This file is called kliqfind.par. It contains the parameters that tell
kliquefinder how it should go about clustering the data. The user should
change only the values of the parameters, and not any of the labelling or
formatting associated with those values. Each set of parameters consists
of five lines:
1)GENERAL HEADING
2)the actual fortran format used by kliqfind
3)Values of parameters
4)Fortran format layout as a guide for placing parameters
(corresponds to fortran format)
5)labels of parameters (in the appropriate order)
If the user feels the need, he/she may change the fortran format associated
with the parameters.
The user also has the option to specify each parameter interactively by
submitting a null file to kliqfind . Kliqfind can then write a new
file for you which saves the parameters in the proper format. You may then
copy this new file to kliqfind.par .
Explanations of each parameter appear at the bottom of this file and
in the user guide.
START VALUES
(I3,1x,I1,1X,I1,1X,I1,1X,F10.5)
001 2 1 1 4161.41610
123x1x1x1x1234.12345
numdyad dydtriad startgrp,noattach,raseed
MATRIX MULTIPLICATION
(F10.5,1X,F10.5,1X,I4,1X,I4)
1.00000 2.00000 9999 1
1234.12345x1234.12345x1234x1234
direct thresht lookt,maxseed
BOUNDARY SPANNING
(4F10.5)
.50000 1.00000 .500 .90000
1234.123451234.123451234.123451234.12345
boundval,fixr,blabound,betwmult
PROXIMITY
(F10.5,1X,F10.5,1X,I1,1X,I1,1X,I1,1X,I1)
-100.00001 3.10000 0 0 2 3
1234.12345x1234.12345X1X1X1X1
nearval,pctile,mutdyad,noneg,halfdyad,dissolve
CONVERGENCE
(F10.5,1X,I8,1X,I1,1X,I1)
.00005 2000 0 1
1234.12345x12345678x1x1
stopval,kcount2,quickend,attachi
OBJECTIVE FUNCTION TO BE MAXIMIZED: STRUCTURAL EQUIVALNCE
(I1,1X,I1,1X,I1)
0 1 1
1x1x1
structeq,network,actrsqr
OBJECTIVE FUNCTION TO BE MAXIMIZED: CONNECTIVITY
(I2,1X,3(I1,1X),F10.5,1X,F10.5,1X,I1)
5 0 1 0 1.00000 1.00000 1
12x1x1x1x1234.12345x1234.12345x1
quantype,squarit,netlev,pergroup,colwt,rowwt,hyperg
MANIPULATION OF DATA
(I1,1X,I1,1X,I1,1X,I1,1X,I1,1X,I1,1X,I1)
0 0 0 0 0 0 1
1x1x1x1x1x1x1
transpos,reweight,symmat,invert,rectmat,gusemarg,tagalong
EVALUATION
(I4,1X,F10.5,1X,F10.5,1X,I8,1X,I1,1X,F10.5)
000 0.00000 1.00000 0 1 10.00000
1234x1234.12345x1234.12345x12345678X1X1234.12345
neval,baseval,topval,numres,newgrps,hiwteval
ANCHORED MULTIDIMENSIONAL SCALING (BETWEEN GROUPS)
(F10.5,1X,I4,1X,F10.5,1X,8(I4,1X),F10.5,1X,I4,1X,F10.5,1X,I4,1X,I4,1X,3F10.5)
1.00000 2 .2000 1 -1 999 0 1 6 4 30 -2.0000 -3 .15 2 0 .75 .25 1.00000
1234.12345X1234X1234.12345X1234X1234X1234X1234X1234X1234X1234X1234X1234.12345X1234X1234.12345X1234X1234X1234.123451234.123451234.12345
IGRATIO,NUMDIM,MINPICT,CENTER,DANCHOR,DANCHOR2,MOVE2,ZSYMMAT,STARTINC,BYINC,MAXINC,KEXP,NORMAL,MINVALG,BYANGLE,BYSCALE,PCTCENG1,PCTCENG2,DRADIUSG
ANCHORED MULTIDIMENSIONAL SCALING (WITHIN GROUPS)
(8(I4,1X),F10.5,1X,I4,1X,F10.5,1X,I4,1X,I4,1X,3F10.5)
1 1 999 0 1 6 4 30 -2.0000 -3 .5 2 0 .75 .25 1.0000
1234X1234X1234X1234X1234X1234X1234X1234X1234.12345X1234X1234.12345X1234X1234X1234.123451234.123451234.12345
CENTERI,DANCHORI,DANCH2I,MOVE2I,ZSYMMATI,STARTINI,BYINCI,MAXINCI,KEXPI,NORMALI,MINVALI,BYANGLEI,BYSCALEI,PCTCENI1,PCTCENI2,DRADIUSI
ROTATION OF ACTORS WITHIN GROUPS
(3(I4,1X))
15 2 1
1234X1234X1234
RINCREM,MEASURE,EXTREME
Explanation of Parameters
START VALUES
numdyad number of dyads to start groups
dydtriad=1 if you want triads
=2 if you want dyads
startgrp = 1 if you want to use triads to initiate clusters
2 if you want a random start (with numdyad groups)
3 if you want to start from a priori groups
4 * special for simulated data: read in a priori placements *
noattach = 1 if you want to run the new way, adding one dyad at a time.
= 0 if you want to choose the number of dyads to start with and
attach only to them.
raseed = seed used for random assignment of actors to groups
MATRIX MULTIPLICATION
direct is the empasis to put on direct choices in findtrid
thresht is the threshold for using a start triad (so that not all triads
must be looked at
lookt is the total number of triads to look at, cannot be greater than 9999.
maxseed is the maximum number of times an actor can be used in a seed group.
BOUNDARY SPANNING
boundval= value above which boundary spanners are flagged for extra print out
blabound= value above which Blau boundary spanners are flagged for extra
print out
fixr = the maximum value of the number of connections initiated by a single
actor as fixed by the measurement process
PROXIMITY
nearval= if an actor's closeness to their best group is less than nearval,
then a new group is started
pctile is the percentile of measures of closeness that should be used to
determine nearval. I.E., if pctile = .5000, then we use the median of
current closeness as the measure of nearval. a value of greater than
one will keep nearval fixed as input.
mutdyad = 0 if dyads can be built while only one actor makes connections
across the network
= 1 if both actors must make connections to build a dyad
noneg = 0 if assignments which result in a negative contribution to the
objective function should be sustained
= 1 if assignments which result in a negative contribution to the
objective function should be invalidated
halfdyad =1 if value contribution of dyad to objective function should be
halved when considering removing an actor from the dyad
=2 if Dyads cannot be formed at all.
=0 otherwise
dissolve : if an actor is removed from a group of size 'dissolve' then all
other actors in that group are reassigned to their next best
groups. dissolve =1 if you do not want to reassign actors to
their best groups.
CONVERGENCE
stopval is the minimum change to end the cluster routine.
kcount2 is the number of reassignments to make in a phase of the ascent
quickend =1 if you don't want to attach isolates or finish ascent --
you just want output on a priori groups
attachi =1 if you want to attach isolates at end of iteration and then reascend
OBJECTIVE FUNCTION TO BE MAXIMIZED: STRUCTURAL EQUIVALNCE
structeq =1 if you want to model via structural equivalence
(based on R-squared)
network =1 if you want diagonals of matrix to be assigned the largest
off-diagonal value in the matrix
actrsqr= use actual change in r-square for structural equivalance
r-squarechange = groupsize/(groupsize+1) * euclidean distance
from mean
OBJECTIVE FUNCTION TO BE MAXIMIZED: CONNECTIVITY
quantype =1 if you want to use standardized scores as measure of distance
=2 if you want to use pearson distance --
(Observed-Expected)/sqrt(Expected)
=3 if you want to use likelihood ratio --
Observed*log(observed/expected)
squarit = 1 if you want to square the distance measure, (but preserve the
initial sign)
netlev = 1 if objective fucntion should be evaluated at the network level
(only applies to Hubert's compactness which is summed at the
network level)
pergroup =1 if you want to maximize the average objective function per group
rowwt = the weight to give to the connections initiated by the actor in
assigning the actor to a group
colwt = the weight to assign to connections directed to an actor in assigning
the actor to a group
hyperg =1 if hypergeometric distribution is assumed, 0 if binomial
distribution is assumed (only affects calculation of variances).
DATA MANIPULATION
transpos =1 if you want to work with the transpose of the original matrix
reweight =the value of the highest weight. All weights higher than reweight
will be trimmed to the value of reweight. Set reweight equal to zero if all
weights should remian as input
symmat is whether or not we want original similarities matrix between
people to be combined and analyzed as a symmetric (1) matrix or not (0).
invert =1 if your raw data indicate distances as opposed to priximities
rectmat = 0 if original matrix is square and intact (rows and columns are same
elememts
= 1 if original matrix is rectangular, and you want to work with X'X (similarity of columns)
= 2 if original matrix is rectangular, and you want to work with XX'
(similarity of rows)
gusemarg =1 if you want to adjust elements from multiplied matrix my row or
column marginal
=0 otherwise
tagalong =1 if actors who are connected to only one other actor in the
network should be removed from analysis and assigned to the subgroup
of the actor with whom they are connected
=0 if actors who are connected to only one other actor should remain
in the analysis
=-1 if actors who are connected to no others in the network should remain in the analysis
EVALUATION
neval = number of samples to use for evaluating final cluster solution
baseval is the base percentage of similarity at which to create new solutions
for evluating the cluster solution
topval = highest percentage of similarity with final solution in evaluation
numres = number of times to process through the ascent-evaluation-residual
cycle (one cycle is assumed).
newgrps = should new groups be created in rival solutions?
Note: one can do a true montecarlo evaluation of a given solution as
described by Hubert if one sets quickend=1, baseval=0.000,topval=0.0
and newgrps=0
hiwteval proportional weight at low end of similarity with final solution
compared with high end. Currently is meaningless -- 6-11-92
ANCHORED MULTIDIMENSIONAL SCALING
igratio = the ratio of distances within subgroups to between subgroups
numdim = the number of dimensions into which the space should be divided.
The current maximim is 2.
minpict is the minimum distance in the plot as a percentage of the maximum
value.
Center = 1 if you want points to be recentered about (0,0) after relative
positions have been identified
Danchor = the position (in order of centrality) of the anchoring actor.
A number equal to zero indicates random assignment, a number less
than zero will be converted to positive, but pctiles will be used
instead of the mean to identify centrality (see pctcen),
a number larger than the number of actors indicates the last actor.
Danchor2 = the position (in order of centrality) of the second anchoring actor.
A number less than zero will be converted to positive,
but pctiles will be used instead of the mean to identify centrality
(see pctcen), a number larger than the number of actors indicates
the last actor.
move2 =1 if the second anchoring actor may be repositioned, 0 if it's angle
remains at 0.
zsymmat =1 if the priximites should be treated as symmetric, 0 otherwise
startinc = the number of increments of the angle space for the first
iteration
byinc = the number of increments of the angle space to increase at each
iteration
maxinc = the maximal number of increments of the angle space should be
divided
kexp = the exponent in the distances, negative implies take absolute value,
0 implies take the log
normal -- for between groups:
=0 if you do not want any normalization
=1 if you want to normalize the non-radius distances by the true distance,
=2 if you want to take negative logs
=3 if you want to take log(hiwt)-log(max distance)
minval is used if normal is greater than 1. It indicates the minimum value
for a between group matrix element
byscale =1 if you want to scale distances by the radii of the relevant actors
=0 otherwise
byangle=1 if you want to use angles as basis of plotting instead
=0 if you want to use Euclidean distances
pctcen = indicates the percentile used for obtaining the anchoring groups if danchor is less than 0.
if value is positive, kliquefinder will first determine most central
subgroup or actor by its degree, and then settle ties based onthe pctcent.
if the value is negative, only the pctcent will be used to identify
the most central subgroups and actors.
dradius = the number that you should divide the radius by to balance out stress
ROTATION OF ACTORS WITHIN SUBGROUPS
rincrem = the number of increments to divide the angle space when rotating
actors within their subgroup space
measure = the measure which should be used to determine the rotation --
1=compactness
2=density
extreme =1 if the radius of an actor to it's anchor should be considered in
determining rotation (the larger the radius, the more the influence).
0 otherwise