Download Case Studies in Data Analysis PDF

TitleCase Studies in Data Analysis
File Size11.6 MB
Total Pages269
Document Text Contents
Page 1

Editorial Policy
for the publication of proceedings of conferences
and other multi-author volumes

Lecture Notes aim to report new developments - quickly, informally, and at a high level. The
following describes criteria and procedures for multi-author volumes. For convenience we refer
throughout to "proceedings" irrespective of whether the papers were presented at a meeting.

The editors of a volume are strongly advised to inform contributors about these points at an early
§ 1. One (or more) expert participant( s) should act as the scientific editor( s) of the volume. They
select the papers which are suitable (cf.§§2-S) for inclusion in the proceedings, and have them
individually refered (as for a journal). It should not be assumed that the published proceedings
must reflect conference events in their entirety. The series editors will normally not interfere with
the editing of a particular proceedings volume - except in fairly obvious cases, or on technical
matters, such as described in §§2-S. The names of the scientific editors appear on the cover and
title-page of the volume.
§2. The proceedings should be reasonably homogeneous i.e. concerned with a limited and well
defined area. Papers that are essentially unrelated to this central topic should be excluded. One or
two longer survey articles on recent developments in the field are often very useful additions. A
detailed introduction on the subject of the congress is desirable.
§3. The final set of manuscripts should have at least 100 pages and preferably not exceed a total
of 400 pages. Keeping the size below this bound should be achieved by stricter selection of articles
and NOT by imposing an upper limit on the length of the individual papers.
§4. The contributions should be of a high mathematical standard and of current interest. Research
articles should present new material and not duplicate other papers already published or due to be
published. They should contain sufficient background and motivation and they should present
proofs, or at least outlines of such, in sufficient detail to enable an expert to complete them. Thus
summaries and mere announcements of papers appearing elsewhere cannot be included, although
more detailed versions of, for instance, a highly technical contribution may well be published
elsewhere later. Contributions in numerical mathematics may be acceptable without formal
theorems/proofs provided they present new algorithms solving problems (previously unsolved or
less well solved) or develop innovative qualitative methods, not yet amenable to a more formal
Surveys, if included, should cover a sufficiently broad topic, and should normally not just review
the author's own recent research. In the case of surveys, exceptionally, proofs of results may not
be necessary.
§S. "Mathematical Reviews" and "Zentralblatt fUr Mathematik" recommend that papers in
proceedings volumes carry an explicit statement that they are in final form and that no similar paper
has been or is being submitted elsewhere, if these papers are to be considered for a review.
Normally, papers that satisfy the criteria of the Lecture Notes in Statistics series also satisfy this
requir~ment, but we strongly recommend that each such paper carries the statement explicitly.
§6. Proceedings should appear soon after the related meeting. The publisher should therefore
receive the complete manuscript (preferably in duplicate) including the Introduction and Table of
Contents within nine months of the date of the meeting at the latest.
§7. Proposals for proceedings volumes should be sent to one of the editors of the series or to
Springer-Verlag New York. They should give sufficient information on the conference, and on
the proposed proceedings. In particular, they should include a list of the expected contributions
with their prospective length. Abstracts or early versions (drafts) of the contributions are helpful.

Page 2

Lecture Notes in Statistics
Edited by S. Fienberg, J. Gani, K. Krickeberg,
1. Oikin, and N. Wermuth


Page 134

Extreme-value Analysis 129
of Canadian Wind Speeds





0 0.60

w 0.50
t= 0.40 en




0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00


FIGURE 4: A Pop plot of the Toronto monthly maxima of the transformed wind speed, illustrating the
goodness of fit of the Gumbel distribution. The diagonal lines represent the upper and lower
10% and 5% critical values for the KS statistic.

goodness-of-fit statistics determined separately for each month. The two month-by-
month analyses of goodness of fit revealed very few differences.

When the parameters were adjusted month by month, the statistics and plots did
reveal a slightly closer fit than in the case when the parameters were held fixed for all
months. However, even in the latter case, little evidence was found that the chosen
model is inappropriate. The results ofthe KS and AD tests were very consistent with
each other, and neither test rejected the null hypothesis more frequently than would
be expected in a set of 60 independent tests (12 tests at each station). The results ofthe
AD test are summarized in Table 4. The actual significance level of the goodness-of-
fit tests should be very close to the nominal level, because the model specified in the
tests is determined from a sample 12 times larger than the sample used in the tests.

In conclusion, it was evident that the Gumbel distribution is a suitable model for
monthly maximum wind speed after the data have been transformed, and that the
same model is appropriate for all months of the year.

Page 135

130 Case Studies in Data Analysis
Case Study No.5, Pages 119-144

TABLE 4: Anderson-Darling goodness-of-fit statistic indicating the difference between the
Gumbel distribution fitted to the monthly maxima of transformed wind speed (all months
combined) and corresponding empirical cumulative distribution functions determined
separately for each month.


Month Vancouver Regina Toronto Montreal Shearwater

Jan. 1.782 2.121" 0.297 0.534 0.898
Feb. 0.861 1.165 1.599 0.628 0.283
Mar. 0.502 2.904 b 0.817 0.605 0.770
Apr. 0.483 0.452 1.224 0.455 0.437
May 0.364 0.750 0.827 0.597 2.238"
Jun. 1.168 0.540 0.923 0.878 0.856
Jul. 0.795 0.967 0.457 0.382 1.977'
Aug. 0.885 0.746 0.879 0.567 0.511
Sep. 0.533 2.787 b 0.692 0.250 2.411 •
Oct. 0.931 0.361 0.684 0.823 0.749
Nov. 0.647 1.044 0.947 3.971 ' 0.758
Dec. 1.095 0.487 1.295 1.392 0.310

·Significant at the 10% level.
bgignificant at the 5% level.
'Significant at the 1% level.


The "raw" data for this analysis, the filtered wind speeds, may be thought of as
fluctuations about a stochastic process varying on long time scales. The extreme-
value analysis must also take the variability of this underlying process, the 365-day
moving average of wind speed, into account. The best way in which to accomplish
this would be to conduct an extreme-value analysis with it as well. However, there
are several difficulties with such an analysis. The 365-day moving averages contain
trends and display the effects of interventions. They also display fluctuations on long
time scales, and thus the record of 38 years (35 years at Shearwater) simply does not
contain enough information to do a reliable extreme-value analysis even if the
nonstationary components of the record could somehow be taken into account.

The fact that the data exhibit trends and interventions due to human activities
implies than an estimate of the variability of the long-time-scale component of the
wind speed (the 365-day moving average) must be more of a statement of policy than
anything else. It is therefore left to policy planners working in concert with climatol-
ogists to make some statement regarding the future annual mean maximum daily
wind speeds.

Lacking a "policy" statement regarding the future effects of urbanization and
interventions on wind speed, the following rather conservative approach was taken.
First it was noted that the trend at all five stations has been decreasing with time.
Secondly, it was assumed that:

(I:) there will be continuing urbanization of the observing sites and hence the
downward trends will not be reversed;

(2) management and standardization of observing sites will continue to improve
(meaning that it is much less likely today than, say, 20 years ago that an instrument
will be placed in an inappropriate location); and

Page 268

Vol. 52: P.K. Goel, T. Ramalingam, The Matching Methodology:
Some Statistical Properties. vili, 152 pages, 1989.

Vol. 53: B.C. Arnold, N. Balakrishnan, Relations, Bounds and
Approximations for Order Statistics. ix, 173 pages, 1989.

Vol. 54: K.R. Shah, B.K. Sinha, Theory of Optimal Designs. viii,
171 pages, 1989.

Vol. 55:L. McDonald,B. Manly,J. Lockwood,J. Logan (Editors),
Estimation and Analysis of Insect Populations. Proceedings,
1988. xiv, 492 pages, 1989.

Vol. 56: J.K. Lindsey, The Analysis of Categorical Data Using
GUM. v,168 pages,1989.

Vol. 57: A. Decarli, B.J. Francis, R. Gilchrist, G.U.H. Seeber
(Editors), Statistical Modelling. Proceedings, 1989. ix, 343 pages,

Vol. 58: O.E.Bamdorl'f-Nielsen,P. Blesild,P.S.Eriksen,Decom-
position and Invariance of Measures, and Statistical Transforma-
tion Models. v, 147 pages, 1989.

Vol. 59: S. Gupta, R. Mukerjee, A Calculus for Factorial Arrange-
ments. vi, 126 pages, 1989.

Vol. 60: L. GyOtfi, W. Hardie, P. Sarda, Ph. Vieu, Nonparametric
Curve Estimation from Time Series. vili, 153 pages, 1989.

Vol. 61: J. Breckling, The Analysis of Directional Time Series:
Applications to Wind Speed and Direction. viii, 238 pages, 1989.

Vol. 62: J.e. Akkerboom, Testing Problems with Linear or Angu-
lar Inequality Constraints. xii, 291 pages, 1990.

Vol. 63: J. Pfanzag!, Estimation in Semiparametric Models: Some
Recent Developments. iv, 112 pages, 1990.

Vol. 64: S. Gabler, Minimax Solutions in Sampling from Finite
Populations. v, 132 pages, 1990.

Vol. 65: A. Janssen, D.M. Mason, Non-Standard Rank Tests. vi,
252 pages, 1990.

Vol. 66: T. Wright, Exact Confidence Bounds when Sampling
from Small Finite Universes. xvi, 431 pages, 1991.

Vol. 67: MA. Tanner, Tools for Statistical Inference: Observed
Data and Data Augmentation Methods. vi, 110 pages, 1991.

Vol. 68: M. Taniguchi,HigherOrder Asymptotic Theory forTime
Series Analysis. viii, 160 pages, 1991.

Vol. 69: NJ.D. Nagelkerke, Maximum Likelihood Estimation of
Functional Relationships. v, 110 pages, 1992.

Vol. 70: K. lida, Studies on the Optimal Search Plan. viii, 130
pages, 1992.

Vol. 71: E.M.R.A. Engel, A Road to Randomness in Physical
Systems. ix, 155 pages, 1992.

Vol. 72: J.K. Lindsey, The Analysis of Stochastic Processes using
GUM. vi, 294 pages, 1992.

Vol. 73: B.C. Arnold, E. Castillo, J.-M. Sarabia, Conditionally
Specified Distributions. xiii, 151 pages, 1992

Vol. 74: P. Barone, A. Frigessi, M. Piccioni (Editors), Stochastic
Models, Statistical Methods, and Algorithms in Image Analysis.
vi, 258 pages, 1992.

Vol. 75: P.K. Goel, N.S. Iyengar (Editors), Bayesian Analysis in
Statistics and Econometrics. xi, 410 pages, 1992.

Vol. 76: L. Bondesson, Generalized Gamma Convolutions and
Related Classes of Distributions and Densities. viii, 173 pages,

Vol. 77: E. Mammen, When Does Bootstrap Work? Asymptotic
Results and, 196 pages, 1992.

Vol. 78: L. Fahnneir, B. Francis, R. Gilchrist, G. Tutz (Editors),
Advances in GUM and Statistical Modelling: Proceedings of the
GUM92 Conference and the 7th International Worlcshop on
Statistical Modelling, Munich, 13-17 July 1992. ix, 225 pages,

Vol. 79: N. Schmitz. Optimal Sequentially Planned Decision
Procedures. xii, 209 pages, 1992.

Vol. 80: M. Fligner,J. Verducci (Editors), Probability Models and
Statistical Analyses for Ranking Data. xxii, 306 pages, 1992

Vol. 81: P. Spirtes, C. Glymour, R. Scheines, Causation, Predic-
tion, and Search. xxiii, 526 pages, 1993.

Vol. 82: A. Koroste1ev and A. Tsyhakov, Minimax Theory of
Image Reconstruction. xii, 268 pages, 1993.

Vol. 83:C. Gatsonis,J.Hodges,R. Kass,N. Singpurwalla (Editors),
Case Studies in Bayesian Statistics. xii, 437 pages, 1993.

Vol. 84: S. Yamada, Pivotal Measures in Statistical Experiments
and Sufficiency. vii, 129 pages, 1994.

Vol. 85: P. Doukhan, Mixing: Properties and Examples. xi, 142
pages, 1994.

Vol. 86: W. Vach, Logistic Regression with Missing Values in the
Covariates. xi, 139 pages, 1994.

Vol. 87: J. M0ller, Lectures on Random Voronoi Tessellations.vii,
134 pages, 1994.

Vol. 88: J.E. Kolassa, Series Approximation Methods in
Statistics. viii, 150 pages, 1994.

Vol. 89:P. Cheeseman,R.W. Oldford (Editors), Selecting Models
From Data: Artificial Intelligence and Statistics IV. x,487 pages,

Vol. 90: A. Csenki, Dependability for Systems with a Partitioned
State Space: Markov and Semi-Marlwv Theory and Computational
Implementation. x, 241 pages, 1994.

Vol. 91: J.D. Malley, Statistical Applications ofJordan Algebras.
viii, 101 pages, 1994.

Vol. 92: M. Eerola, Probabilistic Causality in Longitudinal
Studies. vii, 133 pages, 1994.

Vol. 93: Bernard Van Cutsem (Editor), Classification and
Dissimilarity Analysis. xiv, 238 pages, 1994.

Vol. 94: Jane F. Gentleman and G.A. Whitmore (Editors), Case
Studies in Data Analysis. viii, 262 pages, 1994.

Page 269

General Remarks

Lecture Notes are printed by photo-offset from
the master-copy delivered in camera-ready form
by the authors of monographs, resp. editors of
proceedings volumes. For this purpose Springer-
Verlag provides technical instructions for the
preparation of manuscripts. Volume editors are
requested to distribute these to all contributing
authors of proceedings volumes. Some homo-
geneity in the presentation of the contributions
in a multi-author volume is desirable.

Careful preparation of manuscripts will help
keep production time short and ensure a satis-
factory appearance of the finished book. The
actual production of a Lecture Notes volume
normally takes approximately 8 weeks.

For monograph manuscripts typed or typeset
according to our instructions, Springer-Verlag
can, if necessary, contribute towards the prepa-
ration costs at a fixed rate.

Authors of monographs receive 50 free copies
of their book. Editors of proceedings volumes
similarly receive 50 copies of the book and are
responsible for redistributing these to authors
etc. at their discretion. No reprints of individual
contributions can be supplied. No royalty is
paid on Lecture Notes volumes.

Volume authors and editors are entitled to pur-
chase further copies of their book for their
personal use at a discount of 33.3% and other
Springer mathematics books at a discount of
20% directly from Springer-Verlag. Authors
contributing to proceedings volumes may pur-
chase the volume in which their article appears
at a discount of 20 %.

Springer-Verlag secures the copyright for each

Series Editors:

Professor S. Fienberg
Department of Statistics
Carnegie Mellon University
Pittsburgh, Pennsylvania 15213

Professor J. Gani
Stochastic Analysis Group, SMS
Australian National University
Canberra ACT 2601

Professor K. Krickeberg
3 Rue de L 'Estrapade
75005 Paris

Professor 1. Olkin
Department of Statistics
Stanford University
Stanford, California 94305

Professor N. Wermuth
Department of Psychology
Johannes Gutenberg University
Postfach 3980
D-6500 Mainz

Similer Documents