Download the Full Version of textbook for Fast Typing at textbookfull.

com

Uncertainty Modelling in Data Science Sébastien

Destercke (Ed.)

https://textbookfull.com/product/uncertainty-modelling-in-
data-science-sebastien-destercke-ed/

OR CLICK BUTTON

DOWNLOAD NOW

Download More textbook Instantly Today - Get Yours Now at textbookfull.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Beginning Data Science in R: Data Analysis, Visualization,

and Modelling for the Data Scientist 1st Edition Thomas
Mailund
https://textbookfull.com/product/beginning-data-science-in-r-data-
analysis-visualization-and-modelling-for-the-data-scientist-1st-
edition-thomas-mailund/
textboxfull.com

Soft Computing in Data Science 5th International

Conference SCDS 2019 Iizuka Japan August 28 29 2019
Proceedings Michael W. Berry
https://textbookfull.com/product/soft-computing-in-data-science-5th-
international-conference-scds-2019-iizuka-japan-
august-28-29-2019-proceedings-michael-w-berry/
textboxfull.com

Advanced Data Analysis & Modelling in Chemical Engineering

1st Edition Denis Constales

https://textbookfull.com/product/advanced-data-analysis-modelling-in-
chemical-engineering-1st-edition-denis-constales/

textboxfull.com

Modelling based Teaching in Science Education 1st Edition

John K. Gilbert

https://textbookfull.com/product/modelling-based-teaching-in-science-
education-1st-edition-john-k-gilbert/

textboxfull.com
The Data Science Design Manual 1st Edition Steven S.
Skiena

https://textbookfull.com/product/the-data-science-design-manual-1st-
edition-steven-s-skiena/

textboxfull.com

Research in Data Science Ellen Gasparovic

https://textbookfull.com/product/research-in-data-science-ellen-
gasparovic/

textboxfull.com

Frontiers In Data Science Matthias Dehmer

https://textbookfull.com/product/frontiers-in-data-science-matthias-
dehmer/

textboxfull.com

Uncertainty analysis of experimental data with R 1st

Edition Benjamin David Shaw

https://textbookfull.com/product/uncertainty-analysis-of-experimental-
data-with-r-1st-edition-benjamin-david-shaw/

textboxfull.com

Subsurface Environmental Modelling Between Science and

Policy Dirk Scheer

https://textbookfull.com/product/subsurface-environmental-modelling-
between-science-and-policy-dirk-scheer/

textboxfull.com
Advances in Intelligent Systems and Computing 832

Sébastien Destercke
Thierry Denoeux · María Ángeles Gil
Przemyslaw Grzegorzewski
Olgierd Hryniewicz Editors

Uncertainty
Modelling
in Data
Science
Advances in Intelligent Systems and Computing

Volume 832

Series editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
e-mail: [email protected]
The series “Advances in Intelligent Systems and Computing” contains publications on theory,
applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all
disciplines such as engineering, natural sciences, computer and information science, ICT, economics,
business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the
areas of modern intelligent systems and computing such as: computational intelligence, soft computing
including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms,
social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and
society, cognitive science and systems, Perception and Vision, DNA and immune based systems,
self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics including human-machine
teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent data analysis,
knowledge management, intelligent agents, intelligent decision making and support, intelligent network
security, trust management, interactive entertainment, Web intelligence and multimedia.
The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings
of important conferences, symposia and congresses. They cover significant recent developments in the
field, both of a foundational and applicable character. An important characteristic feature of the series is
the short publication time and world-wide distribution. This permits a rapid and broad dissemination of
research results.

Advisory Board
Chairman
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
e-mail: [email protected]
Members
Rafael Bello Perez, Universidad Central “Marta Abreu” de Las Villas, Santa Clara, Cuba
e-mail: [email protected]
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
e-mail: [email protected]
Hani Hagras, University of Essex, Colchester, UK
e-mail: [email protected]
László T. Kóczy, Széchenyi István University, Győr, Hungary
e-mail: [email protected]
Vladik Kreinovich, University of Texas at El Paso, El Paso, USA
e-mail: [email protected]
Chin-Teng Lin, National Chiao Tung University, Hsinchu, Taiwan
e-mail: [email protected]
Jie Lu, University of Technology, Sydney, Australia
e-mail: [email protected]
Patricia Melin, Tijuana Institute of Technology, Tijuana, Mexico
e-mail: [email protected]
Nadia Nedjah, State University of Rio de Janeiro, Rio de Janeiro, Brazil
e-mail: [email protected]
Ngoc Thanh Nguyen, Wroclaw University of Technology, Wroclaw, Poland
e-mail: [email protected]
Jun Wang, The Chinese University of Hong Kong, Shatin, Hong Kong
e-mail: [email protected]

More information about this series at http://www.springer.com/series/11156

Sébastien Destercke Thierry Denoeux
•

María Ángeles Gil Przemyslaw Grzegorzewski

•

Olgierd Hryniewicz
Editors

Uncertainty Modelling
in Data Science

123
Editors
Sébastien Destercke Przemyslaw Grzegorzewski
CNRS, Heudiasyc Faculty of Mathematics and Information
Sorbonne universités, Université Science
de technologie de Compiègne Warsaw University of Technology
Compiegne, France Warsaw, Poland

Thierry Denoeux and

CNRS, Heudiasyc
Sorbonne universités, Université Systems Research Institute
de technologie de Compiègne Polish Academy of Sciences
Compiegne, France Warsaw, Poland

María Ángeles Gil Olgierd Hryniewicz

Department of Statistics and Operational Department of Stochastic Methods,
Research and Mathematics Didactics Systems Research Institute
University of Oviedo Polish Academy of Sciences
Oviedo, Asturias Warsaw, Poland
Spain

ISSN 2194-5357 ISSN 2194-5365 (electronic)

Advances in Intelligent Systems and Computing
ISBN 978-3-319-97546-7 ISBN 978-3-319-97547-4 (eBook)
https://doi.org/10.1007/978-3-319-97547-4

Library of Congress Control Number: 2018950094

© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

This volume contains the peer-reviewed papers presented at the 9th International
Conference on Soft Methods in Probability and Statistics (SMPS 2018), which
was held in conjunction with the 5th International Conference on Belief Functions
(BELIEF 2018) on 17–21 September 2018 in Compiègne, France. The series of
biannual International Conference on Soft Methods in Probability and Statistics
started in Warsaw in 2002. It then successfully took place in Oviedo (2004), Bristol
(2006), Toulouse (2008), Oviedo/Mieres (2010), Konstanz (2012), Warsaw (2014)
and Rome (2016). SMPS and BELIEF 2018 were organized by the Heudiasyc
laboratory at the Université de Technologie de Compiègne.
Over the last decades, the interest for extensions and alternatives to probability
and statistics has significantly grown in areas as diverse as reliability,
decision-making, data mining and machine learning, optimization, etc. This interest
comes from the need to enrich existing models, in order to include different facets
of uncertainty such as ignorance, vagueness, randomness, conflict or imprecision.
Frameworks such as rough sets, fuzzy sets, fuzzy random variables, random sets,
belief functions, possibility theory, imprecise probabilities, lower previsions,
desirable gambles all share this goal, but have emerged from different needs. By
putting together the BELIEF and SMPS conferences, we hope to increase the
interactions and discussions between the two communities and to converge towards
a more unified view of uncertainty theories.
We also think that the advances, results and tools presented in this volume are
important in the ubiquitous and fast-growing fields of data science, machine
learning and artificial intelligence. Indeed, an important aspect of some of the
learned predictive models is the trust one places in them. Modelling carefully and
with principled methods, the uncertainty associated to the data and the models is
one of the means to increase this trust, as the model will then be able to distinguish
reliable predictions from less reliable ones. In addition, extensions such as fuzzy
sets can be explicitly designed to provide interpretable predictive models,
facilitating user interaction and increasing their trust.

v
vi Preface

The joint event collected 76 submissions, each reviewed by at least two

reviewers. Twenty-nine of these are included in the present volume, which contains
contributions of foundational, methodological and applied nature, on topics as
varied as imprecise data handling, linguistic summaries, model coherence, impre-
cise Markov chains and robust optimization. The resulting proceedings was easily
produced through the use of EasyChair.
We would like to thank all the persons that made this volume and this confer-
ence possible which include all contributing authors, organizers, programme
committee members that help to build such an attractive programme. We are
especially grateful to our three invited speakers, Thomas Augustin (Ludwig-
Maximilians-Universität München) for his talk “Belief functions and valid statistical
inference”, Scott Ferson (University of Liverpool) for his talk “Non-Laplacian
uncertainty: practical consequences of an ugly paradigm shift about how we handle
not knowing” and Ryan Martin (North Carolina State University) for his talk
“Belief functions and valid statistical inference”. We would like to thank all our
generous sponsors: Elsevier and the International Journal of Approximate
Reasoning, the Laboratory of excellence MS2T, the Heudiasyc laboratory, the
International Society of Information Fusion (ISIF), the Compiègne University of
Technology, the city of Compiègne. Furthermore, we would like to thank the editor
of the Springer series of Advances in Soft Computing, Prof. Janusz Kacprzyk, and
Springer-Verlag for their dedication to the production of this volume.

June 2018 Sébastien Destercke

Thierry Denoeux
María Ángeles Gil
Przemyslaw Grzegorzewski
Olgierd Hryniewicz
Organization

Programme Committee

Alessandro Antonucci IDSIA

Thomas Augustin Department of Statistics, Univ. of Munich (LMU)
Giulianella Coletti University of Perugia
Olivier Colot Université Lille 1
Ana Colubi University of Oviedo
Frank Coolen Department of Mathematical Sciences, Durham
University
Inés Couso University of Oviedo
Fabio Cuzzolin Oxford Brookes University
Fabio D’Andreagiovanni Université de Technologie de Compiègne,
UMR CNRS Heudiasyc
Pierpaolo D’Urso Sapienza University of Rome
Bernard De Baets Ghent University
Thierry Denoeux Université de Technologie de Compiègne,
UMR CNRS Heudiasyc
Sébastien Destercke Université de Technologie de Compiègne,
UMR CNRS Heudiasyc
Jean Dezert Onera
Didier Dubois Université de Paul Sabatier, Toulouse,
UMR IRIT
Fabrizio Durante Università del Salento, Lecce
Zied Elouedi Institut Supérieur de Gestion de Tunis
Ramasso Emmanuel Ecole Nationale Supérieure de Mécanique et des
Microtechniques, FEMTO-ST
Maria Brigida Ferraro Department of Statistical Sciences, Sapienza
University of Rome
Maria Angeles Gil Alvarez University of Oviedo

vii
viii Organization

Lluis Godo Artificial Intelligence Research Institute,

IIIA - CSIC
Gil González-Rodríguez University of Oviedo
Michel Grabisch Université Paris I
Przemyslaw Grzegorzewski Systems Research Institute Polish Academy
of Sciences
Olgierd Hryniewicz Polish Academy of Sciences, Systems Research
Institute
Radim Jirousek University of Economics
Anne-Laure Jousselme NATO Centre for Maritime Research
and Experimentation (CMRE)
Frank Klawonn Ostfalia University of Applied Sciences
Vaclav Kratochvil UTIA
Rudolf Kruse University of Magdeburg
Eric Lefevre LGI2A Université d’Artois
Liping Liu The University of Akron
María Asunción Lubiano Universidad de Oviedo
Arnaud Martin Université de Rennes1/IRISA
Ronald W. J. Meester Vrije Universiteit Amsterdam
David Mercier Université d’Artois
Radko Mesiar Slovak University of Technology Bratislava
Rombaut Michele Gipsa-lab
Daniel Milan Institute of Computer Science,
The Czech Academy of Sciences
Enrique Miranda University of Oviedo
Ignacio Montes Carlos III University of Madrid
Susana Montes University of Oviedo
Serafin Moral University of Granada
Frédéric Pichon Université d’Artois
Benjamin Quost Université de Technologie de Compiègne,
UMR CNRS Heudiasyc
Ana Belén Ramos Guajardo University of Oviedo
Johan Schubert Swedish Defence Research Agency
Ferson Scott University of Liverpool, Institute for Risk
and Uncertainty
Prakash P. Shenoy University of Kansas School of Business
Beatriz Sinova University of Oviedo
Martin Stepnicka IRAFM, University of Ostrava
Barbara Vantaggi Sapienza University of Rome
Jirina Vejnarova Institute of Information Theory and Automation
of the AS CR
Paolo Vicig University of Trieste
Liu Zhunga Northwestern Polytechnical University
Contents

Imprecise Statistical Inference for Accelerated Life Testing Data:

Imprecision Related to Log-Rank Test . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Abdullah A. H. Ahmadini and Frank P. A. Coolen
Descriptive Comparison of the Rating Scales Through Different Scale
Estimates: Simulation-Based Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Irene Arellano, Beatriz Sinova, Sara de la Rosa de Sáa,
María Asunción Lubiano, and María Ángeles Gil
Central Moments of a Fuzzy Random Variable Using the Signed
Distance: A Look Towards the Variance . . . . . . . . . . . . . . . . . . . . . . . . 17
Rédina Berkachy and Laurent Donzé
On Missing Membership Degrees: Modelling Non-existence,
Ignorance and Inconsistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Michal Burda, Petra Murinová, and Viktor Pavliska
Characterization of Conditional Submodular Capacities:
Coherence and Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Giulianella Coletti, Davide Petturiti, and Barbara Vantaggi
Some Partial Order Relations on a Set of Random Variables . . . . . . . . 42
Bernard De Baets and Hans De Meyer
A Desirability-Based Axiomatisation for Coherent Choice Functions . . . 46
Jasper De Bock and Gert de Cooman
Cycle-Free Cuts of the Reciprocal Relation Generated by Random
Variables that are Pairwisely Coupled by a Frank Copula . . . . . . . . . . 54
Hans De Meyer and Bernard De Baets
Density Estimation with Imprecise Kernels:
Application to Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Guillaume Dendievel, Sebastien Destercke, and Pierre Wachalski

ix
x Contents

Z-numbers as Generalized Probability Boxes . . . . . . . . . . . . . . . . . . . . . 68

Didier Dubois and Henri Prade
Computing Inferences for Large-Scale Continuous-Time Markov
Chains by Combining Lumping with Imprecision . . . . . . . . . . . . . . . . . 78
Alexander Erreygers and Jasper De Bock
Robust Fuzzy Relational Clustering of Non-linear Data . . . . . . . . . . . . . 87
Maria Brigida Ferraro and Paolo Giordani
Measures of Dispersion for Interval Data . . . . . . . . . . . . . . . . . . . . . . . . 91
Przemyslaw Grzegorzewski
A Maximum Likelihood Approach to Inference Under Coarse Data
Based on Minimax Regret . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Romain Guillaume and Didier Dubois
Monitoring of Time Series Using Fuzzy Weighted
Prediction Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Olgierd Hryniewicz and Katarzyna Kaczmarek-Majer
Control Charts Designed Using Model Averaging Approach for Phase
Change Detection in Bipolar Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Katarzyna Kaczmarek-Majer, Olgierd Hryniewicz, Karol R. Opara,
Weronika Radziszewska, Anna Olwert, Jan W. Owsiński,
and Sławomir Zadrożny
An Imprecise Probabilistic Estimator for the Transition Rate Matrix
of a Continuous-Time Markov Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Thomas Krak, Alexander Erreygers, and Jasper De Bock
Imprecise Probability Inference on Masked Multicomponent System . . . 133
Daniel Krpelik, Frank P. A. Coolen, and Louis J. M. Aslett
Regression Ensemble with Linguistic Descriptions . . . . . . . . . . . . . . . . . 141
Jiří Kupka and Pavel Rusnok
Dynamic Classifier Selection Based on Imprecise Probabilities:
A Case Study for the Naive Bayes Classifier . . . . . . . . . . . . . . . . . . . . . 149
Meizhu Li, Jasper De Bock, and Gert de Cooman
Case Study-Based Sensitivity Analysis of Scale Estimates
w.r.t. the Shape of Fuzzy Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
María Asunción Lubiano, Carlos Carleos, Manuel Montenegro,
and María Ángeles Gil
Compatibility, Coherence and the RIP . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Enrique Miranda and Marco Zaffalon
Contents xi

Estimation of Classification Probabilities in Small Domains

Accounting for Nonresponse Relying on Imprecise Probability . . . . . . . 175
Aziz Omar and Thomas Augustin
Beyond Doss and Fréchet Expectation Sets . . . . . . . . . . . . . . . . . . . . . . 183
Juan Jesus Salamanca
Empirical Comparison of the Performance of Location Estimates
of Fuzzy Number-Valued Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Beatriz Sinova and Stefan Van Aelst
Continuity of the Shafer-Vovk-Ville Operator . . . . . . . . . . . . . . . . . . . . 200
Natan T’Joens, Gert de Cooman, and Jasper De Bock
Choquet Theorem for Random Sets in Polish Spaces and Beyond . . . . . 208
Pedro Terán
Generalising the Pari-Mutuel Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Chiara Corsato, Renato Pelessoni, and Paolo Vicig
A Net Premium Model for Life Insurance Under a Sort of Generalized
Uncertain Interest Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Dabuxilatu Wang
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
 Imprecise Statistical Inference for
Accelerated Life Testing Data:
Imprecision Related to Log-Rank Test

Abdullah A. H. Ahmadini1,2(B) and Frank P. A. Coolen1

1
Department of Mathematics Science, Durham University, Durham, UK
2
Department of Mathematics, Faculty of Science, Jazan University,
Jazan, Saudi Arabia
{abdullah.ahmadini,frank.coolen}@durham.ac.uk

Abstract. In this paper we consider an imprecise predictive inference

method for accelerated life testing. The method is largely nonparametric,
with a basic parametric function to link different stress levels. We discuss
in detail how we use the log-rank test to provide adequate imprecision
for the link function parameter.

1 Introduction

To determine the reliability of a new product in a relatively short period of time,

we use lifespan testing to assess a product, system or component. Accelerated life
testing (ALT) is a methodology that is common in practice, where items tested
under normal use (normal “level”), are not expected to fail for a very long time,
far beyond the time available for testing. In ALT, units are exposed to higher
stress levels (e.g. lightbulbs on at a higher than normal voltage) to induce failure
more rapidly. There are several typical designs for lifespan testing, including
constant-, step- and progressive-stress testing. These types of stress loading in
accelerated testing are explained in detail by Nelson [5].
In this paper, we assume the Arrhenius model for the analysis of ALT with
failure data under a constant level of stress. The Arrhenius model link function
is a standard model for failure time data resulting from ALT. This model is used
primarily in situations when the failure mechanism is driven by temperature,
and has been applied to various maintenance problems in engineering [5]. If the
Arrhenius model provides a realistic link between the different stress levels, then
the observations transformed from the increased stress levels to the normal stress
level should not be distinguishable. According to this model, an observation ti
at the stress level i, subject to stress Ki , can be transformed to an observation
at the normal stress level K0 , by the equation
 γ 
i→0 i e K0
t =t γ (1)
e Ki

c Springer Nature Switzerland AG 2019

S. Destercke et al. (Eds.): SMPS 2018, AISC 832, pp. 1–8, 2019.
https://doi.org/10.1007/978-3-319-97547-4_1
2 A. A. H. Ahmadini and F. P. A. Coolen

where Ki is the accelerated temperature at level i (Kelvin), K0 is the normal

temperature at level 0 (Kelvin), γ is the parameter of the Arrhenius model.
Testing equality of the survival distribution of two or more independent
groups often requires a nonparametric statistical test. There are several non-
parametric test procedures that can be used to test equality of the survival
distributions, a popular one is the log-rank test [4,6]. We use the log-rank test
to find the interval of values of the parameter γ of the Arrhenius link function
for which we do not reject the null hypothesis of two or more groups of fail-
ure data, possibly including right-censored observations, coming from the same
underlying distribution. This can be interpreted such that, for such values of γ,
the combined data at stress level K0 are well mixed. In this paper, to illustrate
our method, we assume that there are no right-censored data. In this case, the
log-rank test is equal to the Wilcoxon test [3]. Note that we also assume to have
failure data at the normal stress level, we comment further on this in the final
section.
In this paper we propose a new log-rank test based method for predictive
inference on a future unit functioning at the normal stress level. We apply the
use of the log-rank test to compare the survival distributions of two groups,
in combination with the Arrhenius model finding the accepted interval of γ
values according to the null hypothesis. The log-rank test is used for the pairwise
comparison of stress levels, leading to an interval of values for γ. This interval is
used to transform the data from the increased stress levels to the normal stress
level. Then, the ultimate aim is inference at the normal stress level. We consider
nonparametric predictive inference at the normal stress level combined with the
Arrhenius model linking observations at different stress levels.
Nonparametric Predictive Inference (NPI) is a statistical method which
allows inferences about future observations to be made based on past data [2]
using imprecise probability [1]. Given ordered observations x1 < x2 < ... < xn ,
and defining x0 = 0 and xn+1 = ∞. The NPI lower and upper survival functions
for a future observation Xn+1 are
n−j
S Xn+1 (t) = , for t ∈ (xj , xj+1 ), j = 0, ..., n. (2)
n+1

n+1−j
S Xn+1 (t) = , for t ∈ (xj , xj+1 ), j = 0, ..., n. (3)
n+1
The difference between the upper and lower survival functions, called impreci-
sion, is non-zero because of the limited inferential assumptions made, and reflects
the amount of information in the data.
This paper is organized as follows. Section 2 introduces the main idea of
imprecise predictive inference based on ALT and log-rank test. The main novelty
of our approach is that the imprecision results from a classical nonparametric
test, which is the log-rank test, integrated with the Arrhenius function to link
different stress levels. In Sect. 3 we explain why we do not use a single log-rank
test on all stress levels. In Sect. 4 our method is illustrated in two examples.
Section 5 presents some concluding remarks.
 Imprecise Statistical Inference for Accelerated Life Testing Data 3

2 Predictive Inference Based on ALT Data and Pairwise

Log-Rank Test

In this section we present new predictive inference based on ALT data and
the log-rank test. The proposed new method consists of two steps. First, the
pairwise log-rank test is used between the stress level Ki and K0 , to get the
intervals [γ i , γ i ] of values γi for which we do not reject the null hypothesis that
the data transformed from level i to level 0, and the original data from level 0,
come from the same underlying distribution, where i = 1, ..., m. With these m
pairs (γ i , γ i ), we define γ = min γ i and γ = max γ i .
Second, we apply the data transformation using γ (γ) for all levels to get
transformed data at level 0 which leads to NPI lower (upper) survival function
S (S). Note that each observation at an increased stress level is transformed
to an interval at level 0, where the interval tends to be larger if a data point
was originally from a higher stress level. If the model fits really well, we expect
most γ i to be quite similar, and also most γ i . The NPI lower survival function is
attained when all data observations at increased stress levels are transformed to
the normal stress level using γ, and the NPI upper survival function results from
the use of γ. If the model fits poorly, γ i are likely to differ a lot, or γ i differ a lot,
or both. Hence, in case of poor model fit, the resulting interval [γ, γ] tends to be
wider than in the case of good model fit. A main novelty of our method is that
imprecision results from pairwise comparisons via a classical test, we comment
further on this in the next section.

3 Why Not to Use a Single Log-Rank Test on All Levels

In our novel method discussed in Sect. 2, we use pairwise log-rank tests between
stress level Ki and K0 . An alternative would be to use one log-rank test for
the data at all stress levels combined. We now explain why this would not lead
to a sensible method of imprecise statistical inference. Suppose we would test
the null hypothesis that data from all stress levels, transformed using parameter
value γa , originate from the same underlying distribution. Let [γ a , γ a ] be the
interval of such values γa for which this hypothesis is not rejected. If the model
fits very well, we would expect γ a to be close to the γ from Sect. 2 and also γ a
to be close to γ. If however, the model fits poorly, the [γ a , γ a ] interval may be
very small or even empty. Therefore, this leads to less imprecision if the model
fits poorly, and that is the reason why we do the pairwise levels and take the
minimum and the maximum of γ i and γ i , respectively. Then, we are interested in
prediction of one future observation at the normal stress level K0 . So, using the
observations transformed from the increased stress levels K1 , ..., Km as well as
the original data obtained at the normal stress level K0 , we apply NPI to derive
lower and upper survival functions for as described in Sect. 1. The examples in
Sect. 4 illustrate the proposed method of Sect. 2 as well as the problem if we
would use the combined approach for all levels.
4 A. A. H. Ahmadini and F. P. A. Coolen

Table 1. Failure times at three temperature levels

Case K0 = 283 K1 = 313 K2 = 353 K1 = 313 (*1.4) K2 = 353 (*0.8)

1 2692.596 241.853 74.557 338.595 59.645
2 3208.336 759.562 94.983 1063.387 75.987
3 3324.788 769.321 138.003 1077.050 110.402
4 5218.419 832.807 180.090 1165.930 144.072
5 5417.057 867.770 180.670 1214.878 144.560
6 5759.910 1066.956 187.721 1493.739 150.176
7 6973.130 1185.382 200.828 1659.535 160.662
8 7690.554 1189.763 211.913 1665.668 169.531
9 8189.063 1401.084 233.529 1961.517 186.823
10 9847.477 1445.231 298.036 2023.323 238.429

Table 2. Accepted γ for log-rank test (Ex1)

Significance level 0.99 0.95 0.90

Stress level Lower γ Upper γ Lower γ Upper γ Lower γ Upper γ
K1 K0 3901.267 6563.545 4254.053 6251.168 4486.491 6017.435
K2 K0 4161.086 5555.130 4499.174 5442.667 4638.931 5353.034
K2 K1 K0 4156.263 5652.662 4464.828 5478.451 4499.174 5419.662

4 Examples
In this section we present two examples. In example 1 we simulated data at
all levels that correspond to the model for the link function we assume for the
analysis. In example 2 we change these data such that the assumed link function
will not provide a good fit anymore. Together, these examples illustrate our novel
imprecise method, from Sect. 2, as well as the problem that could occur if we
used the log-rank test on all stress levels combined, as discussed in Sect. 3.

4.1 Example 1
The method proposed in Sect. 2 is illustrated in an example, which presents
the temperature-accelerated lifespan test. Data are simulated at three temper-
atures. The normal temperature condition was K0 = 283 and the increased
temperatures stress levels were K1 = 313 and K2 = 353 Kelvin. Ten obser-
vations were simulated from a fully specified model, using the Arrhenius link
function in combination with a Weibull distribution at each temperature. The
Arrhenius parameter γ was set at 5200, and the Weibull distribution at K0 had
shape parameter 3 and scale parameter 7000. This model keeps the same shape
parameter at each temperature, but the scale parameter are linked by the Arrhe-
nius relation, which led to scale parameter 1202.942 at K1 and 183.0914 at K2 .
 Imprecise Statistical Inference for Accelerated Life Testing Data 5

Ten units were tested at each temperature, for a total of 30 units used in the
study. The failure times, in hours, are given in Table 1.
To illustrate the log-rank test method using these data, we assume the Arrhe-
nius link function for the data. Note that our method does not assume a paramet-
ric distribution at each stress level. The pairwise log-rank test is used between
K1 and K0 and between K2 and K0 to derive the intervals [γ i , γ i ] of values γi for
which we do not reject the null hypothesis with regard to the well-mixed data
transformation. The resulting intervals [γ i , γ i ] are giving in the first two rows of
Table 2, for three test significance levels. Of course, for larger significance level
the intervals become wider.
According to the accepted intervals in Table 2, we can obtain the NPI lower
and upper survival functions by taking from the pairwise stress level K1 to K0 or
K2 to K0 always the minimum of the γ i and the maximum of the γ i with levels
of significance 0.99, 0.95 and 0.90 values. So, we take γ = min γ i = 3901.267 and
the γ = max γ i = 6563.545 of the pairwise K1 , K0 with 0.99 significance level,
γ = min γ i = 4254.053 and the γ = max γ i = 6251.168 of the pairwise K1 , K0
with 0.95 significance level, and γ = min γ i = 4486.491 and the γ = max γ i =
6017.435 of the pairwise K1 , K0 with 0.90 significance level then transformed
the data to the normal stress level, see Fig. 1(a). In this figure, the lower survival
function S is labeled as S (γ i ) and the upper survival function S is labeled as S
(γ i ). This figure shows that higher significance levels leads to more imprecision
for the NPI lower and upper survival functions.
To illustrate the effect of using the single log-rank test for all stress levels
simultaneously as discussed in Sect. 3, the final row in Table 2 provides the inter-
val [γ a , γ a ] of values γa for all the stress levels together. From this interval we
can again obtain the lower and upper survival functions using NPI, these are
presented in Fig. 1(b). In this example, the data were simulated precisely with
the link function as assumed in our method, so there is not much difference
between the lower and upper survival functions for corresponding significance
levels in Figs. 1(a) and (b). Example 2 will illustrate what happens if the model
does not fit well.

4.2 Example 2

To illustrate our method in case the model does not fit the data well, and also
to show what would have happened if we had used the joint log-rank test in our
method instead of the pairwise tests, we use the same data as in Example 1,
but we change some of these. In Scenario 1 (indicated as Ex 2-1 in Fig. 1), we
multiple the data at level K1 by 1.4. In Scenario 2 (Ex 2-2), we do the same and
in addition we multiply the data at level K2 by 0.8. The resulting data values
are given in the last two columns in Table 1.
For these two scenarios, we have repeated the analysis as described in Exam-
ple 1. The resulting intervals of γ values are given in Tables 3 and 4. Note that
for significance level 0.90 in Scenario 2 the null hypothesis of the joint log-rank
test would be rejected for all values γa , hence we report an empty interval, so
6 A. A. H. Ahmadini and F. P. A. Coolen

clearly our method would not work if we had used this joint test instead of the
pairwise tests.
The NPI lower and upper survival functions in Figs. 1(c) and (e), using our
method as discussed in Sect. 2, have more imprecision. Note that the lower sur-
vival function is identical in both scenarios as the same γ is used, this is because
the increased values at K1 have resulted in smaller values for γ 1 and γ 1 and the
γ in our method is equal to the γ 1 in these cases. In Scenario 2, the observations
at level K2 have decreased, leading to larger γ 2 and γ 2 values, and this leads to
the upper survival functions increasing in comparison to Scenario 1.
If we would have used the joint long-rank test instead of the pairwise tests,
as discussed in Sect. 3, then imprecision would have decreased in these two sce-
narios, as can be seen from Figs. 1(d) and (f). Note that in Fig. 1(f) there are no
lower and upper survival functions corresponding to the use of the joint log-rank
test for significance level 0.90, as this leads to an empty interval of γa values. As
mentioned in Sect. 3, if the model does not fit well, then we are going to sooner
reject the null hypothesis for all the three levels together, see Tables 3 and 4. So
we have a smaller range of values for which we do not reject the null hypoth-
esis. But if the model fits poorly, we actually want a larger range of values, so
increased imprecision. It is obvious that this is achieved by taking the minimum
of the γ i and the maximum of the γ i of the pairwise tests, hence this is our
proposed method in Sect. 2. This is illustrated by Figs. 1(a), (c) and (e).

Table 3. Accepted γ for log-rank test (Ex2-1)

Significance level 0.99 0.95 0.90

Stress level Lower γ Upper γ Lower γ Upper γ Lower γ Upper γ
K1 ∗ 1.4, K0 2907.787 5570.065 3260.574 5257.689 3493.011 5023.956
K2 K0 4161.086 5597.978 4499.174 5442.667 4638.931 5353.034
K2 K1 K0 4455.573 5568.468 4638.930 5368.780 4742.958 5257.689

Table 4. Accepted γ for log-rank test (Ex2-2)

Significance level 0.99 0.95 0.90

Stress level Lower γ Upper γ Lower γ Upper γ Lower γ Upper γ
K1 ∗ 1.4, K0 2907.787 5570.065 3260.574 5257.689 3493.011 5023.956
K2 ∗ 0.8, K0 4479.541 5916.433 4817.629 5761.121 4957.386 5671.488
K2 K1 K0 5031.547 5676.311 5220.356 5531.731 Empty set
Imprecise Statistical Inference for Accelerated Life Testing Data 7

Fig. 1. The NPI lower and upper survival functions.

8 A. A. H. Ahmadini and F. P. A. Coolen

5 Concluding Remarks
This paper has presented an exploration of the use of a novel statistical method
providing imprecise semi-parametric inference for ALT data, where the impreci-
sion is related to the log-rank test statistics. The proposed method applies the
use of the log-rank test to compare the survival distribution of pairwise stress
levels, in combination with the Arrhenius model finding the accepted interval of
γ values according to the null hypothesis. We explored imprecision through the
use of nonparametric test for the parameter of the link function between different
stress levels, which enabled us to transform the observations at increased stress
levels to interval-valued observations at the normal stress level and achieve fur-
ther robustness. We consider nonparametric predictive inference at the normal
stress level combined with the Arrhenius model linking observations at different
stress levels. We showed why, in our method, we use the imprecision from com-
bined pairwise log-rank tests, and not from a single log-rank test on all stress
levels together. The latter would lead to less imprecision if the model fits poorly,
while our proposed method then leads to more imprecision. In this paper, to illus-
trate basic idea of our novel method, we assumed that failure data are available
at all stress levels including the normal stress level. This may not be realistic. If
there are no failure data at the normal stress level, or only right-censored obser-
vations, then we can apply our method using a higher stress level as the basis
for the combinations, so transform data to that stress level. Then the combined
data at that level could be transformed all together to the normal stress level.
The log-rank test in this approach could be replaced by other comparison tests,
where even the use of tests based on imprecise probability theory [7] could be
explored. This is left as an interesting topic for future research.

Acknowledgements. Abdullah A.H. Ahmadini gratefully acknowledges the financial

support received from Jazan University in Saudi Arabia and the Saudi Arabian Cultural
Bureau (SACB) in London for pursuing his Ph.D. at Durham University. The authors
thank two reviewers of this paper for supportive comments and suggestions.

References
1. Augustin, T., Coolen, F., de Cooman, G., Troffaes, M.: Introduction to Imprecise
Probabilities. Wiley, Chichester (2014)
2. Coolen, F.: Nonparametric predictive inference. In: International Encyclopedia of
Statistical Science, pp. 968–970. Springer, Berlin (2011)
3. Gehan, E.: A generalized Wilcoxon test for comparing arbitrarily singly-censored
samples. Biometrika 52, 203–224 (1965)
4. Mantel, N.: Evaluation of survival data and two new rank order statistics arising in
its consideration. Cancer Chemother. Rep. 50, 163–170 (1966)
5. Nelson, W.: Accelerated Testing: Statistical Models, Test Plans, and Data Analysis.
Wiley, New Jersey (1990)
6. Peto, R., Peto, J.: Asymptotically efficient rank invariant test procedures. J. R.
Stat. Soc. Ser. A 135, 185–207 (1972)
7. Benavoli, A., Mangili, F., Corani, G., Zaffalon, M., Ruggeri, F.: A Bayesian Wilcoxon
signed-rank test based on the Dirichlet process. In: Proceedings of the 30th Inter-
national Conference on Machine Learning (ICML 2014), pp. 1–9 (2014)
 Descriptive Comparison of the Rating
Scales Through Different Scale Estimates:
Simulation-Based Analysis

Irene Arellano1 , Beatriz Sinova1 , Sara de la Rosa de Sáa2 ,

Marı́a Asunción Lubiano1 , and Marı́a Ángeles Gil1(B)
1
Departamento de Estadı́stica, I.O. y D.M., Universidad de Oviedo,
C/ Federico Garcı́a Lorca 18, 33007 Oviedo, Spain
{uo239511,sinovabeatriz,lubiano,magil}@uniovi.es
2
Oficina de Evaluación de Tecnologı́as Sanitarias,
Servicio de Salud del Principado de Asturias, Asturias, Spain
[email protected]

Abstract. In dealing with intrinsically imprecise-valued magnitudes, a

common rating scale type is the natural language-based Likert. Along the
last decades, fuzzy scales (more concretely, fuzzy linguistic scales/varia-
bles and fuzzy ratig scales) have also been considered for rating values
of these magnitudes. A comparative descriptive analysis focussed on the
variability/dispersion associated with the magnitude depending on the
considered rating scale is performed in this study. Fuzzy rating responses
are simulated and associated with Likert responses by means of a ‘Lik-
ertization’ criterion. Then, each ‘Likertized’ datum is encoded by means
of a fuzzy linguistic scale. In this way, with the responses available in the
three scales, the value of the different dispersion estimators is calculated
and compared among the scales.

Keywords: Fuzzy linguistic scale · Fuzzy rating scale · Likert scale

Scale estimates

1 Introduction
The Likert-type scales are frequently used in designing questionnaires to rate
characteristics or attributes that cannot be numerically measured (like satis-
faction, perceived quality, perception...). Although they are easy to answer and
they do not require a special training to use them, respondents often do not
find accurate answers to items and the available statistical methodology to ana-
lyze the data from these questionnaires is rather limited. This is mainly due to
the fact that Likert scales are discrete with a very small number of responses
to choose for each item (often 4 to 7). To overcome this concern, Hesketh et
al. [5] proposed the so-called fuzzy rating scale to allow a complete freedom and
expressiveness in responding, without respondents being constrained to choose
among a few pre-specified responses.
c Springer Nature Switzerland AG 2019
S. Destercke et al. (Eds.): SMPS 2018, AISC 832, pp. 9–16, 2019.
https://doi.org/10.1007/978-3-319-97547-4_2
10 I. Arellano et al.

By drawing the fuzzy number that best represents the respondent’s valua-
tion, the fuzzy rating scale captures the logical imprecision associated with such
variables. Moreover, this fuzzy rating scale allows us to have a rich continuous
scale of measurement, unlike the case of a posterior numerical or fuzzy encoding
(the latter encoding Likert points with fuzzy numbers from a linguistic scale,
and usually made by trained experts).
In previous studies (see Gil et al. [3], Lubiano et al. [6–8]) we have con-
firmed that the results when fuzzy rating scales are considered sometimes differ
importantly from the conclusions drawn from numerically or fuzzy linguistically
encoded Likert values.
As differences can often be even clearer from the dispersion than for the
location perspective, this paper aims to examine, by means of simulation devel-
opments, how location-based ‘scale’ estimates are affected by the considered scale
of measurement.

2 Preliminaries
A (bounded) fuzzy number is a mapping U  : R → [0, 1] such that for all
 
α ∈ [0, 1], the α-level set Uα = {x ∈ R : U (x) ≥ α} if α ∈ (0, 1], and U 0 =

cl{x ∈ R : U (x) > 0} (with ‘cl’ denoting the closure of the set) is a nonempty
compact interval.
In dealing with fuzzy number-valued data, distances will be computed by
considering two different metrics introduced by Diamond and Kloeden [1]: the
2-norm metric ρ2 and the 1-norm metric ρ1 , which for fuzzy numbers U  and V
are given by
 
1  
 
ρ2 (U , V ) = (inf Uα − inf Vα )2 + (sup Uα − sup Vα )2 dα,
2 [0,1]
  
  1 α − inf Vα | + | sup U
α − sup Vα | dα.
ρ1 (U , V ) = | inf U
2 [0,1]

3 Scales Measures for Fuzzy Data

In developing statistics with fuzzy data coming from intrinsically imprecise-
valued attributes, random fuzzy numbers constitute a well-formalized model
within the probabilistic setting for the random mechanisms generating such data.
Let X be a random fuzzy number (as defined by Puri and Ralescu [9]) associ-
ated with a probability space, i.e., a fuzzy number-valued mapping X associated
with a probability space and such that, for each α, the α-level interval-valued
mapping is a random interval associated with the probability space.
Let xn = ( n ) be a sample of observations from X . The sample
x1 , . . . , x
Aumann-type mean is the fuzzy number such that for each α
 n n
 
(xn )α = inf(
xi )α /n, sup(
xi )α /n ,
i=1 i=1
 Descriptive Comparison of the Rating Scales 11

and the sample 1-norm median is the fuzzy number such that for each α

xn ))α = [Mei inf(

(Me( xi )α , Mei sup(
xi )α ] .

In De la Rosa de Sáa et al. [2] one can find together the most commonly
used location-based scale estimates, namely: the sample Fréchet-type ρ2 -
xn , Me(
Standard Deviation and, for D ∈ {ρ1 , ρ2 } and M ∈ { xn )}, the sam-
ple D-Average Distance Deviation and the sample D-Median Distance
Deviation, which are respectively given by

1  2
n
xn ) = 
ρ2 -SD( n ) ,
xi , x
ρ2 (
n i=1

1
n  
 xn , M ) =
D-ADD( D(  xn , M ) = Mei D(
xi , M ), D-MDD( xi , M ) .
n i=1

4 Generating Fuzzy Data for Simulation Studies

In this work, simulations have been inspired by real-life datasets in connection

with fuzzy rating scale-based experiments.
To generate fuzzy data from a trapezoidal-valued random fuzzy number X =
Tra(inf X0 , inf X1 , sup X1 , sup X0 ), Sinova et al. [10] suggest to use an alternative
characterization, X = TraX1 , X2 , X3 , X4 , where (see Fig. 1)

X1 = midX1 = (inf X1 + sup X1 )/2, X2 = sprX1 = (sup X1 − inf X1 )/2,

X3 = lsprX0 = inf X1 − inf X0 , X4 = usprX0 = sup X0 − sup X1 ,

(i.e., X1 = core mid-point, X2 = core radius, X3 = ‘left distance’ between core
and support, X4 = ‘right distance’ between core and support) whence

X = TraX1 , X2 , X3 , X4  = Tra(X1 −X2 −X3 , X1 −X2 , X1 +X2 , X1 +X2 +X4 ).

In fact, fuzzy data will be generated by simulating the four real-valued ran-
dom variables X1 , X2 , X3 and X4 , so that the R×[0, ∞)×[0, ∞)×[0, ∞)-valued
random vector (X1 , X2 , X3 , X4 ) will provide us with the 4-tuples (x1 , x2 , x3 , x4 )
with x1 = center and x2 = radius of the core, and x3 = lower and x4 = upper
spread of the fuzzy number. To each generated 4-tuple (x1 , x2 , x3 , x4 ) we asso-
ciate the fuzzy number Trax1 , x2 , x3 , x4 .
According to the simulation procedure, data have been generated from ran-
dom fuzzy numbers with a bounded reference set and abstracting and mimicking
what we have observed in real-life examples employing the fuzzy rating scale
(FRS). More concretely, fuzzy data have been generated such that
12 I. Arellano et al.

Fig. 1. 4-Tuples to be generated for the simulation procedures

– 100·ω1 % of the data have been obtained by first considering a simulation from
a simple random sample of size 4 from a beta β(p, q) distribution, ordering the
corresponding 4-tuple, and finally computing the values xi . The values of p
and q vary in most cases to cover different distributions (namely, symmetrical
weighting central values, symmetrical weighting extreme values, and asym-
metric ones). In most of the comparative studies involving simulations, the
values from the beta distribution are re-scaled and translated to an interval
[l0 , u0 ] different from [0, 1].
– 100 · ω2 % of the data have been obtained considering a simulation of four
random variables Xi = (u0 − l0 ) · Yi + l0 as follows:
Y1 ∼ β(p, q),  
Y2 ∼ Uniform0, min{1/10, Y1 , 1 − Y 1 } ,
Y3 ∼ Uniform0, min{1/5, Y1 − Y2 } , 
Y4 ∼ Uniform 0, min{1/5, 1 − Y1 − Y2 } .
– 100 · ω3 % of the data have been obtained considering a simulation of four
random variables Xi = (u0 − l0 ) · Yi + l0 as follows:

Y1 ∼ β(p, q),
⎧
⎨ Exp(200) if Y1 ∈ [0.25, 0.75]
Y2 ∼ Exp(100 + 4 Y1 ) if Y1 < 0.25
⎩
Exp(500 − 4 Y1 ) otherwise

γ(4, 100) if Y1 − Y2 ≥ 0.25
Y3 ∼
γ(4, 100 + 4 Y1 ) otherwise

γ(4, 100) if Y1 + Y2 ≥ 0.25
Y4 ∼
γ(4, 500 − 4 Y1 ) otherwise.

5 Results
First, FRS data will be simulated in accordance with the above described realistic
simulation procedure. Later, fuzzy data based on a fuzzy rating scale can fairly be
associated/classified in accordance with labels in a Likert scale (more concretely,
with their numerical encoding). This process is to be called “Likertization”.
Furthermore, the associated Likert values could also be later encoded by means
of values from a fuzzy linguistic scale.
 Descriptive Comparison of the Rating Scales 13

For carrying out the Likertization, the “minimum distance Likertization cri-
terion” will be employed (see Fig. 2):

Fig. 2. Minimum distance criterion scheme when the reference interval equals [1, k]

In this way, if the considered Likert scale is a k-point one, given a metric
D between fuzzy data and U  the free fuzzy response to be classified, then U
 is

associated with the integer κ(U ) such that
 ) = arg
κ(U min  , 1{j} ).
D(U
j∈{1,...,k}

Each FRS-based datum will be first Likertized by means of the minimum

distance criterion, and it will later be encoded by means of a fuzzy linguistic scale.
We have chosen the most usual (see, for instance, Herrera et al. [4]) balanced
semantic representations of the linguistic hierarchies of k = 4 and k = 5 levels
(Fig. 3).
FLS34 FLS35

1 1

0 0

0 20 40 60 80 100 0 20 40 60 80 100

Fig. 3. Usual balanced semantic representation of the linguistic hierarchies of k = 4

and k = 5 levels

Simulations-based tables (Tables 1 and 2) collect the percentages of Euclidean

distances between the sample scale estimates D  for the FRS-simulated data and
for their numerically (NEL) and fuzzy linguistically (FLS) encoded ρ1 Likerti-
zation that are over ε ∈ {1, 5, 10, 15}.
14 I. Arellano et al.

Table 1. % of simulated samples of size n for which the Euclidean distance between
the sample scale estimate D  associated with the FRS and the one associated with
either the NEL (numerically encoded Likert) or the FLS (fuzzy linguistic scale) with
k = 4 different values is greater than ε ∈ {1, 5, 10, 15} and (from top to bottom)
β(p, q) ≡ β(1, 1), β(.75, .75), β(4, 2), and β(6, 1)
 
  
% D(FRS) − D(S)  > ε (k = 4, β(p, q) ≡ β(1, 1))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2 n )
xn , x
ADD(


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1
MDD(
xn , Me(

 
  
% D(FRS) − D(S) > ε (k = 4, β(p, q) ≡ β(0.75, 0.75))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2
ADD( n )
xn , x


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1 xn , Me(
MDD(

 
  
% D(FRS) − D(S)  > ε (k = 4, β(p, q) ≡ β(4, 2))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2
ADD( n )
xn , x


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1
MDD(
xn , Me(

 
  
% D(FRS) − D(S)  > ε (k = 4, β(p, q) ≡ β(6, 1))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2
ADD( n )
xn , x


 xn ))
ρ1 xn , Me(
ADD(

ρ2
MDD( n )
xn , x


 xn ))
ρ1 xn , Me(
MDD(
 Descriptive Comparison of the Rating Scales 15

Table 2. % of simulated samples of size n for which the Euclidean distance between
the sample scale estimate D  associated with the FRS and the one associated with
either the NEL (numerically encoded Likert) or the FLS (fuzzy linguistic scale) with
k = 5 different values is greater than ε ∈ {1, 5, 10, 15} and (from top to bottom)
β(p, q) ≡ β(1, 1), β(.75, .75), β(4, 2), and β(6, 1)
 
  
% D(FRS) − D(S)  > ε (k = 5, β(p, q) ≡ β(1, 1))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2 n )
xn , x
ADD(


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1
MDD(
xn , Me(

 
  
% D(FRS) − D(S) > ε (k = 5, β(p, q) ≡ β(0.75, 0.75))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2
ADD( n )
xn , x


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1 xn , Me(
MDD(

 
  
% D(FRS) − D(S)  > ε (k = 5, β(p, q) ≡ β(4, 2))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2
ADD( n )
xn , x


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1
MDD(
xn , Me(

 
  
% D(FRS) − D(S)  > ε (k = 5, β(p, q) ≡ β(6, 1))

D ε = 1 ε = 5 ε = 10 ε = 15
n S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS S = NEL S = FLS

ρ
2 SD(
xn )

ρ2
ADD( n )
xn , x


 xn ))
ρ1
ADD(
xn , Me(

ρ2
MDD( n )
xn , x


 xn ))
ρ1 xn , Me(
MDD(
16 I. Arellano et al.

The percentages have been quantified over 1000 samples of n ∈ {10, 30, 100}
FRS simulated (with different betas) data with reference interval [0, 100] (this
last fact being irrelevant for the study). On the basis of Tables 1 and 2 we cannot
get very general conclusions, but we can definitely assert that scale measures
mostly vary more from the FRS-based data to the encoded Likert ones.
Furthermore, one can state some approximate behaviour patterns, such as
– for almost all situations, the robust scale estimate (the last one) provides
us with much higher percentages than non-robust ones; more concretely,
ρ1
-MDD( xn , Me(
xn )) is almost generally more sensitive to the change in the
rating scale type; this is especially clear for small samples;
– distances are uniformly lower for k = 5 than for k = 4 when the midpoint
of the 1-level is beta distributed with (p, q) ∈ {(1, 1), (0.75, 0.75), (4, 2)};
when (p, q) = (6, 1) such a conclusion is appropriate for robust estimates
and ε ∈ {1, 5}, but there is no clear conclusion for non-robust estimates or
greater values of ε.

Acknowledgements. The research is this paper has been partially supported by the
Spanish Ministry of Economy, Industry and Competitiveness Grant MTM2015-63971-
P. Its support is gratefully acknowledged.

References
1. Diamond, P., Kloeden, P.: Metric spaces of fuzzy sets. Fuzzy Sets Syst. 35, 241–249
(1990)
2. De la Rosa de Sáa, S., Lubiano, S., Sinova, S., Filzmoser, P.: Robust scale estimators
for fuzzy data. Adv. Data Anal. Classif. 11, 731–758 (2017)
3. Gil, M.A., Lubiano, M.A., De la Rosa de Sáa, S., Sinova, B.: Analyzing data from
a fuzzy rating scale-based questionnaire: a case study. Psicothema 27, 182–191
(2015)
4. Herrera, F., Herrera-Viedma, E., Martı́nez, L.: A fuzzy linguistic methodology to
deal with unbalanced linguistic term sets. IEEE Trans. Fuzzy Syst. 16(2), 354–370
(2008)
5. Hesketh, T., Pryor, R., Hesketh, B.: An application of a computerized fuzzy graphic
rating scale to the psychological measurement of individual differences. Int. J. Man-
Mach. Stud. 29, 21–35 (1988)
6. Lubiano, M.A., De la Rosa de Sáa, S., Montenegro, M., Sinova, B., Gil, M.A.:
Descriptive analysis of responses to items in questionnaires. Why not using a fuzzy
rating scale? Inf. Sci. 360, 131–148 (2016)
7. Lubiano, M.A., Montenegro, M., Sinova, B., De la Rosa de Sáa, S., Gil, M.A.:
Hypothesis testing for means in connection with fuzzy rating scale-based data:
algorithms and applications. Eur. J. Oper. Res. 251, 918–929 (2016)
8. Lubiano, M.A., Salas, A., Gil, M.A.: A hypothesis testing-based discussion on the
sensitivity of means of fuzzy data with respect to data shape. Fuzzy Sets Syst.
328, 54–69 (2017)
9. Puri, M.L., Ralescu, D.A.: Fuzzy random variables. J. Math. Anal. Appl. 114,
409–422 (1986)
10. Sinova, B., Gil, M.A., Colubi, A., Van Aelst, S.: The median of a random fuzzy
number. The 1-norm distance approach. Fuzzy Sets Syst. 200, 99–115 (2012)
 Central Moments of a Fuzzy Random
Variable Using the Signed Distance:
A Look Towards the Variance

Rédina Berkachy(B) and Laurent Donzé

Applied Statistics and Modelling, Department of Informatics,

Faculty of Economics and Social Sciences, University of Fribourg,
Fribourg, Switzerland
{Redina.Berkachy,Laurent.Donze}@unifr.ch
http://diuf.unifr.ch/asam

Abstract. The central moments of a random variable are extensively

used to understand the characteristics of distributions in classical statis-
tics. It is well known that the second central moment of a given random
variable is simply its variance. When fuzziness in data occurs, the situ-
ation becomes much more complicated. The central moments of a fuzzy
random variable are often very difficult to be calculated because of the
analytical complexity associated with the product of two fuzzy numbers.
An estimation is needed. Our research showed that the so-called signed
distance is a great tool for this task. The main contribution of this paper
is to present the central moments of a fuzzy random variable using this
distance. Furthermore, since we are interested in the statistical measures
of the distribution, particularly the variance, we put an attention on its
estimation using the signed distance. Using this distance in approximat-
ing the square of a fuzzy difference, we can get an unbiased estimator of
the variance. Finally, we prove that in some conditions our methodology
related to the signed distance returns an exact crisp variance.

Keywords: Central moments · Estimation of the variance

Fuzzy variance · Fuzzy statistics · Signed distance · Unbiased estimator

1 Introduction and Motivation

The central moments of a given random variable are always needed for specify-
ing the distribution or in calculating different statistical measures. The variance
is known as the second central moment. The classical approaches are very clear
on the ways of computing these measures. When the data are exposed to fuzzi-
ness, fuzzy methods are well suited for such situations. Thus, calculating central
moments of a fuzzy random variable can be of good use. But, computing them
is not totally understood, especially regarding the multiplication of two fuzzy
sets. Computational difficulties arise and are not evident to overcome. Since we
know that the product of two fuzzy numbers given by the extension principle of
c Springer Nature Switzerland AG 2019
S. Destercke et al. (Eds.): SMPS 2018, AISC 832, pp. 17–24, 2019.
https://doi.org/10.1007/978-3-319-97547-4_3
Exploring the Variety of Random
Documents with Different Content
as Hama. It is the duty of the women alone to pitch and to strike the
tents, and they acquit themselves with surprising address and
rapidity. All the labours of the encampment generally fall to the lot of
the women. The men take charge of the flocks, kill and skin the
beasts. The costume of the women is very simple; they wear a large
blue chemise, a black machlas, and a sort of black silk scarf, which,
after covering the head, passes twice round the neck, and falls over
the back. They have no covering for the legs; except the wives of the
sheiks, who wear yellow boots. Their great ambition and luxury is to
have a great many bracelets; they have them of glass, of coral, coin,
and amber.
The plain on which we rested was called El Makram. It is not far from
Hama. The place is rather agreeable, and its rich pasturage renders
it suitable to the Bedouins.
The fourth day we had an alarm: four hours after noon the
shepherds came running in haste, crying “To arms! the enemy are
seizing our flocks.” It was the tribe of Daffir, who, watching the
opportunity to revenge themselves on Nasser, had sent a thousand
horse to carry off the flocks at nightfall, to allow no time for a pursuit.
Our men, expecting an attack, were prepared; but it was necessary
to find out on which side the enemy were. Night coming on, four men
dismounted from their horses, took opposite directions, and
crouching down, their ears close to the ground, heard at a great
distance the steps of the plunderers. Night passed without being
able to reach them; but in the morning the troop of Hassné (that of
Mehanna) having joined them, they gave battle. After four hours’
fighting, half the flocks were recaptured; but five hundred camels
remained in the hands of the tribe El Daffir. We had ten men killed
and several wounded. At the return, the affliction was general; the
Bedouins murmured, attributing all that had happened to the caprice
and vanity of Nasser.
Mehanna sent off a courier to his son, who immediately returned
from Damascus, accompanied by a chokedar, (an officer of the
pacha,) in order to make an impression on the Bedouins. On his
arrival he read a letter from the pacha, to the following effect:—
 “We make known to all the emirs and sheiks of the desert,
great and small, encamped on the territory of Damascus,
that we have appointed our son, Nasser Ebn Mehanna,
Emir of all the Anazes (Bedouins of the desert,) inviting
them to obey him. The tribe that shall have the misfortune
to show itself rebellious, shall be destroyed by our
victorious troops, and, as an example, their flocks shall be
slaughtered, and their women delivered up to the soldiers.
Such is our will.
Signed,
“Soliman, Pacha of Damascus and Acre.”
Nasser, proud of his new dignity, affected to read the order to every
body, and to talk Turkish with the officer of the pacha, which still
further increased the disgust of the Bedouins. One day whilst we
were with him, there arrived a very handsome young man, named
Zarrak, the chief of a neighbouring tribe. Nasser, as usual, spoke of
his appointment, vaunted the greatness and power of the vizier of
Damascus, and of the sultan of Constantinople “of the long sabre.”[C]
Zarrak, who listened with impatience, changing colour, rose and
said, “Nasser Aga,[D] learn that all the Bedouins detest thee: if thou
art dazzled by the magnificence of the Turks, go to Damascus; adorn
thy forehead with the caouk;[E] become the minister of the vizier;
dwell in his palace; perhaps thou mayest strike terror into the
Damascenes; but we Bedouins care no more for thee, thy vizier and
thy sultan, than camel dung. I shall depart for the territory of Bagdad,
where I shall find the Drayhy[F] Ebn Chahllan; him will I join.”
Nasser, in his turn growing pale with anger, transmitted the
conversation in Turkish to the chokedar, who thought by violent
menaces to alarm Zarrak. But he, looking at him fiercely, said, “It is
enough: though you have Nasser on your side, I could, if I would,
prevent you from ever eating bread more.” In spite of these offensive
words, all three preserved their coolness; and Zarrak, mounting his
horse, said to Nasser, “Salam aleik (peace to thee); display all thy
power; I await thee.” This challenge caused Nasser much trouble;
but he still persevered in his alliance with the Turks.
The following day we learned that Zarrak had set out with his tribe
for the country of Geziri, and a combination of the Bedouins against
Nasser was talked of in all quarters. Mehanna, having learned what
was passing, called his son to him, and said, “Nasser, will you then
break the pillars of the tent of Melkhgem?” and taking his beard in
his hand, “Will you,” added he, “bring contempt upon this beard at
the end of my days, and tarnish the reputation I have acquired?
Unhappy man, thou hast not invoked the name of God. What I had
foreseen, has happened. All the tribes will unite with the Drayhy.
What then will become of us? It will only remain for us to humble
ourselves before Ebn Sihoud, that enemy of our race, who styles
himself king of the Bedouins; he alone can defend us from the
terrible Drayhy.”
Nasser endeavoured to tranquillize his father, assuring him that
matters were not so bad as he feared. However, the Bedouins began
to take part with one or the other; but the greater part sided with the
father, who was in their true interest.
Sheik Ibrahim was very dissatisfied; he wished to penetrate farther
into the desert, and proceed as far as Bagdad; and he found himself
bound to a tribe that remained between Damascus and Homs. He
thus lost all the summer, being able to remove only with danger of
his life. He desired me to obtain some knowledge respecting the
Drayhy, to learn his character, the places where he passed the
summer, where he wintered, if he received strangers, and many
other particulars; in short, he told me he had the greatest interest in
being rightly informed.
These details were difficult to obtain without exciting suspicion: it
was necessary to find some one not of the tribe of El Hassnnée. At
length I became acquainted with a man named Abdallah el Chahen
(the poet.) Knowing that poets are sought after by the great, I asked
him about all the tribes he had visited, and learned with pleasure that
he had been for a long time with the Drayhy. I obtained from him all
the information I had desired.
One day Nasser made me write to Sheik Saddad, and him of
Corietain, to demand from each a thousand piastres and six
machlas. This claim is called right of fraternity: it is an arrangement
between the sheiks of villages and the more powerful chiefs of the
Bedouins, to be protected against the ravages of the other tribes.
This is an annual tax. These unhappy villages are ruined to satisfy
two tyrants—the Bedouins and the Turks.
Mehanna holds this fraternity with all the villages of the territories of
Damascus, Homs, and Hama, which brings him in a revenue of
about fifty thousand piastres. The pacha of Damascus pays him
twelve thousand five hundred, and the cities of Homs and Hama
furnish him besides a certain quantity of corn, rice, dried grapes, and
stuffs. The small tribes bring him butter and cheese. In spite of all, he
never has any money, and is often in debt, without having any
expenses to incur; which greatly astonished us. We learned that he
gave all away in presents to the most distinguished warriors, either
of his own tribe or to others, and that he had thus raised for himself a
powerful party. He is always ill-clothed, and when he receives a
handsome pelisse or other article for a present, he gives it to the
person who happens to be near him at the moment. The Bedouin
proverb, that generosity covers all defects, is amply verified in
Mehanna, whose liberality alone renders the conduct of Nasser
bearable.
A short time after this event we went to encamp, three hours from
the Orontes, upon lands called El Zididi, on which there are many
springs.
Mehanna having one day been with ten horsemen to visit the Aga of
Homs, returned loaded with presents from all the merchants, who
cultivate his friendship, because, wheresoever dissatisfied with them,
he intercepts their commerce and plunders the caravans.—
Immediately upon his return, Nasser set forth on an expedition
against the tribe Abdelli, which is commanded by the Emir El
Doghiani, and encamped near Palmyra, on two small hills of equal
size, called Eldain (the breast;) he returned after three days with five
hundred camels and two hundred sheep. In this affair we lost three
men, and Zamel’s mare was killed under him. On the other hand, we
took three mares, killed ten men, and wounded twenty more.
Notwithstanding this success, the Bedouins were indignant at
Nasser’s want of faith, who had no cause of hatred against this tribe.
On all sides measures were taken with the Drayhy, to destroy the
tribe El Hassnnée. The news reached the Emir Douhi, the chief of
the tribe Would Ali, a kinsman and intimate friend of Mehanna, and
who, as well as himself, is charged with the escort of the grand
caravan; and he came with thirty horsemen to make known the
danger with which he was threatened. The heads of the tribe went
out to meet Douhi: having entered the tent, Mehanna ordered coffee;
the emir stopped him and said, “Mehanna, thy coffee is drunk
already! I come here neither to eat nor drink, but to inform thee that
the behaviour of thy son Nasser Pacha (for so he styled him in
derision) is bringing down destruction upon thee and thine: know that
all the Bedouins have leagued together, and are about to declare
against thee a war of extermination.” Mehanna, changing colour,
exclaimed, “Well, art thou now satisfied, Nasser? Thou wilt be the
last of the race of Melkghem.”
Nasser, still obstinate, replied that he should make head against all
the Bedouins; and that he should have the support of twenty
thousand Osmanlis, as well as that of Mola Ismael, the chief of the
Kurdish cavalry, who bears the schako. Douhi passed the night in
endeavouring to turn Nasser from his projects, but without
succeeding: the day following, he departed, saying, “My conscience
forbids me to join you. Our relationship, and the bread we have
eaten together, prevent me from declaring war against you. Farewell;
I leave you with sorrow.”
From this moment our time passed very disagreeably with the
Bedouins. We could never quit them, for all the people who went to a
distance from the tents were massacred. There were continual
attacks on both sides, sudden changes of the encampment for
greater security, alarms, reprisals, incessant disputes between
Mehanna and his son; but the old man was so kind and so
credulous, that Nasser always succeeded in persuading him that he
was in the right.
We were told a thousand traits of his simplicity: amongst others, that
being at Damascus whilst Yousouf Pacha, the grand vizier of the
Porte, was holding his court there on his return from Egypt after the
departure of the French, Mehanna was presented to him, as well as
the other grandees; but, being little acquainted with Turkish
etiquette, he accosted him without ceremony and with the Bedouin
mode of salutation, and placed himself on the divan by his side
without being invited. Yousouf, equally unaccustomed to the usages
of the Bedouins, and ignorant of the dignity of the little shabby old
man who treated him with such familiarity, ordered him to be taken
from his presence and beheaded. The slaves took him out, and were
preparing to execute the order, when the Pacha of Damascus cried
aloud, “Hold! what is it you are doing? If there should fall a hair of his
head, with all your power, you will never send another caravan to
Mecca.” The vizier instantly had him brought back, and placed him
by his side; he gave him coffee, had him invested with a rich
Cachemire turban, a rich gombaz (robe,) and a pelisse of honour,
and presented him with a thousand piastres. Mehanna, deaf, and
besides not understanding Turkish, knew nothing of what was
passing; but taking off the fine clothes, he gave them to three of his
slaves who accompanied him. The vizier asked him, through the
dragoman, if he was not satisfied with the present. Mehanna replied,
“Tell the vizier of the sultan, that we Bedouins seek not to distinguish
ourselves by fine clothes: I am ill clad, but all the Bedouins know me;
they know that I am Mehanna el Zadel, the son of Melkghem.” The
pacha, not daring to offend him, affected to smile, and to be much
pleased.
The summer passed away. By the month of October the tribe was in
the vicinity of Aleppo. My heart beat on finding myself so near my
home; but, according to our agreement, I could not even send news
of myself to my friends. Sheik Ibrahim desired to pass the winter at
Damascus—no Bedouin durst conduct us. We obtained with great
difficulty an escort as far as a village, two days from Aleppo, called
Soghene (the hot.) The hospitable inhabitants contended for the
pleasure of receiving us. A natural warm bath accounts for the name
of the village; and the beauty of its inhabitants may probably be
attributable to its warm springs. From thence we reached Palmyra,
but with a difficulty for which we were indemnified by the pleasure of
seeing Sheik Ragial again. After passing a fortnight with our friends,
we went back to Corietain, where Sheik Selim and the curate Moussi
welcomed us with genuine kindness;—they were never tired of
hearing our accounts of the Bedouins.—Sheik Ibrahim satisfied their
friendly concern about our affairs, by saying that our speculation was
wonderfully advantageous; that we had gained more than we had
expected; whilst in reality, between presents and losses, we only had
remaining the goods deposited with Moussi.—We lost thirty days at
Corietain in preparing for our departure. Winter was rapidly coming
on, and no one durst furnish us with cattle, being convinced we
should be plundered on the road. At last, Sheik Ibrahim bought a bad
horse, I hired an ass, and in miserable weather, with a freezing wind,
we set off, accompanied by four men on foot, for the village of Dair
Antié. After some hours we arrived at a defile between two
mountains, named Beni el Gebelain. At this spot, twenty Bedouin
horsemen came upon us. Our guides, far from defending us, hid
their guns and remained spectators of our disaster. The Bedouins
stripped us, and left us nothing but our shirts. We implored them to
kill us rather than expose us to the cold. At last, touched at our
condition, they had the generosity to leave each of us a gombaz. As
for our beasts, they were too sorry to tempt them. Being hardly able
to walk, they would have only uselessly detained them. Night came
on, and the cold was excessive, and deprived us of the use of
speech. Our eyes were red, our skin blue; at the end of some time I
fell to the ground, fainting and frozen. Sheik Ibrahim in despair made
gesticulations to the guides, but was unable to speak. One of them,
a Syrian Christian, took pity upon me and the grief of Sheik Ibrahim;
he threw down his horse, which was also half dead with cold and
fatigue, killed it, opened the belly, and placed me without
consciousness in the skin, with only my head out. At the end of half
an hour, I regained my senses, quite astonished at finding myself
alive again, and in so strange a position. Warmth restored my
speech; and I earnestly thanked Sheik Ibrahim and the good Arab. I
took courage, and found strength to proceed. A little after, our guides
cried out, “Here’s the village!” and we entered the first house. It
belonged to a farrier, named Hanna el Bitar. He showed a lively
sympathy in our situation, set about covering us both with camel-
dung, and gave us a little wine—a few drops at a time: having thus
restored our strength and warmth, he withdrew us from our dunghill,
put us to bed, and made us take some good soup. After a sleep,
which was indispensable, we borrowed two hundred piastres to pay
our guides and carry us to Damascus, which we reached the 23d
December, 1810.
M. Chabassan, a French physician, the only Frank at Damascus,
received us; but as we were to pass the winter here, we afterwards
took up our quarters in the Lazarist Convent, which was abandoned.
I will not describe the celebrated city of Sham (Damascus), the Gate
of Glory (Bab el Cahbé), as the Turks style it. Our long residence has
enabled us to know it minutely; but it has been too often visited by
travellers to offer any new interest. I return to my narrative.
One day, being at the bazaar to pass away the time in the Turkish
fashion, we saw running towards us a Bedouin, who embraced us,
saying, “Do you not recollect your brother Hettall, who ate your bread
at Nuarat el Nahaman?” Delighted with meeting him, we took him
home, and having regaled him, and asked him many questions, we
learned that the affairs of the tribe Hassnnée were in a bad
condition, and that the league against them was extending daily.
Hettall told us that he was of the tribe of Would Ali, whose chief,
Douhi, was known to us. This tribe winters in the territory of Sarka
and Balka; it reaches from the country of Ismael to the Dead Sea,
and returns to Horan in the spring. He proposed to us to visit it,
promising a good sale for our merchandise. Having consented, it
was agreed that he should come for us in the month of March.
Sheik Ibrahim having received, through the intervention of M.
Chabassan, a group of a thousand tallaris from Aleppo, desired me
to make new purchases. When they were completed, I showed them
to him, and asked whether any thing would remain for us at our
return? “My dear son,” he replied, “the gratitude of every chief of a
tribe brings me more than all my merchandise.—Be under no
concern. You also shall receive your return in money and in
reputation. You shall be renowned in your time; but I must know all
the tribes and their chiefs. I depend upon you to get to the Drayhy,
and for that purpose you must absolutely pass for a Bedouin. Let
your beard grow, dress like them, and imitate their usages. Ask no
explanations—remember our terms.” My only reply was, “May God
give us strength!”
Twenty times was I on the point of abandoning an enterprise of
which I perceived all the dangers without knowing the object. This
silence, this blind obedience, became insupportable. However, my
wish to come to the issue, and my attachment to M. Lascaris, gave
me patience.
At the time agreed, Hettall arrived with three camels and two guides,
and we set out the 15th March, 1811, one year and twenty-eight
days after our first departure from Aleppo. The tribe was at a place
called Misarib, three days from Damascus. Nothing remarkable
happened on the road. We passed the nights under a starry sky; and
on the third day, by sunset, we were in the midst of the tents of
Would Ali. The coup d’œil was delightful. Every tent was surrounded
by horses, camels, goats, and sheep, with the lance of the horseman
planted at the entrance: that of the Emir Douhi arose in the centre.
He received us with all possible consideration, and made us sup with
him. He is a man of understanding, and is equally loved and feared
by his people. He commands five thousand tents, and three tribes,
which are joined to his; those of Benin Sakhrer, of El Serhaan, and
El Sarddié. He had divided his soldiers into companies or divisions,
each commanded by one of his kinsmen.
The Bedouins are fond of hearing stories after supper. This is one
that the emir told us: it depicts the extreme attachment they have for
their horses, and the self-love they manifest with regard to their own
qualities.
One of his tribe, named Giabal, possessed a very celebrated mare.
Hassad Pacha, then vizier of Damascus, made him on various
occasions all sorts of offers to part with it, but in vain, for a Bedouin
loves his horse as he does his wife. The pacha then employed
threats, but with no better success. At length, another Bedouin,
named Giafar, came to the pacha, and asked what he would give
him if he brought him Giabal’s mare? “I will fill thy barley sack with
gold,” replied Hassad, who felt indignant at his want of success. This
took place without transpiring; and Giabal fastened his mare at night
by the foot with an iron ring, the chain of which passed into his tent,
being held by a picket fixed in the ground under the very felt which
served him and his wife as a bed. At midnight, Giafar creeps into the
tent on all-fours, and, insinuating himself between Giabal and his
wife, gently pushes first the one, and then the other: the husband
thought his wife was pushing, the wife thought the same of the
husband; and each made more room. Giafar then, with a knife well
sharpened, makes a slit in the felt, takes out the picket, unties the
mare, mounts her, and, grasping Giabal’s lance, pricks him slightly
with it, calling out, “It is I, Giafar, who have taken thy noble mare; I
give thee early notice!” and off he goes. Giabal instantly darts from
the tent, calls his friends, mounts his brother’s mare, and pursues
Giafar for four hours. Giabal’s brother’s mare was of the same blood
as his own, though not so good. Outstripping all the other horsemen,
he was on the point of overtaking Giafar, when he cried out, “Pinch
her right ear, and give her the stirrup.” Giafar did so, and flew like
lightning. The pursuit was then useless: the distance between them
was too great. The other Bedouins reproached Giabal with being
himself the cause of the loss of his mare.[G] “I would rather,” said he,
“lose her, than lower her reputation. Would you have me let it be said
in the tribe of Would Ali, that any other mare has outrun mine? I have
at least the satisfaction of saying that no other could overtake her.”
He returned with this consolation, and Giafar received the price of
his address.
Some one else related that in the tribe of Nedgde there was a mare
of equal reputation with that belonging to Giabal, and that a Bedouin
of another tribe, named Daher, was almost mad with longing to
possess her. Having in vain offered all his camels and his riches, he
determined to stain his face with the juice of an herb, to clothe
himself in rags, to tie up his neck and legs like a lame beggar, and,
thus equipped, to wait for Nabee, the owner of the mare, in a road by
which he knew he must pass. When he drew near, he said to him in
a feeble voice: “I am a poor stranger: for three days I have been
unable to stir from this to get food: help me, and God will reward
you.” The Bedouin offered to take him on his horse, and carry him
home; but the rogue replied: “I am not able to rise, I have not
strength.” The other, full of compassion, dismounted, brought the
mare close, and placed him on her with great difficulty. As soon as
he found himself in the saddle, Daher gave her a touch with the
stirrup, and went off, saying—“It is I, Daher, who have got her and
am carrying her off.”
The owner of the mare called out to him to listen: sure that he could
not be pursued, he returned, and stopped at a short distance, for
Nabee was armed with his lance. He then said to him, “Thou hast my
mare; since it pleases God, I wish thee success: but, I conjure thee,
tell no one how thou hast obtained her.” “Why not?” said Daher.
“Because some one really ill might remain without aid: you would be
the cause why no one would perform an act of charity more, from the
fear of being duped as I have been.” Struck with these words, Daher
reflected a moment, dismounted from the horse and gave her back
to her master, and embraced him. Nabee took him home. They
remained together three days, and swore fraternity.
Sheik Ibrahim was enchanted with these stories, which gave him to
understand the character and the generosity of the Bedouins. The
tribe of Douhi is richer and more rapacious than that of Mehanna;
their horses are finer. We stayed with them a fortnight. Sheik Ibrahim
gave presents to all the chiefs, and sold some articles to the women,
to keep up our character of merchants. We then went to visit the
three tributary sheiks of the Emir Douhi.
Sheik Ibrahim told me that he had no other object in staying among
the Bedouins than that of giving me an opportunity of studying more
closely their language and their customs; that it was necessary for
his own purposes to get to the Drayhy; but that I must avail myself of
our roamings amidst the tribes to take exact notes of their names
and their numbers, which it was most important to him to know.
Their manner of speech is extremely difficult to acquire, even for an
Arab, although in fact it is the same language. I applied myself with
success. I also learned, in the course of our long wanderings, the
names of all the sheiks and the numbers of all the tribes, a thing
which had never been accomplished before: I shall give the list at the
end of my journal.
The numerous tribes are often obliged to divide themselves into
detachments from two hundred to five hundred tents, and to occupy
a large space, in order to procure water and pasturage for their
flocks. We went successively through their encampments, until we
could find means to transport ourselves to the Drayhy, who was at
war with the tribes of the territory of Damascus. We were universally
well received.
In one tribe it was a poor widow who showed us hospitality. In order
to regale us, she killed her last sheep and borrowed bread. She
informed us that her husband and her three sons had been killed in
the war against the Wahabees, a formidable tribe in the
neighbourhood of Mecca. Expressing our astonishment that she
should rob herself on our account:—“He that enters the house of the
living,” said she, “and does not eat, it is as though he were visiting
the dead.”
One tribe already considerable had been lately formed in the
following manner:—A Bedouin had a very beautiful daughter, whom
the chief of his tribe demanded in marriage; but he would not give
her, and in order to avoid his solicitations he went away furtively with
all his family. The sheik being told of what had happened, some one
said: “Serah (he is gone).” “Serhan[H] (he is a wolf),” replied he;
meaning by that, that he was a savage. From that time the tribe of
which this Bedouin became chief, has been always called the tribe El
Serhan (the wolf.) Whenever the Bedouins are courageous and have
good horses, they in a short time become powerful.
At last we heard of the arrival of the Drayhy in Mesopotamia. At this
period Sheik Ibrahim was obliged to go to Damascus for
merchandise and money, which were both equally wanting. We had
made acquaintance there with a Bedouin of one of the tribes near
the Euphrates, which had preserved a neutrality in the affair of
Nasser. This Bedouin, whose name was Gazens el Hamad, had
come with others to Damascus to sell butter. He engaged to carry
our goods on his camels, and take us to the Drayhy: but alas! we
were not destined to reach him so easily. Scarcely were we come to
Corietain, to take back our goods left at the depot, when we received
news of a victory gained by Zaher, the son of the Drayhy, over
Nasser; a victory which gave renewed violence to the war. All the
tribes ranged themselves on one side or the other. That of Salkeh,
our guide’s tribe, had been attacked by the Drayhy, who was
following up his advantages with great inveteracy, and no one dared
to cross the desert. M. Lascaris was in despair. He could neither eat
nor sleep: in short, exasperated to the highest degree at finding
himself stopped in his projects, he even found fault with me. I then
said to him,—“It is now time we should understand one another. If
you wish to get to the Drayhy for the purpose of trading, it is utter
madness, and I decline to follow you. If you have other projects or
motives adequate to the exposure of your life, let me know them,
and you shall find me ready to sacrifice myself to serve you.” “Well
then, my dear son!” said he, “I will trust you: know that this
commerce is merely a pretence to conceal a mission with which I
was charged at Paris. These are my instructions, reduced to ten
heads.
1. To set out from Paris to Aleppo.
2. To find a zealous Arab, and to attach him to me as interpreter.
3. To acquire a knowledge of the language.
4. To go to Palmyra.
5. To penetrate amongst the Bedouins.
6. To become acquainted with all the chiefs, and to gain their
friendship.
7. To unite them together in the same cause.
8. To induce them to break off all alliance with the Osmanlis.
9. To get acquainted with the whole desert, the halting-places, and
where water and pasturage are to be found, as far even as the
frontiers of India.
10. To return to Europe, safe and sound, after having accomplished
my mission.”
“And after that?” said I;—but he imposed silence. “Recollect our
conditions,” added he; “I will let you know all by degrees. At present
let it suffice to know that I must reach the Drayhy, even though it
should cost me my life.”
This half-confidence vexed me, and prevented sleep in my turn: to
find difficulties almost insurmountable, and to perceive but very
confusedly the advantages of my devotion, was sufficiently
disheartening. However, I took the resolution of persevering to the
end, as I was so far engaged, and I dwelt only on the means of
success. My beard had sprouted; I was perfectly versed in the
language of the Bedouins; I resolved to go alone and on foot to the
Drayhy: it was the only possible chance to be attempted. I went to
seek my friend Wardi, who had recalled me to life by putting me into
his horse’s belly, and communicated to him my intention. After
having endeavoured to divert me from it, by telling me that the
fatigue would be great; that there would be ten painful nights’ march;
that we must hide ourselves by day, not to be seen on the road; that
we should be unable to carry with us what was strictly necessary:
seeing that nothing could make me retract, he engaged to go with
me as guide, in consideration of a large sum of money. Having told
my resolution to M. Lascaris, he also made many friendly objections
on the score of the dangers I should incur; but, in reality, I perceived
that he was well pleased with me.
We settled all our matters; I agreed to write to him by the return of
my guide, after having arrived at the Drayhy; and the night was far
advanced when we threw ourselves on our beds. I was very much
agitated; my sleep evinced it, and I soon woke M. Lascaris by my
cries. I dreamt that being at the top of a steep rock, at the foot of
which flowed a rapid river that I was unable to pass, I had lain down
at the brink of the precipice, and that all at once a tree had taken root
in my mouth; that it grew, and spread its branches like a green tent,
but in growing it tore my throat, and its roots penetrated into my
entrails; and I uttered violent cries. Having related my dream to
Sheik Ibrahim, he was in great wonder at it, and declared that it was
an excellent omen, and prognosticated after many difficulties
important results.
It was essential that I should be covered with rags, in order not to
excite suspicion or cupidity if we were discovered on the road. This
was my costume for the journey: a coarse cotton shirt pieced; a dirty
torn gombaz; an old caffié, with a bit of linen, once white, for a
turban; a sheep-skin cloak with half the wool off, and shoes mended
to the weight of four pounds: besides these, a leather belt, from
which hung a knife worth two paras, a steel, a little tobacco in an old
bag, and a pipe. I blackened my eyes, and dirtied my face, and then
presented myself to Sheik Ibrahim to take my leave. On seeing me,
he shed tears:—“May God,” said he, “give you strength enough to
accomplish your generous design! I shall owe every thing to your
perseverance. May the Almighty be with you and preserve you from
all danger! may he blind the wicked, and bring you back that I may
reward you!” I could hardly refrain from tears in my turn. At last,
however, the conversation becoming more cheerful, Sheik Ibrahim
said smilingly, that if I were to go to Paris in this costume, I should
get much money by showing myself.—We supped; and at sunset we
departed. I walked without fatigue till midnight: but then my feet
began to swell. My shoes hurt me, and I took them off; the thorns of
the plants the camels browse on pricked me, and the small stones
wounded me. I tried to put on my shoes; and in continual suffering I
walked on till morning. A little grotto gave us shelter for the day. I
wrapped my feet in a piece of my cloak that I tore off, and slept
without having strength to take any nourishment. I was still asleep
when my guide called me to depart: my feet were much swollen—my
heart failed me—I wished to wait till the following day. My conductor
reproached me for my weakness:—“I knew well enough,” said he,
“that you were too delicate for such a journey. I before told you that it
was impossible to stop here: if we pass the night, we must also pass
the next day; our provisions will be consumed, and we shall die of
hunger in the desert. We had better give up our undertaking while
there is yet time.”
These words reanimated me, and we set off. I dragged myself along
with difficulty till near midnight, when we came to a plain, in which
the sand rose and fell in undulations: here we rested ourselves till
day. The first dawn enabled us to perceive at a distance two objects,
which we took for camels. My guide, alarmed, dug a hole in the
sand, to conceal us; we got in up to the neck, leaving only our heads
out. In this painful situation we remained with our eyes fixed on the
supposed camels, when, about noon, Wardi exclaimed, “God be
praised! they are only ostriches.” We got out of our grave with joy,
and for the first time since our departure I ate a little cake and drank
a drop of water. We remained there till night, awaiting the time to
move forward. Being then in the midst of the sands, I suffered less in
walking. We passed the next day in sleeping. We were opposite
Palmyra, to the south. Daybreak, after the fourth night, overtook us
at the bank of a large river called El Rahib, running from south to
north; my guide stripped, and carried me on his back to the other
side, and then returned for his clothes. I wished to rest myself, but he
told me it would not be prudent to stop where the river was fordable.
In fact, we had not proceeded half an hour, when we saw five
hundred well-mounted Bedouins approaching the river, going from
the east to the west. Having found some low bushes, we halted
amongst them till night. The sixth night brought us within some hours
of the Euphrates. The seventh day, the great difficulty was over; and
if I had not suffered so much in my feet, I could have forgotten all my
fatigues at the sight of the sunrise on the banks of that magnificent
river. Some hospitable Bedouins, whose occupation it is to take
people over from one side to the other, took us into their tents, where
for the first time we made a hearty meal. We obtained intelligence
respecting the Drayhy: he was at three days’ distance between Zaite
and Zauer. He had made peace with the Emir Fahed, imposing
tribute on him; they spoke to me of his military talents and his
extreme courage, of his intention to annihilate Mehanna and Nasser,
and to return to his desert near Bassora and Bagdad. These details
were just as I was wishing: I took my resolution immediately. I asked
for a guide to take me to the Drayhy, telling the Bedouins that I was a
merchant of Aleppo, having a correspondent at Bagdad, who owed
me twenty-five thousand piastres, and who had just become
bankrupt: that the war between the Bedouins had intercepted the
communications, and I had no other resource than to risk myself
alone, and put myself under the protection of the Drayhy to get to
Bagdad, where all my fortune was at stake. These kind Bedouins
offered vows to Allah that I might recover my money; and Wardi
himself took more interest in my journey, when he understood its
importance. After having passed the day in examining the tribe Beni
Tay, we departed the next day well escorted; and nothing interesting
occurred on our march. We saw the setting sun of the third day gild
the five thousand tents of the Drayhy, which covered the plain as far
as the eye could reach. Surrounded by camels, horses, and flocks,
which concealed the earth, never had I seen such a spectacle of
power and wealth. The emir’s tent in the centre was a hundred and
sixty feet long. He received me very politely, and without any
question proposed to me to sup with him. After supper, he said to
me: “Whence do you come, and whither are you going?” I replied as
I had done to the Bedouins of the Euphrates. “You are welcome,
then,” said he; “your arrival will cause a thousand benedictions.
Please God you will succeed; but, according to our custom, we
cannot speak of business till after three days devoted to hospitality
and repose.” I made the customary thanks, and retired. The next day
I despatched Wardi to M. Lascaris.
The Drayhy is a man of fifty, tall, and of a handsome countenance,
with a small beard entirely white; his aspect is stern; he is
considered as the most able of all the chiefs of tribes; he has two
sons, Zaer and Sahdoun: they are married, and dwell in the same
tent as himself. His tribe, called El Dualla, is numerous and very rich.
Chance favoured me wonderfully from the first day of my arrival. The
emir was in want of a secretary; I offered to assist him for the
moment, and I soon gained his confidence by the hints and the
information I was able to give him regarding the tribes I had studied.
When I spoke to him of my own business, he expressed so much
regret at seeing me about to depart, that I feigned to yield to his
wishes. He said, “If you will remain with me, you shall be like my son;
all that you say shall be done.” I availed myself of this confidence, to
induce him to pass over the Euphrates, in order to bring him nearer
to Sheik Ibrahim: I suggested to him all the influence he might gain
over the tribes of the country, by withdrawing them from Nasser: I
represented to him all the presents they would be obliged to offer
him; the terror with which he would inspire the Osmanlis, and the
mischief he would do his enemies by consuming their pasturage. As
it was the first time he was quitting the desert of Bagdad to come into
Mesopotamia, my advice and my information were a great resource
to him, and he followed them. The departure was superb to witness.
The horsemen before, on horses of high pedigree; women on
howdahs magnificently draped, and on dromedaries, surrounded by
negress slaves. Men, loaded with provisions, were running
throughout the caravan, calling out: “Who is hungry?” and
distributing bread, dates, &c. Every three hours, a halt was made, to
take coffee; and at night the tents were raised as if by enchantment.
We followed the banks of the Euphrates, whose clear waters
gleamed like silver: I myself was mounted on a mare of pure blood;
and the whole journey appeared like a triumphal march, presenting a
strong contrast with my former passage over the same country, in
my rags and with my tortured feet.
On the fourth day, the Emir Zahed met us with a thousand horse. All
sorts of sports ensued on horseback and with lance. At night, the
Drayhy, his sons, and myself, went to sup in the tribe of Zahed. The
day following, we crossed the river, encamped on the Damascus
territory, and kept advancing westward. We then encamped at El
Jaffet, in the pachalik of Aleppo. The report of the arrival of the
Drayhy was quickly spread, and he received from Mehanna a letter
beginning with their respective titles, and continuing thus:
“In the name of God most merciful, health! We have learnt with
surprise that you have passed the Euphrates, and are advancing into
the provinces left us by our fathers. Do you then think that you alone
can devour the pasture of all the birds? Know that we have so many
warriors that we are unable to number them. Besides, we shall be
supported by the valiant Osmanlis, whom nothing can resist: we
counsel you to return by the road by which you came; otherwise all
conceivable misfortunes will befall you, and repentance will come too
late.”
On reading this letter, I saw the Drayhy growing pale with anger; his
eyes flashed fire. After a momentary silence, “Katib,” said he, in a
terrible voice, “take your pen, and write to this dog!”
This was the answer:—“I have read your menaces, which with me do
not weigh a grain of mustard. I shall lower your flag, and purify the
earth from you and your renegade of a son, Nasser. As for the
territory you claim, the sword shall decide it. Soon will I set forward
to exterminate you. Prepare yourself. War is declared.”
Then addressing myself to the Drayhy: “I have some advice to give
you,” said I. “You are a stranger here; you know not which party the
tribes of the country will espouse. Mehanna is loved by the
Bedouins, and supported by the Turks; you are about to undertake a
war, without knowing the number of your enemies. If you experience
a single defeat, all will combine against you, and you will not be
strong enough to resist. Send then a message to the neighbouring
sheiks to tell them that you are come to destroy the tents of
Melkghem, in order to free them from the yoke of the Osmanlis; and
demand of them to declare themselves. Thus, being aware of your
force, you may compare it with theirs, and act in
consequence.”—“You are truly a man of sage counsel,” said the
Drayhy, delighted with my suggestion. “I am nothing of myself,” I
replied: “it is by the favour of my master, if I know any thing: it is he
who is the man of wisdom and knowledge, and well skilled in affairs:
he alone is capable of giving you advice. You would be enchanted
with him, if you could know him. I am sure that if you had him with
you, and were aided by his sagacity, you would become chief of all
the Bedouins of the desert.” “I will instantly send a hundred
horsemen to bring him,” said the Drayhy, with alacrity. “We are still
too far off,” said I; “the journey would be painful; when we get nearer
to Corietain I will bring him to you.”
I was fearful of some untoward accident to the Sheik Ibrahim; and
wished to be near him, to present him myself. I was so attached to
him, that I could have sacrificed myself a thousand times to do him
service.
But to return to our council of war. The Drayhy gave me a list, to
write to ten of the principal sheiks of the tribes. This was his letter:
“I have left my country to come and deliver you from the tyranny of
Nasser, who wishes to become your master by the power of the
Turks; to change your usages, destroy your manners, and subject
you to the Osmanlis. I have declared war against him; tell me frankly
if you are for him or for me; and let those who will aid me come and
join me.—Health!”
The next day, having despatched ten horsemen with these letters,
we advanced into the extensive and beautiful territory of Chaumerie,
thirty hours from Hama. After a short absence, our messengers
returned. The Emir Douhi, and the Sheik Sellame, answered that
they should preserve a neutrality; Sheik Cassem, the kinsman of
Mehanna, declared for him; the remaining seven tribes came and
encamped around us, their sheiks promising the Drayhy to partake
his dangers for life or for death. However, our spies brought
intelligence that Mehanna in alarm had sent Nasser to Hama to
obtain assistance from the Turks. The Drayhy immediately
assembled his army, eight thousand strong, six thousand horse, and
a thousand deloulmardoufs,—that is to say, a thousand camels,
each carrying two men armed with matchlocks,—and began to
march on the fourth day; leaving orders for the rest of the tribe to
follow the second day after, in order the more to stimulate the
courage of his warriors in the battle, by the vicinity of their wives and
children. I remained with the latter, and we went to encamp at El
Jamié, one hour from the tribe El Hassnnée, and two days from
Hama. On the fifth day, the Drayhy announced to us a brilliant
victory; and shortly afterwards arrived the camels, sheep, horses and
arms, taken from the enemy. The men, who had been obliged to
remain at the tents in charge of the baggage, went out to meet the
conquerors, and demand their share of the spoil, to which they are
entitled; and the army soon appeared in triumph.
The Drayhy had taken Mehanna rather by surprise, during the
absence of Nasser; but the tribe of Hassnnée having shouted their
war-cry, the combatants proved nearly equal in numbers; the battle
lasted till night. Our warriors lost twenty-two men; but they killed
twice as many of the enemy, and took possession of their flocks.
Zaher also took the mare of Fares, the son of Mehanna, which
amongst the Bedouins is reckoned a glorious exploit.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

textbookfull.com