Professional Documents
Culture Documents
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.
Biometrika Trust is collaborating with JSTOR to digitize, preserve and extend access to Biometrika.
http://www.jstor.org
Nonparametrics
BiometrikaCentenary:
BY PETER HALL
Centrefor and itsApplications,
Mathematics AustralianNationalUniversity,
CanberraACT 0200,Australia
peter.hall@anu.edu.au
SUMMARY
Contributions in Biometrikato nonparametricstatisticsare surveyed,withparticular
emphasison morerecentwork.It is arguedthateven in thejournal's veryearlyyearsit
had interestsin smoothingmethodsfordensityestimation, and in techniquesfornonpara-
metricregression, in additionto its moreobviousfocuson 'classical'nonparametricstat-
isticssuch as rankingmethods.Biometrika's earlyinvolvement withpermutationtesting
has linksto its contemporary workon bootstrapmethods.Emphases on different types
of nonparametric statisticshave evolved over time,however.A rapidlyincreasingpro-
portionofthejournal'spages is devotedto papersthatincludesome use ofnonparametric
ideas or methods.
Somekeywords:Bandwidth;Bayesianmethods;Bootstrap;Categoricaldata; Classification;
Curveestimation;
Density estimation;Discrimination;Empirical likelihood;Jackknife;Kernel methods;Local polynomial
methods;Maximumpenalisedlikelihood;Nonparametriclikelihood;Nonparametricregression; Permutation
test;Rank test;Resampling;Smoothing;Survivalanalysis;Waveletmethods.
1. INTRODUCTION
It is perhapsimpossibleto formulatea formaldefinition of what constitutes'nonpara-
not least because thetermencompassessuch a wide varietyofmethods.
metricstatistics',
Mathematically, thedefinition ofnonparametric methodsas thosewherecharacterisation
oftheallowableclass ofmodelsrequiresan infinite numberofparametersis attractiveon
some grounds.It highlights the inappropriatenessof the term'nonparametric'; nonpara-
metricmodelsare notthosethatinvolveno parameter, butratherthosewherethenumber
ofparametersis extremely large.Such a definition
failsto distinguish between'nonpara-
metric'and 'semiparametric' methods,however.Furthermore, it does not helpin any way
among membersof the verydisparateclass of nonparametric
to differentiate techniques,
whichrangefrom rank and permutation teststo wavelet methods for curve estimation,
and fromthe bootstrapto conceptsof'nonparametriclikelihood'thatare at least halfa
centuryold.
In thelast quarter-century thisextraordinary has beenfostered
diversity by thedevelop-
mentofpowerful, high-speedcomputingtechnology, whichnowherein statisticshas been
feltmoreprofoundly thanin nonparametric methods.The developmenthas allowedsome
older nonparametrictechniques,formerly studiedlargelyfroma theoreticalviewpoint,
to be applied as practical,adaptivetools; and it has spawneda particularly wide range
of new statisticalmethodsin curveestimation,sample reuse techniques,nonparametric
conceptsoflikelihood,and otherareas.
2. RANK-BASED METHODS
Papers in Biometrikaplayeda major role in theearlydevelopmentofrankand permu-
tationtests,and includedmanyofM. G. Kendall's contributions.For example,Kendall's
4. RESAMPLING METHODS
4 1. Thejackknife
The jackknife,introducedby Quenouille(1949) in the Journalof theRoyal Statistical
Society,was an earlysamplereusemethod,inspiringa substantialliteraturein Biometrika.
In thatfirstpaper,about reducingthebias ofan estimatorofserialcorrelation,Quenouille
suggesteddividingthe sample into two half-samples. In a subsequentBiometrikapaper
(Quenouille,1956) he proposedthemethodthatwe knowtodayas thejackknife, so named
by Tukey. Quenouille (1956) listedfour'desirableproperties'of an estimator,the last
beingthatit be unbiased.He thenwenton to show thateach successiveapplicationof
his methodreducesthe order of asymptoticbias by a factorof n-', wheren denotes
sample size, and to prove that asymptoticvarianceis not affected.He exploredsmall-
sample propertiesof his methodin a primitive'simulationstudy',workingthroughthe
calculationsand theirimplicationsfor a single sample drawn froma normal N(2, 1)
population when n= 10.
A significant
portionof earlyresearchon thejackknifeand its properties, forexample
Tukey's(1958) discoverythatvariancecould be estimatedusingthejackknife,was pub-
lishedin NorthAmerica.To thepresentday,however,Biometrika contributes substantially
to the developmentof variantsof the method,as well as to applicationsto a verywide
range of problemsin statistics.In particular,Durbin (1959) was the firstto apply
'Quenouille'smethod',as he termedit,to ratioestimation.
Rao (1965) and Rao & Webster(1966) also used 'Quenouille'smethod'forratio esti-
mation,showingthat the optimal samplingfractionis 1. Hutchison(1971) compared
'Quenouille'smethod'forratioestimationwithfiveotherprocedures,includingthenaive
ACKNOWLEDGEMENT
BIOMETRIKA REFERENCES
ABRAMOVICH, F. & SILVERMAN, B. W. (1998). Waveletdecompositionapproachesto statisticalinverseprob-
lems.Biometrika 85, 115-29.
ABRAMSON, I. & MULLER, H. G. (1994). Estimating directionfieldsin autonomousequationmodels,withan
applicationto systemidentification fromcross-sectional data. Biometrika 81, 663-72.
AITCHISON, J. & AITKEN, C. G. G. (1976). Multivariate binarydiscrimination bythekernelmethod.Bioinetr-ika
63, 413-20.
ANSLEY, C. F. & KOHN, R. (1982). A geometricalderivationof the fixedintervalsmoothingalgorithm.
Biometrika 69, 486-7.
ANSLEY, C. F. & KOHN, R. (1987). Efficient generalizedcross-validation forstatespace models.Bioinetrika
74, 139-48.
ANSLEY, C. F., KOHN, R. & WONG, C.-M. (1993). Nonparametric splineregressionwithpriorinformation.
Bioinetrika 80, 75-88.
ANTONIADIS, A., GIJBELS, I. & GRPGOIRE, G. (1997). Model selectionusingwaveletdecompositionand appli-
cations.Biometrika 84, 751-63.
AZZALINI, A. (1981). A note on the estimationof a distribution functionand quantilesby a kernelmethod.
Bioinetr-ika68, 326-8.
AZZALINI, A. (1984). Estimationand hypothesis testingof autoregressive timeseries.Biometr-ika 71, 85-90.
AZZALINI, A., BOWMAN, A. W. & HARDLE, W. (1989). On the use of nonparametric regressionformodel
checking.Biometr-ika 76, 1-11.
BAGGERLY, K. A. (1998). Empiricallikelihoodas a goodness-of-fit measure.Biornetirika 85, 535-47.
BANKS, D. L. (1988). Histosplinesmoothingthe Bayesianbootstrap.Biometrika 75, 673-84.
BERAN, R. (1987). Prepivoting to reducelevelerrorof confidencesets.Biomet7ika 74, 457-68.
BESAG, J. E. & CLIFFORD, P. (1989). GeneralizedMonte Carlo significance tests.Bioinetrika 76, 633-42.
BESAG, J. E. & CLIFFORD, P. (1991). SequentialMonte Carlo p-values.Biometrika 78, 301-4.
BLOM, G. (1976). Some properties of incompleteU-statistics. Biomnet7ika63, 573-80.
B6HNING, D., SCHLATTMANN, P. & DIETZ, E. (1996). Intervalcensoreddata: A note on the nonparametric
maximumlikelihoodestimatorof thedistribution function.Biometrika 83, 462-6.
Boos, D. D. & MONOHAN, J. F. (1986). Bootstrapmethodsusingpriorinformation. Biometrikca 73, 77-83.
BOOTH, J. G. & HALL, P. (1994). Monte Carlo approximationand the iteratedbootstrap.Biometrika 81,
331-40.
BOWMAN, A. W. (1980). A note on consistencyof the kernelmethodforthe analysisof categoricaldata.
Biometrika 67, 682-4.
BOWMAN, A. W. (1984). An alternative methodof cross-validationforthe smoothingof densityestimates.
Biometrika 71, 353-60.
BOWMAN, A. W., HALL, P. & PRVAN, T. (1998). Bandwidthselectionforthesmoothing ofdistribution functions.
Biometrika 85, 799-808.
BOWMAN, A. W., HALL, P. & TITTERINGTON, D. M. (1984). Cross-validation in nonparamnetricestimationof
probabilitiesand probabilitydensities.Biometrika 71, 341-51.
BRETH, M., MARITZ, J. S. & WILLIAMS, E. J.(1978). On distribution-freelower confidence limits for the mean
of a nonnegativerandomvariable.Biometrika 65, 529-34. Comment (1979), 66, 689.
BROWN, M. (1984). On thechoice of varianceforthelog ranktest.Biometr-ika 71, 65-74.
BRUCE, A. G. & GAO, H.-Y. (1996). Understanding WaveShrink:Varianceand bias estimation.Biometrika
83, 727-45.
BRUNK, H. D. (1978). Univariatedensityestimationby orthogonalseries.Biornetr-ika 65, 521-8.
BUCKLEY, M. J. (1991). Detectinga smooth signal: optimalityof cusum based procedures.Biornetrika 78,
253-62.
BUCKLEY, M. J. (1994). Fast computationof a discretizedthin-platesmoothingspline for image data.
Biomet7-ika 81, 247-58.
BUCKLEY, M. J. & EAGLESON, G. K. (1986). Assessinglarge sets of rankcorrelations. Biomnetrika 73, 151-7.
BUCKLEY, M. J. & EAGLESON, G. K. (1989). A graphicalmethodfor estimatingthe residual variance in
nonparametric regression.Biometrika 76, 203-10.
BUCKLEY, M. J.,EAGLESON, G. K. & SILVERMAN, B. W. (1988). The estimationof residualvariancein non-
parametricregression.Bioinetrika 75, 189-99.
BURMAN, P. (1989). A comparativestudyofordinary cross-validation,v-foldcross-validation and therepeated
learning-testing methods.Biomet7ika 76, 503-14.
BURMAN, P., CHOW,E. & NOLAN, D. (1994). A cross-validation methodfor dependentdata. Biometrika
81, 351-8.
BURMAN, P. & NOLAN, D. (1995). A generalAkaike-typecriterionformodel selectionin robustregression.
Biometrika 82, 877-86.
CHAMBERS, E. A. (1946). Statisticaltechniquesin applied psychology. Bio,netrika 33, 269-73.
OTHER REFERENCES
CLEVELAND, E. S. & LOADER, C. L. (1996). Smoothingbylocal regression:
principlesand methods.In Statistical
Theoryand Computational Aspectsof Smoothing, Ed. W. Hardle and M. G. Schimek,pp. 10-49. Vienna:
Physica-Verlag.
Cox, D. R. (1972). Regressionmodelsand lifetables(withDiscussion).J. R. Statist.Soc. B 74, 187-200.
EFRON, B. (1979). Bootstrapmethods:anotherlook at thejackknife.Ann.Statist.7, 1-26.