Graphics Reference
In-Depth Information
Instead of visualizing the structure of samples or variables in a given dataset, re-
searchersmaybeinterestedinvisualizingimagescollectedwithcertainformats.Usu-
ally the target images are collected with various types of noise pattern and it is neces-
sary to apply statistical or mathematical modelling to remove or diminish the noise
structurebeforethepossiblegenuineimagescanbevisualized.JörgPolzehlandVlad-
imirSpokoiny present one such novel adaptive smoothing procedurein reconstruct-
ing noisy images for better visualization.
he continuing increase in computer power has had many different impacts on
statistics. Computationally intensive smoothing methods are now commonplace, al-
thoughtheywereimpossibleonlyafewyearsago.AdrianBowmangives anoverview
oftherelationsbetweensmoothingandvisualization.Yuan-chinChang,Yuh-JyeLee,
Hsing-Kuo Pao, Mei-Hsien Lee and Su-Yun Huang investigate the impact of kernel
machinemethodsonanumberofclassicaltechniques:principalcomponents,canon-
ical correlation and cluster analysis. hey use visualizations to compare their results
with those from the original methods.
Cluster analyses have oten been a bit suspect to statisticians. he lack of formal
models in the past and the di
culty of judging the success of the clusterings were
both negative factors. Fritz Leisch considers the graphical evaluation of clusterings
andsomeofthepossibilitiesforasoundermethodologicalapproach.
Multivariate categorical data were di
cult to visualize in the past. he chapter by
David Meyer, Achim Zeileis and Kurt Hornik describes fairly classical approaches
for low dimensions and emphasizes the link to model building. Heike Hofmann de-
scribes the powerful tools of interactive mosaicplots that have become available in
recent years, not least through her own efforts, and discusses how different varia-
tions of the plot form can be used for gaining insight into multivariate data features.
Alfred Inselberg, the original proposerof parallel coordinate plots, offers an over-
view of this approach to multivariate data in his usual distinctive style. Here he con-
siders in particular classification problems and how parallel coordinate views can be
adaptedandamendedtosupportthiskindofanalysis.
Most analyses using graphics make use of a standard set of graphical tools, for
example, scatterplots, barcharts, and histograms. Han-Ming Wu, ShengLi Tzeng and
Chun-houh Chen describe a different approach, built around using colour approxi-
mations for individual values in a data matrix and applying cluster analyses to order
the matrix rows and columns in informative ways.
For many years Bayesians were primarily theoreticians. hanks to MCMC meth-
ods they are now able to also apply their ideas to great effect. his has led to new
demands in assessing model fit and the quality of the results. Jouni Kerman, An-
drew Gelman, Tian Zheng and Yuejing Ding discuss graphical approaches for tack-
ling these issues in a Bayesian framework.
Without sotware todrawthe displays,graphic analyis isalmost impossible nowa-
days. Junji Nakano, Yamamoto Yoshikazu and Keisuke Honda are working on Java-
based sotware to provide support for new developments, and they outline their ap-
proach here. Many researchers are interested in providing tools via the Web.Yoshiro
Yamamoto, Masaya Iizuka and Tomokazu Fujino discuss using XML for interactive
statistical graphics and explain the issues involved.
