Uncertainty Modelling in Data Science Sébastien

Destercke (Ed.) download

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

Download full version ebook from https://textbookfull.com

We believe these products will be a great fit for you. Click
the link to download now, or visit textbookfull.com
to discover even more!

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/

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/

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/

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

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

Research in Data Science Ellen Gasparovic

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

Frontiers In Data Science Matthias Dehmer

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

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/

Subsurface Environmental Modelling Between Science and

Policy Dirk Scheer

https://textbookfull.com/product/subsurface-environmental-
modelling-between-science-and-policy-dirk-scheer/
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
Discovering Diverse Content Through
Random Scribd Documents
continued western winds. The lake, or that sea, may thus have been
so raised as to overflow the low lands adjacent to it, as those of
Egypt and Armenia, which, according to a tradition of the Egyptians
and Hebrews, were overflowed about two thousand three hundred
years before the Christian era; those of Attica, said to have been
overflowed in the time of Ogyges, about five hundred years later;
and those of Thessaly, in the time of Deucalion, still three hundred
years posterior. But such deluges as these will not account for the
shells found in the higher lands. A second opinion has been
entertained, which is, that in times anterior to the records either of
history or tradition, the bed of the ocean, the principal residence of
the shelled tribe, has, by some great convulsion of nature, been
heaved to the heights at which we now find shells and other marine
animals. The favorers of this opinion do well to suppose the great
events on which it rests to have taken place beyond all the eras of
history; for within these, certainly, none such are to be found; and
we may venture to say farther, that no fact has taken place, either in
our own days, or in the thousands of years recorded in history,
which proves the existence of any natural agents, within or without
the bowels of the earth, of force sufficient to heave, to the height of
fifteen thousand feet, such masses as the Andes. The difference
between the power necessary to produce such an effect, and that
which shuffled together the different parts of Calabria in our days, is
so immense, that from the existence of the latter, we are not
authorized to infer that of the former.
M. de Voltaire has suggested a third solution of this difficulty. (Quest.
Encycl. Coquilles.) He cites an instance in Touraine, where, in the
space of eighty years, a particular spot of earth had been twice
metamorphosed into soft stone, which had become hard when
employed in building. In this stone shells of various kinds were
produced, discoverable at first only with a microscope, but
afterwards growing with the stone. From this fact, I suppose, he
would have us infer, that, besides the usual process for generating
shells by the elaboration of earth and water in animal vessels, nature
may have provided an equivalent operation, by passing the same
materials through the pores of calcareous earths and stones; as we
see calcareous drop-stones generating every day, by the percolation
of water through limestone, and new marble forming in the quarries
from which the old has been taken out. And it might be asked,
whether is it more difficult for nature to shoot the calcareous juice
into the form of a shell, than other juices into the forms of crystals,
plants, animals, according to the construction of the vessels through
which they pass? There is a wonder somewhere. Is it greatest on
this branch of the dilemma; on that which supposes the existence of
a power, of which we have no evidence in any other case; or on the
first, which requires us to believe the creation of a body of water
and its subsequent annihilation? The establishment of the instance,
cited by M. de Voltaire, of the growth of shells unattached to animal
bodies, would have been that of his theory. But he has not
established it. He has not even left it on ground so respectable as to
have rendered it an object of inquiry to the literati of his own
country. Abandoning this fact, therefore, the three hypotheses are
equally unsatisfactory; and we must be contented to acknowledge,
that this great phenomenon is as yet unsolved. Ignorance is
preferable to error; and he is less remote from the truth who
believes nothing, than he who believes what is wrong.
There is great abundance (more especially when you approach the
mountains) of stone, white, blue, brown, &c., fit for the chisel, good
mill-stone, such also as stands the fire, and slate stone. We are told
of flint, fit for gun-flints, on the Meherrin in Brunswick, on the
Mississippi between the mouth of the Ohio and Kaskaskia, and on
others of the western waters. Isinglass or mica is in several places;
loadstone also; and an Asbestos of a ligneous texture, is sometimes
to be met with.
Marle abounds generally. A clay, of which, like the Sturbridge in
England, bricks are made, which will resist long the violent action of
fire, has been found on Tuckahoe creek of James' river, and no
doubt will be found in other places. Chalk is said to be in Botetourt
and Bedford. In the latter county is some earth believed to be
gypseous. Ochres are found in various parts.
In the lime-stone country are many caves, the earthy floors of which
are impregnated with nitre. On Rich creek, a branch of the Great
Kanhaway, about sixty miles below the lead mines, is a very large
one, about twenty yards wide, and entering a hill a quarter or half a
mile. The vault is of rock, from nine to fifteen or twenty feet above
the floor. A Mr. Lynch, who gives me this account, undertook to
extract the nitre. Besides a coat of the salt which had formed on the
vault and floor, he found the earth highly impregnated to the depth
of seven feet in some places, and generally of three, every bushel
yielding on an average three pounds of nitre. Mr. Lynch having made
about ten hundred pounds of the salt from it, consigned it to some
others, who have since made ten thousand pounds. They have done
this by pursuing the cave into the hill, never trying a second time the
earth they have once exhausted, to see how far or soon it receives
another impregnation. At least fifty of these caves are worked on the
Greenbriar. There are many of them known on Cumberland river.
The country westward of the Alleghany abounds with springs of
common salt. The most remarkable we have heard of are at Bullet's-
lick, the Big-bones, the Blue-licks, and on the north fork of Holston.
The area of Bullet's-lick is of many acres. Digging the earth to the
depth of three feet the water begins to boil up, and the deeper you
go and the drier the weather, the stronger is the brine. A thousand
gallons of water yield from a bushel to a bushel and a half of salt,
which is about eighty pounds of water to one pound of salt. So that
sea-water is more than three times as strong as that of these
springs. A salt spring has been lately discovered at the Turkey foot
on Yohogany, by which river it is overflowed, except at very low
water. Its merit is not yet known. Dunning's lick is also as yet
untried, but it is supposed to be the best on this side the Ohio. The
salt springs on the margin of the Onondago lake are said to give a
saline taste to the waters of the lake.
There are several medicinal springs, some of which are indubitably
efficacious, while others seem to owe their reputation as much to
fancy and change of air and regimen, as to their real virtues. None
of them having undergone a chemical analysis in skilful hands, nor
been so far the subject of observations as to have produced a
reduction into classes of the disorders which they relieve; it is in my
power to give little more than an enumeration of them.
The most efficacious of these are two springs in Augusta near the
first sources of James' river, where it is called Jackson's river. They
rise near the foot of the ridge of mountains generally called the
Warm spring mountains, but in the maps Jackson's mountains. The
one distinguished by the name of the Warm spring, and the other of
the Hot spring. The Warm spring issues with a very bold stream,
sufficient to work a grist mill and to keep the waters of its basin,
which is thirty feet in diameter, at the vital warmth, viz. 96° of
Fahrenheit's thermometer. The matter with which these waters is
allied is very volatile; its smell indicates it to be sulphureous, as also
does the circumstance of its turning silver black. They relieve
rheumatisms. Other complaints also of very different natures have
been removed or lessened by them. It rains here four or five days in
every week.
The Hot spring is about six miles from the Warm, is much smaller,
and has been so hot as to have boiled an egg. Some believe its
degree of heat to be lessened. It raises the mercury in Fahrenheit's
thermometer to 112 degrees, which is fever heat. It sometimes
relieves where the Warm fails. A fountain of common water, issuing
within a few inches of its margin, gives it a singular appearance.
Comparing the temperature of these with that of the Hot springs of
Kamschatka, of which Krachininnikow gives an account, the
difference is very great, the latter raising the mercury to 200° which
is within 12° of boiling water. These springs are very much resorted
to in spite of a total want of accommodation for the sick. Their
waters are strongest in the hottest months, which occasions their
being visited in July and August principally.
The Sweet springs are in the county of Botetourt, at the eastern foot
of the Alleghany, about forty-two miles from the Warm springs. They
are still less known. Having been found to relieve cases in which the
others had been ineffectually tried, it is probable their composition is
different. They are different also in their temperature, being as cold
as common water; which is not mentioned, however, as a proof of a
distinct impregnation. This is among the first sources of James' river.
On Potomac river, in Berkley county, above the North mountain, are
medicinal springs, much more frequented than those of Augusta.
Their powers, however, are less, the waters weakly mineralized, and
scarcely warm. They are more visited, because situated in a fertile,
plentiful, and populous country, better provided with
accommodations, always safe from the Indians, and nearest to the
more populous States.
In Louisa county, on the head waters of the South Ann branch of
York river, are springs of some medicinal virtue. They are not much
used however. There is a weak chalybeate at Richmond; and many
others in various parts of the country, which are of too little worth,
or too little note, to be enumerated after those before mentioned.
We are told of a sulphur spring on Howard's creek of Greenbriar, and
another at Boonsborough on Kentucky.
In the low grounds of the Great Kanhaway, seven miles above the
mouth of Elk river, and sixty-seven above that of the Kanhaway itself,
is a hole in the earth of the capacity of thirty or forty gallons, from
which issues constantly a bituminous vapor, in so strong a current as
to give to the sand about its orifice the motion which it has in a
boiling spring. On presenting a lighted candle or torch within
eighteen inches of the hole it flames up in a column of eighteen
inches in diameter, and four or five feet height, which sometimes
burns out within twenty minutes, and at other times has been
known to continue three days, and then has been still left burning.
The flame is unsteady, of the density of that of burning spirits, and
smells like burning pit coal. Water sometimes collects in the basin,
which is remarkably cold, and is kept in ebullition by the vapor
issuing through it. If the vapor be fired in that state, the water soon
becomes so warm that the hand cannot bear it, and evaporates
wholly in a short time. This, with the circumjacent lands, is the
property of His Excellency General Washington and of General Lewis.
There is a similar one on Sandy river, the flame of which is a column
of about twelve inches diameter, and three feet high. General Clarke,
who informs me of it, kindled the vapor, staid about an hour, and left
it burning.
The mention of uncommon springs leads me to that of Syphon
fountains. There is one of these near the intersection of the Lord
Fairfax's boundary with the North mountain, not far from Brock's
gap, on the stream of which is a grist mill, which grinds two bushel
of grain at every flood of the spring; another near Cow-pasture river,
a mile and a half below its confluence with the Bull-pasture river,
and sixteen or seventeen miles from Hot springs, which intermits
once in every twelve hours; one also near the mouth of the north
Holston.
After these may be mentioned the Natural Well, on the lands of a Mr.
Lewis in Frederick county. It is somewhat larger than a common
well; the water rises in it as near the surface of the earth as in the
neighboring artificial wells, and is of a depth as yet unknown. It is
said there is a current in it tending sensibly downwards. If this be
true, it probably feeds some fountain, of which it is the natural
reservoir, distinguished from others, like that of Madison's cave, by
being accessible. It is used with a bucket and windlass as an
ordinary well.
A complete catalogue of the trees, plants, fruits, &c., is probably not
desired. I will sketch out those which would principally attract notice,
as being first, Medicinal; second, Esculent; third, Ornamental; or
four, useful for fabrication; adding the Linnæan to the popular
names, as the latter might not convey precise information to a
foreigner. I shall confine myself too to native plants.
1. Senna. Cassia ligustrina.
Arsmart. Polygonum Sagittatum.
Clivers, or goose-grass. Galium spurium.
Lobelia of several species.
Palma Christi. Ricinus.
(3,) Jamestown weed. Datura Stramonium.
Mallow. Malva rotundafolia.
Syrian mallow. Hibiscus moschentos.
Hibiscus Virginicus.
Indian mallow. Sida rhombifolia.
Sida abutilon.
Virginia marshmallow. Napæa hermaphrodita.
Napæa dioica.
Indian physic. Spirea trifoliata.
Euphorbia Ipecacuanhæ.
Pleurisy root. Asclepias decumbens.
Virginia snake-root. Aristolochia serpentaria.
Black snake-root. Actæa racemosa.
Seneca rattlesnake-root. Polygala Senega.
Valerian. Valeriana locusta radiata.
Gentiana, Saponaria, Villosa & Centaurium.
Ginseng. Panax quinquefolius.
Angelica. Angelica sylvestris.
Cassava. Jatropha urens.
2. Tuckahoe. Lycoperdon tuber.
Jerusalem artichoke. Helianthus tuberosus.
Long potatoes. Convolvulus batatas.
Granadillas. Maycocks, Maracocks, Passiflora incarnata.
Panic. Panicum of many species.
Indian millet. Holcus laxus.
Indian millet. Holcus striosus.
Wild oat. Zizania aquatica.
Wild pea. Dolichos of Clayton.
Lupine. Lupinus perennis.
Wild hop. Humulus lupulus.
Wild cherry. Prunus Virginiana.
Cherokee plum. Prunus sylvestris fructu majori.
Clayton.
Wild plum. Prunus sylvestris fructu minori. Clayton.
Wild crab apple. Pyrus coronaria.
Red mulberry. Morus rubra.
Persimmon. Diospiros Virginiana.
Sugar maple. Acer saccarinum.
Scaly bark hiccory. Juglans alba cortice squamoso.
Clayton.
Common hiccory. Juglans alba, fructu minore rancido.
Clayton.
Paccan, or Illinois nut. Not described by Linnæus,
Millar, or Clayton. Were I to venture to describe
this, speaking of the fruit from memory, and of
the leaf from plants of two years' growth, I
should specify it as Juglans alba, foliolis
lanceolatis, acuminatis, serratis, tomentosis,
fructu minore, ovato, compresso, vix insculpto,
dulci, putamine tenerrimo. It grows on the
Illinois, Wabash, Ohio, and Mississippi. It is
spoken of by Don Ulloa under the name of
Pacanos, in his Noticias Americanas. Entret. 6.
Black walnut. Juglans nigra.
White walnut. Juglans alba.
Chesnut. Fagus castanea.
Chinquapin. Fagus pumila.
Hazlenut. Corylus avellana.
Grapes. Vitis. Various kinds; though only three
described by Clayton.
Scarlet strawberries. Fragaria Virginiana of Millar.
Whortleberries. Vaccinium uliginosum.
Wild gooseberries. Ribes grossularia.
Cranberries. Vaccinium oxycoccos.
Black raspberries. Rubus occidentalis.
Blackberries. Rubus fruticosus.
Dewberries. Rubus cæsius.
 Cloudberries. Rubus Chamæmorus.
3. Plane tree. Platanus occidentalis.
Poplar. Liriodendron tulipifera.
Populus heterophylla.
Black poplar. Populus nigra.
Aspen. Populus tremula.
Linden, or lime. Telia Americana.
Red flowering maple. Acer rubrum.
Horse-chesnut, or buck's-eye. Æsculus pavia.
Catalpa. Bignonia catalpa.
Umbrella. Magnolia tripetala.
Swamp laurel. Magnolia glauca.
Cucumber-tree. Magnolia acuminata.
Portugal bay. Laurus indica.
Red bay. Laurus borbonia.
Dwarf-rose bay. Rhododendron maximum.
Laurel of the western country. Qu. species?
Wild pimento. Laurus benzoin.
Sassafras. Laurus sassafras.
Locust. Robinia pseudo-acacia.
Honey-locust. Gleditsia. 1. b
Dogwood. Cornus florida.
Fringe, or snow-drop tree. Chionanthus Virginica.
Barberry. Barberis vulgaris.
Redbud, or Judas-tree. Cercis Canadensis.
Holly. Ilex aquifolium.
Cockspur hawthorn. Cratægus coccinea.
Spindle-tree. Euonymus Europæus.
Evergreen spindle-tree. Euonymus Americanus.
Itea Virginica.
Elder. Sambucus nigra.
Papaw. Annona triloba.
Candleberry myrtle. Myrica cerifera.
Dwarf laurel. Kalmia angustifolia} called ivy with us.
Kalmia latifolia }
Ivy. Hedera quinquefolia.
 Trumpet honeysuckle. Lonicera sempervirens.
Upright honeysuckle. Azalea nudiflora.
Yellow jasmine. Bignonia sempervirens.
Calycanthus floridus.
American aloe. Agave Virginica.
Sumach. Rhus. Qu. species?
Poke. Phytolacca decandra.
Long moss. Tillandsia Usneoides.
4. Reed. Arundo phragmitis.
Virginia hemp. Acnida cannabina.
Flax. Linum Virginianum.
Black, or pitch-pine. Pinus tæda.
White pine. Pinus strobus.
Yellow pine. Pinus Virginica.
Spruce pine. Pinus foliis singularibus. Clayton.
Hemlock spruce Fir. Pinus Canadensis.
Arbor vitæ. Thuya occidentalis.
Juniper. Juniperus Virginica (called cedar with us.)
Cypress. Cupressus disticha.
White cedar. Cupressus Thyoides.
Black oak. Quercus nigra.
White oak. Quercus alba.
Red oak. Quercus rubra.
Willow oak. Quercus phellos.
Chesnut oak. Quercus prinus.
Black jack oak. Quercus aquatica. Clayton.
Ground oak. Quercus pumila. Clayton.
Live oak. Quercus Virginiana. Millar.
Black birch. Betula nigra.
White birch. Betula alba.
Beach. Fagus sylvatica.
Ash. Fraxinus Americana.
Fraxinus Novæ Angliæ. Millar.
Elm. Ulmus Americana.
Willow. Salix. Qu. species?
Sweet gum. Liquidambar styraciflua.
The following were found in Virginia when first visited by the
English; but it is not said whether of spontaneous growth, or by
cultivation only. Most probably they were natives of more southern
climates, and handed along the continent from one nation to
another of the savages.
Tobacco. Nicotiana.
Maize. Zea mays.
Round potatoes. Solanum tuberosum.
Pumpkins. Cucurbita pepo.
Cymlings. Cucurbita verrucosa.
Squashes. Cucurbita melopepo.

There is an infinitude of other plants and flowers, for an

enumeration and scientific description of which I must refer to the
Flora Virginica of our great botanist, Dr. Clayton, published by
Gronovius at Leyden, in 1762. This accurate observer was a native
and resident of this State, passed a long life in exploring and
describing its plants, and is supposed to have enlarged the botanical
catalogue as much as almost any man who has lived.
Besides these plants, which are native, our farms produce wheat,
rye, barley, oats, buck-wheat, broom corn, and Indian corn. The
climate suits rice well enough, wherever the lands do. Tobacco,
hemp, flax, and cotton, are staple commodities. Indigo yields two
cuttings. The silk-worm is a native, and the mulberry, proper for its
food, grows kindly.
We cultivate, also, potatoes, both the long and the round, turnips,
carrots, parsnips, pumkins, and ground nuts (Arachis.) Our grasses
are lucerne, st. foin, burnet, timothy, ray, and orchard grass; red,
white, and yellow clover; greensward, blue grass, and crab grass.
The gardens yield musk-melons, water-melons, tomatoes, okra,
pomegranates, figs, and the esculant plants of Europe.
The orchards produce apples, pears, cherries, quinces, peaches,
nectarines, apricots, almonds, and plums.
Our quadrupeds have been mostly described by Linnæus and Mons.
de Buffon. Of these the mammoth, or big buffalo, as called by the
Indians, must certainly have been the largest. Their tradition is, that
he was carnivorous, and still exists in the northern parts of America.
A delegation of warriors from the Delaware tribe having visited the
Governor of Virginia, during the revolution, on matters of business,
after these had been discussed and settled in council, the Governor
asked them some questions relative to their country, and among
others, what they knew or had heard of the animal whose bones
were found at the Saltlicks on the Ohio. Their chief speaker
immediately put himself into an attitude of oratory, and with a pomp
suited to what he conceived the elevation of his subject, informed
him that it was a tradition handed down from their fathers, "That in
ancient times a herd of these tremendous animals came to the Big-
bone licks, and began an universal destruction of the bear, deer,
elks, buffaloes, and other animals which had been created for the
use of the Indians; that the Great Man above, looking down and
seeing this, was so enraged that he seized his lightning, descended
on the earth, seated himself on a neighboring mountain, on a rock
of which his seat and the print of his feet are still to be seen, and
hurled his bolts among them till the whole were slaughtered, except
the big bull, who presenting his forehead to the shafts, shook them
off as they fell; but missing one at length, it wounded him in the
side; whereon, springing round, he bounded over the Ohio, over the
Wabash, the Illinois, and finally over the great lakes, where he is
living at this day." It is well known, that on the Ohio, and in many
parts of America further north, tusks, grinders, and skeletons of
unparalleled magnitude, are found in great numbers, some lying on
the surface of the earth, and some a little below it. A Mr. Stanley,
taken prisoner near the mouth of the Tennessee, relates, that after
being transferred through several tribes, from one to another, he
was at length carried over the mountains west of the Missouri to a
river which runs westwardly; that these bones abounded there, and
that the natives described to him the animal to which they belonged
as still existing in the northern parts of their country; from which
description he judged it to be an elephant. Bones of the same kind
have been lately found, some feet below the surface of the earth, in
salines opened on the North Holston, a branch of the Tennessee,
about the latitude of 36½° north. From the accounts published in
Europe, I suppose it to be decided that these are of the same kind
with those found in Siberia. Instances are mentioned of like animal
remains found in the more southern climates of both hemispheres;
but they are either so loosely mentioned as to leave a doubt of the
fact, so inaccurately described as not to authorize the classing them
with the great northern bones, or so rare as to found a suspicion
that they have been carried thither as curiosities from the northern
regions. So that, on the whole, there seem to be no certain vestiges
of the existence of this animal farther south than the salines just
mentioned. It is remarkable that the tusks and skeletons have been
ascribed by the naturalists of Europe to the elephant, while the
grinders have been given to the hippopotamus, or river horse. Yet it
is acknowledged, that the tusks and skeletons are much larger than
those of the elephant, and the grinders many times greater than
those of the hippopotamus, and essentially different in form.
Wherever these grinders are found, there also we find the tusks and
skeleton; but no skeleton of the hippopotamus nor grinders of the
elephant. It will not be said that the hippopotamus and elephant
came always to the same spot, the former to deposit his grinders,
and the latter his tusks and skeleton. For what became of the parts
not deposited there? We must agree then, that these remains belong
to each other, that they are of one and the same animal, that this
was not a hippopotamus, because the hippopotamus had no tusks,
nor such a frame, and because the grinders differ in their size as
well as in the number and form of their points. That this was not an
elephant, I think ascertained by proofs equally decisive. I will not
avail myself of the authority of the celebrated[4] anatomist, who,
from an examination of the form and structure of the tusks, has
declared they were essentially different from those of the elephant;
because another[5] anatomist, equally celebrated, has declared, on a
like examination, that they are precisely the same. Between two
such authorities I will suppose this circumstance equivocal. But, 1.
The skeleton of the mammoth (for so the incognitum has been
called) bespeaks an animal of five or six times the cubic volume of
the elephant, as Mons. de Buffon has admitted. 2. The grinders are
five times as large, are square, and the grinding surface studded
with four or five rows of blunt points; whereas those of the elephant
are broad and thin, and their grinding surface flat. 3. I have never
heard an instance, and suppose there has been none, of the grinder
of an elephant being found in America. 4. From the known
temperature and constitution of the elephant, he could never have
existed in those regions where the remains of the mammoth have
been found. The elephant is a native only of the torrid zone and its
vicinities; if, with the assistance of warm apartments and warm
clothing, he has been preserved in the temperate climates of
Europe, it has only been for a small portion of what would have
been his natural period, and no instance of his multiplication in them
has ever been known. But no bones of the mammoth, as I have
before observed, have been ever found further south than the
salines of Holston, and they have been found as far north as the
Arctic circle. Those, therefore, who are of opinion that the elephant
and mammoth are the same, must believe, 1. That the elephant
known to us can exist and multiply in the frozen zone; or, 2. That an
eternal fire may once have warmed those regions, and since
abandoned them, of which, however, the globe exhibits no
unequivocal indications; or, 3. That the obliquity of the ecliptic, when
these elephants lived, was so great as to include within the tropics
all those regions in which the bones are found; the tropics being, as
is before observed, the natural limits of habitation for the elephant.
But if it be admitted that this obliquity has really decreased, and we
adopt the highest rate of decrease yet pretended, that is, of one
minute in a century, to transfer the northern tropic to the Arctic
circle, would carry the existence of these supposed elephants two
hundred and fifty thousand years back; a period far beyond our
conception of the duration of animal bones less exposed to the open
air than these are in many instances. Besides, though these regions
would then be supposed within the tropics, yet their winters would
have been too severe for the sensibility of the elephant. They would
have had, too, but one day and one night in the year, a circumstance
to which we have no reason to suppose the nature of the elephant
fitted. However, it has been demonstrated, that, if a variation of
obliquity in the ecliptic takes place at all, it is vibratory, and never
exceeds the limits of nine degrees, which is not sufficient to bring
these bones within the tropics. One of these hypotheses, or some
other equally voluntary and inadmissible to cautious philosophy,
must be adopted to support the opinion that these are the bones of
the elephant. For my own part, I find it easier to believe that an
animal may have existed, resembling the elephant in his tusks, and
general anatomy, while his nature was in other respects extremely
different. From the 30th degree of south latitude to the 30th degree
of north, are nearly the limits which nature has fixed for the
existence and multiplication of the elephant known to us. Proceeding
thence northwardly to 36½ degrees, we enter those assigned to the
mammoth. The farther we advance north, the more their vestiges
multiply as far as the earth has been explored in that direction; and
it is as probable as otherwise, that this progression continues to the
pole itself, if land extends so far. The centre of the frozen zone, then,
may be the acme of their vigor, as that of the torrid is of the
elephant. Thus nature seems to have drawn a belt of separation
between these two tremendous animals, whose breadth, indeed, is
not precisely known, though at present we may suppose it about 6½
degrees of latitude; to have assigned to the elephant the regions
south of these confines, and those north to the mammoth, founding
the constitution of the one in her extreme of heat, and that of the
other in the extreme of cold. When the Creator has therefore
separated their nature as far as the extent of the scale of animal life
allowed to this planet would permit, it seems perverse to declare it
the same, from a partial resemblance of their tusks and bones. But
to whatever animal we ascribe these remains, it is certain such a
one has existed in America, and that it has been the largest of all
terrestrial beings. It should have sufficed to have rescued the earth
it inhabited, and the atmosphere it breathed, from the imputation of
impotence in the conception and nourishment of animal life on a
large scale; to have stifled, in its birth, the opinion of a writer, the
most learned, too, of all others in the science of animal history, that
in the new world, "La nature vivante est beaucoup moins agissante,
beaucoup moins forte:"[6] that nature is less active, less energetic
on one side of the globe than she is on the other. As if both sides
were not warmed by the same genial sun; as if a soil of the same
chemical composition was less capable of elaboration into animal
nutriment; as if the fruits and grains from that soil and sun yielded a
less rich chyle, gave less extension to the solids and fluids of the
body, or produced sooner in the cartilages, membranes, and fibres,
that rigidity which restrains all further extension, and terminates
animal growth. The truth is, that a pigmy and a Patagonian, a
mouse and a mammoth, derive their dimensions from the same
nutritive juices. The difference of increment depends on
circumstances unsearchable to beings with our capacities. Every race
of animals seems to have received from their Maker certain laws of
extension at the time of their formation. Their elaborate organs were
formed to produce this, while proper obstacles were opposed to its
further progress. Below these limits they cannot fall, nor rise above
them. What intermediate station they shall take may depend on soil,
on climate, on food, on a careful choice of breeders. But all the
manna of heaven would never raise the mouse to the bulk of the
mammoth.

The opinion advanced by the Count de Buffon,[7] is 1. That the

animals common both to the old and new world are smaller in the
latter. 2. That those peculiar to the new are on a smaller scale. 3.
That those which have been domesticated in both have degenerated
in America; and 4. That on the whole it exhibits fewer species. And
the reason he thinks is, that the heats of America are less; that more
waters are spread over its surface by nature, and fewer of these
drained off by the hand of man. In other words, that heat is friendly,
and moisture adverse to the production and development of large
quadrupeds. I will not meet this hypothesis on its first doubtful
ground, whether the climate of America be comparatively more
humid? Because we are not furnished with observations sufficient to
decide this question. And though, till it be decided, we are as free to
deny as others are to affirm the fact, yet for a moment let it be
supposed. The hypothesis, after this supposition, proceeds to
another; that moisture is unfriendly to animal growth. The truth of
this is inscrutable to us by reasonings à priori. Nature has hidden
from us her modus agendi. Our only appeal on such questions is to
experience; and I think that experience is against the supposition. It
is by the assistance of heat and moisture that vegetables are
elaborated from the elements of earth, air, water, and fire. We
accordingly see the more humid climates produce the greater
quantity of vegetables. Vegetables are mediately or immediately the
food of every animal; and in proportion to the quantity of food, we
see animals not only multiplied in their numbers, but improved in
their bulk, as far as the laws of their nature will admit. Of this
opinion is the Count de Buffon himself in another part of his work;[8]
"en general il paroit ques les pays un peu froids conviennent mieux á
nos boeufs que les pays chauds, et qu'ils sont d'autant plus gross et
plus grands que le climat est plus humide et plus abondans en
paturages. Les boeufs de Danemarck, de la Podolie, de l'Ulkraine et
de la Tartarie qu habitent les Calmouques sont les plus grands de
tous." Here then a race of animals, and one of the largest too, has
been increased in its dimensions by cold and moisture, in direct
opposition to the hypothesis, which supposes that these two
circumstances diminish animal bulk, and that it is their contraries
heat and dryness which enlarge it. But when we appeal to
experience we are not to rest satisfied with a single fact. Let us,
therefore, try our question on more general ground. Let us take two
portions of the earth, Europe and America for instance, sufficiently
extensive to give operation to general causes; let us consider the
circumstances peculiar to each, and observe their effect on animal
nature. America, running through the torrid as well as temperate
zone, has more heat collectively taken, than Europe. But Europe,
according to our hypothesis, is the dryest. They are equally adapted
then to animal productions; each being endowed with one of those
causes which befriend animal growth, and with one which opposes
it. If it be thought unequal to compare Europe with America, which
is so much larger, I answer, not more so than to compare America
with the whole world. Besides, the purpose of the comparison is to
try an hypothesis, which makes the size of animals depend on the
heat and moisture of climate. If, therefore, we take a region so
extensive as to comprehend a sensible distinction of climate, and so
extensive too as that local accidents, or the intercourse of animals
on its borders, may not materially affect the size of those in its
interior parts, we shall comply with those conditions which the
hypothesis may reasonably demand. The objection would be the
weaker in the present case, because any intercourse of animals
which may take place on the confines of Europe and Asia, is to the
advantage of the former, Asia producing certainly larger animals than
Europe. Let us then take a comparative view of the quadrupeds of
Europe and America, presenting them to the eye in three different
tables, in one of which shall be enumerated those found in both
countries; in a second, those found in one only; in a third, those
which have been domesticated in both. To facilitate the comparison,
let those of each table be arranged in gradation according to their
sizes, from the greatest to the smallest, so far as their sizes can be
conjectured. The weights of the large animals shall be expressed in
the English avoirdupois and its decimals; those of the smaller, in the
same ounce and its decimals. Those which are marked thus *, are
actual weights of particular subjects, deemed among the largest of
their species. Those marked thus †, are furnished by judicious
persons, well acquainted with the species, and saying, from
conjecture only, what the largest individual they had seen would
probably have weighed. The other weights are taken from Messrs.
Buffon and D'Aubenton, and are of such subjects as came casually to
their hands for dissection. This circumstance must be remembered
where their weights and mine stand opposed; the latter being stated
not to produce a conclusion in favor of the American species, but to
justify a suspension of opinion until we are better informed, and a
suspicion, in the meantime, that there is no uniform difference in
favor of either; which is all I pretend.
A comparative view of the Quadrupeds of Europe and of America.

I. ABORIGINALS OF BOTH.
Europe. America.
lb. lb.
Mammoth
Buffalo. Bison *1800
White Bear. Ours blanc
Carribou. Renne
Bear. Ours 153.7 *410
Elk. Elan. Original palmated
Red deer. Cerf 288.8 *273
Fallow Deer. Daim 167.8
Wolf. Loup 69.8
Roe. Chevreuil 56.7
Glutton. Glouton. Carcajou
Wild cat. Chat sauvage †30
Lynx. Loup cervier 25.
Beaver. Castor 18.5 *45
Badger. Blaireau 13.6
Red fox. Renard 13.5
Gray fox. Isatis
Otter. Loutre 8.9 †12
Monax. Marmotte 6.5
Vison. Fouine 2.8
Hedgehog. Herisson 2.2
Marten. Marte 1.9 †6
oz.
Water rat. Rat d'eau 7.5
Weasel. Belette 2.2 oz.
Flying squirrel. Polatouche 2.2 †4
Shrew mouse. Musaraigne 1.

II. ABORIGINALS OF ONE ONLY.

EUROPE.
lb.
Sanglier. Wild boar 280.
Mouflon. Wild sheep 56.
Bouquetin. Wild goat
Lievre. Hare 7.6
Lapin. Rabbit 3.4
Putois. Polecat 3.3
Genette 3.1
Desman. Muskrat oz.
Ecureuil. Squirrel 12.
Hermine. Ermin 8.2
Rat. Rat 7.5
Loirs 3.1
Lerot. Dormouse 1.8
Taupe. Mole 1.2
Hampster .6
Zisel
Leming
Souris. Mouse .6

AMERICA
lb.
Tapir 534.
Elk, round horned †450.
Puma
Jaguar 218.
Cabiai 109.
Tamanoir 109.
Tammandua 65.4
Cougar of North-America 75.
Cougar of South-America 59.4
Ocelot
Pecari 46.3
Jaguaret 43.6
Alco
Lama
Paco
Paca 32.7
Serval
Sloth. Unau 27.25
Saricovienne
Kincajou
Tatou Kabassou 21.8
Urson. Urchin
Raccoon. Raton 16.5
Coati
Coendou 16.3
Sloth. Aï 13.
Sapajou Ouarini
Sapajou Coaita 9.8
Tatou Encubert
Tatou Apar
Tatou Cachiea 7.
Little Coendou 6.5
Opossum. Sarigu
Tapeti
Margay
Crabier
Agouti 4.2
Sapajou Saï 3.5
Tatou Cirquinçon
Tatou Tatouate 3.3
Mouffette Squash
Mouffette Chinche
Mouffette Conepate
Scunk
Mouffette. Zorilla
Whabus. Hare. Rabbit
Aperea
Akouchi
Ondatra. Muskrat
Pilori
Great gray squirrel †2.7
Fox squirrel of Virginia †2.625
Surikate 2.
Mink †2.
Sapajou. Sajou 1.8
Indian pig. Cochon d'Inde 1.6
Sapajou Saïmiri 1.5
Phalanger
Coqualain
Lesser gray squirrel †1.5
Black squirrel †1.5
oz.
Red squirrel 10.
Sagoin Saki
Sagoin Pinche
Sagoin Tamarin
Sagoin Ouistiti 4.4
Sagoin Marakine
Sagoin Mico
Cayopollin
 Fourmillier
Marmose
Sarigue of Cayenne
Tucan
Red mole
Ground squirrel 4.

III. DOMESTICATED IN BOTH.

Europe. America.
lb. lb.
Cow 765. *2500
Horse *1366
Ass
Hog *1200
Sheep *125
Goat *80
Dog 67.6
Cat 7.

I have not inserted in the first table the Phoca,[9] nor leather-winged
bat, because the one living half the year in the water, and the other
being a winged animal, the individuals of each species may visit both
continents.
Of the animals in the first table, Monsieur de Buffon himself informs
us, [XXVII. 130, XXX. 213,] that the beaver, the otter, and shrew
mouse, though of the same species, are larger in America than in
Europe. This should therefore have corrected the generality of his
expressions, XVIII. 145, and elsewhere, that the animals common to
the two countries, are considerably less in America than in Europe,
"et cela sans aucune exception." He tells us too, [Quadrup. VIII.
334, edit. Paris, 1777,] that on examining a bear from America, he
remarked no difference, "dans la forme de cet ours d'Amerique
comparé a celui d'Europe," but adds from Bartram's journal, that an
American bear weighed four hundred pounds, English, equal to three
hundred and sixty-seven pounds French; whereas we find the
European bear examined by Mons. D'Aubenton, [XVII. 82,] weighed
but one hundred and forty-one pounds French. That the palmated
elk is larger in America than in Europe, we are informed by Kalm,[10]
a naturalist, who visited the former by public appointment, for the
express purpose of examining the subjects of natural history. In this
fact Pennant concurs with him. [Barrington's Miscellanies.] The same
Kalm tells us[11] that the black moose, or renne of America, is as
high as a tall horse; and Catesby,[12] that it is about the bigness of a
middle-sized ox. The same account of their size has been given me
by many who have seen them. But Monsieur D'Aubenton says[13]
that the renne of Europe is about the size of a red deer. The weasel
is larger in America than in Europe, as may be seen by comparing its
dimensions as reported by Monsieur D'Aubenton[14] and Kalm. The
latter tells us,[15] that the lynx, badger, red fox, and flying squirrel,
are the same in America as in Europe; by which expression I
understand, they are the same in all material circumstances, in size
as well as others; for if they were smaller, they would differ from the
European. Our gray fox is, by Catesby's account,[16] little different in
size and shape from the European fox. I presume he means the red
fox of Europe, as does Kalm, where he says,[17] that in size "they do
not quite come up to our foxes." For proceeding next to the red fox
of America, he says, "they are entirely the same with the European
sort;" which shows he had in view one European sort only, which
was the red. So that the result of their testimony is, that the
American gray fox is somewhat less than the European red; which is
equally true of the gray fox of Europe, as may be seen by comparing
the measures of the Count de Buffon and Monsieur D'Aubenton.[18]
The white bear of America is as large as that of Europe. The bones
of the mammoth which has been found in America, are as large as
those found in the old world. It may be asked, why I insert the
mammoth, as if it still existed? I ask in return, why I should omit it,
as if it did not exist? Such is the economy of nature, that no instance
can be produced, of her having permitted any one race of her
animals to become extinct; of her having formed any link in her
great work so weak as to be broken. To add to this, the traditionary
testimony of the Indians, that this animal still exists in the northern
and western parts of America, would be adding the light of a taper
to that of the meridian sun. Those parts still remain in their
aboriginal state, unexplored and undisturbed by us, or by others for
us. He may as well exist there now, as he did formerly where we
find his bones. If he be a carnivorous animal, as some anatomists
have conjectured, and the Indians affirm, his early retirement may
be accounted for from the general destruction of the wild game by
the Indians, which commences in the first instant of their connection
with us, for the purpose of purchasing match-coats, hatchets, and
firelocks, with their skins. There remain then the buffalo, red deer,
fallow deer, wolf, roe, glutton, wild cat, monax, bison, hedgehog,
marten, and water-rat, of the comparative sizes of which we have
not sufficient testimony. It does not appear that Messieurs de Buffon
and D'Aubenton have measured, weighed, or seen those of America.
It is said of some of them, by some travellers, that they are smaller
than the European. But who were these travellers? Have they not
been men of a very different description from those who have laid
open to us the other three quarters of the world? Was natural
history the object of their travels? Did they measure or weigh the
animals they speak of? or did they not judge of them by sight, or
perhaps even from report only? Were they acquainted with the
animals of their own country, with which they undertake to compare
them? Have they not been so ignorant as often to mistake the
species? A true answer to these questions would probably lighten
their authority, so as to render it insufficient for the foundation of an
hypothesis. How unripe we yet are, for an accurate comparison of
the animals of the two countries, will appear from the work of
Monsieur de Buffon. The ideas we should have formed of the sizes
of some animals, from the information he had received at his first
publications concerning them, are very different from what his
subsequent communications give us. And indeed his candor in this
can never be too much praised. One sentence of his book must do
him immortal honor. "J'aime autant une personne qui me releve
d'une erreur, qu'une autre qui m'apprend une verité, parce qu'en
effet une erreur corrigée est une verité."[19] He seems to have
thought the cabiai he first examined wanted little of its full growth.
"Il n'etoit pas encore tout-a-fait adulte."[20] Yet he weighed but
forty-six and a half pounds, and he found afterwards,[21] that these
animals, when full grown, weigh one hundred pounds. He had
supposed, from the examination of a jaguar,[22] said to be two years
old, which weighed but sixteen pounds twelve ounces, that when he
should have acquired his full growth, he would not be larger than a
middle-sized dog. But a subsequent account[23] raises his weight to
two hundred pounds. Further information will, doubtless, produce
further corrections. The wonder is, not that there is yet something in
this great work to correct, but that there is so little. The result of this
view then is, that of twenty-six quadrupeds common to both
countries, seven are said to be larger in America, seven of equal
size, and twelve not sufficiently examined. So that the first table
impeaches the first member of the assertion, that of the animals
common to both countries, the American are smallest, "et cela sans
aucune exception." It shows it is not just, in all the latitude in which
its author has advanced it, and probably not to such a degree as to
found a distinction between the two countries.
Proceeding to the second table, which arranges the animals found in
one of the two countries only, Monsieur de Buffon observes, that the
tapir, the elephant of America, is but of the size of a small cow. To
preserve our comparison, I will add, that the wild boar, the elephant
of Europe, is little more than half that size. I have made an elk with
round or cylindrical horns an animal of America, and peculiar to it;
because I have seen many of them myself, and more of their horns;
and because I can say, from the best information, that, in Virginia,
this kind of elk has abounded much, and still exists in smaller
numbers; and I could never learn that the palmated kind had been
seen here at all. I suppose this confined to the more northern
latitudes.[24] I have made our hare or rabbit peculiar, believing it to
be different from both the European animals of those
denominations, and calling it therefore by its Algonquin name,
Whabus, to keep it distinct from these. Kalm is of the same opinion.
[25] I have enumerated the squirrels according to our own
knowledge, derived from daily sight of them, because I am not able
to reconcile with that the European appellations and descriptions. I
have heard of other species, but they have never come within my
own notice. These, I think, are the only instances in which I have
departed from the authority of Monsieur de Buffon in the
construction of this table. I take him for my ground work, because I
think him the best informed of any naturalist who has ever written.
The result is, that there are eighteen quadrupeds peculiar to Europe;
more than four times as many, to wit, seventy four, peculiar to
America; that the[26] first of these seventy-four weighs more than
the whole column of Europeans; and consequently this second table
disproves the second member of the assertion, that the animals
peculiar to the new world are on a smaller scale, so far as that
assertion relied on European animals for support; and it is in full
opposition to the theory which makes the animal volume to depend
on the circumstances of heat and moisture.
The third table comprehends those quadrupeds only which are
domestic in both countries. That some of these, in some parts of
America, have become less than their original stock, is doubtless
true; and the reason is very obvious. In a thinly-peopled country, the
spontaneous productions of the forests, and waste fields, are
sufficient to support indifferently the domestic animals of the farmer,
with a very little aid from him, in the severest and scarcest season.
He therefore finds it more convenient to receive them from the hand
of nature in that indifferent state, than to keep up their size by a
care and nourishment which would cost him much labor. If, on this
low fare, these animals dwindle, it is no more than they do in those
parts of Europe where the poverty of the soil, or the poverty of the
owner, reduces them to the same scanty subsistence. It is the
uniform effect of one and the same cause, whether acting on this or
that side of the globe. It would be erring, therefore, against this rule
of philosophy, which teaches us to ascribe like effects to like causes,
should we impute this diminution of size in America to any imbecility
or want of uniformity in the operations of nature. It may be affirmed
with truth, that, in those countries, and with those individuals in
America, where necessity or curiosity has produced equal attention,
as in Europe, to the nourishment of animals, the horses, cattle,
sheep, and hogs, of the one continent are as large as those of the
other. There are particular instances, well attested, where individuals
of this country have imported good breeders from England, and
have improved their size by care in the course of some years. To
make a fair comparison between the two countries, it will not answer
to bring together animals of what might be deemed the middle or
ordinary size of then species; because an error in judging of that
middle or ordinary size, would vary the result of the comparison.
Thus Mons. D'Aubenton[27] considers a horse of 4 feet five inches
high and 400 lb. weight French, equal to 4 feet 8.6 inches and 436
lb. English, as a middle-sized horse. Such a one is deemed a small
horse in America. The extremes must therefore be resorted to. The
same anatomist[28] dissected a horse of 5 feet 9 inches height,
French measure, equal to 6 feet 1.7 English. This is near 6 inches
higher than any horse I have seen; and could it be supposed that I
had seen the largest horses in America, the conclusion would be,
that ours have diminished, or that we have bred from a smaller
stock. In Connecticut and Rhode Island, where the climate is
favorable to the production of grass, bullocks have been slaughtered
which weighed 2,500, 2,200, and 2,100 lbs. nett; and those of 1,800
lbs. have been frequent. I have seen a hog[29] weigh 1,050 lbs. after
the blood, bowels, and hair had been taken from him. Before he was
killed, an attempt was made to weigh him with a pair of steel yards,
graduated to 1,200 lbs., but he weighed more. Yet this hog was
probably not within fifty generations of the European stock. I am
well informed of another which weighed 1,100 lbs. gross. Asses have
been still more neglected than any other domestic animal in
America. They are neither fed or housed in the most rigorous season
of the year. Yet they are larger than those measured by Mons.
D'Aubenton,[30] of 3 feet 7¼ inches, 3 feet 4 inches, and 3 feet 2½
inches, the latter weighing only 215.8 lbs. These sizes, I suppose,
have been produced by the same negligence in Europe, which has
produced a like diminution here. Where care has been taken of them
on that side of the water, they have been raised to a size bordering
on that of the horse; not by the heat and dryness of the climate, but
by good food and shelter. Goats have been also much neglected in
America. Yet they are very prolific here, bearing twice or three times
a year, and from one to five kids at a birth. Mons. de Buffon has
been sensible of a difference in this circumstance in favor of
America.[31] But what are their greatest weights, I cannot say. A
large sheep here weighs 100 lbs. I observe Mons. D'Aubenton calls a
ram of 62 lbs. one of the middle size.[32] But to say what are the
extremes of growth in these and the other domestic animals of
America, would require information of which no one individual is
possessed. The weights actually known and stated in the third table
preceding will suffice to show, that we may conclude on probable
grounds, that, with equal food and care, the climate of America will
preserve the races of domestic animals as large as the European
stock from which they are derived; and, consequently, that the third
member of Mons. de Buffon's assertion that the domestic animals
are subject to degeneration from the climate of America, is as
probably wrong as the first and second were certainly so.
That the last part of it is erroneous, which affirms that the species of
American quadrupeds are comparatively few, is evident from the
tables taken together. By these it appears that there are an hundred
species aboriginal in America. Mons. de Buffon supposes about
double that number existing on the whole earth.[33] Of these
Europe, Asia, and Africa, furnish suppose one hundred and twenty-
six; that is, the twenty-six common to Europe and America, and
about one hundred which are not in America at all. The American
species, then, are to those of the rest of the earth, as one hundred
to one hundred and twenty-six, or four to five. But the residue of the
earth being double the extent of America, the exact proportion
would have been but as four to eight.
Hitherto I have considered this hypothesis as applied to brute
animals only, and not in its extension to the man of America,
whether aboriginal or transplanted. It is the opinion of Mons. de
Buffon that the former furnishes no exception to it.[34]
 "Quoique le sauvage du nouveau monde soit à peu près de même stature que l'homme de notre
monde, cela ne suffit pas pour qu'il puisse faire une exception au fait général du rapetissement de
la nature vivante dans tout ce continent; le sauvage est foible et petit par les organes de la
génération; il n'a ni poil, ni barbe, and nulle ardeur pour sa femelle. Quoique plus léger que
l'Européen, parce qu'il a plus d'habitude à courir, il est cependant beaucoup moins fort de corps; il
est aussi bien moins sensible, et cependant plus craintif et plus lâche; il n'a nulle vivacité, nulle
activité dans l'ame; celle du corps est moins un exercise, un mouvement volontaire qu'une
nécessité d'action causée par le besoin; ôtez lui la faim et la soif, vous détruirez en même tems le
principe actif de tous ses mouvemens; il demeurera stupidement en repos sur ses jambes ou
couché pendant des jours entiers. Il ne faut pas aller chercher plus loin à cause de la vie dispersée
des sauvages et de leur éloignement pour la société; la plus précieuse étincelle du feu de la nature
leur a été refusée; ils manquent d'ardeur pour leur femelle, et par consequent d'amour pour leur
semblables; ne connoissant pas l'attachment le plus vif, le plus tendre de tous, leurs autres
sentimens de ce genre, sont froids et languissans; ils aiment foiblement leurs pères et leurs enfans;
la société la plus intime de toutes, celle de la même famille, n'a donc chez eux que de foibles liens;
la société d'une famille à l'autre n'en a point de tout; dès lors nulle réunion, nulle république, nulle
état social. La physique de l'amour fait chez eux le moral des mœurs; leur cœur est glacé, leur
societé et leur empire dur. Ils ne regardent leurs femmes que comme des servantes de peine ou
des bêtes de somme qu'ils chargent, sans ménagement, du fardeau de leur chasse, et qu'ils
forcent, sans pitié, sans reconnoissance, à des ouvrages qui souvent sont au dessus de leurs
forces; ils n'ont que peu d'enfans; ils en out peu de soin; tout se ressent de leur premier defaut; ils
sont indifférents parce qu'ils sont peu puissants, et cette indifference pour le sexe est la tache
originelle qui flétrit la nature, qui l'empeche de s'épanouir, et qui detruisant les germes de la vie,
coupe en même temps la racine de société. L'homme ne fait donc point d'exception ici. La nature
en lui refusant les puissances de l'amour l'a plus maltraité et plus rapetissé qu'aucun des animaux."
An afflicting picture, indeed, which for the honor of human nature, I am glad to believe has no original.
Of the Indian of South America I know nothing; for I would not honor with the appellation of
knowledge, what I derive from the fables published of them. These I believe to be just as true as the
fables of Æsop. This belief is founded on what I have seen of man, white, red, and black, and what has
been written of him by authors, enlightened themselves, and writing among an enlightened people. The
Indian of North America being more within our reach, I can speak of him somewhat from my own
knowledge, but more from the information of others better acquainted with him, and on whose truth
and judgment I can rely. From these sources I am able to say, in contradiction to this representation,
that he is neither more defective in ardor, nor more impotent with his female, than the white reduced to
the same diet and exercise; that he is brave, when an enterprise depends on bravery; education with
him making the point of honor consist in the destruction of an enemy by stratagem, and in the
preservation of his own person free from injury; or, perhaps, this is nature, while it is education which
teaches us to[35] honor force more than finesse; that he will defend himself against a host of enemies,
always choosing to be killed, rather than to surrender,[36] though it be to the whites, who he knows will
treat him well; that in other situations, also, he meets death with more deliberation, and endures
tortures with a firmness unknown almost to religious enthusiasm with us; that he is affectionate to his
children, careful of them, and indulgent in the extreme; that his affections comprehend his other
connections, weakening, as with us, from circle to circle, as they recede from the centre; that his
friendships are strong and faithful to the uttermost[37] extremity; that his sensibility is keen, even the
warriors weeping most bitterly on the loss of their children, though in general they endeavor to appear
superior to human events; that his vivacity and activity of mind is equal to ours in the same situation;
hence his eagerness for hunting, and for games of chance. The women are submitted to unjust
drudgery. This I believe is the case with every barbarous people. With such, force is law. The stronger
sex imposes on the weaker. It is civilization alone which replaces women in the enjoyment of their
natural equality. That first teaches us to subdue the selfish passions, and to respect those rights in
others which we value in ourselves. Were we in equal barbarism, our females would be equal drudges.
The man with them is less strong than with us, but their women stronger than ours; and both for the
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