE, CACM,12(3),1969,185-7.
!!!!END PROLOGUE
INTEGER I, IJ, IL(21), IU(21), J, K, KFLAG, KK, L, M, N, NN
DOUBLE PRECISION R, T, TT, TTY, TY, X(N), Y(N)
!!!!FIRST EXECUTABLE STATEMENT SSORT
NN = N
KK = IABS(KFLAG)
!
! ALTER ARRAY X TO GET DECREASING ORDER IF NEEDED
!
15 IF (KFLAG.GE.1) GO TO 30
DO 20 I=1,NN
20 X(I) = -X(I)
30 GO TO (100,200),KK
!
! SORT X ONLY
!
100 CONTINUE
M = 1
I = 1
J = NN
R = .375
110 IF (I .EQ. J) GO TO 155
115 IF (R .GT. .5898437) GO TO 120
R = R+3.90625E-2
GO TO 125
120 R = R-.21875
125 K = I
! SELECT A CENTRAL ELEMENT OF THE
! ARRAY AND SAVE IT IN LOCATION T
IJ = I + IDINT( DBLE(J-I) * R )
T = X(IJ)
! IF FIRST ELEMENT OF ARRAY IS GREATER
! THAN T, INTERCHANGE WITH T
IF (X(I) .LE. T) GO TO 130
X(IJ) = X(I)
X(I) = T
T = X(IJ)
130 L = J
! IF LAST ELEMENT OF ARRAY IS LESS THAN
! T, INTERCHANGE WITH T
IF (X(J) .GE. T) GO TO 140
X(IJ) = X(J)
X(J) = T
T = X(IJ)
! IF FIRST ELEMENT OF ARRAY IS GREATER
! THAN T, INTERCHANGE WITH T
IF (X(I) .LE. T) GO TO 140
X(IJ) = X(I)
X(I) = T
T = X(IJ)
GO TO 140
135 TT = X(L)
X(L) = X(K)
X(K) = TT
! FIND AN ELEMENT IN THE SECOND HALF OF
! THE ARRAY WHICH IS SMALLER THAN T
140 L = L-1
IF (X(L) .GT. T) GO TO 140
! FIND AN ELEMENT IN THE FIRST HALF OF
! THE ARRAY WHICH IS GREATER THAN T
145 K = K+1
IF (X(K) .LT. T) GO TO 145
! INTERCHANGE THESE ELEMENTS
IF (K .LE. L) GO TO 135
! SAVE UPPER AND LOWER SUBSCRIPTS OF
! THE ARRAY YET TO BE SORTED
IF (L-I .LE. J-K) GO TO 150
IL(M) = I
IU(M) = L
I = K
M = M+1
GO TO 160
150 IL(M) = K
IU(M) = J
J = L
M = M+1
GO TO 160
! BEGIN AGAIN ON ANOTHER PORTION OF
! THE UNSORTED ARRAY
155 M = M-1
IF (M .EQ. 0) GO TO 300
I = IL(M)
J = IU(M)
160 IF (J-I .GE. 1) GO TO 125
IF (I .EQ. 1) GO TO 110
I = I-1
165 I = I+1
IF (I .EQ. J) GO TO 155
T = X(I+1)
IF (X(I) .LE. T) GO TO 165
K = I
170 X(K+1) = X(K)
K = K-1
IF (T .LT. X(K)) GO TO 170
X(K+1) = T
GO TO 165
!
! SORT X AND CARRY Y ALONG
!
200 CONTINUE
M = 1
I = 1
J = NN
R = .375
210 IF (I .EQ. J) GO TO 255
215 IF (R .GT. .5898437) GO TO 220
R = R+3.90625E-2
GO TO 225
220 R = R-.21875
225 K = I
! SELECT A CENTRAL ELEMENT OF THE
! ARRAY AND SAVE IT IN LOCATION T
IJ = I + IDINT( DBLE(J-I) * R )
T = X(IJ)
TY = Y(IJ)
! IF FIRST ELEMENT OF ARRAY IS GREATER
! THAN T, INTERCHANGE WITH T
IF (X(I) .LE. T) GO TO 230
X(IJ) = X(I)
X(I) = T
T = X(IJ)
Y(IJ) = Y(I)
Y(I) = TY
TY = Y(IJ)
230 L = J
! IF LAST ELEMENT OF ARRAY IS LESS THAN
! T, INTERCHANGE WITH T
IF (X(J) .GE. T) GO TO 240
X(IJ) = X(J)
X(J) = T
T = X(IJ)
Y(IJ) = Y(J)
Y(J) = TY
TY = Y(IJ)
! IF FIRST ELEMENT OF ARRAY IS GREATER
! THAN T, INTERCHANGE WITH T
IF (X(I) .LE. T) GO TO 240
X(IJ) = X(I)
X(I) = T
T = X(IJ)
Y(IJ) = Y(I)
Y(I) = TY
TY = Y(IJ)
GO TO 240
235 TT = X(L)
X(L) = X(K)
X(K) = TT
TTY = Y(L)
Y(L) = Y(K)
Y(K) = TTY
! FIND AN ELEMENT IN THE SECOND HALF OF
! THE ARRAY WHICH IS SMALLER THAN T
240 L = L-1
IF (X(L) .GT. T) GO TO 240
! FIND AN ELEMENT IN THE FIRST HALF OF
! THE ARRAY WHICH IS GREATER THAN T
245 K = K+1
IF (X(K) .LT. T) GO TO 245
! INTERCHANGE THESE ELEMENTS
IF (K .LE. L) GO TO 235
! SAVE UPPER AND LOWER SUBSCRIPTS OF
! THE ARRAY YET TO BE SORTED
IF (L-I .LE. J-K) GO TO 250
IL(M) = I
IU(M) = L
I = K
M = M+1
GO TO 260
250 IL(M) = K
IU(M) = J
J = L
M = M+1
GO TO 260
! BEGIN AGAIN ON ANOTHER PORTION OF
! THE UNSORTED ARRAY
255 M = M-1
IF (M .EQ. 0) GO TO 300
I = IL(M)
J = IU(M)
260 IF (J-I .GE. 1) GO TO 225
IF (I .EQ. 1) GO TO 210
I = I-1
265 I = I+1
IF (I .EQ. J) GO TO 255
T = X(I+1)
TY = Y(I+1)
IF (X(I) .LE. T) GO TO 265
K = I
270 X(K+1) = X(K)
Y(K+1) = Y(K)
K = K-1
IF (T .LT. X(K)) GO TO 270
X(K+1) = T
Y(K+1) = TY
GO TO 265
!
! CLEAN UP
!
300 IF (KFLAG.GE.1) RETURN
DO 310 I=1,NN
310 X(I) = -X(I)
RETURN
END "