Approximating Point Process Likelihoods with GLIM

Author(s): Mark Berman and T. Rolf Turner

Reviewed work(s):
Source: Journal of the Royal Statistical Society. Series C (Applied Statistics), Vol. 41, No. 1
(1992), pp. 31-38
Published by: Wiley for the Royal Statistical Society
Stable URL: http://www.jstor.org/stable/2347614 .
Accessed: 08/01/2013 07:54

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
data=spruces
. n <- data$n
xx <- quadscheme(unmark(data),nd=100,method="grid")
Xn <- as.ppp(xx,win=data$window)
Xn
w <- dirichlet.weights(Xn)
length(w)
sum(w)-area.owin(data$windo)
#fitting homogeneous Poisson process
log.lik.poiss <- function(par){
n*log(par[1])- area.owin(data$window)* par[1]
}

optim(0.5,log.lik.poiss, control=list(fnscale=-1),hessian=TRUE)
#fitting inhomogeneous Poisson process see page 97 of spatstat
log.lik.i.poiss <- function(par){
sum(par[1]+par[2]*data$x+par[3]*data$y)-sum(w*exp( par[1]+par[2]*Xn$x+par[3]*Xn$y ))
}
opt <- optim(c(0,0,0),log.lik.i.poiss, control=list(fnscale=-1),hessian=T)
eigen(opt$hessian)

Wiley and Royal Statistical Society are collaborating with JSTOR to digitize, preserve and extend access to
Journal of the Royal Statistical Society. Series C (Applied Statistics).

http://www.jstor.org

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
AppL Statist. (1 992)
41, No. 1, pp. 31-38

ApproximatingPoint Process Likelihoods

with GLIM
By MARK BERMANt and T. ROLF TURNER

CSIRO, Sydney,Australia

[Received August 1987. Final revisionJuly1990]

SUMMARY
This paper shows how approximatemaximumlikelihoodestimationforfairlygeneralpointprocesses
on the line can be performedwith GLIM. The approximation is based on a weighted sum
approximationto an integralin the likelihood.Various weightingschemes are brieflyexamined. The
methodology is illustratedwith an example, and its extension to Poisson processes in higher
dimensions is brieflydescribed.
Keywords: GLIM; Likelihood;Pointprocess; Poisson process; Weights

1. Introduction
Althoughmodelsforpointprocesseson thelineand theirstatisticalanalysisabound
(Cox and Lewis, 1966;Cox and Isham, 1980),therehas been,untilrecently, relatively
littlemodellingof the dependenceof pointprocesseson covariatesand theirasso-
ciatedstatisticalanalysis.Probablytheearliestmajor modelsof thistypewereCox's
(1972) modulatedPoisson processesand theirgeneralization,and Berman's (1981)
modulated gamma processes. Both papers only seriouslyaddressed questions of
hypothesistesting(via the likelihoodratiotest),but did not satisfactorily
solve the
problemof parameterestimation.More recently,therehave been several papers
devotedto seismologicalapplications,e.g. Ogata and Akaike (1982), Ogata (1983,
1988), Ogata and Katsura(1986) and Vere-Jonesand Ozaki (1982). In thesepapers,
therelationshipbetweenthepointprocessand thecovariates,whilebeingnumerically
convenient,can giveriseto likelihoodswhichcan be negativeforcertainparameter
values. To overcomethisproblem,most of the above-mentioned researchersuse a
penalizedlikelihoodapproachto estimatetheparameters.As a result,the statistical
propertiesof theseestimatorsare not clear.
Anotherrecentdevelopmentin this fieldis the use of generalizedlinearmodels
(GLMs). Becker(1983, 1986) and Brillinger and Preisler(1986) use theGLM frame-
workforspecificinhomogeneousPoisson processes.However,theydo notexploitthe
fullpotentialofthisframework. In thispaper,we shallshowhowa fairlylargeclass of
point processes on the line, whichincorporatedependenceon covariates,can be
modelledas GLMs; thisclass includesmodels(or slightvariantsof them)used in all

tAddressfor correspondence:CSIRO Divison of Mathematicsand Statistics,PO Box 218, Lindfield,New South

Wales 2070, Australia.

C) 1992 RoyalStatisticalSociety 0035-9254/92/41031 $2.00

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
32 BERMAN AND TURNER

theabove-mentioned papers,exceptforthemodulatedgammaprocess.The practical

advantage of using a properlydefinedGLM formulationis that it enables the
maximizationof an approximationto the likelihoodby means of the well-known
computerpackage GLIM (Payne, 1985), ratherthan for instancemaximizinga
penalizedlikelihoodas is done in theabove-mentioned seismologicalpapers.This, in
turn,enablestheuse of theasymptotically optimalstatisticalpropertiesof maximum
likelihood(ML) estimationforlarge classes of pointprocesses(Ogata, 1978). Our
hope is that,notonlywillpointprocessspecialistssee how certaintechnicalproblems
can be overcomewiththe aid of GLIM, but also that otherstatisticians,already
familiarwithGLIM, will feelmorecomfortablewhenpresentedwithpointprocess
problems.
At thispoint,we mustalso mentiontheworkof Hurley(1990) whichhas developed
concurrently withours,has strongconnectionswiththepresentpaperand has cometo
our attentionduringtherefereeing process.
The presentpaper is organizedas follows. Section 2 containssome basic point
process theory,and in particulara generalformulafor the likelihoodof a point
processon the line. This is illustratedwithan exampletakenfromOgata (1983). In
Section3, we showhow GLIM can be usedto approximatethelikelihood,and we use
thisapproximationto analysea modifiedversionof Ogata's model. The theoryof
Section3 is extendedto spatialPoissonprocessesinSection4, whileSection5 contains
a summaryand conclusions.

2. Basic Theoryand an Example

Suppose thatwe observean orderlypointprocess(Cox and Isham, 1980) on (0, T1
withconditionalintensity
function(CIF)
Xo(x IFx) = lim(Pr{1 eventin [x,x + A)I FX} /A), (2.1)
whereFX representsthe historyof the process (and possibly other explanatory
stochastic processes) prior to x, and 0 = (01, . . ., Op) is a vector of unknown
parametersrequiringestimation.Assumethatthe CIF is of the form
,~p
Xo(xIFx)= g E OkQk(XIFx)3 (2.2)

whereg( ) is a monotoniccontinuousfunctionand Qk(xI Fx), k= 1, . . ., p, is a

sequence of covariateswhose values are knownforall x e [0, T]. Each Qk may be
purelya functionofx (e.g. xk) or an observedstochasticprocess.Wheng(y) = exp y
and all the Qk are independentofFX,equation(2.2) represents theintensity
function
of Cox's (1972) modulatedPoisson process.
Let the event times in (0, T1 of the observed point process be xl < x2 < . . . < x.
Suppose thatwe wishto computetheML estimatorsof theOj. The log-likelihoodfor
any orderlypointprocesscan be written
nT
L(H)-L(; xl,. . ., xn;T)= E logXo(xiIFx)- | X(xIFx) dx. (2.3)
i=1 0

To make thistheorymore concrete,we introducean examplestudiedby Ogata

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
 APPROXIMATING POINT PROCESS LIKELIHOODS 33
(1983), section4.2. He examinedtherelationship
between'shallow' and 'deep' earth-
quakes in a regioncoveringthe NorthIsland of New Zealand from1946 to 1980.
Therewere58 shallowearthquakes(at timesxl, .. ., x58)and 84 deep earthquakes(at
times ul, . . ., U84). Ogata modelled the dependence of shallow earthquakes on deep
earthquakesusingthe CIF
J K
Xo(xIFx) = it + Y. oaj0(2x/T-1) + Z 1k (X-UI)k -1 exp{-d(x-ul)}, (2.4)
j=1 k=1 Ul<X

whereJ,KandO = {d, j1,jj= 1, .. ., J, k, k= 1, ... ., K} areunknownand4j( )is

the jth Legendre polynomial. The firstsummationin equation (2.4) represents
orthogonaltimetrendcomponents,whilethe second representsdirectlythe depen-
denceof shallowearthquakeson deep earthquakes.
For fixedJ,K and d, equation(2.4) is of theform(2.2) withg(y) =y. If we wishto
computethe ML estimatorsfor a model of the form(2.2), it is at firstglance
numericallyconvenientto use g(y) =y, for thenthe unknownparameterscan be
takenoutsidetheintegralsignin equation(2.3) and onlytheintegralsIT Qk (x IFx) dx,
k= 1, .. ., p, need be computed. Unfortunately, the use of g(y)=y does not
guaranteethenon-negativity of theCIF and hencethenon-negativity of itslikelihood
(see equations (2.1)-(2.3)). Therefore,for each (J, K, d) combinationexamined,
Ogata maximized
Lp(9, J,K) = L(O, J,K) - P(O, J,K), (2.5)
whereP(O, J,K) is a non-negative penaltyfunctiondesignedto keep the CIF non-
negativeforx E [0, T1. The non-linearparameterd was variedovera gridofcandidate
values,and equation(2.5) was maximizedoverthelinearparametersforeach d. The
ordersJ and K of themodelwerechosento minimizea modifiedAkaikeinformation
criterion(AIC), namely
AICp =-2 LP(9J,K, J,K) + 2(J+K+ 1), (2.6)
where9J,Ki thevalueof0 maximizing equation(2.5) givenJandK; theusual AIC has
L insteadofLp in equation(2.6). The modelchosenusingAICp had J= 7, K= 0. Thus
themodelchosencontainedonlytrendcomponents;thissuggeststhatshallowearth-
quakes are independentof deep earthquakesoccurringbeforehand.
Most of the seismologicallyorientedpapersmentionedearlierconsideredmodels
witha CIF of theform(2.2) and withg(y) =y. Because of theconsequentpossibility
of the CIF becoming negative,they also maximizeda penalized log-likelihood.
Unfortunately, thisobscuresboththenumericaland thestatisticalpropertiesof the
estimators.In thenextsection,we showhow GLIM can be used in conjunctionwitha
non-negative g( ) (e.g. g(y) = exp y) to maximizedirectlyan approximationto the
log-likelihood.The methodwill be applied to Ogata's data and comparedwithhis
originalanalysis.

3. Approximating
LikelihoodswithGLIM
There are two closelyrelatedways of using GLIM to maximizeequation (2.3)
approximately.Both involvea discretization
of thetimeaxis. One is a probabilistic
approximation,whiletheotheris numerical.In theprobabilisticapproximation,the

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
34 BERMAN AND TURNER

timeaxis is dividedintointervalsof equal width,thewidthbeingchosensufficiently

small to ensurethatnone of the intervalscontainsmorethan one event.The point
processis thenapproximatedby a 0-1 timeseries.GLIM can thenbe applied using
binomial ERRORS withsample size 1. Sometimesthese are replaced by Poisson
ERRORS. Details will not be given here. Recent examples of the use of such
probabilisticapproximationscan be foundin Becker(1983, 1986) and Brillinger and
Preisler(1986).
The numericalapproximation,whichwe introducehere,is moreflexiblethanthe
probabilisticapproximation.In particular,itallowsfortheuse ofintervalsof unequal
widths.The generalizationof thisidea is particularlyimportantwhen considering
spatialPoisson processes,whichwillbe discussedin Section4. The idea is similarto,
butgoes further than,an approximation usedbyClayton(1983) forsurvivaldata. The
firststepis to approximatetheintegralin equation (2.3) by a weightedsum:
T M
Xo(x IFx) dx- E wjXo(sjIFsj), (3.1)
0 k=1

wherethewjareweightsdetermined
bysomequadratureruleand 0 s, < s2 < . . .

SM= T. Choose thesj so thatX- {xl,. ..., x} C {s1,S2, . ., SM} S. We shall call
pointsin S, X and S - X the designpoints,the data pointsand the dummypoints
respectively. Let Nj = 1 if sj E X and Nj = 0 otherwise.Then substitution
of equation
(3.1) into equation (2.3) gives
M
L(O) _ Wj{(Nj/wj) log Xo(sj FsJ)- Xo(sj
I1)}. (3.2)
j=1
The right-hand side of equation (3.2) is (modulo a constantnot dependingon 0) the
weightedlog-likelihoodof a familyof independentPoisson variateswithparameters
X0(sj IFsj). The maximizationof equation(3.2) can thusbe accomplishedin GLIM by
declaring{Nj/wj} as theYVARIATE vectorwithPoisson ERRORS, WEIGHT given
by { wj}, and theLINK functionto be g (LOG is thedefaultforPoisson errors).
Note the followingpoints:
GLIM does not object to the factthatthe 'Poisson' variablesNj/wjand the
(a)
weightswjare non-integer;
(b) GLIM does objectto the wjbeingnegative;
(c) theparameterestimates,likelihoodvaluesand covariancematricesproduced
by GLIM when maximizingthe approximatelog-likelihood(3.2) are valid
approximationsto analogous functionsof thetruelog-likelihood(2.3) in the
sensethat,ifM --+oo in such a waythatapproximation(3.1) becomesexact,
theapproximating functionsof equation(3.2) willconvergeto theanalogous
functionsof equation (2.3).
Often,anydummypointswillbe chosento be equispaced,thenumberbeingchosen
accordingto the(numerical)accuracyof estimationrequired.The necessityof using
thedata pointswillimpose a non-uniform spacingof thedesignpoints,however.
Simpleschemesforweightsare obtainedeitherby
(a) a linear approximation to X(x) on [sj, sj? j,i]= 1, . ., M- 1 or
(b) a quadratic approximationto Xo(x) on [s2j-1, S2j 1], j= 1, * *, I (M- 1)

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
 APPROXIMATING POINT PROCESS LIKELIHOODS 35

(assuming that M is odd).

The linear approximation will always produce positive weights. However, ifs2j -
s2j-1 and s2j+1 - s2j are verydifferent,the quadratic approximation leads to negative
weightsto which GLIM objects. This is easily overcome by adding an odd number of
equispaced dummy points in each interval(xi, xi+1), i = 0, 1, ..., n (xO-=O,x,11 = T).
Such a schemeensuresthats2j+1 - s2j = s2j - s2j-1 forall j.

3.1. Return to Example

We generalized Ogata's CIF (2.4) to
J K
Xo(xIFx) = g t, + Z cvj j(2xI T-1) + E E (x-ul) exp d(x-ul) (3.3)
L =I k=1 ul<x

We tried both the IDENTITY link g(y) =y (i.e. Ogata's model but without any
penalty function) and the LOG link g(y) = exp y. Because of its complexity,we have
only considered versions of equation (3.3) withthe value of d (1.07 x 10-4) chosen by
Ogata. For a range of (J, K) values, we used GLIM to maximize log-likelihood (3.2)
withX0(x IFx) given by equation (3.3). We used both linear and quadratic weights. The
total number M of design points used varied between the minimumpossible, 60, and
1093. With the IDENTITY link, we found that GLIM stopped for some (J, K, M)
combinations because X (sj IFj) was negative for somej. However, no problems were
encountered using the LOG link. With this link function,we calculated the AIC for
each (J, K) combination (i.e. equation (2.6) with Lp replaced by L). Often there are
several models with AIC values very close to the minimum. It is important in such
cases to estimate the log-likelihood accurately; however, if our model is fixed and we
are only interested in parameter estimates and their standard errors, accurate
estimation of the log-likelihood may be less important. A CIF such as equation (3.3)
will tend to oscillate more as J or K increases, by analogy with ordinarypolynomials.
Hence, as a general rule, the larger J or K is, the more design points will be needed to
estimateL(O) accurately. This is borne out by the example considered here. In Table 1,
we give the AIC values forvarious (J, K) combinations using the LOG link and linear
weightswith M= 60 and M= 1061.
Table 1 shows clearly how the two approximations to the AIC for each (J, K)
combination become more discrepant broadly speaking as J or K increases. For all

TABLE 1
AIC valuesfor a model withCIF givenbyequation (3.3) usingg(x) = exp x, d= 1.07x 10 4and linear
weightswithM= 60 and M= 1061

K AIC valuefor thefollowingvaluesof J and M:

J
0 2 4 8 12
M
60 1061 60 1061 60 1061 60 1061 60 1061

0 743.9 743.9 727.6 727.4 730.4 729.5 730.5 723.6 735.5 727.9
1 725.7 725.3 729.5 729.1 732.1 731.2 732.0 724.5 737.4 729.6
2 727.6 727.1 731.5 731.1 730.6 729.3 733.9 725.8 738.7 731.6

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
36 BERMAN AND TURNER

(J,K) combinationsexamined, the approximationsbased on both linear and

quadraticweightsappearedto stabilizeas M approached1000.
For themodelwitha LOG link,itwas foundthattheminimumAIC value achieved
was 723.6, and thatthiswas achievedwhenJ= 8, K= 0. Thus, as in Ogata's analysis,
the model chosen containsonly timetrendcomponents.A graph of the resulting
estimatedCIF, togetherwithOgata's estimateand a yearlyhistogramof thedata, is
shownin Fig. 1. The two estimatedCIFs are mostlysimilarin generalappearance.
Exceptforthespuriousverticalasymptotes at theendpoints,we feelthatourestimate
reflectsthe behaviourof the data-as encapsulatedby the histogram-somewhat
betterthan does Ogata's estimate.This can be seen forinstancein the levelling-off
around1000 daysand thepeaks near6000 and 11000days.The leastappealingaspect
of our fitis thepresenceof thespuriousverticalasymptotes.Thisis largelydue to the
factthat&!8 ( = 1.821)is positiveand so is 48 inbothtails.Hence,&848 dominatesinthe
tails,and exponentiating magnifiesthisdominance.(We confirmedthisby plotting
thefittedcurvewithouttheterm&!848. The fitwas muchmorelikeOgata's in thetails
and theasymptotesdisappeared.However,theminimumAIC valuewhenJ= 7, K= 0
was 731.7, far largerthan that for J= 8, K= 0.) This is the unfortunateprice we
appearto haveto pay iflikelihoodmethodsare to be used. We believethatitis worth
payingbecause theseare only edge effects(effectswhichalso arise in manyother
branchesof statisticalmodelling)and because the inferencesto be drawnfromthe
data analysis rest more securelyon the asymptoticstatisticalpropertiesof ML
estimationthando thosebased on penalizedlikelihoodmethods.

4. Spatial Poisson Processes

The ideas in Section3 can also be appliedto spatialPoisson processes,but not,in

0
o

C)

W0- t

l I

0 2000 4000 6000 8000 10000 12000 14000

days

Fig. 1. Ogata (.)... and Berman-Turner

(-----) estimatesof theCIF forOgata's (1983) earthquake
data, togetherwitha yearlyhistogram( .) of thedata

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
 APPROXIMATING POINT PROCESS LIKELIHOODS 37
general,to otherspatialpointprocesses.The reasonforthisis thatthelog-likelihood
formostnon-Poissonspatialpointprocessescannotbe written in a relatively
simple
formsuchas equation(2.3). In thissection,we shallbrieflyoutlinethemethodology
forthePoisson case. For simplicity,
we shallonlyconsiderthetwo-dimensional case;
themethodologyextends,in principle,to higherdimensionsin an obviousway. LetA
denote the region on which the Poisson process is observed and let xl,. . ., x" denote
theeventlocationsin A.
By analogywithequation (2.2), assumethattheintensity
functionof theprocess,
X4(x),is of theform

X (x) = g _ Ok Qk (X) (4.1)

where g( ) is again a monotonic continuous function and Qk(x), k= 1, . ., p, is a

sequenceof covariateswhosevaluesare knownforall x e A. Unlikein equation(2.2),
X0(x)and Qk(x)are independentofthepointprocesseverywhere else,and so thepoint
processis Poisson. The log-likelihoodforthePoisson processis givenby
n
L(O;x,, . . .,xn;A)= E logXo(xi)- | (x) dx. (4.2)
i=l A

Note thesimilarity of equation(4.2) to equation(2.3). The onlyexampleof fitting a

spatialPoisson processwithintensity functionoftheform(4. 1) ofwhichwe areaware
is due to Vere-Jonesand Musmeci(1986). As in the papersmentionedearlier,they
assumethatg(y) =y and circumvent theproblemof a negativeintensity functionby
usingpenalized likelihood estimation as describedearlier.
Because of the similarityof equations (2.3) and (4.2), the theoryof Section 3
leadingto theuse of GLIM extendseasilyto spatialPoisson processes.In particular,
log-likelihood(3.2) holdswithXG(sjIFj) beingreplacedbyXo(sj), wheres, . . ., sMis a
set of designpointsin A whichincludethedata points,and the wj are non-negative
weightsobtainedfroman appropriatetwo-dimensional quadraturerule.The various
GLIM declarationsare just as in the one-dimensionalcase, and the parameter
estimates,likelihoodvalues and covariancematricesproducedby GLIM are valid in
thesensedescribedin Section 3.
We turnbrieflyto a discussionof appropriatetwo-dimensional quadraturerules.
Perhapsthemostobvioussuchrule(especiallyifA is a rectangle)is the'productrule'
(Davis and Rabinowitz,1975)in whichA is dividedintoa gridof smallerrectangles,
theverticesof therectanglesbeingthedesignpoints.The associatedquadraturerule
has weightswhichare typicallythe productsof weightsobtained fromunivariate
quadraturerules.Unfortunately, therequirement thatthe designpointsincludethe
data pointsimpliesthat,fora productrule,thereare at least(n + 2)2 designpointsifA
is a rectangle.This quickly strainsboth centralprocessorunit (CPU) time and
GLIM's storagerequirements.
Attractivealternativesto productrules are obtained using eitherthe Dirichlet
tessellationor the Delaunay triangulation(Green and Sibson, 1978; Ahuja and
Schachter, 1983), especiallyif only approximateparameterestimatesand their
standarderrorsare required.Typically,we woulduse then data pointstogetherwitha
coarserectangulargridof dummypointsas a base forconstructing thetessellationor
triangulation.Two simpleassociatedweightingschemesare as follows.In the first

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions
38 BERMAN AND TURNER

scheme,approximateXG(x)byXG(sj)on theDirichlettilesurrounding
sj. In thesecond
scheme,approximateX0(x)by a plane on the Delaunay trianglecontainingx. Both
schemesare guaranteedto producenon-negativeweights.
Our experiencehas been thatDirichletor Delaunay-basedquadraturerulesgive
answerscomparable withthose fromproductrules using far fewerdesign points
withconsequentsubstantialsavingsin CPU time. Details are available fromthe
authors.

5. SummarizingRemarks
In thispaper, we have shownhow approximateML estimationforfairlygeneral
point processeson the line and Poisson processesin space can be performedwith
GLIM. In addition,we have demonstrated how theuse of penaltyfunctionscan be
avoided,and how, forspatialdata, storageand CPU timeproblemscan be overcome
by usingDirichlettessellationsor Delaunay triangulations
insteadof productrules.
We hopethatourpaperwillencouragethegreateruse ofpointprocessmodellingwith
theaid of GLIM.

References
Ahuja, N. and Schachter,B. J. (1983) PatternModels, sect. 1.3. New York: Wiley.
Becker,N. G. (1983) Analysisof data froma singleepidemic.Aust. J. Statist.,25, 191-197.
(1986) A generalizedlinearmodellingapproachto theanalysisof data froma singleepidemic.In
Proc. Pacific StatisticalCongr.,Auckland,pp. 464-467. Amsterdam:North-Holland.
Berman,M. (1981) Inhomogeneousand modulatedgammaprocesses.Biometrika,68, 143-152.
Brillinger,D. R. and Preisler,H. K. (1986) Two examplesof quantaldata analysis:(a) multivariate
point
process,(b) puredeathprocessin an experimental design.13thInt. BiometricsConf., Seattle.
Clayton,D. G. (1983) Fittinga generalfamilyof failure-time distributions
usingGLIM. Appl. Statist.,
32, 102-109.
Cox, D. R. (1972) The statisticalanalysis of dependenciesin point processes. In StochasticPoint
Processes(ed. P. A. W. Lewis), pp. 55-66. New York: Wiley.
Cox, D. R. and Isham, V. (1980) Point Processes. London: Chapman and Hall.
Cox, D. R. and Lewis,P. A. W. (1966) The StatisticalAnalysisof Seriesof Events.London: Methuen.
Davis, P. J. and Rabinowitz,P. (1975) Methods of NumericalIntegration,sect. 5.6. New York:
AcademicPress.
Green, P. J. and Sibson, R. (1978) ComputingDirichlettessellationsin the plane. Comput. J., 21,
168-173.
Hurley,M. A. (1990) Fittingsome inhomogeneouspointprocesses.To be published.
Ogata, Y. (1978) Asymptoticbehaviourof themaximumlikelihoodestimatorsforthestationarypoint
processes.Ann. Inst. Statist.Math., 30, 243-261.
(1983) Likelihoodanalysisof pointprocessesand itsapplicationsto seismologicaldata. Bull. Int.
Statist.Inst., 50, 943-961.
(1988) Statisticalmodelsforearthquakeoccurrencesand residualanalysisforpointprocesses.J.
Am. Statist.Ass., 83, 9-27.
Ogata, Y. and Akaike, H. (1982) On linearintensity models formixeddoublystochasticPoisson and
self-excitingpointprocesses.J. R. Statist.Soc. B, 44, 102-107.
Ogata, Y. and Katsura, K. (1986) Point-processmodels with linearlyparameterizedintensityfor
applicationto earthquakedata. J. Appl. Probab. 23A, 291-310.
Payne, C. D. (1985) The GLIM SystemRelease 3.77 Manual. Oxford:NumericalAlgorithmsGroup.
Vere-Jones,D. and Musmeci,F. (1986) Orthogonalpolynomialestimatesforpointprocessdensities.
TechnicalReport.Instituteof Statisticsand OperationsResearch,VictoriaUniversity of Wellington.
Vere-Jones, D. and Ozaki, T. (1982) Some examplesof statisticalinferenceappliedto earthquakedata.
Ann. Inst. Statist.Math., 34, 189-207.

This content downloaded on Tue, 8 Jan 2013 07:54:34 AM

All use subject to JSTOR Terms and Conditions