Handbook of Diagnostic Classification Models

Models and Model Extensions Applications

Software Packages Matthias Von Davier download

https://textbookfull.com/product/handbook-of-diagnostic-
classification-models-models-and-model-extensions-applications-
software-packages-matthias-von-davier/

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!

Generalized linear models and extensions Fourth Edition

Hardin

https://textbookfull.com/product/generalized-linear-models-and-
extensions-fourth-edition-hardin/

Advances in DEA theory and applications : with

extensions to forecasting models First Edition Tone

https://textbookfull.com/product/advances-in-dea-theory-and-
applications-with-extensions-to-forecasting-models-first-edition-
tone/

Generalized Linear Models and Extensions 4th Edition

James W. Hardin

https://textbookfull.com/product/generalized-linear-models-and-
extensions-4th-edition-james-w-hardin/

Mixture Models and Applications Nizar Bouguila

https://textbookfull.com/product/mixture-models-and-applications-
nizar-bouguila/
Behavioural Models From Modelling Finite Automata to
Analysing Business Processes 1st Edition Matthias Kunze

https://textbookfull.com/product/behavioural-models-from-
modelling-finite-automata-to-analysing-business-processes-1st-
edition-matthias-kunze/

Conn's handbook of models for human aging 2nd Edition

Ram

https://textbookfull.com/product/conns-handbook-of-models-for-
human-aging-2nd-edition-ram/

Measuring agreement : models, methods, and applications

1st Edition Choudhary

https://textbookfull.com/product/measuring-agreement-models-
methods-and-applications-1st-edition-choudhary/

Computational Models of Reading A Handbook 1st Edition

Erik D Reichle

https://textbookfull.com/product/computational-models-of-reading-
a-handbook-1st-edition-erik-d-reichle/

Computational Models of Reading A Handbook 1st Edition

Erik D Reichle

https://textbookfull.com/product/computational-models-of-reading-
a-handbook-1st-edition-erik-d-reichle-2/
Methodology of Educational Measurement and Assessment

Matthias von Davier

Young-Sun Lee Editors

Handbook of
Diagnostic
Classification
Models
Models and Model Extensions,
Applications, Software Packages
Methodology of Educational Measurement
and Assessment

Series Editors
Bernard Veldkamp, Research Center for Examinations and Certification (RCEC),
University of Twente, Enschede, The Netherlands
Matthias von Davier, National Board of Medical Examiners (NBME), Philadelphia,
USA
This book series collates key contributions to a fast-developing field of education
research. It is an international forum for theoretical and empirical studies exploring
new and existing methods of collecting, analyzing, and reporting data from
educational measurements and assessments. Covering a high-profile topic from
multiple viewpoints, it aims to foster a broader understanding of fresh developments
as innovative software tools and new concepts such as competency models and skills
diagnosis continue to gain traction in educational institutions around the world.
Methodology of Educational Measurement and Assessment offers readers reliable
critical evaluations, reviews and comparisons of existing methodologies alongside
authoritative analysis and commentary on new and emerging approaches. It will
showcase empirical research on applications, examine issues such as reliability,
validity, and comparability, and help keep readers up to speed on developments in
statistical modeling approaches. The fully peer-reviewed publications in the series
cover measurement and assessment at all levels of education and feature work
by academics and education professionals from around the world. Providing an
authoritative central clearing-house for research in a core sector in education, the
series forms a major contribution to the international literature.

More information about this series at http://www.springer.com/series/13206

Matthias von Davier • Young-Sun Lee
Editors

Handbook of Diagnostic
Classification Models
Models and Model Extensions, Applications,
Software Packages

123
Editors
Matthias von Davier Young-Sun Lee
National Board of Medical Teachers College
Examiners (NBME) Columbia University
Philadelphia, PA, USA New York, NY, USA

ISSN 2367-170X ISSN 2367-1718 (electronic)

Methodology of Educational Measurement and Assessment
ISBN 978-3-030-05583-7 ISBN 978-3-030-05584-4 (eBook)
https://doi.org/10.1007/978-3-030-05584-4

© 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

The Handbook of Diagnostic Classification Models represents a collection of chap-

ters reviewing diagnostic models, their applications, and descriptions of software
tool, written by leading experts in the field. This volume covers most (one can never
claim completeness) of the current major modeling families and approaches as well
as provides a resource that can be used for self-study, teaching, or research that
applies or extends the materials included in the book.
While virtually any project of this type takes longer than expected, and many will
be tempted to remind us that Murphy’s law strikes almost surely, we were amazed
by the willingness of all contributors to put in the hours to finish their chapters and
to review other chapters and, finally, to revise their contributions in order to help
putting together a coherent volume. We hope that this process, together with some
occasional assistance from the editors and the publisher, helped to compile a multi-
authored work together that covers most aspects of doing research around diagnostic
modeling.
We also want to remind readers as well as ourselves of colleagues who passed
away and who leave a void in the research community. We lost Kikumi Tatsuoka, of
whom one can truthfully say that her rule space approach is one of the major roots,
maybe even the most important one, of this field. In her long career, she shaped
many aspects of diagnostic modeling, and we should recall that, among these, the
Q-matrix is one of the central building blocks present in the vast majority of these
methods. The rule space method is described along with other early approaches in
Chap. 1.
We furthermore would like to remember Lou DiBello, who made important
contributions to the field, notably in his modified rule space work, and his work
on the unified model together with colleagues. The work around extensions of the
unified model is described in Chap. 3. We also want to remind readers of Wen-
Chung Wang who just recently passed away. Wen-Chung and his coauthors worked
on many topics around diagnostic models and other psychometric approaches. His
work around DIF methods for use with diagnostic modeling approaches is found in
Chap. 18. We hope that the friends we lost would have liked this volume.

v
vi Preface

Ending on a more positive note: working in a dynamic field that produces

new knowledge every day, we are aware that the handbook is one stepping stone
on the long path to fully understanding the potential of these powerful modeling
approaches. We are expecting to see books that extend the material we have put
together here; moreover, we expect to see this handbook be replaced or superseded
by a new edition in a couple of years. If we are lucky, we may be involved in putting
together some of the chapters of these future collections describing what will then
be the state of the art in diagnostic modeling.

Philadelphia, PA, USA Matthias von Davier

New York, NY, USA Young-Sun Lee
Contents

1 Introduction: From Latent Classes to Cognitive

Diagnostic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Matthias von Davier and Young-Sun Lee

Part I Approaches to Cognitive Diagnosis

2 Nonparametric Item Response Theory and Mokken Scale
Analysis, with Relations to Latent Class Models and Cognitive
Diagnostic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
L. Andries van der Ark, Gina Rossi, and Klaas Sijtsma
3 The Reparameterized Unified Model System: A Diagnostic
Assessment Modeling Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
William Stout, Robert Henson, Lou DiBello, and Benjamin Shear
4 Bayesian Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Russell G. Almond and Juan-Diego Zapata-Rivera
5 Nonparametric Methods in Cognitively Diagnostic Assessment . . . . . . 107
Chia-Yi Chiu and Hans-Friedrich Köhn
6 The General Diagnostic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Matthias von Davier
7 The G-DINA Model Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Jimmy de la Torre and Nathan D. Minchen
8 Loglinear Cognitive Diagnostic Model (LCDM) . . . . . . . . . . . . . . . . . . . . . . . . 171
Robert Henson and Jonathan L. Templin
9 Diagnostic Modeling of Skill Hierarchies and Cognitive
Processes with MLTM-D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Susan E. Embretson

vii
viii Contents

10 Explanatory Cognitive Diagnostic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

Yoon Soo Park and Young-Sun Lee
11 Insights from Reparameterized DINA and Beyond. . . . . . . . . . . . . . . . . . . . . 223
Lawrence T. DeCarlo

Part II Special Topics

12 Q-Matrix Learning via Latent Variable Selection
and Identifiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Jingchen Liu and Hyeon-Ah Kang
13 Global- and Item-Level Model Fit Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Zhuangzhuang Han and Matthew S. Johnson
14 Exploratory Data Analysis for Cognitive Diagnosis: Stochastic
Co-blockmodel and Spectral Co-clustering. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Yunxiao Chen and Xiaoou Li
15 Recent Developments in Cognitive Diagnostic Computerized
Adaptive Testing (CD-CAT): A Comprehensive Review . . . . . . . . . . . . . . . 307
Xiaofeng Yu, Ying Cheng, and Hua-Hua Chang
16 Identifiability and Cognitive Diagnosis Models . . . . . . . . . . . . . . . . . . . . . . . . . 333
Gongjun Xu
17 Measures of Agreement: Reliability, Classification Accuracy,
and Classification Consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Sandip Sinharay and Matthew S. Johnson
18 Differential Item Functioning in Diagnostic Classification Models . . . 379
Xue-Lan Qiu, Xiaomin Li, and Wen-Chung Wang
19 Bifactor MIRT as an Appealing and Related Alternative
to CDMs in the Presence of Skill Attribute Continuity. . . . . . . . . . . . . . . . . 395
Daniel M. Bolt

Part III Applications

20 Utilizing Process Data for Cognitive Diagnosis. . . . . . . . . . . . . . . . . . . . . . . . . . 421
Hong Jiao, Dandan Liao, and Peida Zhan
21 Application of Cognitive Diagnostic Models to Learning
and Assessment Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Benjamin Deonovic, Pravin Chopade, Michael Yudelson,
Jimmy de la Torre, and Alina A. von Davier
22 CDMs in Vocational Education: Assessment and Usage
of Diagnostic Problem-Solving Strategies in Car Mechatronics. . . . . . . 461
Stephan Abele and Matthias von Davier
Contents ix

23 Applying the General Diagnostic Model to Proficiency Data

from a National Skills Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Xueli Xu and Matthias von Davier
24 Reduced Reparameterized Unified Model Applied to Learning
Spatial Rotation Skills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
Susu Zhang, Jeff Douglas, Shiyu Wang,
and Steven Andrew Culpepper
25 How to Conduct a Study with Diagnostic Models. . . . . . . . . . . . . . . . . . . . . . . 525
Young-Sun Lee and Diego A. Luna-Bazaldua

Part IV Software, Data, and Tools

26 The R Package CDM for Diagnostic Modeling. . . . . . . . . . . . . . . . . . . . . . . . . . 549
Alexander Robitzsch and Ann Cathrice George
27 Diagnostic Classification Modeling with flexMIRT . . . . . . . . . . . . . . . . . . . . . 573
Li Cai and Carrie R. Houts
28 Using Mplus to Estimate the Log-Linear Cognitive
Diagnosis Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
Meghan Fager, Jesse Pace, and Jonathan L. Templin
29 Cognitive Diagnosis Modeling Using the GDINA R Package . . . . . . . . . . 593
Wenchao Ma
30 GDM Software mdltm Including Parallel EM Algorithm . . . . . . . . . . . . . . 603
Lale Khorramdel, Hyo Jeong Shin, and Matthias von Davier
31 Estimating CDMs Using MCMC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
Xiang Liu and Matthew S. Johnson

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
Contributors

Stephan Abele Institute of Vocational Education and Vocational Didactics, Tech-

nische Universität Dresden, Dresden, Germany
Russell G. Almond Department of Educational Psychology and Learning Sys-
tems, Florida State University, Tallahassee, FL, USA
Daniel M. Bolt Department of Educational Psychology, University of Wisconsin –
Madison, Madison, WI, USA
Li Cai University of California, Los Angeles, CA, USA
Vector Psychometric Group, LLC, Chapel Hill, NC, USA
Hua-Hua Chang Department of Educational Studies, Purdue University, IN, USA
Yunxiao Chen London School of Economics and Political Science, London, UK
Ying Cheng Department of Psychology, University of Notre Dame, Notre Dame,
IN, USA
Chia-Yi Chiu Department of Educational Psychology, Rutgers, The State Univer-
sity of New Jersey, New Brunswick, NJ, USA
Pravin Chopade ACTNext ACT Inc., Iowa City, IA, USA
Steven Andrew Culpepper Department of Statistics, University of Illinois at
Urbana-Champaign, Champaign, IL, USA
Jimmy de la Torre Division of Learning, Development and Diversity, University
of Hong Kong, Hong Kong, China
Lawrence T. DeCarlo Department of Human Development, Teachers College,
Columbia University, New York, NY, USA
Benjamin Deonovic ACTNext ACT Inc., Iowa City, IA, USA

xi
xii Contributors

Lou DiBello Department of Statistics, University of Illinois at Urbana-Champaign,

Urbana, IL, USA
Jeff Douglas Department of Statistics, University of Illinois at Urbana-Champaign,
Champaign, IL, USA
Susan E. Embretson School of Psychology, Georgia Institute of Technology,
Atlanta, GA, USA
Meghan Fager National University, Precision Institute, La Jolla, CA, USA
Ann Cathrice George Federal Institute for Educational Research, Innovation and
Development of the Austrian School System, Salzburg, Austria
Zhuangzhuang Han Department of Human Development, Teachers College,
Columbia University, New York, NY, USA
Robert Henson Educational Research Methodology (ERM) Department, The
University of North Carolina at Greensboro, Greensboro, NC, USA
Carrie R. Houts Vector Psychometric Group, LLC, Chapel Hill, NC, USA
Hong Jiao Department of Human Development and Quantitative Methodology,
University of Maryland, College Park, MD, USA
Matthew S. Johnson Educational Testing Service, Princeton, NJ, USA
Hyeon-Ah Kang Department of Educational Psychology, University of Texas,
Austin, TX, USA
Lale Khorramdel National Board of Medical Examiners (NBME), Philadelphia,
PA, USA
Hans-Friedrich Köhn Department of Psychology, University of Illinois at Urbana-
Champaign, Champaign, IL, USA
Young-Sun Lee Teachers College, Columbia University, New York, NY, USA
Xiaomin Li The University of Hong Kong, Pokfulam, Hong Kong
Xiaoou Li School of Statistics, University of Minnesota, Minneapolis, MN, USA
Dandan Liao American Institutes for Research, Washington, DC, USA
Jingchen Liu Department of Statistics, Columbia University, New York, NY, USA
Xiang Liu Department of Human Development, Teachers College, Columbia
University, New York, NY, USA
Diego A. Luna-Bazaldua School of Psychology, National Autonomous University
of Mexico, Mexico City, Mexico
Wenchao Ma The University of Alabama, Tuscaloosa, AL, USA
Nathan D. Minchen Pearson, Bronx, NY, USA
Contributors xiii

Jesse Pace University of Kansas, Lawrence, KS, USA

Yoon Soo Park Department of Medical Education, College of Medicine, Univer-
sity of Illinois at Chicago, Chicago, IL, USA
Xue-Lan Qiu The University of Hong Kong, Pokfulam, Hong Kong
Alexander Robitzsch Department of Educational Measurement, IPN Leibniz
Institute for Science and Mathematics Education, Kiel, Germany
Centre for International Student Assessment, Munich, Germany
Gina Rossi Research Group Personality and Psychopathology, Vrije Universiteit
Brussel, Brussels, Belgium
Benjamin Shear Research and Evaluation Methodology, University of Colorado,
Boulder, CO, USA
Hyo Jeong Shin Educational Testing Service, Princeton, NJ, USA
Klaas Sijtsma Department of Methodology and Statistics, TSB, Tilburg Univer-
sity, Tilburg, The Netherlands
Sandip Sinharay Educational Testing Service, Princeton, NJ, USA
William Stout Department of Statistics, University of Illinois at Urbana-
Champaign, Urbana, IL, USA
Jonathan L. Templin Educational Measurement and Statistics Program, Univer-
sity of Iowa, Iowa City, IA, USA
L. Andries van der Ark Research Institute of Child Development and Education,
University of Amsterdam, Amsterdam, The Netherlands
Alina A. von Davier ACTNext ACT Inc., Iowa City, IA, USA
Matthias von Davier National Board of Medical Examiners (NBME), Philadel-
phia, PA, USA
Shiyu Wang Department of Educational Psychology, University of Georgia,
Athens, GA, USA
Wen-Chung Wang The University of Hong Kong, Pokfulam, Hong Kong
Gongjun Xu Department of Statistics, University of Michigan, Ann Arbor, MI,
USA
Xueli Xu Educational Testing Service, Princeton, NJ, USA
xiv Contributors

Xiaofeng Yu Department of Psychology, University of Notre Dame, Notre Dame,

IN, USA
Jiangxi Normal University, Nanchang, China
Michael Yudelson ACTNext ACT Inc., Iowa City, IA, USA
Juan-Diego Zapata-Rivera Educational Testing Service, Princeton, NJ, USA
Peida Zhan Department of Psychology, College of Teacher Education, Zhejiang
Normal University, Zhejiang, China
Susu Zhang Department of Statistics, Columbia University, New York, NY, USA
Chapter 1
Introduction: From Latent Classes
to Cognitive Diagnostic Models

Matthias von Davier and Young-Sun Lee

Abstract This chapter provides historical and structural context for models and
approaches presented in this volume, by presenting an overview of important
predecessors of diagnostic classification models which we will refer to as DCM
in this volume, or alternatively cognitive diagnostic models (CDMs). The chapter
covers general notation and concepts central to latent class analysis, followed by an
introduction of mastery models, ranging from deterministic to probabilistic forms.
The ensuing sections cover knowledge state and rule space approaches, which can
be viewed as deterministic skill-profile models. The chapter closes with a section on
the multiple classification latent class model and the deterministic input noisy and
(DINA) model.

1.1 Introduction

This chapter provides historical and structural context for models and approaches
presented in this volume, by presenting an overview of important predecessors of
diagnostic classification models which we will refer to as DCM in this volume, or
alternatively cognitive diagnostic models (CDMs). We are attempting to organize
the growing field somewhat systematically to help clarify the development and
relationships between models. However, given the fact that DCMs have been
developed based on at least two, if not three traditions, not all readers may
necessarily agree with the order in which we put the early developments. While
there is a multitude of approaches that can be considered predecessors of current

M. von Davier ()

National Board of Medical Examiners (NBME), Philadelphia, PA, USA
e-mail: [email protected]
Y.-S. Lee
Teachers College, Columbia University, New York, NY, USA
e-mail: [email protected]

© Springer Nature Switzerland AG 2019 1

M. von Davier, Y.-S. Lee (eds.), Handbook of Diagnostic Classification Models,
Methodology of Educational Measurement and Assessment,
https://doi.org/10.1007/978-3-030-05584-4_1
2 M. von Davier and Y.-S. Lee

approaches to cognitive diagnostic modeling, there are many connections between

these seemingly different approaches, while several different lines of development
may be later understood as variants of one common more general approach (e.g.,
von Davier, 2013, 2014). In essence, any attempt to organize the many different
approaches that exist today may lead to simplifications, and potentially omissions
of related approaches.
The aim of all these approaches, however, can be summarized as the attempt to
provide powerful tools to help researchers learn about how observed behaviors, such
as responding to test items, can be used to derive information about generalizable
behavioral tendencies.
We begin the chapter with a section on general notation and concepts central to
latent class analysis, followed by an introduction of mastery models, ranging from
deterministic to probabilistic forms. The ensuing sections cover knowledge state and
rule space approaches, which can be viewed as deterministic skill-profile models.
The chapter closes with a section on the multiple classification latent class model
and the deterministic input noisy and (DINA) model.

1.2 Notation, Log-linear Models, and Latent Class Analysis

This section introduces notation used in subsequent chapters. We use the case of
binary observed variables as a standard example but note that all definitions can
be directly extended to polytomous nominal or ordinal response variables. Let
X = (X1 , . . . , XK ) denote K binary (or polytomous) response variables and let
xn = (xn1 , . . . , xnK ) denote the observed responses for test takers n = 1, . . . , N. Let
G denote a grouping variable with gn ∈ {1, . . . , M} for all test takers. In the case of
discrete mixture (or latent class) models, gn is unobserved, while for multiple group
models, gn is completely or partially observed (von Davier & Yamamoto, 2004; von
Davier & Carstensen, 2007).
The probability of observing x = (x1 , . . . , xK ) will be denoted by

P (X = x) = P (x1 , . . . , xK ) .

Obviously, these probabilities are unknown, while we may have some idea which
observed variables have higher or lower probability of exhibiting certain values. For
cognitive tasks, we may have some idea about the order of items with respect to the
likelihood of successful completion, but typically, there is no complete knowledge
about the joint distribution of response variables.
The aim of modeling response data is to provide a hypothesis of how this
unknown joint distribution can be constructed in a systematic way, either through
associations and interactions between observables, or by means of predictors, or
through assumed unobserved variables, or a combination of these.
1 Introduction: From Latent Classes to Cognitive Diagnostic Models 3

1.2.1 Log-linear Models

One customary way to model the joint distribution of the responses x1 , . . . , xK

is using log-linear models (e.g., Haberman, 1979; Hagenaars, 1993). Log-linear
models can be used with or without assuming latent variables. Log-linear models
describe transformed probabilities, using the natural logarithm. We can write

lnP (x1 , . . . , xK ) = f (x1 , . . . , xK ) ,

where f (x1 , . . . , xK ) is a function of the observed responses. One possible

assumption is that the log of the response probabilities can be expressed as
 

K   
K
f (x1 , . . . , xK ) = λ0 + λ1i xi + λ2ij xi xj + . . . + λKi...k xv .
i=1 {i=j} {i=...=k} v=1

Log-linear models in the context of CDMs have been discussed for example by
von Davier and Yamamoto (2004) and Xu and von Davier (2008) for dichotomous
and ordinal skill attribute variables. von Davier (2018) showed how certain log-
linear models used in the context of CDMs can be considered generalizations of
models recently discussed under the term network psychometrics (e.g., Marsman
et al., 2018; von Davier, 2018). In the example above, all products of any possible
subset of observed variables are included, however, it is customary to also consider
models that only include terms up to a certain degree D, assuming for higher degrees
E > D that λEi...,iE = 0.
One central issue when estimating log-linear models for large num-
 of observables is that 
bers a normalization factor is needed. Since, 1 =
(x1 ,...,xK ) P (x1 , . . . , xK ) = (x1 ,...,xK ) exp f (x1 , . . . , xK ), it follows that
⎡ ⎡  K ⎤⎤
 K   
λ0 =log ⎣ exp ⎣ λ1i xi + λ2ij xi xj + . . . + λKi...k xv ⎦⎦ .
(x1 ,...,xK ) i=1 {i=j} {i=...=k} v=1

This normalization factor involves a sum over all possible configurations

(x1 , . . . , xK ). For K binary variables, this is a sum involving 2K terms, for K = 30
items this is a summation over 1,073,741,824 terms. von Davier (2018) describes
how noise contrastive estimation (e.g., Guttmann & Hyvärinen, 2010, 2012) can
be used for estimation of log-linear and network psychometrics models, as well as
extended log-linear models for polytomous and dichotomous CDMs.
Log-linear models can be extended by assuming latent variables (Haberman,
1979; Hagenaars, 1993) so that the distribution of observed response variables
depends on an unobserved variable α,
K 
K  
f (x1 , . . . , xK |α) = λ0 (α) + λ1i (α) xi + . . . + λKi...k (α) xv
i=1 {i=...=k} v=1
4 M. von Davier and Y.-S. Lee

and by definition
⎡  ⎤

K  
K
P (x1 , . . . , xK |α) = exp ⎣λ0 (α) + λ1i (α) xi + . . . + λKi...k (α) xv ⎦ .
i=1 {i=...=k} v=1

If the latent variable is discrete, it can be written as α ∈ {g1 , . . . , gG }, with G sets

of each parameter type λdgi1 ...,id = λdi1 ...,id (g) for g = g1 , . . . , gG and d = 0, . . . ,
K. With this modification, the model becomes more complex. All parameters
may depend on some unobserved quantity α, some grouping variable g, or some
combination of both.

1.2.2 Latent Class Analysis

Latent Class Analysis (LCA) can be understood as an approach that assumes the
dependence of response probabilities on an unobserved discrete variable, which we
denote by c. In this sense, LCA is a direct application of the definition of conditional
response probabilities, as introduced above. LCA assumes a latent categorical
variable that cannot be directly observed. The LCA model equation follows from
a set of three assumptions, some of which match assumptions commonly used in
other latent variable models:
1. Class dependent response probabilities: For response variables xi , LCA assumes
class specific response probabilities. While there is no direct constraint that
imposes

P (xi |cv ) = P (xi |cw ) ,

it is a prerequisite for class separation so that respondents who are mem-

bers of different classes cv = cw can be reliably classified given their observed
responses.
2. Local independence: A central assumption is local independence given class
membership c,

K
P (x1 , . . . , xK |c) = P (xi |c) .
i=1

In LCA the class membership variable c is the latent variable that is expected
to ‘explain’ the dependencies between observed responses. Once conditional prob-
abilities are considered, the dependencies between observed variables vanish, under
this assumption.
1 Introduction: From Latent Classes to Cognitive Diagnostic Models 5

3. Classes are mutually exclusive and exhaustive: For each examinee v there is one,
and only one, ‘true’ latent class membership cv ∈ {1, . . . , G}. While the latent
variable in LCA is nominal, this assumption is analogous to the assumption of
a true (but unobserved expected) score in classical test theory (CTT) or a true
ability θ in item response theory (IRT).
These three assumptions make the LCA a discrete mixture distribution model,
since it follows from this set of assumptions that the marginal probability of a
response pattern is given by


G 
G K
P (x1 , . . . , xK ) = πc P (x1 , . . . , xK |c) = πc P (xi |c)
i=1
c=1 c=1

with mixing proportions (class sizes) π c = P(C = c). A logarithmic transform

following assumption 2 above yields,


K 
K
ln P (x1 , . . . , xK |c) = ln P (xi |c) = [xi ln P (Xi = 1|c)
i=1 i=1

+ (1 − xi ) ln P (Xi = 0|c)]

and further, using standard rules for the logarithm,


K 
K
P (Xi = 1|c)
ln P (x1 , . . . , xK |c) = ln P (Xi = 0|c) + xi ln .
P (Xi = 0|c)
i=1 i=1

As such, LCA can be understood as a log-linear model without interactions (as

local independence is assumed), conditional
 on a nominal latent variable. This can
P (Xi =1|c)
be seen by setting λ1ci = ln P (Xi =0|c) (a term that represents the log-odds for item

i conditional on class membership c) and λ0c = K i=1 ln P (Xi = 0|c) (an intercept
term) and observing that


K
ln P (x1 , . . . , xK |c) = λ0c + xi λ1ci .
i=1

Note that the log-odds λ1ci and the conditional response probabilities have the
following relationship:

exp (λ1ci )
= P (Xi = 1|c) .
1 + exp (λ1ci )

While the within-class model of LCA is rather restrictive, as independence of

all responses is assumed, the LCA is a very flexible model, since the number of
6 M. von Davier and Y.-S. Lee

classes C is not specified a priori. Any dependence between observed variables can
be modeled by increasing the number of classes, however, identifiability may be an
issue (e.g., Goodman, 1974; Allman, Matias, & Rhodes, 2009; Xu, this volume).
Therefore, this flexibility is also a weakness of the LCA. With the addition of
classes to the model the fit between model predictions and observed data will
always improve, which may result in a LCA solution that overfits the observed
dependencies. In addition, the increase in number of classes leads to a substantial
increase in the number of parameters to be estimated. For additional details on
applications of LCA, see the volumes by Langeheine and Rost (1988), Rost and
Langeheine (1997), and Hagenaars and McCutcheon (2002), as well as the chapter
by Dayton and Macready (2006).
Confirmatory approaches to LCA constrain the number of classes and often also
impose inequality or equality constraints on class specific response probabilities
(e.g., Croon, 1990). Most DCMs covered in this volume can be written as
constrained variants of LCA (von Davier, 2009). Some constrained versions of
LCA share many interesting similarities with (M-)IRT models (e.g., Haberman, von
Davier, & Lee, 2008) and can be used to replace these models.

1.3 Mastery Models

Mastery models assume a skill domain for which we can sort any person into one of
two classes: expert versus novice, master versus non-master, or professional versus
amateur. This may not be adequate for most domains, even if there is a distinct ‘can
do’ versus ‘cannot do’; there are often gradual differences in the ‘can do’. In this
section, however, we use this notion of mastery and assume all respondents can be
classified into two groups without further distinction.
While these types of distinctions may be oversimplifications, can they still be
useful categories to describe how test takers respond to a test? If we consider
ideas from developmental psychology (e.g., Piaget, 1950; Wilson, 1989), we find
that some things in life are thought of as being acquired or learned in terms of
qualitative jumps. We may want to entertain the idea of mastery learning for a
while and examine where this leads us in terms of how a latent variable model may
represent this concept. For example, young children cannot perform or solve task
X until they mature and ‘get it’, after which the same task becomes quite easy for
them.
The mastery-state can be represented by a random variable that takes on two
values: ‘1’ = mastery and ‘0’ = non-mastery. Formally, we define a latent variable
A, with av ∈ {0, 1} for all respondents v = 1, . . . , n, and with

av = 1 if person v masters the skill of interest

1 Introduction: From Latent Classes to Cognitive Diagnostic Models 7

and

av = 0 if person v does not master the skill.

The two mastery levels are expected to differ with respect to the probabilities of
success, just as in assumption 1 presented in the section on LCA above. However,
in mastery models, there is an order expectation, or even an order restriction in place:
it is expected (and potentially specified directly in the model) that for all response
variables the probability of success is larger for masters than for non-masters. More
formally,

P (Xi = 1|a = 1) = 1 − si ≥ gi = P (Xi = 1|a = 0)

may be assumed for all response variables X1 , . . . , XK . For each item, there are
four probabilities to consider, the conditional probabilities of success and failure
under mastery and non-mastery. These are often denoted as follows (e.g., Dayton &
Macready, 1977):
• Guessing correctly by non-masters: gi = P(Xi = 1| a = 0)
• Incorrect response by non-masters: 1 − gi = P(Xi = 0| a = 0)
• Slipping = unexpected incorrect response by masters: si = P(Xi = 0
| a = 1)
• Correct response by masters: 1 − si = P(Xi = 1| a = 1)
A variety of constraints on these parameters have been suggested in the literature,
some examples are discussed by Macready and Dayton (1977). Nowadays, the
term ‘slipping’ is often used instead of ‘unexpected error’ while ‘guessing’ is
still in use (Junker & Sijtsma, 2001). Just like LCA, mastery models also assume
local independence and that masters and non-masters are mutually exclusive and
exhaustive. Based on the equivalency shown in the previous section, a mastery
model with two levels can be written either in the form of a 2-class LCA or as a
log-linear model with latent variables:


K
P (Xi = 1|a) xi
P (x1 , . . . , xK |a) = P (Xi = 0|a)
P (Xi = 0|a)
i=1

and with the definitions above, we have P (Xi =1|a) =(1−si )a gi[1−a] , and for the
complement we have P (Xi = 0|a) = sia (1 − gi )[1−a] . A logarithmic transforma-
tion and insertion of the definitions yields the following:


K 
K
(1−si ) a
gi [1−a]
ln P (x1 , . . . , xK |a) = ln sia (1−gi )[1−a] + xi ln .
si (1−gi )
i=1 i=1
8 M. von Davier and Y.-S. Lee

 a [1−a]
gi
As before, by setting K i=1 ln si (1 − gi )
a [1−a]
= λ0a and ln (1−s
si
i)
(1−gi ) =
λ1ai , the equivalency of the mastery model to a log-linear model with a binary latent
variable is obtained. Note that λ1ai can be written as

P (Xi = 1|a = 1) P (Xi = 1|a = 0)

a ln + [1−a] ln =λ10i +a [λ11i −λ10i ] ,
P (Xi = 0|a = 1) P (Xi = 0|a = 0)

which again contains the log-odds for masters and non-masters, multiplied by the
mastery status.

1.4 Located Latent Class or Multi State Mastery Models

The additional model specifications needed to move from LCA, which is character-
ized by a nominal latent class variable, to located classes are easily introduced.
The last section that examined mastery models provides the basis for these
developments. For a correct response xi = 1, the term λ1ai = λ10i + a[λ11i − λ10i ]
is part of the sum. This term is linear in the mastery level a ∈ {0, 1} and if λ11i > λ10i
or equivalently, P(Xi = 1| a = 1) > P(Xi = 1| a = 0), the term λ1ai is monotone
increasing over the (in the case of mastery models: two) ordered mastery levels.
With more than two levels of mastery, for example an ordinal variable that
represents non-mastery as zero, but allows multiple levels of mastery represented

as successive integer, i.e., a ∈ {0, 1, 2, . . . M}, a model can be defined as

  
K
ln P x1 , . . . , xK |a  = λ0a + xi λ1ai
i=1

with

λ1ai < λ1ai for all a  < a  .

This ensures that

   
P Xi = 1|a  < P Xi = 1|a  for all a  < a  .

This produces a monotone increasing sequence of response probabilities over


a ∈ {0, 1, 2, . . . M}. Note, however, that this type of constraints (still) produces a
comparably large number of quantities that need to be estimated. However, this
model includes equality constraints (e.g., Formann, 1985, 1992) which may be
imposed via additional assumptions about how model parameters relate to the
ordered levels of mastery. Essentially, each latent class in this model becomes an
ordered mastery level, but the distances between classes differ by item i and class
1 Introduction: From Latent Classes to Cognitive Diagnostic Models 9


level a ∈ {0, 1, 2, . . . M}. This model requires (M + 1)K parameters one set of K
item parameters for each class. As before, probabilities can be derived using the
equivalency

  exp (λ1ai )
P Xi = 1|a  = .
1 + exp (λ1ai )

A more parsimonious model can be implemented by imposing the following

constraint

λ1ai = βi + γi θa

which requires 2K item location β i and slope parameters γ i and M + 1 ordered

 
class specific locations θ a < θ a for a < a ∈ {0, . . . , M}. With the transformation

βi + γi θa = a (θ − b)

it can be easily observed that located latent class models define the class specific
response probabilities as

exp (a (θ − b))
P (Xi = 1|θa = θ ) =
1 + exp (a (θ − b))

which is very similar to IRT (Lord & Novick, 1968), while assuming a discrete
latent variable with located latent classes (e.g., Formann, 1992; Haberman et al.,
2008).

1.5 Rule Space Methodology and Knowledge Spaces

Rule space (RS; e.g., Tatsuoka, 1983, 1990, 2009) and knowledge spaces (KS;
Doignon & Falmagne, 1985, 1998; Albert & Lukas 1999) are independently
developed approaches to the question of how the association between performance
on heterogeneous tasks and multiple skills can be conceptualized. Much like
mastery models, RS and KS assume that a respondent who masters a certain number
of skills is on a regular basis capable of solving tasks that require these skills. In
contrast to the first generation of mastery models, both RS and KS assume that
there are multiple skills to be considered, and that each respondent is characterized
by a skill pattern or attribute pattern – or a knowledge state – and that every task
requires a subset of the skills represented in the skill space of respondents.
Consider an example with two skills, addition and multiplication, ignoring for
a moment that there is an additional skill required that tells us in what order these
operations have to be executed. If asking examinees to solve tasks of the type
10 M. von Davier and Y.-S. Lee

(a) 3 + 4 =?
(b) 4 * 5 =?
(c) 3 * 3 + 2 =?
one could argue that there are four potential groups of test takers. Group 1 does
neither master addition nor multiplication and cannot solve any of the task types;
Group 2 masters only addition and can solve tasks of type (a) only; Group 3 only
masters multiplication (no matter how unlikely that may seem to a math educator)
and hence can solve only tasks of type (b); and Group 4 masters both addition and
multiplication, and hence can solve tasks of type (a), (b), and (c) on a regular basis.
More formally, for tasks that require a subset of D skills, we can assign to each
task i = 1, . . . , K a vector of skill requirements qi = (qi1 , . . . , qiD ) ∈ {0, 1}D that
indicates which skill (or attribute) is required for that task. The matrix
⎛ ⎞
q11 . . . q1D
Q = ⎝ ... ... ... ⎠
qK1 . . . qKD

is referred to as Q-matrix and represents a hypothesized relationship of how a skill

vector (skill state) or attribute pattern a = (a1 , . . . , aD ) is connected to expected
performance on each task. The ideal (the most likely, or expected given a skill
pattern) response on item i given can be written as

  D
q
xi[I ] q i , a = ad id ∈ {0, 1}
d=1

which equals one if the attribute mastery pattern a matches or exceeds non-zero
entries of the skill requirements qi , i.e., if at least all required skills are mastered,
and is zero otherwise. The above equation can be applied to all items to construct
an ideal response pattern
D 
 q 
D
qKd
x (a) =
[I ] 1d
ad , . . . , ad
d=1 d=1

for each attribute mastery pattern a. The observed response pattern xv produced
by respondent v can then be compared to each of these ideal response vectors,
and the closest match determined. This can be done in a variety of ways; for
example, Tatsuoka (1983, 1985) discussed methods based on distance measures,
but also presents classification based on IRT ability estimates and person fit. von
Davier, DiBello, and Yamamoto (2008) provide a summary of the IRT and fit based
approach. A simple measure of agreement can be defined as
K  
i=1 xvi ∗ xi[I ] q i , a
sim (x v , a) = 
    2

K 2
i=1 xvi
K
i=1 xi[I ] q i , a
1 Introduction: From Latent Classes to Cognitive Diagnostic Models 11

which equals the cosine similarity of the observed and ideal vectors. The cosine
similarity is a correlation related measure commonly used in data mining, machine
learning and natural language processing (Tan, Steinbach, & Kumar, 2005). Respon-
dents can be assigned to the attribute pattern that produces the largest similarity
measure relative to the observed vector xv .
Tatsuoka’s RS has demonstrated its utility in many applications over the years.
Recently, the method gained new interest under the name ‘attribute hierarchy
method’ (AHM; Leighton, Gierl, & Hunka, 2004). The authors describe the AHM as
being an instantiation of rule space that differs from Tatsuoka’s (1983, 1985, 1990,
2009) methodology in that it allows attribute hierarchies. Attribute hierarchies limit
the permissible attribute space, as some attributes have to be mastered before other
can be mastered, by definition of what a hierarchy encompasses. von Davier and
Haberman (2014) show how the assumption of hierarchical attributes restricts the
number and type of parameters of diagnostic classification and multiple mastery
models.
Both RS and KS were initially conceptualized as deterministic classification
approaches. Respondents would be classified according to their similarity to ideal
response patterns, regardless of the observation that only very few respondents will
produce exactly the ‘ideal’ patterns that can be expected based on the Q-matrix.
Attempts to produce a less deterministic version of these approaches have been
made, and Schrepp (2005) describes similarities between KS approaches and latent
class analysis. The next section describes models that share many of the features
of RS and KS approaches, but provide a structured latent attribute space, and a
probabilistic approach to define how multiple mastery levels relate to response
probabilities in a systematic way, rather than by means of unstructured class profiles
as used in LCA.

1.5.1 Multiple Classification Models and Deterministic Input

Noisy and (DINA) Models

Latent class models with multiple latent variables (Haberman, 1979; Haertel, 1989)
or multiple classification latent class models (MCLCM; Maris, 1999) extend latent
class analysis (LCA) in such a way that multiple nominal or ordinal latent variables
can be identified simultaneously. This approach retains the defining properties
of LCA, local independence given latent class, assumption of an exhaustive and
disjunctive latent classification variable, and distinctness of conditional probabilities
across classes.
The MCLCM approach can be viewed as a non-parametric precursor to many of
the diagnostic models introduced in subsequent chapters. For a MCLCM with two
latent variables c1 ∈ {0, . . . , C1 }, c2 ∈ {0, . . . , C2 } denote the joint distribution of
these with πc1 ,c2 and define


C1 
C2 
K
P (x1 , . . . , xK ) = πc1 ,c2 P (xi |c1 , c2 ) .
c1 =0 c2 =0 i=1
12 M. von Davier and Y.-S. Lee

This is a well-defined LCA that can be rewritten as a single latent1 variable

LCA with ‘attribute’ a = {c1 , c2 } and MNCL = (C1 + 1) (C2 + 1) latent classes
representing all possible combinations. However, one may introduce additional
structure – constraints on the response probabilities – for the two-variable case to
specify whether the conditional probabilities may for some items depend on only
one or the other component c1 or c2 . More specifically, one may assume

P (xi |c1 , c2 ) = P (xi |fi1 (c1 ) , fi2 (c2 )) .

As a special case with specific relevance to diagnostic models, we will consider the
following form of these constraints in the example
q
fid (cd ) = cdid

for d = 1, 2 and with qi1 , qi2 ∈ {0, 1}.

Basically, if one or the other qi∗ is zero, the dependency on that component of
the multiple classification LCM variable vanishes from the conditional probability
of item response xi . This is true because

cd0 = 1

for all levels of cd whenever qi1 = 0. With this constraint, the conditional
probabilities of a response variable may depend on both c1 , c2 in MNCL levels
for some items, on c1 only in (C1 + 1) levels for some other items, or on c2 with
(C2 + 1) levels for a third set of items, or on neither one of them in a fourth group
of response variables.
Two additional restrictions lead to the model that is commonly known as the
DINA (Deterministic Input, Noisy And) model (Macready & Dayton, 1977; Junker
& Sijtsma, 2001). First, all components of the latent skill pattern a are assumed to
be binary (and as before, we use ad for binary attributes, while for nominal classes,
we use c1 , c2 , . . . ), that is

a = (a1 , . . . , aD ) ∈ {0, 1}D

and for the conditional probabilities we assume

 

D
q
P (xi |a1 , . . . , aD ) = P xi | ad id .
d=1

1 Classvariables are represented as integers, but the use of integers do not imply any ordering here;
only equivalence classes are used in the context of LCA.
1 Introduction: From Latent Classes to Cognitive Diagnostic Models 13

Note that the conditional probability depends on a binary variable ξaqi =

D qid
d=1 ad ∈ {0, 1} which is a function of the skill pattern a and one row of
the Q-matrix, a vector that specifies the skill requirements for a specific item. Just
as in the section on mastery models, applying this definition leads to the following
expressions:
 
P Xi = 1|ξaqi = 1 = 1 − si

and
 
P Xi = 1|ξaqi = 0 = gi .

The DINA model is said to be conjunctive because

 it reduces the respondent-skill

qid qid
by item-attribute comparison to only two levels D a
d=1 d = 1 or D d=1 ad = 0.
With this, we can write
xi 1−xi
P (xi |a1 , . . . , aD ) = (1 − si )ξaqi gi 1−ξaqi si ξaqi (1 − gi )1−ξaqi .

Only those respondents who possess all necessary skills have a “high” probability
1 − si of solving an item, while respondents who lack at least one of the required
skills have a “low” probability gi —the same “low” probability no matter whether
only one or all required skills are not mastered.
Note that the gi and the si denote the item parameters in the DINA model,
so that there are two parameters per item in this model. In addition, the skill
vectors av = (av1 , . . . avK ) are unobserved, so we typically have to assume that
the distribution of skills P (A = (a1 , . . . aK )) = π(a1 ,...aK ) is unknown. Therefore,
there are {0, 1}K  − 1 = 2K − 1 independent skill pattern probabilities with

(a1 ,...aK ) π(a1 ,...aK ) = 1.0 if an unconstrained estimate of the skill distribution is
attempted. There may be fewer parameters if a constrained distribution over the skill
space (von Davier & Yamamoto, 2004; Xu & von Davier, 2008) is used. For model
identification, no constraints are needed on the guessing and slipping parameters
(even though it is desirable that 1 − si > gi for somewhat sensible results).
While de la Torre (2009) does not make statements about identifiability of the
DINA model and the uniqueness of the model parameters, Junker and Sijtsma
(2001) discuss (a lack of) empirical identification in the context of their data
example used in conjunction with Markov chain Monte Carlo (MCMC) estimation.
Haertel (1989) describes identification of latent class skill patterns in the DINA
model, and notes that “it may be impossible to distinguish all these classes
empirically using a given set of items. Depending upon the items’ skill requirements,
latent response patterns for two or more classes may be identical (p.303).” One
of the remedies Haertel (1989) suggests is the combination of two or more latent
classes that cannot be distinguished. In subsequent chapters, identifiability of
diagnostic models is discussed in more detail (Xu, this volume; Liu & Kang, this
14 M. von Davier and Y.-S. Lee

volume; DeCarlo, this volume) and von Davier (2014) provides an example of how
the (lack of) empirical identifiability of diagnostic models can be checked.
The DINA model is a very restrictive model as it assumes only two parameters
per item, and skill attributes only enter the item functions through conjunction
qid
function ξaqi = D a
d=1 d . This restricts the probability space so that different
attribute mastery patterns, in particular those that are not a perfect match of the
Q-matrix for an item, are all mapped onto the same low “guessing” probability.
There are several issues with the assumption made in the DINA model. Formally,
this assumption is equivalent to assuming a log-linear model (see Eq. 1) in which
all parameters are set to zero except the one that parameterizes the highest order
interaction term. Additionally, from the point of view of most applications of
skills, compensation happens: Multiplication can be replaced by repeated addition,
a lack of vocabulary when acquiring a new language, or even learning disabilities
can be compensated for (and eventually remedied) by higher general intelligence
(e.g., Reis, McGuire, & Neu, 2000), etc. In total darkness, hearing can be used
to, admittedly poorly, compensate for lack of vision. For diagnostic models and
compensatory and non-compensatory MIRT models, it was found that real data
examples are often fit better (in terms of item fit, or overall goodness of fit assessed
with information criteria or similar) with additive/compensatory models rather than
conjunctive models (de la Torre & Minchen, this volume; von Davier, 2013).
In addition, it was found that the DINA model may be affected by model
identification issues. DeCarlo (2011) and Fang, Liu, and Ying (2017) show that the
DINA model is not identified unless there are what some may call ‘pure’ items in the
Q-matrix, that is, items that only measure a single attribute. DeCarlo (2011) shows
that the DINA model with the Q-matrix provided for the Fraction Subtraction data
(Tatsuoka, 1985) is not able to identify all attribute patterns. Fang, Liu, and Ying
(2017) provide more general results on the requirements for the Q-matrix. Xu (this
volume) and Liu and Kang (this volume) provide further results and more recent
examples.

1.6 Summary

The notation and models introduced in this chapter form the basis for many of the
subsequent chapters. Most, if not all DCMs can be written as constrained latent class
models or alternatively, log-linear models with discrete latent variables.
This introduction does not provide an in-depth coverage of how to evaluate the
different approaches. However, all models presented in this volume are approaches
that provide marginal probability distributions for multivariate discrete observables.
This means that methods from categorical data analysis can be used to compare
models and to evaluate model data fit.
While some of the models introduced above may be considered approaches for
diagnostic classification and may have been used as such, many more sophisticated
1 Introduction: From Latent Classes to Cognitive Diagnostic Models 15

approaches have been developed since, based on these initial modeling attempts.
The aim of the current volume is providing a systematic overview of these more
recent approaches.
The Handbook Diagnostic Classification Models aims at capturing the current
state of research and applications in this domain. While a complete overview of this
broad area of research would require a multi-volume effort, we tried to capture a
collection of major research streams that have been developed over several years
and that continue to produce new results.
The first part of the volume covers major developments of diagnostic models
in the form of chapters that introduce the models formally, provide information
on parameter estimation and on how to test model-data fit, and applications or
extensions of the approach.
The second part of the volume describes special topics and applications. Special
topics such as Q-matrix issues are covered, including the data driven improvement
and construction, as well as issues around model identifiability. The third part
presents applications of diagnostic models, as these are a centerpiece to reasons why
not only methodologists but also applied researchers may want to study the volume.
These applications show how diagnostic models can be used to derive more fine-
grained information about respondents than what traditional methods such as CTT
or IRT can provide.
The fourth part of the book includes a range of available software packages,
including the use of general purpose statistical software, specialized add-on pack-
ages, and available stand-alone software for estimation and testing of CDMs.
In many cases, latent class analysis, customary IRT, and other latent variable
models can directly be considered alternatives to diagnostic models, as these are
often more parsimonious (in the case of IRT) or do not make as strong (parametric)
assumptions about the latent structures and how these structures are related to the
conditional response probabilities in the levels of the latent variables. Standard
procedure should therefore be used as a comparison of more complex modeling
approaches with customary standard examples of latent variable models such as
IRT or LCA. Such a practice will ensure that researchers can compare their findings
to those obtained from less complex models to check whether the increased model
complexity provides added value, through improved model-data fit, and by means
of more useful derived quantities such as estimated mastery states.

References

Albert, D., & Lukas, J. (1999). Knowledge spaces: Theories, empirical research and applications.
Mahwah, NJ: Erlbaum.
Allman, E. S., Matias, C., & Rhodes, J. A. (2009). Identifiability of parameters in latent
structure models with many observed variables. The Annals of Statistics, 37(6A), 3099–3132.
https://doi.org/10.1214/09-AOS689
Croon, M. A. (1990). Latent class analysis with ordered classes. British Journal of Mathematical
and Statistical Psychology, 43, 171–192.
Other documents randomly have
different content
of construction or creation as the basis of reasoning upon the works
of Nature, we resort to a being in whom that power is the highest of
which we have direct evidence. In the works of man we have direct
and palpable proof that the phenomena—the products of human skill
and human force—are brought about by the faculties of an
intelligent and reasoning being. If we dig into the earth and find
there a statue, an implement, or a weapon, we do not hesitate to
conclude that the spot was once inhabited by men, just as surely as
we should conclude the same thing if we found there human bones.
The world, above-ground and below-ground, is full of concrete
objects that we know must have been fashioned by human skill,
guided by human intelligence. This intelligence, this intellect, is not
matter; it is a being; it is a person. It is not a force, acting without
consciousness; it is a being wielding a force which is under the
control of volition. The force and the volition are both limited, but
within the limitations they constitute the power of man. Pass, then,
to the works of Nature, or to what you call the power of Nature. As,
in the case of man, you can not conclude that he created for himself
his own faculties, that he prescribed for himself the limitations of his
power over matter, or that he formed those limitations as mere
matters of habit, or that it was from habit alone that he derived his
great constructive powers, so, in studying the works of Nature, you
must conclude that some intelligent being made the laws of matter
and motion, prescribed the unvarying order and method of action,
laid down the limitations, originated the properties, and, in so doing,
acted by volition, choice, and design. The distinction, as I conceive,
between man and Nature is, that there has been bestowed on man,
in a very inferior degree, a part of the original power of creation. On
Nature there has been bestowed none of this power. As we find that
the existence of man as an intelligent being, endowed with certain
high faculties, among which is a certain degree of the power of
creating new objects, can not be accounted for without the
hypothesis of a creator, still less can we account for the existence
and phenomena of Nature, which has in itself no degree of the
creating power, without the same hypothesis.
Kosmicos. Stop where you are. Why do you separate man from
Nature? Have you yet to learn that man is a part of Nature? I
suspect you have, after all, been reading the book of Genesis for
something more than a hypothesis, and that you have adopted the
notion that God made Adam a living soul. Put away all the nursery-
stories, and come down to the "hard-pan" of actual facts, which
show by an overwhelming array of evidence that man had a very
different origin.
Sophereus. You know, my friend, that I never learned any nursery-
stories, and therefore I have none to unlearn. It may be my
misfortune, but I find myself here in the world in mature years,
studying the phenomena of life, without having had any early
teaching, but with such reasoning as I can apply to what I observe,
and to what science, history, and philosophy can furnish to me. I
belong to no church, to no sect, to no party, and I have not even a
country. I am a citizen of the world, on my travels through it,
learning what I can. Now, what are your facts? Let us get down, as
you say, upon the "hard-pan," and make it as hard as you please.
Kosmicos. First answer my question: Why do you separate man from
Nature?
Sophereus. I know very well that in a certain sense man is a part of
Nature. But it is necessary to contemplate man apart from all the
rest of Nature, because we find that he is endowed with intellect,
and we have very good and direct evidence that his intellect is an
actor; and we know that he is endowed with consciousness, and we
have very good and direct evidence that, by introspection, he
becomes aware of his own consciousness, and what it is.
Kosmicos. Very well, assume all that if you choose. Now let me show
you an origin of man, with his intellect and consciousness, which will
entirely overthrow the idea that he was a special creation in the
sense to which you seem to be drifting, namely, that of miraculous
interposition by a being called God. You must be aware, as you have
read so much, that modern science has made great discoveries, and
that there are certain conclusions on this subject which are drawn
from very numerous and important data. Those data involve the
origin of all the different animals, man included. They are all to be
accounted for in the same way and by the same reasoning. Now, if
we go back to a period when none of them existed, we find a
method of accounting for them that is infinitely superior as a
hypothesis to any idea of their special creation as an act or as a
series of acts of divine and direct interposition. I will take this
method as it is given by Herbert Spencer, because, as he has
reasoned it, it accounts for both intellect and consciousness; and Mr.
Spencer is allowed to be one of the leading minds of this age. Mark
the starting-point of his whole philosophy on this subject of organic
life. Darwin, as you know, supposes some one very low form of
organic life, an aquatic grub, and out of it he evolves all the other
animal organisms, by the process of natural and sexual selection,
through successive generations, ending in man. This hypothesis
leaves the original organism to be accounted for, and, although
Darwin does not expressly assert that it was the Creator who
fashioned the first organism, he leaves it to be implied. Spencer, on
the other hand, explicitly denies the absolute commencement of
organic life on the globe. Observe that the terms of his theory of
evolution are much more complete than Darwin's, for he says that
"the affirmation of universal evolution is in itself a negation of an
absolute commencement of anything. Construed in terms of
evolution, every kind of being is conceived as a product of
modifications wrought by insensible gradations on a pre-existing
being; and this holds as fully of the supposed commencement of
organic life, or a first organism, as of all subsequent developments
of organic life."[106]
You will see, therefore, that the idea of a Creator, fashioning a type
of animal organism, or making a commencement of organic life, is
excluded by this great philosopher, although he does concur in the
main in Darwin's general explanation of the mode in which one
organism is evolved out of a pre-existing organism. He goes much
farther, because his system of universal evolution embraces the
elements out of which any organic life whatever has been
developed, and negatives the idea of any absolute commencement
of anything whatever. He begins with the original molecules of
organizable matter. By modifications induced upon modifications
these become formed, by their inherent tendencies, into higher
types of organic molecules, as we see in the artificial evolution
effected by chemists in their laboratories; who, although they are
unable to form the complex combinations directly from their
elements, can form them indirectly through successive modifications
of simpler combinations, by the use of equivalents. In Nature, the
more complex combinations are formed by modifications directly
from the elements, and each modification is a change of the
molecule into equilibrium with its environment, subjecting it, that is
to say, to new conditions. Then, larger aggregates, compound
molecules, are successively generated; more complex or
heterogeneous aggregates arise out of one another, and there
results a geometrically increasing multitude of these larger and more
complex aggregates. So that by the action of the successive higher
forms on one another, joined with the action of the environing
conditions, the highest forms of organic molecules are reached. Thus
in the early world, as in the modern laboratory, inferior types of
organic substances, by their mutual actions under fit conditions,
evolved the superior types of organic substances, and at length
ended in organizable protoplasm. Now, let me read to you Mr.
Spencer's description of the mode in which the substance called
"protein" becomes developed into organic life. "And it can hardly be
doubted," he says, "that the shaping of organizable protoplasm,
which is a substance modifiable in multitudinous ways with extreme
facility, went on after the same manner. As I learn from one of our
first chemists, Prof. Frankland, protein is capable of existing under
probably at least a thousand isomeric forms; and, as we shall
presently see, it is capable of forming, with itself and other
elements, substances yet more intricate in composition, that are
practically intricate in their varieties of kind. Exposed to those
innumerable modifications of conditions which the earth's surface
afforded, here in amount of light, there in amount of heat, and
elsewhere in the mineral quality of its aqueous medium, this
extremely changeable substance must have undergone, now one,
now another, of its countless metamorphoses. And to the mutual
influences of its metamorphic forms, under favoring conditions, we
may ascribe the production of the still more composite, still more
sensitive, still more variously-changeable portions of organic matter,
which, in masses more minute and simpler than existing protozoa,
displayed actions varying little by little into those called vital actions,
which protein itself exhibits in a certain degree, and which the
lowest known living things exhibit only in a greater degree. Thus,
setting out with inductions from the experiences of organic chemists
at the one extreme, and with inductions from the observations of
biologists at the other extreme, we are enabled to deductively bridge
the interval—are enabled to conceive how organic compounds were
evolved, and how, by a continuance of the process, the nascent life
displayed in these becomes gradually more pronounced."[107]
It is in this way that Spencer accounts for the formation of the cell
which becomes developed into a living organism, out of which are
successively evolved all the higher forms of animal organisms, until
we reach man.
Sophereus. And is this put forward as something which rational people
are to believe?
Kosmicos. Undoubtedly it is put forward as something that is to be
believed, because it is supported by a vast array of evidence; and let
me tell you that this conception of Nature as a whole is the
consummate flower of this nineteenth century in the domain of
philosophic speculation.
Sophereus. Perhaps it is. But although this nineteenth century has
witnessed many great scientific discoveries, and has produced
extraordinary inventions, I do not find that among the speculative
philosophers of this age there are such very superior powers of
reasoning displayed that we ought to regard them as authorities
entitled to challenge our acceptance of their theories without
examination. I must say that among your scientific people of the
present day, and especially among the philosophers of the class of
which Mr. Spencer is the leading representative, there are certain
tendencies and defects which surprise me. One of their defects is
that they do not obviate remote difficulties, perhaps because they
have not been trained, as other men have, to foresee where such
difficulties must arise. This is sometimes apparent even when the
difficulties are not very remote, but are quite obvious. One of their
tendencies is to arrive at a theory from some of the phenomena, and
then to strain the remaining phenomena to suit the theory; and
sometimes they proceed to the invention or imagination of
phenomena which are necessary to the completion of a chain of
proof. This last process is called bridging the interval. I will now
apply this criticism to Mr. Spencer's philosophy of the origin of man.
In the first place he has not obviated a fundamental difficulty,
whether it be a near or a remote one. Where did the molecules get
their tendency or capacity to arrange themselves into higher and
more complex forms? Whence came the auxiliary or additional force
of their surrounding environment? What endowed protein with its
capacity to assume a thousand isomeric forms? What made the
favoring conditions which have helped on the influence of its
metamorphic tendencies, so as to produce still more sensitive and
variously-changeable portions of organic matter? These questions
must have an answer; and, when we ask them, we see the
significance of the inquiry, "Where wast thou (man) when I laid the
foundations of the world?" For these things, on the evolution theory,
are the foundations of the world. It is no answer to say, as Mr.
Spencer does, that these tendencies, or capacities of matter, and
these laws of the favoring conditions, came from the Unknown
Cause. Known or unknown, did they have a cause, or did they make
themselves? Did these, the foundations of the world, have an origin,
or were they without any origin? If they had an origin, was it from
the will and power of a being capable of giving existence to them
and prescribing their modes of action? If they had no origin, if they
existed from all eternity, how came it that they formed this
extraordinary habit of invariable action in a certain method, which
amid all its multiformity shows an astonishing persistency? If we
deny, with Mr. Spencer, the absolute commencement of organic life
on the globe, we must still go back of all the traces of organic life,
and inquire whence matter, molecules, organized or unorganized,
derived the capacities or tendencies to become organized, and how
the favoring conditions became established as auxiliary or subsidiary
forces. And therefore it is that this difficulty, whether remote or near
at hand, is not met by Mr. Spencer: for whether we call the cause an
unknown or a known cause, the question is, Was there a cause, or
did the foundations of the world lay themselves? The reasoning
powers of mankind, exercised by daily observation of cause and
effect, of creative power and created product, are equal to the
conception of a First Cause as a being who could have laid the
foundations of the world, but they are utterly unequal to the
conception that they had no origin whatever. Again, consider how
numerous are the missing links in the chain of evolution, how many
gaps are filled up by pure inventions or assumptions. The evolution
of one distinct and perfect animal, or being, out of a pre-existing
animal or being of a different type, has never been proved as a fact.
Yet whole pedigrees of such generation of species have been
constructed upon the same principles as we should construct the
pedigree of an individual. Furthermore, if we regard the facts about
which there can be no controversy, we find not only distinct species
of animals, but we find the same species divided into male and
female, with a system of procreation and gestation established for
the multiplication of individuals of that species. Now go back to the
imaginary period when protein began to form itself into something
verging toward organic life, and then there became evolved the
nascent life of an organized being. How did the division of the sexes
originate? Did some of the molecules or their progressive forms, or
their aggregates, or masses, under some conditions, tend to the
production of the male, and others under certain conditions tend to
the development of the female, so that the sexes were formed by a
mere habit of arrangement without any special intervention? Here is
one of the most serious difficulties which the doctrine of evolution,
whether it be the Darwinian or the Spencerian theory, has to
encounter. There is a division into male and female: there is a law of
procreation by the union of the two sexes. This is a fact about which
there can be no dispute. It is one of the most remarkable facts in
Nature. It is the means by which species are continued, and the
world is peopled with individuals of each species. Is it conceivable
that this occurred without any design, that it had no origin in a
formative will, that it had, properly speaking, no origin at all, but
that it grew out of the tendencies of organized matter to take on
such a diversity in varying conditions? And if the latter was all the
origin that it had, whence came the tendencies and whence the
favoring conditions that helped them on toward the result? It seems
to me that the Spencerian theory, so far as it suggests a mode in
which the two sexes of animals came to exist, is hardly less fanciful
than what Plato has given us in his "Timæus." I have studied them
both.
If you will hand me Mr. Spencer's work from which you have just
quoted, I will point out a passage which fully justifies my criticism. It
is this: "Before it can be ascertained how organized beings have
been gradually evolved, there must be reached the conviction that
they have been gradually evolved." He says this in praise of De
Maillet, one of the earliest of the modern speculators who reached
this conviction, and whose "wild notions" as to the way should not
make us, says Mr. Spencer, "forget the merit of his intuition that
animals and plants were produced by natural causes."[108] That is to
say, first form to yourself a theory, and have a thorough conviction
of it. Then investigate, and shape the facts so as to support the
theory. Is it not plain that an inquiry into the mode in which
organized beings have been gradually evolved must precede any
conclusion or conviction on the subject? It is one of those cases in
which the how a thing has been done lies at the basis of the inquiry
whether it has probably been done at all. If a suggested mode turns
out to be wild and visionary, what is the value of any "intuition" of
the main fact? But, what is still more extraordinary in this kind of
deduction, which is no deduction, is the way in which, according to
Mr. Spencer, the first conviction is to be reached before one looks for
the facts. The process of the evolution of organisms, according to
Mr. Spencer's philosophy, is contained as a part in the great whole of
evolution in general. We first convince ourselves that evolution
obtains in all the other departments of Nature, and is the
interpretation of all their phenomena. Then we conclude that it has
obtained in the animal kingdom, and so we have the conviction
necessary to be acquired before we examine the phenomena; and
then we make that investigation so as to reconcile the facts with the
supposed universal laws of matter and motion. I do not exaggerate
in the least. Here is what he says: "Only when the process of
evolution of organisms is affiliated on the process of evolution in
general can it be truly said to be explained. The thing required is to
show that its various results are corollaries from first principles. We
have to reconcile the facts with the universal laws of the
redistribution of matter and motion."[109] What would Bacon have
thought of this method of establishing the probable truth of a
theory? It leaves out of consideration a multitude of facts, and one
of them at least is of the utmost importance. It is that in the domain
of animated matter, in organized beings, and most signally in the
animal kingdom, there is a principle of life; and, whatever may be
the universal laws of the redistribution of matter and motion, in their
operation upon or among the products which are not endowed with
this principle, when we come to reason about products that are
endowed with it we are not entitled to conclude that this principle of
animal life is itself a product of the operation of those laws because
they have resulted in products which do not possess life, or life of
the same kind. In order to reach the conviction that animal
organisms have resulted solely from the operation of the laws of
matter and motion, we must not undertake to reconcile the facts
with those laws, but we must have some evidence that those laws
have produced living beings with complex and diversified organisms,
and this evidence must at least tend to exclude every other
hypothesis. It is not enough to flout at all other hypotheses, or to
pronounce them ex cathedra to be idle tales.
Kosmicos. You must not catch at single expressions and make
yourself a captious critic. That would be unworthy of such an
inquirer as you profess to be, and as I believe you are. Mr. Spencer
did not mean, by reconciling the facts with the laws of matter and
motion, that we are to distort the facts. He meant that we are to
discover the correspondence between the facts and the operation of
those laws. Now, let me show you more explicitly that he is quite
right. There are certain laws of matter and motion, discoverable and
discovered by scientific investigation, which prevail throughout all
Nature. The phenomena which they produce, although not yet fully
understood, justify the assumption of their universality and their
modes of operation. It is perfectly legitimate, therefore, to reason
that the same laws which have produced the observable phenomena
in other departments of Nature have had a like potency as causes by
which the phenomena in the animal kingdom have been produced.
Using this legitimate mode of reasoning, Mr. Spencer traces the
operation of those laws upon the primal molecules, which are
peculiarly sensitive to their effects. He follows them through the
successive aggregations of higher combinations until he arrives at
the protoplasmic substance, out of which, from its capability of
assuming an infinity of forms, aided by the environing conditions,
the simplest organic forms become evolved, and thus what you call
the principle of life gradually arose through a vast extent of time. He
is therefore perfectly consistent with himself in denying the absolute
commencement of organic life on the globe; for you must
understand that he means by this to deny that there was any point
of time, or any particular organism, at or in which animal life can be
said to have had its first commencement, without having been
preceded by some other kind of being, out of which the more highly
organized being has been produced by modifications wrought by
insensible gradations. If you will attend closely to his reasoning, you
will see that you have small cause for criticising it as you have; and,
if you will look at one of his illustrations, you will see the strength of
his position. Hear what he says: "It is no more needful to suppose
an absolute commencement of organic life or a 'first organism' than
it is needful to suppose an absolute commencement of social life and
a first social organism. The assumption of such a necessity in this
last case, made by early speculators with their theories of 'social
contracts' and the like, is disproved by the facts; and the facts, so far
as they are ascertained, disprove the assumption of such a necessity
in the first case."[110] That is to say, as the social facts, the social
phenomena, disprove the "social contract" as an occurrence taking
place by human design and intention, so the phenomena of animal
life disprove the assumption of such an occurrence as its
commencement by divine intervention, or its commencement at all.
Sophereus. I think I understood all this before, just as you put it, but I
am not the less obliged to you for the restatement. In regard to
society, I know not why the family, the institution of marriage, is not
to be regarded as the first social organism, and the union of two or
more families in some kind of mutual league is certainly the first
society in a more comprehensive sense. I care very little about the
theory of the social contract, as applied to more complex societies,
although, as a kind of legal fiction, it is well enough for all the uses
which sound reasoners nowadays make of it. But the institution of
marriage, the family, is no fiction at all; it is a fact, however it was
first established, and it was the absolute commencement of social
life. But I do not hold to this sort of analogies, or to this mode of
reasoning from what happens in a department, in which the actions
of men have largely or exclusively influenced the complex
phenomena, to a department in which human influence has had
nothing to do with the phenomena. But now let us come back to the
proposition that there never was any absolute commencement of
organic life on the globe. I will take Mr. Spencer's meaning—his
denial, as you put it—and will test it by one or two observations
upon his own explanation, as given in the elaborate paper in which
he replied to a critic in the "North American Review" a little more
than four years ago.[111] In the first place, then, as to time. It will
not do to say that there never was a time when such a product as
life, animated or organized life, had its first existence. To whatever it
owed its existence, it must at some time have begun to exist. It
matters not how far back in the ages of the globe you place it: you
must contemplate a time when it did not exist, and a point of time at
which it began to exist. It matters not that you can not fix this time.
There was such a time, whether you can fix it chronologically or not.
In the next place, however minute the supposed gradations which
you trace backward from a recognizable organism to the primal
protoplasmic substance, out of which you suppose it to have been
gradually evolved, and through whatever extent of time you imagine
these gradations to have been worked out by the operation of the
forces of Nature, modifying successive beings, you must find an
organism to which you can attribute life. Whatever that organism
was, it was the commencement of organic life; for, when you go
back of it in the series, you come to something that was not organic
life, but was merely a collection of molecules or a product of
aggregated molecules, that had a capacity to be developed into an
animated organism under favorable conditions. "It is," says Mr.
Spencer, "by the action of the successively higher forms on one
another, joined with the action of environing conditions, that the
highest forms are reached." Some one, then, of those highest forms,
something that can be called an animal organism, some being
endowed with life, was the commencement of organic life on the
globe; and it is just as correct and necessary to speak of it as the
"absolute" commencement as it is when we speak of Darwin's
aquatic grub, or of the Mosaic account of the creation of the
different animals by the hand and will of God. Neither Mr. Spencer
nor any other man can construct a chain of animated existence back
into the region of its non-existence without showing that it began to
have an existence. He can say that the affirmation of universal
evolution is in itself a negation of an absolute commencement of
anything. And so it is theoretically. But this does not get over the
difficulty. On his own explanation of the mode in which organisms
have been evolved, there must have been a first organism, and in
that first organism life began. So that I am not yet prepared to yield
my criticism, or to yield my convictions to a writer who is so much
carried away by his theory.
Kosmicos. But you will allow that the theory is perfect in itself; and
why, then, do you say that he is carried away by it? You ought either
to give up your criticism, or to show that there is a superior
hypothesis by which to account for the origin of organisms, and one
that is supported by stronger proofs and better reasoning. You have
nothing to oppose to Mr. Spencer's explanation of the origin of
organic life, excepting the fable which you find in the book of
Genesis.
Sophereus. Undoubtedly the opposite hypothesis is that which
attributes to a Creator the production of organic life; and whether
the Mosaic account, as it stands, be a fable or a true narrative of an
actual occurrence, what we have to do is to ascertain, upon correct
principles of reasoning, whether the creating power can be
dispensed with. Mr. Spencer dispenses with it altogether. He gives it
a direct negative in the most absolute manner. But the perfection of
his theory depends upon its ability to sustain itself as an explanation
of the existence of organisms without the intervention of a creating
power anywhere at any time. I have already suggested the serious
defect of his whole philosophic scheme as applied to the existence of
organisms, namely, that the foundation of the theory, the existence
of the molecules with their properties and capacities tending to
rearrangement under the laws of matter and motion, those laws
themselves, and the environing conditions which assist the process
of adjustment and combination, must all have had an origin, or a
cause. If we can get along without that origin, without any cause,
without any actor laying the foundations of the world, we can make
a theory. But that theory can not sustain itself by such a negation if
all experience, observation, and reflection amount to anything; for
these all point in one direction. They all tend to show that every
existing thing must have had a cause, that every product must have
had an origin, and, if we place that origin in the operation of certain
laws of matter and motion upon and among the primal molecules of
matter, we still have to look for the origin of those laws and of the
molecules on which they have operated. If we say that these things
had no origin, that they existed without having been caused to exist,
we end in a negation at which reason at once rebels. If, on the other
hand, we reject, as we must reject, this negation, then the same
power which could establish the laws of matter and motion, and give
origin to the molecules and the favoring conditions by which their
aggregated higher forms are supposed to have been developed, was
alike capable of the direct production of species, the creation of the
sexes, and the establishment of the laws of procreation and
gestation. So that it becomes a question of probability, of the weight
of evidence, as to whether we can explain the phenomena of
species, of the sexual division and the sexual union, with all that
they involve, without the hypothesis of direct intervention, design,
and formative skill of a boundless character. I have seen no
explanation of the origin of species and of the sexual distinction,
with its concomitant methods of reproduction, that does not end in
an utter blank, whenever it undertakes to dispense with that kind of
direct design to which is derisively given the name of "miraculous
interposition," but which in truth implies no miracle at all.
Kosmicos. I have to be perpetually recalling you to the first principles
of Mr. Spencer's philosophy. You seem to think it enough to point to
the existence of species and the sexual division, as if his philosophy
did not afford the means of accounting for them by the operation of
natural causes. Let me put to you, then, this question: If natural
causes have produced a crystal, by successive new combinations of
molecules of matter through gradations rising successively into
higher forms, why should not natural causes, acting upon other
molecules in a corresponding way, have produced organic life, or
animated organisms? If natural causes have evolved out of certain
molecules the substance known as organizable protein, why should
not the continued operation of the same or similar causes have
modified organizable protein into some distinct and recognizable
animated organism? If you admit this as a possible or highly
probable result, why should not natural causes have produced, in
the course of millions of years, the division of the sexes and the
methods of procreation and multiplication?
Sophereus. I will assign the reasons for not adopting the conclusions
to which you expect me to arrive, in a certain order. In the first
place, the capacity of certain molecules to result in the formation of
a crystal, under the operation of what you call natural causes,
requires that the molecules, their capacity, and the natural causes
should all have had an origin, call it known or unknown. The cause
was of equal potency to produce the crystal directly, or anything else
that exists in Nature. The same thing is true of certain other
molecules which, under the operation of the so-called natural
causes, have resulted in organizable protein. There must have been
an origin to the molecules, to their capacity, and to the laws which
effect their combinations; and this cause could equally fashion an
organism and fashion it in the related forms of male and female by
direct intervention, for to such a power there is no assignable limit.
In the next place, the distinction between inanimate and animated
matter, between beings endowed and beings not endowed with
animal life, is a distinction that can not be overlooked; for, although
we find this distinction to be a fact that has resulted after the
operation of whatever causes may have produced it, we must still
note that there is a distinction, and a very important one. It may be
that the dividing line is very difficult of detection; that it is impossible
to determine in all cases just where organizable matter passes from
dead matter into a living organism. But that at some point there has
arisen a living organism, however produced, is certain. Now,
suppose that what you call natural causes have operated to bring
organizable matter up to this dividing line, the question is, whether
we can conclude that they have had the potency to pass that line,
and to lead of themselves to all the varying and manifold results of
species, the division of the sexes, and all that follows that division.
Certain great facts seem to me to negative this conclusion. The first
is, that we have species, which differ absolutely from each other as
organisms, in their modes of life, and their destinies, however strong
may be the resemblances which obtain among them in certain
respects. The second fact is, that each of the true species is divided
into the related forms of male and female, and is placed under a law
of procreation, by the sexual union, for the multiplication of
individuals of that species. The third fact is, that no crosses take
place in Nature between different species of animals—between the
true species—resulting in a third species, or a third animal. It is true
that multiplication of individuals of some of the lowest organisms
takes place without the bisexual process of procreation, as where, in
the severance of a part of an organism the severed part grows,
under favorable conditions, into a perfect organism of the same kind,
as in the analogous phenomenon of a plant propagated by a branch
or a slip from the parent stem. But this occurrence does not take
place among the animals which are placed for their multiplication
under the law of the sexual union and the sexual procreation. The
sexual division, therefore, the law of sexual procreation, and all that
they involve, have to be accounted for. Can they be accounted for by
the theory of evolution? Wherever you place their first occurrence,
you have to find a process adequate to their production. What, then,
entitles you to say that the hypothesis of their production, by the
capacity and tendency of organizable substances, when they have
reached certain combinations, is superior to the hypothesis of a
direct interposition and a formative will? At the outset, you must
begin with some interposition and some formative will; you must
account for the existence of the very capacities of matter to become
organized under the laws of the redistribution of matter and motion,
or you will end nowhere whatever. If you assume, as you must, that,
in laying "the foundations of the world," there was exercised some
interposition and some formative will, you have a power which was
just as adequate to the production of species, and their sexual
division, as it was to the endowment of matter with certain
properties and capacities, and the establishment of any laws for the
redistribution of matter and motion. If you deny the existence and
potency of the original power in the one production you must deny
them in the other. If you concede them in the one case, you must
concede them in the other. Now, although the original power was
equal to the endowment of organizable matter with its capacities for
and tendencies to organization, and may be theoretically assumed to
have made that endowment, the question is, whether these
capacities and tendencies, without special formative interposition,
and by the mere force of what you call natural causes, were equal to
the production of such phenomena as the division of the sexes and
all that follows that division. Can it with any truth he said that the
so-called natural causes have produced any phenomena which can
be compared, on the question of special design, to the phenomena
of the sexual division, the law of sexual procreation, and the whole
system of the multiplication of individuals of distinct and true
species? When I can see any facts which will warrant the belief that
the origin of the sexes is to be attributed to the capacity of
organizable protein to form itself into new compounds, to the
capacity of these new compounds to become living organisms, and
to the capacity of these living organisms, without the intervention of
any formative will specially designing the result, to divide themselves
into related forms of male and female, to establish for themselves
the law of procreation, and to limit that procreation to the same
species, I shall, perhaps, begin to see some ground for the superior
claims of the evolution hypothesis. I should like, by-the-by, to see a
system of classification of animal organisms, based exclusively on
the distinction between the bisexual and the unisexual, or the non-
sexual, methods of reproduction, and without running it out into the
analogies of the vegetable world. I fancy that it would be found
extremely difficult to account for the bisexual division without
reaching the conclusion that it required and was effected by a
special interposition. At all events, I should like to see it explained
how the asexual and the unisexual construction passed into the
bisexual by the mere operation of what you call natural causes.
Kosmicos. You said, a while ago, that you had never learned any
nursery-stories. Yet, all along, you seem to me to have been under
the influence of the Mosaic account of the creation. Of course you
have read it, and, although you did not learn anything about it in
childhood, and now try to treat it solely as a hypothesis, without any
regard to its claims as a divinely inspired narrative, it is certainly
worth your while to see how completely it becomes an idle tale of
the nursery when scientific tests are applied to it. Hear what Spencer
says about the creation of man, as given by Moses: "The old Hebrew
idea that God takes clay and molds a new creature, as a potter
might mold a vessel, is probably too grossly anthropomorphic to be
accepted by any modern defender of special creations."
Sophereus. Let us see about this. Let us discard all idea of the source
from which Moses received his information of the occurrences which
he relates, and put his account upon the same level with Plato's
description of the origin of animals, and with the Darwinian or
Spencerian theory of that origin; regarding all three of them, that is
to say, as mere hypotheses. Whatever may be the supposed conflict
between the Mosaic account of the creation and the conclusions of
geologists concerning the periods during which the earth may have
become formed as we now find it, the question is, on the one hand,
whether the Hebrew historian's account of the process of creation is
a conception substantially the same as that at which we should have
arrived from a study of Nature if we had never had that account
transmitted to us from a period when the traditions of mankind were
taking the shapes in which they have reached us from different
sources; or whether, on the other hand, it is so "grossly
anthropomorphic" and absurd that it is not worthy of any
consideration as an occurrence that it will bear the slightest test of
scientific scrutiny. Let any one take the Mosaic narrative, and,
divesting himself of all influence of supposed inspiration or divine
authority speaking through the chosen servant of God, and
disregarding the meaning of those obscure statements which divide
the stages of the work into the first and the second "day," etc., let
him follow out the order in which the Creator is said by Moses to
have acted. He will find in the narrative an immense condensation,
highly figurative expressions, and many elliptical passages. But he
will also find that the Creator is described as proceeding in the
exertion of his omnipotent power in a manner which we should be
very likely to deduce from a study of his works without this
narrative. We have, first, the reduction of the earth from its chaotic
condition—"without form and void"—to the separation of its
elemental substances; then the creation of light; the separation of
earth and water; the productive capacity of the dry land; the
establishment of the vegetable kingdom, each product "after its
kind"; the formation of the heavenly bodies as lights in the
firmament, to make the division of day and night, seasons and
years. It is obviously immaterial, so far as this order of the work is
concerned, down to the stage when the formation of the first
animals took place, in what length of time this first stage of the work
was accomplished; whether it was done by an Omnipotence that
could speak things into existence by a word, or whether the process
was carried on through periods of time of which we can have no
measure, and by the operation of infinitely slow-moving agencies
selected and employed for the accomplishment of a certain result.
Confining our attention to the first stage of the work as we find it
described, we have the formation of the earth, light, air, the
heavenly bodies, alternations of day and night, seasons and years,
and the vegetable kingdom, before any animal creation. We then
come to the formation of animals which are to inhabit this
convenient abode, and which are described as taking place in the
following order: first the water animals, the fowls of the air, and the
beasts of the field, "each after its kind"; then, and finally, the
creation of man. Respecting his creation, we are told that it was the
purpose of the Almighty to make a being after a very different
"image" from that of any other creature on the earth; and whatever
may be the true interpretation of the language employed, whether
man was created literally "in our image, after our likeness," or
according to an image and a likeness of which his Creator had
conceived, there can be no doubt that what Moses described as the
purpose of God was to make a being differing absolutely from all the
other animals by a broad line of demarkation which is perfectly
discoverable through all the resemblances that obtain between him
and all the other living creatures. To this new being there was given,
we are told, dominion over all the other animals, and the fruits of
the earth were assigned to him for food; he was formed out of the
dust of the earth, the breath of life was breathed into his nostrils,
and he became "a living soul." Let us now see if this statement of
the creation of man is so "grossly anthropomorphic" as is supposed.
You are aware that Buffon, who was certainly no mean naturalist or
philosopher, and who was uninfluenced by the idea that the book of
Genesis was an inspired production, reached the conclusion that a
study of nature renders the order of man's creation as described by
Moses a substantially true hypothesis. "We are persuaded," said
Buffon, "independently of the authority of the sacred books, that
man was created last, and that he only came to wield the scepter of
the earth when that earth was found worthy of his sway."[112] You
evolutionists will say that this may be very true upon your
hypothesis of his gradual development out of other animals, through
untold periods of time. But now let us see whether Moses was so
grossly unscientific, upon the supposition that God created man as
he describes. If man was created, or molded, by the Deity, he was
formed, in his physical structure, out of matter; and all matter may
be figuratively and even scientifically described as "the dust of the
earth," or as "clay," or by any other term that will give an idea of a
substance that was not spirit. If Moses had said that man's body was
formed out of the constituent elements of matter, or some of them,
he would have said nothing that a modern believer in special
creations need shrink from, for he would have stated an indisputable
fact. He stated in one form of expression the very same fact that a
modern scientist would have to state in another form, whatever
might have been the mode, or the power, or the time in or by which
the constituent elements were brought together and molded into the
human body. So that the derisive figure of God taking clay and
molding it into the human form, as a potter would mold a vessel,
does not strike me as presenting any proof that the account given by
Moses is so destitute of scientific accuracy, or as rendering his
statements a ridiculous hypothesis.
Kosmicos. Well, then, it comes at last to this: that you consider the
substance of the Mosaic account of the creation, independent of its
authority as an inspired statement, to be entitled to stand as a
hypothesis against the explanations given to us by the scientists of
the great modern school of evolution, notwithstanding those
explanations are in one form or another now accepted by the most
advanced scientific thinkers and explorers?
Sophereus. I certainly do. But understand me explicitly. As, after my
study of the probable origin of the solar system, and our discussion
of that subject, I expressed my conclusion that the phenomena
called for and manifested the exercise of a formative will by some
acts of special creation, so now, in reference to the animal kingdom,
I have reached the same conclusion, for reasons which I have
endeavored to assign. I can see that the operation of the process
which you call evolution may have caused certain limited
modifications in the structure and habits of life of different animals;
or rather, that limited modifications of structure and habits of life
have occurred, and hence you deduce what you call the process of
evolution. But to me this entirely fails to account for, or to suggest a
rational explanation of, the distinct existence of species, their
division into male and female, and the establishment of the laws of
procreation by which individuals of a species are multiplied—a
process which does not admit of the production of individuals of an
essentially different type from the parents, and which, so far as we
have any means of knowledge, has never commenced in one species
and ended in another, in any length of time that can be imagined, or
through any series of modifications.
Kosmicos. Let us postpone the farther discussion of the origin of
species to some future time, when I will endeavor to convince you
that both Darwin and Spencer have satisfactorily accounted for
them.
Sophereus. Very well; I shall be glad to be enlightened.

THE SINGLE-CELL HYPOTHESIS.

Note.—It will readily occur to the reader that Sophereus might
most pertinently have asked: Whence did the primal cell
originate? It is conceived of as the ultimate unit of organizable
matter; invisible to the naked eye, perhaps incapable of being
reached by the microscope, but consisting of an infinitesimally
small portion of matter, more or less organized in itself, and
possessing a capacity to unite with itself other minute particles
of matter, and so to form larger aggregates of molecules. The
hypothesis is, that this single cell has given origin to all
animated organisms, and, through an indefinite series of such
organisms, to the human race. The single cell, then, having this
capacity and this extraordinary destiny, was either the first and
only one of its kind, or it was one of many of the same kind. If
we select any supposed point of time in the far antecedent
history of matter, the question may be asked whether there
existed at first but one such cell, or many. If there were many of
such cells, how came they to exist? If one only was selected out
of many, for this extraordinary destiny of giving origin to all the
animated organisms, who or what made the selection for this
transcendent office of the one cell? If there never was but one
such cell, how did it come to exist? As these questions are
clearly pertinent, the effort to answer them inevitably conducts
us to the idea of creation, or else to the conclusion that the
numerous cells and the selected one had no origin; that the
selection was not made, but was accidental; or that the one cell,
if there never was but one, was not a created thing. Human
reason can not accept this conclusion.
 CHAPTER X.
"Species," "races," and "varieties"—Sexual division—Causation.

The two friendly disputants have again met. Sophereus begins their
further colloquy, in an effort to reach a common understanding of
certain terms, so that they may not be speaking of different things.
Sophereus. I have more than once referred to the fact that Nature
does not permit crosses between the true species of animals, in
breeding, and that we have no reason to suppose it ever did. This is
a very important fact to be considered in weighing the claims of your
theory of evolution. I have been looking into Darwin, and I find it
somewhat uncertain in what sense he uses the terms "species,"
"races," and "varieties." In his "Descent of Man," he devotes a good
deal of space to the discussion of the various classifications made by
different naturalists under these respective terms; and there is no
small danger of confusion arising from the use of these terms unless
they are defined. The possibility of the process of evolution, as a
means of accounting for the existence of any known animal,
depends in some degree upon the animals among which, by sexual
generation, the supposed transition from one kind of animal to
another kind has taken place. Darwin speaks of the difficulty of
defining "species"; and yet it is obvious (is it not?) that the theory of
the graduation of different forms into one another depends for its
possibility upon the forms which have admitted of interbreeding.
While, therefore, the term "species" is in one sense arbitrary, as
used by different naturalists, and there is no definition of it common
to them all, it is still necessary to have a clear idea of the limits
within which crosses can take place in breeding, because there are
such limits in nature. Thus, in the case of man, as known to us in
history and by observation, there are different families, which are
classed as "races." Darwin speaks of the weighty arguments which
naturalists have, or may have, for "raising the races of man to the
dignity of species." Whether this would be anything more than a
matter of scientific nomenclature, is perhaps unnecessary to
consider. Whether we call the "races" of men "species," or speak of
them as families of one race, we know as a fact that interbreeding
can take place among them all, and that between man and any
other animal it can not take place. The same thing is true of the
equine and the bovine races and their several varieties. Whether, in
speaking of the different families or races of men, we consider them
all as one "species," or as different species—and so of the varieties
of the equine or the bovine races—the important fact is, that there
are limits within which interbreeding can take place, and out of
which it can not take place. Do you admit or deny that the barriers
against sexual generation between animals of essentially different
types, which are established in nature, are important facts in judging
of the hypothesis of animal evolution?
Kosmicos. Take care that you have an accurate idea of what the
theory of evolution is. Apply it, for example, to the origin of man, as
an animal, proceeding "by a series of forms graduating insensibly
from some ape-like creature to man as he now exists." This
expresses the whole theory as applied to one animal, man, without
going behind his ape-like progenitors. It does not suppose a crossing
between the ape-like creature and some other creature that was not
an ape. It supposes a gradual development of the ape-like creature
into the man as he now exists; and, of course, the interbreeding
took place between the males and the females of that ape-like race
and their descendants—the descendants, through a long series of
forms, being gradually modified into men, by the operation of the
laws of natural and sexual selection, which I need not again explain
to you.
Sophereus. Very well, I have always so understood the theory. But
then I have also understood it to be a part of the same theory that
there is important auxiliary proof of the supposed process of
evolution to be derived from what is known to take place in the
interbreeding of different races or families of the same animal.
Whatever value there may be in this last fact, as auxiliary evidence
of the supposed process of evolution, there must have been a time,
in the development of the long series of forms proceeding from the
ape-like progenitor, when an animal had been produced which could
propagate nothing but its own type, and between which and the
surrounding other animals no propagation could take place, if we are
to judge by what all nature teaches us. You may say that the laws of
natural and sexual selection would still go on operating among the
numerous individuals of this animal which had become in itself a
completed product, and that to their descendants would be
transmitted newly acquired organs and powers, new habits of life,
and all else that natural and sexual selection can be imagined to
have brought about. But at some time, somewhere in the series, you
reach an animal of a distinct character, in which natural and sexual
selection have done all that they can do; in which there can be no
propagation of offspring but those of a distinct and peculiar type,
and the invincible barrier against a sexual union with any other type
becomes established. For this reason, we must recognize the limits
of possible interbreeding. It is best for us, therefore, to come to
some understanding of the sense in which we shall use the term
"species." For I shall press upon you this consideration—that animals
differ absolutely from each other; that there can be no interbreeding
between animals which so differ; and yet that, without interbreeding
between animals having distinct organizations, natural and sexual
selection had not the force necessary to produce, in any length of
time, such a being as man out of such a being as the ape.
Kosmicos. I will let Darwin answer you, in a passage which I will read.
"Whether primeval man," he observes, "when he possessed but few
arts, and those of the rudest kind, and when his power of language
was extremely imperfect, would have deserved to be called man,
must depend on the definition which we employ. In a long series of
forms graduating insensibly from some ape-like creature to man as
he now exists, it would be impossible to fix on any definite time
when the term 'man' ought to be used. But this is a matter of very
little importance." That is to say, in the long series of forms
descending from the ape-like creature, we can not fix on any one of
the modified descendants which we can pronounce to be separated
from the family of apes, and to have become the new family, man,
because to do this requires a definition of man. Man as he now
exists we know, but the primeval man we do not know. He may have
been an animal capable of sexual union with some of his kindred
who stood nearest to him, but yet remained apes, or he may not. It
is not important what he was, or whether we can find the time when
he ceased to belong to the family of apes and became the primeval
man. The hypothesis of his descent remains good, notwithstanding
we can not find that time, because it is supported by a great
multitude of facts.
Sophereus. I have never seen any facts which I can regard as giving
direct support to the theory. But, waiving this want of evidence,
doubtless it is not important to find the time, chronologically, when
the modified descendants, supposed to have proceeded from the
ape-like creature, became the primeval man; but it is of the utmost
importance to have some satisfactory grounds for believing that
there ever was such an occurrence as the development of the animal
man, primeval or modern man, out of such an animal as the ape.
And therefore, without reference to the sense in which naturalists
use the term "species," I shall give you the sense in which I use it. I
use it to designate the animals which are distinct from each other, as
the man, the horse, the ape, and the dog are all distinct from each
other. Speaking of man as one true species, I include all the races of
men. Speaking of the apes as another species, I include all the
families of apes. Speaking of the bovine, the equine, or the canine
species, I include in each their respective varieties. Now, as crosses
in interbreeding can take place between the different varieties or
families of these several species, and can not take place between
the species themselves—between those which I thus class as species
—the limits of such crosses become important facts in considering
the theory of evolution, because they narrow the inquiry to the
possibility of effecting a propagation of one species out of another
species. Take any animal which has become a completed and final
product—a peculiar and distinct creature—whether made so by
aboriginal creation or produced by what you call evolution. The
reproductive faculty of the males and the females of this distinct and
peculiar animal is limited to the generative reproduction of
individuals of the same type, by a sexual union of two individuals of
that type. Their progeny, in successive generations, may be marked
by adventitious and slowly acquired peculiarities; but unless there
can be found some instance or instances in which the process of
modification has resulted in an animal which we must regard as an
'essentially new creature—a new species—what becomes of the
auxiliary evidence which is supposed to be derived from the effects
of interbreeding between those individuals which can interbreed? I
lose all hold upon the theory of evolution, unless I can have some
proof that natural and sexual selection have overcome the barriers
against a sexual union among animals which are divided into males
and females of the several species, each of which is placed under a
law of procreation and gestation peculiar to itself, and never
produces any type but its own.
Kosmicos. You wander from the principle of evolution. I have to be
perpetually restating it. Observe, then, that there are multitudes of
facts which warrant the belief that, starting with any one kind of
animal organism, however peculiar and distinct, the struggle for
existence among the enormous number of individuals of that animal
becomes most intense, and a furious battle is constantly going on.
The best-appointed males, in the fierceness of the strife for
possession of the females, develop new organs and powers, or their
original organs and powers are greatly enhanced. Their descendants
share in these modifications; and the modifications go on in a
geometrical ratio of increase through millions of years, until at some
time there is developed an animal which differs absolutely from its
remote progenitors which were away back in the remote past, and
which began the struggle for individual life and the continuation of
their species or their race in a condition of things which left the
fittest survivors the sole or nearly the sole propagators of new
individuals. This struggle for existence may have begun—probably it
did begin—before the separation of the sexes, when the organism
was unisexual or even asexual. That is to say, there may have been,
and there probably was, an organism which multiplied with
enormous rapidity, without the bisexual method of reproduction. The
vast multitude of such individuals would lead to the destruction of
the weakest; the strong survivors would continue to give rise to
other individuals, modified from the original type, until at length, by
force of this perpetual exertion and struggle and the survival of the
fittest, modifications of the method of reproduction would ensue,
and the bisexual division would be developed and perpetuated.
Sophereus. I confess I did not expect to hear you go quite so far. I
will yield all the potency to natural and sexual selection that can be
fairly claimed for them as modifying agencies operating after the
sexual division has come about; but I have, I repeat, seen no facts
which justify the hypothesis that they have led to distinct organisms
between which no propagation can take place. But now you expect
me to accept the startling conclusion that at some time the asexual
or the unisexual method of reproduction passed into the bisexual,
without any formative will or design of a creating power, and without
any act of direct creation. We know what Plato imagined as the
origin of the sexual division, and that he could not get along without
the intervention of the gods. What modern naturalist has done any
better? I have examined Darwin's works pretty diligently, and I can
not get from them any solution of the origin of the bisexual division.
I am left to reason upon it as I best can. We know, then, that in the
higher animal organisms the individuals of each species are divided
into the related forms of male and female, and that for each species
there exists the one invariable method of the sexual union, and a
law of gestation peculiar to itself. One hypothesis is that this system
was produced by the operation of natural causes, like those which
are supposed to have differentiated the various kinds of organisms;
the other hypothesis is that it was introduced with special design, by
an act of some creative will. If we view the phenomena of the sexual
division and the sexual genesis in the highest animal in which they
obtain, we find that they lead to certain social results, which plainly
indicate that in this animal they exist for a great and comprehensive
moral purpose, which far transcends all that can be imagined as the
moral purpose for which they exist in the other animals. To a
comparatively very limited extent, certain social consequences flow
from the law of sexual division and genesis among the other
animals. But there is no animal in which the moral and social effects
of this law are to be compared to those which it produces in the
human race. Not only does the same law of multiplication obtain
among the human race; not only does it lead to love of the offspring
far more durable and powerful than in the case of any other animal;
not only is it the origin of a society far more complex, more lasting,
and more varied in its conditions than any that can be discovered in
the associations of other animals which appear to have some social
habits and to form themselves into communities, but in the human
race alone, so far as we have any means of knowledge, has the
passion of sexual love become refined into a sentiment. You may
remember the passage in the "Paradise Lost" in which Raphael, in
his conversation with Adam, touches so finely the distinction
between sexual love in the human race and in all the other animals.
The angel reminds Adam that he shares with the brutes the physical
enjoyment which leads to propagation; and then tells him that there
was implanted in his nature a higher and different capacity of
enjoyment in love. The conclusion is:—

"... for this cause

Among the beasts no mate for thee was found."

In the human being alone, even when there is not much else to
distinguish the savage from the beasts around him, the passion of
love is often something more nearly akin to what might be looked
for in an elevated nature, than it can be among the brutes. What do
the poetry and romance of the ruder nations show, but that this
passion of sexual love in the human being is one in which physical
appetite and sentimental feeling are so "well commingled" that their
union marks the compound nature of an animal and a spiritual
being? How human society has resulted from this passion, how in
the great aggregate of its forces it moves the world, how in its
highest development it gives rise to the social virtues, and in its
baser manifestations leads to vice, misery, and degradation, I do not
need to remind you. How, then, is it possible to avoid the conclusion
that in man the sexual passion was implanted by special design and
for a special purpose, which extends far beyond the immediate end
of a continuation of the race?
Kosmicos. Why do you resort to a special purpose in the constitution
of one animal, and to the absence of a similar purpose from the
constitution of another animal? In both, the consequences make a
case of the post hoc just as plainly as they make a case of the
propter hoc. It is just as rational to conclude that they only show the
former as it is to conclude that they establish the latter. In man, we
have the physical fact of the sexual division, and all you can say is
that it is followed by certain great and varied moral phenomena. In
the other animals, we have the same physical fact, followed by
moral phenomena less complex and varied, and not so lasting. In
neither case can you say that there was a special and separate
design, according to which the same physical fact was intended to
produce the special consequences which we observe in each. Why,
as the species called man became developed into beings of a higher
order than the primates of the race or than their remote progenitors,
should not this passion of sexual love have become elevated into a
sentiment and been followed by the effects of that elevation, just as
the gratification of another appetite, that for food, par exemple, has
been refined by the intellectual pleasures of the social banquet and
the interchange of social courtesies? Is there anything to be proved
by the institution or the practice of marriage, beyond this—that it
has been found by experience to be of great social utility, and is
therefore regulated by human laws and customs, which vary in the
different races of mankind? Monogamy is the rule among some
nations, polygamy is at least allowed in others. You can predicate
nothing of either excepting that each society deems its own practice
to be upon the whole the most advantageous. You can not say that
there is any fixed law of nature which renders it unnatural for one
man to have more than one wife. In many ages of the world there
have been states of society in which the family has had as good a
foundation in polygamous as it has had in monogamous unions.
Looking, then, at these undeniable facts, and also at the fact that
marriage, whether monogamous or polygamous, is an institution
regulated by human law and custom, we have to inquire for the
reason why human law and custom take any cognizance of the
relation. We find that, among some of the other animals, the sexes
do not pair excepting for a single birth. The connection lasts no
longer than for a certain period during which the protection of both
parents is needed by the offspring, and not always so long even as
that. It has become the experience of mankind that the connection
of the parents ought to be formed for more than one birth; shall be
of indefinite duration; and this because of the physical and social
benefits which flow from such a permanency of the union. This has
given rise to certain moral feelings concerning the relation of
husband and wife. But we have no more warrant, from anything that
we can discover in nature, for regarding the permanency of marriage
among the human race as a divine institution than we have for
regarding its temporary continuance among the other animals as a
divinely appointed temporary arrangement. In the one case, the
permanency of the union has resulted from experience of its utility.
In the other case, the animal perceives no such utility, and therefore
does not follow the practice. Upon the hypothesis that all the
animals, man included, had a common origin, it is very easy to
account for the difference which prevails between man and the other
animals in this matter of marriage, or the pairing of the sexes. As
man became by insensible gradations evolved out of some pre-
existing organism, and as moral sentiments became evolved out of
his superior and more complex relations with his fellows, from his
experience of the practical utility of certain kinds of conduct and
practice, the sentiments became insensibly interwoven with his
feelings about the most important of his social relations, the union of
the sexes in marriage. This is quite sufficient to account for the
difference between man and the other animals in regard to the
duration of such unions, without resorting to any intentional or
divine or superhuman origin of that difference.
Sophereus. For the purpose of the argument, I concede that this is a
case of either the post hoc or the propter hoc. I have been pretty
careful, however, in all my investigations, not to lose sight of this
distinction in reasoning on the phenomena of nature or those of
society. I think I can perceive when there is a connection between
cause and effect, when that connection evinces an intelligent design,
and when the phenomena bear no relation to a certain fact beyond
that of sequence in time. What, then, have we to begin with? We
have the fact that the human race is divided into the two forms of
male and female, and that the passion or appetite of sexual love
exists in both sexes, and that its gratification is the immediate cause
of a production of other individuals of the same species. We next
have the fact that this union of the sexes is followed by an
extraordinary amount of moral and social phenomena that are
peculiar to the human race. This sequence proves to me an
intentional design that the moral and social phenomena shall flow
from the occurrence of the sexual union, for it establishes not only a
possibility, but an immensely strong probability, that the phenomena
were designed to flow from this one occurrence among this
particular species of animal. If this connection between the original
physiological fact and the moral and social phenomena be
established to our reasonable satisfaction, it is the highest kind of
moral evidence of a special design in the existence of the sexual
division and the sexual passion among the human race. You
remember old Sir Thomas Browne's suggestion, that men might
have been propagated as trees are. But they are not so propagated.
If they were, no such consequences would have followed as those
which do follow from the mode in which they are in fact propagated.
These consequences are most numerous and complex, and they are
capable of being assigned to nothing but the sexual division and the
sexual union as the means of continuing the race. Turn now to some
of the other animals among whom there prevail the same bisexual
division and the same method of procreation and multiplication. You
find they result in sexual unions of very short duration, and that, if it
is followed by phenomena that in some feeble degree resemble
those which are found in human society, they bear no comparison in
point of complexity and character to those which in the human race
mark the family, the tribe, and the nation. And here there occurs
something which is closely analogous to what I pointed out to you in
considering the supposed development of the first animal organism.
I said that although you may theoretically suppose that the first
animal organism was formed by the spontaneous union of molecular
aggregates, and that the higher organisms were evolved out of the
lower solely by the operation of causes which you call "natural," yet
that when you come to account for the existence of true and distinct
species, each with its sexual division and its law of procreation and
gestation, you must infer a special design and a formative will,
because there has never been suggested any method by which the
so-called natural causes could have produced this division of the
sexes and this invariable law of the sexual procreation among
individuals of the same species. Here, then, we arrive at a distinct
moral purpose; for, when we compare the different social
phenomena which follow the operation of the sexual division and
procreation in man with the social phenomena which follow in the
case of the other animals, we find a difference that is not simply one
of degree, but is one of kind. We find the origin of the family, the
tribe, and the nation: the source of the complex phenomena of
human society. We may therefore rationally conclude that in man the
sexual division and the sexual passion were designed to have effects
that they were not designed to have in the other animals. To
suppose that these vastly superior consequences in the case of man
are the mere results of his perception of their utility will not account
for the fact that when he does not recognize the utility—when he
departs from the law of his human existence—human society can
not be formed and continued. Although it is possible for human
society to exist with polygamous marriages, and even to have some
strength and duration, yet human society without the family, with
promiscuous sexual intercourse, with no marriages and no ties
between parents and children, never has existed or can exist.
Compare Plato's curious constitution of the body of "guardians," in
his "Republic," and the strange method of unions, the offspring of
which were not allowed to know their parents or the parents to
know their own children. This was not imagined as a form of human
society, but was entirely like a breeding-stud. Among the brutes,
permanent marriages, families, do not exist, not because the
animals do not perceive their social utility, but because the purposes
of their lives, their manifest destinies, show that there was no
reason for endowing them with any higher capacity for the sexual
enjoyment than that which leads to the very limited consequences
for which the division of the sexes was in their cases ordained. But
in the case of man there is a further and higher capacity for the
sexual enjoyment, which becomes the root of his social happiness,
and which distinguishes him from the brute creation quite as
palpably as the superiority of his intellectual faculties. In all this we
must recognize a moral purpose.
Kosmicos. Pray tell me why it is not just as rational to conclude that
these moral phenomena, as results of the human passion of love,
have become, in all their complex and diversified aspects, the
consequences of a progressive elevation of the human animal to a
higher plane of existence than that occupied by the inferior species,
or than that occupied by the primeval man. When man had become
developed into an animal in whom the intellect could become what it
is, he could begin to perceive the social utility of certain modes of
life, and from this idea of their utility would result certain maxims of
conduct which would be acted on as moral obligations. Thus,
commencing with a consciousness that the race exists with the
sexual division into male and female, there would begin to be
formed some ideas of the superior social utility of a regulated sexual
union of individuals and of permanent marriages. These ideas would
become refined as the progressive elevation of the race went on,
and that which we recognize as the sentimental element in the
passion of love would become developed out of the perceptions of a
superior utility in the permanent devotion and consecration of two
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