Applied Psychometrics using SPSS and AMOS 1st

Edition Holmes Finch pdf download

https://ebookgate.com/product/applied-psychometrics-using-spss-
and-amos-1st-edition-holmes-finch/

Get the full ebook with Bonus Features for a Better Reading Experience on ebookgate.com
Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Applied Psychometrics Using SPSS and AMOS 1st Edition

Holmes Finch

https://ebookgate.com/product/applied-psychometrics-using-spss-and-
amos-1st-edition-holmes-finch-2/

ebookgate.com

Multilevel Modeling Using R 1st Edition Edition W. Holmes

Finch

https://ebookgate.com/product/multilevel-modeling-using-r-1st-edition-
edition-w-holmes-finch/

ebookgate.com

Multilevel Modeling Using R 1st Edition Edition W. Holmes

Finch

https://ebookgate.com/product/multilevel-modeling-using-r-1st-edition-
edition-w-holmes-finch-2/

ebookgate.com

Introduction to Structural Equation Modeling Using IBM

SPSS Statistics and AMOS 2nd Edition Niels J. Blunch

https://ebookgate.com/product/introduction-to-structural-equation-
modeling-using-ibm-spss-statistics-and-amos-2nd-edition-niels-j-
blunch/
ebookgate.com
Discovering Statistics Using IBM SPSS Statistics 6th
Edition Andy Field

https://ebookgate.com/product/discovering-statistics-using-ibm-spss-
statistics-6th-edition-andy-field/

ebookgate.com

Using SPSS for Windows and Macintosh Analyzing and

Understanding Data 4th Edition Samuel B. Green

https://ebookgate.com/product/using-spss-for-windows-and-macintosh-
analyzing-and-understanding-data-4th-edition-samuel-b-green/

ebookgate.com

SPSS Advanced 15 0 Manual 1st Edition Inc. Spss

https://ebookgate.com/product/spss-advanced-15-0-manual-1st-edition-
inc-spss/

ebookgate.com

Statistics Without Maths for Psychology Using Spss for

Windows 4th Edition Christine P. Dancey

https://ebookgate.com/product/statistics-without-maths-for-psychology-
using-spss-for-windows-4th-edition-christine-p-dancey/

ebookgate.com

Using IBM SPSS Statistics An Interactive Hands On Approach

2nd ed. Edition James O. Aldrich

https://ebookgate.com/product/using-ibm-spss-statistics-an-
interactive-hands-on-approach-2nd-ed-edition-james-o-aldrich/

ebookgate.com
Applied Psychometrics
Using SPSS and AMOS

W. Holmes Finch
Ball State University

Jason C. Immekus
University of Louisville

Brian F. French
Washington State University

INFORMATION AGE PUBLISHING, INC.

Charlotte, NC • www.infoagepub.com
 Library of Congress Cataloging-in-Publication Data
A CIP record for this book is available from the Library of Congress
  http://www.loc.gov
ISBN: 978-1-68123-526-4 (Paperback)
978-1-68123-527-1 (Hardcover)
978-1-68123-528-8 (ebook)

Copyright © 2016 Information Age Publishing Inc.

Printed in the United States of America

 Contents

Preface............................................................................................ ix

1 Introduction to Psychometric Concepts.......................................... 1

Measurement Basics........................................................................... 1
Classical Test Theory......................................................................... 2
Item Response Theory....................................................................... 5
Dichotomous Items............................................................................. 5
Polytomous Items............................................................................. 10

2 Item Analysis..................................................................................15
Introduction..................................................................................... 15
Classical Test Theory Item Difficulty............................................. 15
CTT Item Difficulty for Dichotomous Items....................................15
CTT Item Difficulty for Polytomous Items......................................... 27
Classical Test Theory Item Discrimination.................................... 29
Extreme Groups Calculation Method................................................ 30
Biserial/Point Biserial Correlations................................................. 40
Chapter Summary............................................................................ 46

3 Reliability.......................................................................................47
Introduction..................................................................................... 47
Measures of Internal Consistency................................................... 48
KR-20............................................................................................. 49
Cronbach’s α for Ordinal Data........................................................ 61
 Split-Half Reliability........................................................................ 64
Test–Retest Reliability...................................................................... 70
Chapter Summary............................................................................ 73

4 Generalizability Theory..................................................................75
Introduction..................................................................................... 75
G-Studies/D-Studies........................................................................ 76
Variance Components..................................................................... 77
Generalizability Coefficient/Phi Coefficient................................. 80
Example 1: One Facet Crossed Design.............................................. 81
Example 2: Two Facet Crossed Design.............................................. 88
Chapter Summary............................................................................ 93

5 Validity........................................................................................... 95
Introduction..................................................................................... 95
Types or Sources of Validity Evidence............................................ 97
Concurrent Validity....................................................................... 101
Considerations in Concurrent Validity Assessment.......................... 112
Predictive Validity...........................................................................114
Discriminant Groups Validity........................................................116
Construct Validity.......................................................................... 127
Exploratory Factor Analysis as a Tool for Investigating Construct
Validity................................................................................... 127
Confirmatory Factor Analysis as a Tool for Investigating Construct
Validity................................................................................... 150
Fitting a CFA Model Using AMOS................................................ 153
Chapter Summary.......................................................................... 166

6 Issues in Scoring...........................................................................169
Introduction................................................................................... 169
Types of Scores............................................................................... 170
Raw Scores.................................................................................... 170
Weighted Scores............................................................................. 171
Percentile Scores............................................................................. 171
Standard Scores............................................................................. 173
Calculation of Raw Scores Using SPSS..........................................176
Calculation of Weighted Scores Using SPSS................................ 183
Calculation of Percentiles Using SPSS......................................... 188
 Calculation of Standardized Scores Using SPSS......................... 191
Chapter Summary.......................................................................... 193

7 Differential Item Functioning.......................................................195

Introduction................................................................................... 195
DIF Versus Impact.......................................................................... 199
Mantel–Haenszel Test.................................................................... 199
Logistic Regression........................................................................ 203
Examples........................................................................................ 206
Logistic Regression Example.......................................................... 215
Chapter Summary.......................................................................... 222

8 Equating.......................................................................................225
Introduction................................................................................... 225
Equating Sampling Designs.......................................................... 226
Mean Equating............................................................................... 229
Linear Equating............................................................................. 237
Equipercentile Equating............................................................... 251
Chapter Summary.......................................................................... 271
References....................................................................................273
 Preface

T his book was developed to provide a “one-stop” resource for many

common analyses that an applied researcher might complete when
working with various instruments to measure educational and psychologi-
cal traits. We have developed examples, collected our favorite examples and
resources, and provided explanations of analyses in one easily digestible
text. Many of the analyses presented assist in providing the recommended
evidence to support the inferences drawn from scores from such instru-
ments. That is, the results from applying these techniques assist providing
score reliability and validity evidence.
Through our years as graduate students and the first segment of our
academic lives, we have explored the use of various programs for scale de-
velopment and the study of the psychometric properties of the scores to
provide such evidence. We have had the experience, as I am sure many of
you have had, of turning to multiple books for instructions and examples
to complete analyses in the scale development and validation process. For
those readers just beginning on the psychometric ride, you too will soon ex-
perience this. By no means will this book stop the need for multiple sourc-
es, in fact, that is always encouraged. However, this book should allow the
reader to use this as a main guide and supplement to experience analyses
described in major text books. Our examples are intended to be clear and
concise with SPSS examples that can be easily adapted to fit many situa-
tions, as the reader learns and uses various techniques.
 The anticipated audience for this book includes researchers, practi-
tioners, and graduate students searching for a guide to perform common
psychometric analyses on various types of assessment data. We assume a
basic level of statistical knowledge but review concepts throughout. We en-
vision that this text will (a) patiently wait on some office shelves begging
to be handed to a student as a resource, (b) have a permanent home on
desks where it continually rises to the top of the stacks for daily use of the
applied researcher, (c) be happily carried in bags to and from work and
class by the graduate student learning techniques, (d) be listed proudly as a
reference text on syllabi, and finally (e) as an occasional drink coaster while
deep thoughts are pondered about how to solve measurement problems.
We hope that through such uses, particularly the latter, that we have pro-
vided some insight and assistance to the user in appropriately applying the
techniques and concepts discussed.
We cover major topics such as item analysis, score reliability and valid-
ity, generalizability theory, differential item functioning, equating, and so
on. Under each topic we present foundational ideas and give examples of
how to apply these ideas immediately in ones work. Chapter 7, for instance,
contains information on differential item functioning (DIF). We discuss
DIF, its importance in the score validation process, and provide three tech-
niques using SPSS to detect DIF, including how to handle clustered data
in such analyses. The caveat is we do not provide a detailed discussion of
each topic but rather the essence of each topic and several resources for
additional reading. Again, we remind you that this is just one resource to
promote your psychometric knowledge and practice.
We do assume the user has some basic knowledge and skill level in
operating SPSS. At the same time, we do attempt to present material in a
very understandable language avoiding or explaining jargon as we go. You
may find the occasional joke thrown in from time to time to spice it up. But
remember we are researchers, not comedians, even though students and
family seem to laugh often at us for the things we think about!
We do ask that if you have feedback, efficiency techniques, improve-
ments, or just plain find mistakes to please notify us. We welcome user feed-
back and will incorporate this into a revision, if demanded by the reader!
So with that, let us get started on our SPSS adventure in applying psy-
chometric techniques. In the words of Dr. Seuss, “Take it slowly. This book
is dangerous.”
Enjoy.
—W. Holmes Finch, Brian F. French,
and Jason C. Immekus,
 1
Introduction to Psychometric Concepts

Measurement Basics
Measurement is a mainstay of educational and psychological practice.
Teachers and schools measure student performance through tests, psy-
chologists measure client mood through scales such as the Beck Depres-
sion Inventory, and colleges and universities use measurements of scholas-
tic aptitude in making admissions decisions. In all of these cases, obtained
scores on the measurements plays a critical role in decision making about
individuals and groups. Therefore, these scores must be well understood
and carefully studied to ensure that they provide the best information pos-
sible. Over the last roughly 100 years, a subspecialty combining statistics
and educational psychology has developed in order to study such measures.
This field, known as psychometrics, focuses on the development, and vet-
ting of educational and psychological assessments using a wide variety of
tools. Together, these tools represent a wide array of statistical analyses that
can provide the researcher with a great deal of information regarding the
performance of a particular measure. We will cover many of these tools to-
gether in this book, focusing on how the SPSS software system can be used
to obtain information about individual items as well as the scale as a whole.
2  Applied Psychometrics Using SPSS and AMOS

In the course of reading this book, you will become familiar with
methods for analyzing data involving an entire scale (i.e., the collection of
items), as well as the individual items themselves. In addition, you will learn
about differences and similarities in studying both dichotomous items,
which have two possible outcomes, and polytomous items, which have more
than two potential outcomes. We will discuss methods for understanding
performance of an instrument at the scale level, including assessment of
reliability and validity. We will also learn about item analysis, which will pro-
vide information regarding the difficulty of individual items (i.e., how likely
an individual is to endorse the item), as well as its ability to differentiate
among examinees with different standings on the measured trait, known
as discrimination. Throughout the text we will refer to the individuals com-
pleting the items as examinees, for convenience sake. Similarly, we may re-
fer to the instruments as tests, though in fact they may not always be tests in
the sense that we often think about them. It is important to note that virtu-
ally all of the topics that we study together in this text are equally applicable
to traditional tests of achievement or aptitude, as well as to affective assess-
ments of mood, and other non-cognitive constructs. Finally, throughout
the text we will discuss the notion of the latent trait being measured. This
simply refers to the thing that we believe our instrument is assessing, be that
intelligence, depression, or political outlook. The score obtained from the
instrument will typically serve as the manifest indication of this unobserved,
or latent variable. Prior to getting into the nuts and bolts of how to analyze
data using SPSS, let us first discuss the two primary paradigms that underlie
nearly all of these analyses, classical test theory and item response theory.

Classical Test Theory

In many ways, classical test theory (CTT) serves as the basis for much of
what we think of as psychometrics and measurement. Developed over the
last 100 years or so, it underlies the notion of instrument reliability, and
much of validity assessment. In addition, although the form of the CTT
model differs substantially from that of the later developed item response
theory (IRT) model, which we will discuss shortly, they share many of the
same basic concepts. At its heart, CTT is simply a way to link an observed
score on an instrument to the unobserved entity that we are hopefully mea-
suring. Thus, for example, if we give a class of 5th graders a math exam, we
rely on individual scores to tell us how much math the students know. Ideal-
ly we would directly assess this knowledge, but for reasons that will soon be-
come clear, this isn’t possible. However, if our test is well designed, the score
should be a reasonably accurate and reliable estimate of that knowledge.
 Introduction to Psychometric Concepts   3

In this section, we will discuss the ideas underlying CTT, and their implica-
tions for educational and psychological measurement.
The basic equation in CTT is simply X = T + E, where X is the observed
score on the measure for some individual, T is the individual’s true score
on the construct of interest, and E is random error. Put into words, this
equation states that the observed score an individual receives on a test is a
function of their true knowledge of the subject (assuming we’re discussing
some type of cognitive or achievement measure) and a set of other factors
that are random in nature. In a general sense, we can think of T as a stable
characteristic inherent to the individual that would remain unchanged
over repeated administrations of the instrument, if that could be done so
that after each test examinees forgot that they had taken it (Haertel, 2006).
Error, on the other hand, is generally conceptualized as ephemeral and
unstable factors influencing examinee performance on the measure. One
way to consider error is by classifying it into four distinct types or categories,
including (a) natural variation in an individual’s performance due to fac-
tors specific to them on the day of testing (e.g., fatigue, hunger, mood); (b)
environmental factors present during test administration (e.g., room tem-
perature, ambient noise); (c) scoring variation (e.g., ratings by evaluators);
and (d) test items selected (Feldt & Brennan, 1989).
The random nature of error leads to a number of interesting proper-
ties of the CTT model. First of all, if a particular examinee could be given
the test repeatedly over a very large number of times, and each time forget
that (s)he had taken it, the mean of the errors across those test adminis-
trations would be 0 (i.e., the population mean, mE = 0). Furthermore, the
random nature of error leads to the conclusion that it is completely uncor-
related with T. In other words, if we had a group of students taking our test
a large number of times, and calculated Pearson’s r between the true score
and error, it would come out to be 0 (i.e., rT,E = 0). In addition, if we had
multiple forms of the same exam, the errors across those forms would also
be uncorrelated, again because their errors are random. Thus, rE 1,E 2 = 0.
While these results are interesting in and of themselves, they lead to a re-
lationship that is key in CTT. In general, whenever we have one variable that
is the composite of two other variables, like X = T + E, we express the variance
of the composite as σ 2X = σT2 + σ 2E + 2cov(T , E ). Given that we know T and E
are uncorrelated, we also know that the covariance between them (cov) is
also 0. Therefore, we can rewrite the composite variance of X as σ 2X = σT2 + σ 2E .
This relationship is central to the idea of test reliability, which we discuss in
some detail in Chapter 3. For the moment, we can simply define the concept
of reliability as the ratio of variance in T to the variance in X, or
4  Applied Psychometrics Using SPSS and AMOS

σT2
rxx = . (1.1)
σ + σ 2E
2
T

In Chapter 3, we will discuss in detail how to estimate reliability for a scale

using a set of data.
Often in practice, consumers of test information are interested in learn-
ing something about the likely value of an examinee’s true score on the trait
being measured. Though T can never be known exactly, using some of the
basic elements of CTT it is possible to construct a confidence interval within
which we have a certain level of confidence (e.g., 95%) that T exists. In order
to construct such an interval, we first need to understand the standard error
of measurement (SEM). Theoretically, if we could give the same individual
a measure many times, and each time they would forget they had taken the
measure, we would obtain a distribution for X. With such a distribution, we
could then calculate the standard deviation, and if there were multiple such
examinees who each had taken the test many times, then we could get stan-
dard deviations (σ 2Ei ) for each of them as well. For a given examinee, this
standard deviation would be a reflection of the variability in his/her scores.
Given that we assume T is stable for an individual, these standard deviations
would actually reflect the error variation for each individual. If we were to
average these standard deviations across all of the individual examinees in a
given sample, we would obtain the SEM. In other words,

ΣiN=1σ 2Ei
SEM = . (1.2)
N

Of course, in practice we will not be able to have individual examinees take

a test many times while forgetting that they’ve done so before, so conse-
quently we will not have access to σ 2Ei . However, using a bit of algebra, it
is possible to obtain a method for estimating SEM that is based on easily
obtained statistics such as a reliability estimate (ρXX) and the standard devia-
tion of the observed test scores (σX). In this formulation,

= σ 1 − rXX . (1.3)

As we will see in Chapter 3, the SEM can be used to construct a confidence

interval around T for examinees in the sample.
Prior to completing our discussion of CTT, it is important to consider the
issue of parallel forms of a measure. The idea of parallel forms is important in
CTT, particularly with regard to the estimation of reliability. Multiple forms
of an instrument are said to be strictly parallel when they are developed using
 Introduction to Psychometric Concepts   5

identical test specifications, produce the same distributions of X (same mean

and standard deviation), have equal covariances between any pairs of forms,
and covary equally with other measures. In addition, individual examinees
will have the same value for T for strictly parallel forms. Finally, given the
random nature of E, the covariance between error terms for two strictly paral-
lel forms will be 0. We will revisit the concept of parallel forms in Chapter 3,
when we discuss the estimation of scale reliability.

Item Response Theory

Another approach to thinking about psychometrics and measurement comes
in the form of a series of statistical models known collectively as item response
theory (IRT) (van der Linden & Hambleton, 1997; Yen & Fitzpatrick, 2006).
Whereas the focus of CTT is typically (though by no means exclusively) at the
scale level with issues such as reliability and validity, the focus of IRT is at the
item level. Indeed, the set of models that make up the suite of IRT tools all have
in common a focus on the relationship among characteristics of the items, the
examinees and the probability of the examinee providing a particular response
to the item (e.g., correct or incorrect). As we will see, IRT models are available
for both dichotomous and polytomous item responses. In this section, we will
first focus on models for dichotomous data, such as is common for items scored
as correct/incorrect, and then on models for polytomous data that might be
seen with rating scales and graded responses.

Dichotomous Items
Probably the single most common family of models in IRT is based in
the logistic framework. For dichotomous items there are three such models
that are commonly used, each differing from the others in terms of how
much information they contain about the items. The simplest such model,
which is known as the 1-parameter logistic (1PL) model, will serve as our
general (and hopefully gentle) introduction to IRT. The 1PL model can be
expressed mathematically as:

a(θ −b )
e i j
P (x j = 1 θi , a, b j ) = a(θ −b ) (1.4)
1+ e i j

where xj is the response to item j, where we code correct as 1 and incorrect

as 0. The variable θi is the value of the latent trait being measured by the test
(e.g., reading aptitude) for examinee i. The 1PL model also contains two item
parameter values: aj , item discrimination; and bj , item difficulty. For this model
6  Applied Psychometrics Using SPSS and AMOS

it is assumed that aj is constant across items, while bj is allowed to vary across

items. In terms of their meaning, we can view bj as an indicator of how likely an
individual with low proficiency is to answer the item correctly, when discussing
proficiency exams. It is important to note that item difficulty and examinee
level on the latent trait are on the same scale, which is centered at 0 and theo-
retically ranges from –∞ to ∞, though in practice it typically lies between –3 and
3 (de Ayala, 2009). An item with a lower difficulty is relatively easier than those
with higher difficulty values, such that examinees with lower levels of θi are
more likely to answer it correctly. Item discrimination, aj , reflects how well the
item can differentiate among those with greater or lesser amounts of θi , with
larger values indicating greater such discriminatory power.
For the 1PL model, all items are assumed to have equal values for item
discrimination, which is what differentiates it from other IRT models as we
will see. In some cases, researchers want to assume that the value of aj = 1, thus
defining a special case of the 1PL known as the Rasch model. Therefore, we
can think of the Rasch model as a special case of the 1PL. However, it should
be noted that in the broader measurement community the two models carry
with them very different implications for practice. We will not discuss the issues
surrounding the Rasch model further in this book, but do encourage the inter-
ested reader to investigate them. Interesting and brief summaries can be found
in Embretson and Reise (2000) and de Ayala (2009), among others.
The item characteristic curve (ICC) is a common tool used in examin-
ing the qualities of individual items. It relates the latent trait being mea-
sured (on the X axis), with the probability of a correct response (in the
case of dichotomous items) based on the particular model selected on the
Y axis. As an example, consider two items based on the 1PL model where
b1 = –0.4, b2 = 0.7, and a = 1.2. The ICC’s for these two items appear in Fig-
ure 1.1. We can see that while the shape of the items is the same, Item 2 is
shifted to the right of Item 1, because it has a higher difficulty parameter
value. In addition, we could use the ICC to determine the probability of
a correct response for an individual with a given value of θ by drawing a
straight line up from the X axis until it reaches the curve, at which point
we would draw a second line from that point on the curve to the Y axis to
obtain the probability of a correct item response.
When we cannot assume that item discrimination values are equal
across the items, we can use the 2-parameter logistic (2PL) model, which
has very similar form to the 1PL:

a (θ −b )
e j i j
P (x j = 1 θi , a j , b j ) = a (θ −b ) . (1.5)
1+ e j i j
 Introduction to Psychometric Concepts   7

Figure 1.1 Item characteristic curves for two hypothetic items based on 1PL Model.

The difference is that now aj is specific to item j. By allowing item discrimi-

nation to vary, it is possible to identify those items that are best able to
differentiate among individuals based on the level of the latent trait be-
ing measured, with higher values indicating an item that is more discrimi-
nating. The ICC’s for a pair of 2PL items with item parameters a1 = 1.2,
b1 = –0.4, a2 = 0.8, and b2 = 0.7, respectively, appear in Figure 1.2. As with the
1PL model, the more difficult item (2) is shifted to the right of the easier
item (1). In addition, the higher discrimination value of Item 1 is reflected
in its steeper ICC as compared to that of Item 2.
A third variation on the logistic framework for dichotomous items is the
3-parameter logistic model (3PL), which incorporates the probability that
an examinee will provide a correct item response simply due to chance,
perhaps by guessing. The 3PL model is expressed as

a (θ −b )
e j i j
P (x j = 1 θi , a j , b j ) = c j +(1 − c j ) a (θ −b ) , (1.6)
1+ e j i j

where cj is the pseudo-chance (or pseudo-guessing) parameter, represent-

ing the probability of a correct response for an examinee whose value of θi
8  Applied Psychometrics Using SPSS and AMOS

Figure 1.2 Item characteristic curves for two hypothetic items based on 2PL Model.

approaches –∞. As an example, imagine a multiple choice test item with

five options to select from. An individual with an extremely low value of
θi would also have an extremely low probability of answering the item cor-
rectly, based on their knowledge of the subject being tested. However, for
a multiple choice type item, there is the possibility that such an examinee
could randomly select the correct answer. This possibility is captured in the
item parameter cj . It is important to note that cj is referred to as a pseudo-
chance or pseudo-guessing parameter (rather than simply the guessing pa-
rameter) because it reflects not merely the probability of totally random
selection of the correct answer, but also differences in how well the options
in a multiple choice exam might be worded (making them relatively more
or less attractive), and the propensities of different examinees to guess
and guess well. Figure 1.3 includes ICC’s for two 3PL items, which have
identical difficulty and discrimination parameter values to those in the 2PL
example, and where c1 = 0.2, while c2 = 0.1. The difference between these
ICC’s and those in Figure 1.2 is the lower asymptote for each item. Whereas
in the 2PL case, the probability of a correct response converges to 0 as θi
approaches –∞, for the 3PL models, the probability of a correct response
converges to the value of c.
 Introduction to Psychometric Concepts   9

Figure 1.3 Item characteristic curves for two hypothetic items based on 3PL Model.

There are three common assumptions that underlie these logistic mod-
els. The first of these, unidimensionality, is that only a single latent trait is
being measured by a set of items. Thus, a test designed to measure reading
fluency in young children should only measure the construct of reading flu-
ency, and not other, additional constructs. A second, related assumption is
that of local independence, which states that responses to any pair of items
should be completely uncorrelated if we hold θi constant. Another way to
view this assumption is that the only factor that should influence an exam-
inee’s item response is her/his proficiency on the trait being measured.
The third assumption underlying IRT models is that the model form is cor-
rect for the data being analyzed. In other words, if we select the 2PL model
for a set of item responses, we are assuming that this functional form is cor-
rect, such that the probability of a correct response increases concomitantly
with increases in θi , that items have different values of aj , and that there is
no possibility of obtaining a correct answer due to chance.
10  Applied Psychometrics Using SPSS and AMOS

Polytomous Items
In many contexts, items on an instrument are not scored dichotomous-
ly, but rather can take one of several values. Examples in achievement test-
ing include graded responses where scores may include categories such as
poor, fair, good, and excellent. For many affective measures in psychology,
such as personality assessments, the item responses are on a Likert Scale
taking the values 1, 2, 3, 4, or 5, for example. In all of these cases, the IRT
models described previously would not be appropriate for modeling item
responses because they only account for dichotomous outcomes. A number
of more general models have been developed, however, which do accom-
modate such polytomous items. One such model, the generalized partial
credit model (GPCM), is analogous to the 2PL model in that it models
item responses using item discrimination and location parameters, and an
individual’s propensity on the measured latent trait (Muraki, 1992). The
GPCM takes the form:
k
Σh =j 1a j (θi −b jh )
e
P (X jk θi , a j , b jk ) = m Σch =1a j (θi −b jh )
, (1.7)
Σc =j1e

where θi and aj are the latent trait and discrimination parameters, as de-
fined previously. The parameter bjh is a threshold that reflects the level of
the latent trait necessary for an individual to go to item response h from
response h – 1. The value of mj represents the number of categories (e.g., 5)
possible for item j. As an aside, this suggests that the items can have dif-
ferent numbers of categories. If we assume that the values of aj = 1 for all
items, then we have the partial credit model, which is simply the Rasch ver-
sion of the GPCM.
As an example of how the GPCM works, imagine for a specific item that
we have four categories (0, 1, 2, 3) from which an individual can endorse.
Further, assume that for Item 1, the threshold values are –1, 0, and 1. This
would mean that individual respondents with θi values below –1 are most
likely to provide an answer of 1 to the item. Similarly, those with –1 ≤ θi < 0
have the highest probability of yielding a response of 2, while those with
0 ≤ θi < 1 have the highest probability of a 3 response. Finally, individuals
with θi ≥ 1 have the greatest probability of producing a value of 3 on the
item. The ICC for this item appears below in Figure 1.4. While for dichoto-
mous items there was only one curve linking θi and the item response prob-
abilities, for polytomous items there are separate curves associated with
each of the response options. Using these curves, we can discern the prob-
ability that an individual with a given level of the latent trait will produce a
 Introduction to Psychometric Concepts   11

Figure 1.4 ICC for item based on Graded Partial Credit Model.

particular response to the item. For example, respondents with θi < –2 are
most likely to respond with a 0, while those with θi = –0.5 will most likely
respond with a 1. It is also clear from viewing these curves that individuals
with, for example, high values of θi could still provide a response of 0 or 1,
but the probability for such is quite low.
An alternative to the GPCM for polytomous data is the graded response
model (GRM) developed by Samejima (1969). Whereas the GPCM com-
pares the likelihood of an individual responding in two adjacent categories
(h versus h – 1), the GRM focuses on the probability of responding in cat-
egory h or higher versus categories less than h. In other words, rather than
focusing as the GPCM on a series of dichotomous choices between adjacent
categories, the GRM instead compares the probability of one category or
higher versus all lower categories. The GRM is expressed as

a (θ −b )
e j i Xj
P (X jk or higher) = a (θ −b ) . (1.8)
1+ e j i X j

Once again, aj and θi are as defined previously. In this case, bX j is the boundary
point between a value on the item of h or higher, versus all responses below h.
12  Applied Psychometrics Using SPSS and AMOS

To illustrate the GRM, let’s consider the example described above for
the GPCM. We have an item with four possible response options to select
from, with thresholds of –1, 0, and 1. The meaning of these threshold values
is slightly different in the GRM than for the GPCM. In the latter case, the
first threshold marked the border between a response of 1 versus 2, whereas
for the GRM it marks the border between a response of 0 versus 1, 2, or 3.
Thus, a respondent with a θi value of 0.5 is more likely to produce a response
of 3 or 4 versus 1 or 2. In order to obtain probabilities for a specific item
response, we would use the following equation: Pix(θ) = Pix(θ) – Pix+1(θ).
Therefore, to obtain the probability of an individual providing a response
of 2, we would simply calculate Pi 2(θ) = Pi 2(θ) – Pi 3(θ). Figure 1.5 contains
an ICC for the GRM of this item. Its general form is very similar to that of
the GPCM, with a separate curve for each response option. However, no-
tice that the curves are shifted somewhat on the X axis, and the maximum
probabilities associated with specific item responses are somewhat lower
for the GRM. For example, while the maximum probability of producing a
response of 2 is approximately 0.52 in the GPCM, it is only 0.4 for the GRM.
This difference in probabilities is reflective of the different model forms.
In addition to the GPCM and GRM, a third general approach for deal-
ing with ordinal item responses is the rating scale model (RSM). While

Figure 1.5 ICC for item based on Graded Response Model.

 Introduction to Psychometric Concepts   13

conceptually similar to both the GPCM and GRM, the RSM has somewhat
different properties, which can be highlighted by looking at the model
formulation:
x
−Σh =j 0 τh +x j (θi −δ j )
e
P (X j θi , δ j , τ) = x . (1.9)
m −Σh =j 0 τh +x j (θi −δ j )
Σh =j 0e

In this model, τh represents the threshold for item response h, and is on the
scale of the latent variable being measured. Thus, for an ordinal item with 1,
2, 3, 4, and 5 as response options, there would be 4 thresholds (number of
categories minus 1), where τ1 separates a response of 1 from 2, τ2 separates
2 from 3, τ3 separates 3 from 4, and τ4 separates 4 from 5. The value xj cor-
responds to the number of parameters that an individual has passed, based
on their value on the latent trait being measured. The δj parameter indicates
an item’s location, or central point, on the latent trait, around which the
thresholds gather. To see how the RSM works, let’s consider an item that
has 5 response options, where δj = 0.5, and the thresholds are as follows:
τ1 = –0.6, τ2 = –0.1, τ3 = 0.7, and τ4 = 1.2. An individual completing the in-
strument who has θi = 0.54 would pass thresholds 1 and 2, but not 3, leading
them to provide a response of 3 on the item. Now that we are familiar with
CTT and IRT models, let us move to item analysis in Chapter 2 with these
ideas in mind. We also want you to be aware that many of our datasets and
code are available to download at the website for this book (https://labs.
wsu.edu/psychometric/resources/).
 2
Item Analysis

Introduction
This chapter introduces statistical procedures to conduct item analyses
based on classical test theory (CTT). These analyses typically serve as a first
step in the investigation of the psychometric properties of scale scores.
Analyses described in this chapter provide ways to estimate the difficulty
and discrimination of dichotomously (e.g., correct/incorrect) and poly-
tomously (e.g., Disagree, Neutral, Agree) scored questionnaire and test
items. Understanding the characteristics of scale items is a necessary step
for deciding which items will remain, be revised, or excluded from the final
version of an educational or psychological measure (e.g., motivation, math-
ematics achievement).

Classical Test Theory Item Difficulty

CTT Item Difficulty for Dichotomous Items
Item difficulty is an index of how examinees answered an item. For di-
chotomously scored items, CTT defines item difficulty as the proportion of
persons who obtained a correct item response (i.e., proportion passing), or
16  Applied Psychometrics Using SPSS and AMOS

the proportion of those agreeing with items measuring agreement to a state-

ment. In this case, item difficulty can range from 0.00 to 1.00, with values ap-
proaching 0.00 indicating more difficult items and those close to 1.00 con-
sidered somewhat easier items. Within applied testing contexts, it has been
argued that values between 0.30 and 0.70, with a mean of 0.50, provide the
most useful information about examinees’ knowledge or skill level (Allen &
Yen, 1979), if the purpose of the test is to produce a normal distribution of
score estimates of ability levels. For criterion assessments, a more restricted
range is generally desired with a mean difficulty value of 0.80.
SPSS provides two approaches for estimating CTT item difficulty. The
first approach involves calculating the item means, which represents the
proportion of examinees obtaining a correct response. The following exam-
ple estimates the CTT item difficulty of a 20-item measure based on the re-
sponses of 2,000 examinees, using the ex2.sav data. To conduct this analysis,
in the menu bar, select Analyze, followed by Descriptive Statistics and then
Descriptives. We will then obtain the SPSS dialogue box shown in Figure 2.1.

Figure 2.1 SPSS Descriptives dialogue box without items assigned to Variable(s)
window.

As shown in Figure 2.1, the items in the dataset appear in the left-hand
side of the dialogue box. To estimate item difficulty, we simply highlight
the items and click the arrow pointing to the Variable(s) window. Figure 2.2
shows what the dialogue box would look like.
 Item Analysis   17

Figure 2.2 Descriptives dialogue with variables assigned to Variable(s) window.

By default, SPSS will provide us with the following information for each
selected item: Minimum, maximum, standard deviation, and the mean. If
we don’t want to change any of these values, we would simply click OK to
obtain the output reported in Figure 2.3.

Descriptive Statistics
N Minimum Maximum Mean Std. Deviation
i1 2000 0 1 .91 .290
i2 2000 0 1 .61 .489
i3 2000 0 1 .51 .500
i4 2000 0 1 .32 .466
i5 2000 0 1 .36 .479
i6 2000 0 1 .67 .469
i7 2000 0 1 .59 .492
i8 2000 0 1 .51 .500
i9 2000 0 1 .79 .405
i10 2000 0 1 .81 .392
i11 2000 0 1 .54 .499
i12 2000 0 1 .48 .500
i13 2000 0 1 .21 .405
i14 2000 0 1 .84 .369
i15 2000 0 1 .78 .412
i16 2000 0 1 .59 .492
i17 2000 0 1 .59 .492
i18 2000 0 1 .36 .479
i19 2000 0 1 .36 .480
i20 2000 0 1 .35 .476
Valid N (listwise) 2000

Figure 2.3 SPSS Descriptive Statistics output window.

18  Applied Psychometrics Using SPSS and AMOS

The output in Figure 2.3 includes descriptive statistics for each of the 20
items, including the item, sample size (N), minimum and maximum values
for each item, the means (which are the item difficulties), and the standard
deviations (Std. Deviation). The analyses were conducted on 20 items based
on 2000 examinees. As shown, Item 1 was the easiest item, with the majority
of the examinees (0.91) obtaining a correct response. That is, 91% of the
examinees answered the item correctly. In contrast, a relatively small propor-
tion of the examinee group (0.21) obtained a correct response on Item 13.
In general, there is a fairly broad range of item difficulty values across the 20
items, between these two extremes. Such a pattern of item difficulty values
is desired when testing a range of examinees (e.g., age) with varied stand-
ing on the measured trait (e.g., reading achievement). The standard devia-
tion column shows that the examinee distribution was most varied when
an item’s proportion correct approximates 0.50 (Crocker & Algina, 1986).
Finally, the Minimum column indicates that the lowest item score was 0 (in-
correct), whereas the Maximum column reports that the highest item score
was 1 (correct) across the item set. This last information is quite useful when
we first screen the items, as a way of detecting miskeyed entries. If the mini-
mum and maximum values for each item were not 0 and 1, respectively, we
would know that there were one or more typos when the data were entered.
Referring to Figure 2.4, if we wish to change the statistics that are dis-
played in the output, we would simply click on the Options button in the
dialogue box.

Figure 2.4 Using the Options button for selecting which descriptive statistics to
display.

Clicking on Options yields the following window:

 Item Analysis   19

Figure 2.5 Options window for descriptive statistics.

We can see that the SPSS default statistics of Mean, Std. Deviation, Minimum,
and Maximum were all selected. If we only wanted to display the mean, we
would simply unclick the boxes for the other statistics, so that the window
appears as follows:

Figure 2.6 Options window for descriptive statistics with only the Mean selected.
20  Applied Psychometrics Using SPSS and AMOS

By clicking Continue in the options box, and then OK in the Descriptives

box, we would obtain the following output.

Descriptive Statistics
N Mean
i1 2000 .91
i2 2000 .61
i3 2000 .51
i4 2000 .32
i5 2000 .36
i6 2000 .67
i7 2000 .59
i8 2000 .51
i9 2000 .79
i10 2000 .81
i11 2000 .54
i12 2000 .48
i13 2000 .21
i14 2000 .84
i15 2000 .78
i16 2000 .59
i17 2000 .59
i18 2000 .36
i19 2000 .36
i20 2000 .35
Valid N (listwise) 2000

Figure 2.7 Descriptive Statistics output window.

As an alternative to using the mean of dichotomous items, difficulty

estimates can also be obtained using the Frequencies menu option under
Analyze and Descriptive Statistics. This approach produces frequency ta-
bles for all variables specified. Specifically, using the previously mentioned
menu sequence, we would obtain the following dialogue box:
 Item Analysis   21

Figure 2.8 Frequencies dialogue box with no items selected.

As with the previous dialogue box examples, our first step is to move the
desired items from the left hand box to the Variable(s) box on the right,
using the arrow button in the middle.

Figure 2.9 Frequencies dialogue box with all items selected.

We will want to leave the Display frequency tables box checked so that the
tables will appear in the output. We can click OK, and obtain the following
individual tables for each item. In order to save space, only tables for the
first 5 items appear below.
22  Applied Psychometrics Using SPSS and AMOS

i1
Frequency Percent Valid Percent Cumulative Percent
Valid 0 185 9.3 9.3 9.3
1 1815 90.8 90.8 100.0
Total 2000 100.0 100.0
i2
Frequency Percent Valid Percent Cumulative Percent
Valid 0 789 39.5 39.5 39.5
1 1211 60.6 60.6 100.0
Total 2000 100.0 100.0
i3
Frequency Percent Valid Percent Cumulative Percent
Valid 0 989 49.5 49.5 49.5
1 1011 50.6 50.6 100.0
Total 2000 100.0 100.0
i4
Frequency Percent Valid Percent Cumulative Percent
Valid 0 1362 68.1 68.1 68.1
1 638 31.9 31.9 100.0
Total 2000 100.0 100.0
i5
Frequency Percent Valid Percent Cumulative Percent
Valid 0 1289 64.5 64.5 64.5
1 711 35.6 35.6 100.0
Total 2000 100.0 100.0

Figure 2.10 Frequency tables for Items 1 through 5.

The output shows the Frequency (number of examinees in the given

item category) and Percent (number of examinees in category/total sam-
ple size) of examinees obtaining correct and incorrect responses, Valid per-
cent (number of examinees in category with no missing data/total sample
size), and the Cumulative Percent. As we can see, 185 of the 2000 examin-
ees (9.3%) answered Item 1 incorrectly, whereas 1815 (90.7%) provided a
correct response, thereby making the item difficulty 0.908. Proportions can
be obtained by dividing the values in the Percent column by 100.
The SPSS Frequencies menu selection offers two additional options that
might be useful for the researcher examining item difficulties. First, it is
possible to obtain the means (and other descriptive statistics) of the item
responses through the Statistics button in the frequencies dialogue box.
 Item Analysis   23

Figure 2.11 Frequencies dialogue box with Item 1 selected, and highlighting the
Statistics button.

Doing so opens the dialogue box shown in Figure 2.12.

Figure 2.12 Statistics dialogue box under Frequencies menu option.

The user is presented with a number of options in terms of statistics that

can be produced for each variable. In this case, we would like to obtain the
mean (item difficulty), and the standard deviation, and therefore will click
24  Applied Psychometrics Using SPSS and AMOS

the boxes next to them. For the sake of brevity, we will only present results
for Item 1, shown in Figure 2.13.

Figure 2.13 Statistics dialogue box under Frequencies menu option with mean
and standard deviation selected.

We then click Continue, followed by OK on the Frequencies dialogue box,

and obtain the following output reported in Figure 2.14.

Frequencies
Statistics
i1
N Valid 2000
Missing 0
Mean .91
Std. Deviation .290

i1
Frequency Percent Valid Percent Cumulative Percent
Valid 0 185 9.3 9.3 9.3
1 1815 90.8 90.8 100.0
Total 2000 100.0 100.0

Figure 2.14 Mean, Standard Deviation, and Frequency table for Item 1.

As shown, this mean matches that obtained using the Descriptives com-
mand, and is an estimate of the CTT item difficulty.
 Item Analysis   25

In addition to the mean (and other descriptive statistics), it is also pos-

sible to obtain graphs representing the distribution of item responses. For
the dichotomous items this will perhaps not be so interesting, but for the
polytomous items to be discussed below, such graphs might prove to be
quite helpful for characterizing the item response patterns. Figure 2.15
shows how to obtain such graphs by again using the menu sequence to pull
up the Frequencies dialogue box.

Figure 2.15 Frequencies dialogue box with Item 1 selected, and highlighting the
Charts button.

When we click on this button, the dialogue shown in Figure 2.16 appears.

Figure 2.16 Charts dialogue box under Frequencies menu option.

26  Applied Psychometrics Using SPSS and AMOS

As shown, we have three choices in terms of the graph type. As an example,

we will select the Bar charts, by clicking on the radio button next to that op-
tion, as demonstrated in Figure 2.17.

Figure 2.17 Charts dialogue box under Frequencies menu option with Bar charts
selected.

Figure 2.18 shows the resulting bar chart reporting the response fre-
quencies for Item 1.

2,000

1,500
Frequency

1,000

500

0
0 1
i1

Figure 2.18 Bar chart for Item 1 with Frequencies on the y-axis.

By default the frequency of item responses appears on the y-axis, with item
responses appearing on the x-axis. By selecting Percentages under Chart
Values (see Figure 2.17), we can place the percentages on the y-axis, shown
in Figure 2.19.
 Item Analysis   27

100

80

Percent 60

40

20

0
0 1
i1

Figure 2.19 Bar chart for Item 1 with Percentages on the y-axis.

CTT Item Difficulty for Polytomous Items

Because item difficulty within CTT is defined as the average perfor-
mance of the examinee group on a particular item, means obtained through
the Descriptive Statistics menu sequence described above can be used to
estimate the item difficulty for polytomously scored items. The following
example involves a scale consisting of 20 polytomous items that are scored
on a 5 point Likert Scale, measuring the extent to which respondents dis-
agree/agree (e.g., 1 = Strongly Disagree, 2 = Disagree, 3 = Neutral, 4 = Agree,
5 = Strongly Agree) with a series of survey statements. The instrument was
administered to 1,533 respondents, and the resulting data are contained in
a file called poly1.sav.
In order to obtain the means for the items, we can use the same se-
quence of menu options that appears in Figures 2.1 and 2.2. The resulting
output appears in Figure 2.20.
28  Applied Psychometrics Using SPSS and AMOS

Descriptive Statistics
N Minimum Maximum Mean Std. Deviation
V1 1533 1 5 4.12 .550
V2 1533 1 5 3.75 .686
V3 1533 2 5 4.00 .573
V4 1533 1 5 3.99 .608
V5 1533 1 5 4.00 .579
V6 1533 1 5 4.06 .545
V7 1533 1 5 3.95 .665
V8 1533 1 5 3.89 .636
V9 1533 1 5 3.90 .622
V10 1533 1 5 3.80 .678
V11 1533 1 5 3.94 .616
V12 1533 2 5 4.06 .586
V13 1533 1 5 3.94 .675
V14 1533 1 5 3.80 .743
V15 1533 2 5 4.04 .593
V16 1533 1 5 3.79 .741
V17 1533 2 5 3.86 .696
V18 1533 1 5 3.60 .776
V19 1533 1 5 4.01 .603
V20 1533 1 5 3.85 .695
Valid N (listwise) 1533

Figure 2.20 Descriptive Statistics output window.

The previous output shows that overall, respondents generally agreed with
each scale item. That is, all items had a mean value above 3.5. Item de-
scriptive statistics also suggest that respondents were most varied in their
responses to Items 18 (SD = .776), 14 (SD = 0.743), and 16 (SD = 0.741).
Furthermore, the range of observed item responses was between Strongly
Disagree (1) and Strongly Agree (5). Note that a few items (i.e., 3, 12, 15, 17)
had a restriction of range. That is, no one selected option 1.
To gain insights into the distribution of all responses to each item, the
Frequencies menu sequence outlined above can be used to obtain the fre-
quency of responses to each option for each item. Such output will help the
researcher understand not only the relative difficulty of each item, but also
the distribution of responses. For brevity, Figure 2.21 provides the output
only for Item 1.
 Item Analysis   29

V1
Frequency Percent Valid Percent Cumulative Percent
Valid 1 1 .1 .1 .1
2 9 .6 .6 .7
3 120 7.8 7.8 8.5
4 1084 70.7 70.7 79.2
5 319 20.8 20.8 100.0
Total 1533 100.0 100.0

Figure 2.21 Frequency of responses to Item 1.

Results in Figure 2.21 show that the majority of individuals in the sample
responded to the item with a 4, and that 91.5% gave either a 4 or a 5, in-
dicating high agreement with the item. Only 10 individuals, or 0.7% indi-
cated that they disagreed or strongly disagreed with the item. We can also
examine the bar chart for the responses to this (or any of the other) item
using the command sequence in Figures 2.15 through 2.17. In this case, we
elected to place the Percent on the y-axis (Figure 2.22).

80

60
Percent

40

20

0
1 2 3 4 5
V1

Figure 2.22 Bar chart for Polytomous Item 1 with Percentages on the y-axis.

Classical Test Theory Item Discrimination

Item discrimination refers to the degree to which an item functions to differ-
entiate respondents with relatively higher levels of the trait being measured
by the scale from those with lower trait levels. Therefore, understanding the
items’ discriminatory power is critical to developing a useful measure. The
underlying notion of item discrimination is that a larger proportion of high
scoring (high total score) persons should respond correctly or more highly
endorse the item in comparison to persons with lower total scores on the
30  Applied Psychometrics Using SPSS and AMOS

scale. There are several ways to estimate item discrimination and we discuss
a few here including the extreme groups calculation method, biserial and
point biserial correlations, and Cronbach’s coefficient alpha. We do not
show how to calculate the biserial correlation in SPSS because it was not an
option at the time of this printing. This section provides the SPSS steps to
calculate item discrimination using several methods. A detailed discussion
of coefficient alpha will appear in Chapter 3, including more discussion on
item discrimination.

Extreme Groups Calculation Method

The extreme groups calculation method, or the index of discrimination
(Crocker & Algina, 1986), is used with dichotomously scored items. It is
calculated as the difference between the proportion of the highest scoring
(upper) and lowest scoring (lower) groups of examinees obtaining a cor-
rect item response. Estimation thus requires first establishing the upper
and lower examinee groups, typically based on the total test score. Discrimi-
nation values can range from –1.00 to 1.00, with positive values indicat-
ing items that favor the upper scoring group and negative values showing
items that favored the lower scoring group. Items are considered to be per-
forming well when they have relatively large positive discrimination values,
meaning a larger portion of the high scoring group responded correctly
compared to the low scoring group.
In order to calculate the extreme groups discrimination index, we first
need to calculate the total score on the instrument. We’ll go back to the
20 dichotomous item test that was discussed earlier in this chapter. To cal-
culate the total score, we would use the Transform and Compute Variable
menu options, shown in Figure 2.23.

Figure 2.23 SPSS menu bar.

 Item Analysis   31

The dialogue box shown in Figure 2.24 will subsequently appear.

Figure 2.24 Compute Variable dialogue box.

In the Target Variable slot, we assign a name for the new variable that will ap-
pear in our dataset. In this example, we call the new variable “score.” Next,
in the numeric expression box, we will sum the items using the statements
provided in Figure 2.25. When we are done, the window will look as follows.
32  Applied Psychometrics Using SPSS and AMOS

Figure 2.25 Compute Variable dialogue box for computing score variable.

When we click OK, a new column containing the score will be added at the
end of our data set.
In order to calculate the extreme groups discrimination values for each
item, we will need to identify the upper and lower scoring groups. In this
case, we will use the top 25% and the bottom 25% of examinees, based on
the total test score. First, we need to identify the 25th and 75th percentile
values. We can do this going through Analyze ► Descriptive Statistics ► Ex-
plore menu sequence, which yields the dialogue box in Figure 2.26.
 Item Analysis   33

Figure 2.26 Explore dialogue box.

As shown in Figure 2.27, we will need to move the score variable to the De-
pendent List window.

Figure 2.27 Explore dialogue box for the score variable.

We then click Statistics and get the window shown in Figure 2.28, in which
we will click Percentiles.
34  Applied Psychometrics Using SPSS and AMOS

Figure 2.28 Explore: Statistics dialogue box for selecting percentiles.

This will give us the percentiles that we need in order to identify the 25th
and 75th percentile values. By clicking Continue and then OK in the previ-
ous window, we obtain the output reported in Figure 2.29. Of particular
interest to us in the current example is the following table.

Percentiles
Percentiles
5 10 25 50 75 90 95
Weighted Average score 5.0000 6.0000 9.0000 11.0000 14.0000 16.0000 18.0000
(Definition 1)
Tukey’s Hinges score 9.0000 11.0000 14.0000

Figure 2.29 Percentiles output from the explore command.

From this output, we can see that the score at the 25th percentile is 9, and
the score at the 75th percentile is 14. Thus, the bottom 25% group consists
of those with scores of 9 or less, and the upper 25% group includes those
with scores of 14 or more.
We now will use these percentiles to create a new variable that catego-
rizes examinees into one of the three groups. This is done using the Trans-
form ► Recode Into Different Variables menu sequence, in order to obtain
the following window (Figure 2.30).
 Item Analysis   35

Figure 2.30 Recode Into Different Variables dialogue box.

We can then recode the variable score into a variable consisting of 3 sepa-
rate groups, those with scores of 9 or below, those with scores between 10
and 13, and those with scores of 14 or more. First, as shown in Figure 2.31,
we move the score variable into the Numeric Variable -> Output Variable
box, and provide the name of the new variable, along with any variable
label that we would like.

Figure 2.31 Recode Into Different Variables dialogue box for score.
36  Applied Psychometrics Using SPSS and AMOS

By clicking the Change button, we will then obtain the window in Figure 2.32.

Figure 2.32 Recode Into Different Variables dialogue box for score with group
variable.

Next, we click the Old and New Values button, which yields the window
shown in Figure 2.33.

Figure 2.33 Recode Into Different Variables: Old and New Values dialogue box.
 Item Analysis   37

In this window, we will want to indicate which values on the score corre-
spond to the grouping category. For instance, we will label the lowest 25%
group with a 1. To do this, we will click the button next to Range, LOWEST
through value. In the box associated with this button, we will put the 25th
percentile value of 9. In the upper right hand corner of the box (identified
by the red arrow) we will put the group identifier value of 1. We will then
click the Add button (Indicated by the blue button) next to the Old –> New
box in order to create the recoded variable. The window in Figure 2.34 will
then appear.

Figure 2.34 Recode Into Different Variables: Old and New Values dialogue box
featuring the Range button.

We will repeat this sequence by clicking the button next to Range, placing
10 and 13 in the accompanying boxes, typing a 2 in the Value box, under
New Value, and clicking Add. Finally, we will follow the same sequence for
the upper group by clicking on the Range, value through HIGHEST button,
typing 14 in the accompanying box, putting 3 in the box next to Value un-
der New Value, and clicking Add. The window in Figure 2.35 should then
appear as follows.
38  Applied Psychometrics Using SPSS and AMOS

Figure 2.35 Recode Into Different Variables: Old and New Values dialogue box
with all group categories included.

When we click Continue, followed by OK, a new variable will be added at the
end of the dataset.
We are now ready to obtain the proportions that we need to calculate
the extreme groups item discrimination values for each item. This can be
done by splitting the file using Data ► Split File menu sequence. The win-
dow in Figure 2.36 will then appear.

Figure 2.36 Split File dialogue box

 Item Analysis   39

We will then click on the Organize output by groups button, and place the
grouping variable in the Groups Based on window, as shown in Figure 2.37.

Figure 2.37 Split File dialogue box with Grouping Variable selected.

We then click OK, after which we will rerun the analysis described in Figures
2.6 and 2.7. The resulting output in Figure 2.38 will appear.

Descriptives
Grouping variable = 1.00 Grouping variable = 2.00 Grouping variable = 3.00
Descriptive Statistics Descriptive Statistics Descriptive Statistics
N Mean N Mean N Mean
i1 677 .76 i1 804 .97 i1 519 1.00
i2 677 .38 i2 804 .64 i2 519 .86
i3 677 .22 i3 804 .53 i3 519 .83
i4 677 .21 i4 804 .29 i4 519 .51
i5 677 .20 i5 804 .27 i5 519 .69
i6 677 .39 i6 804 .75 i6 519 .93
i7 677 .31 i7 804 .64 i7 519 .89
i8 677 .25 i8 804 .52 i8 519 .85
i9 677 .62 i9 804 .85 i9 519 .94
i10 677 .59 i10 804 .89 i10 519 .98
i11 677 .37 i11 804 .52 i11 519 .78
i12 677 .27 i12 804 .45 i12 519 .78
i13 677 .11 i13 804 .16 i13 519 .40
i14 677 .68 i14 804 .88 i14 519 .98
i15 677 .55 i15 804 .86 i15 519 .97
i16 677 .36 i16 804 .62 i16 519 .85
i17 677 .37 i17 804 .61 i17 519 .84
i18 677 .21 i18 804 .34 i18 519 .57
i19 677 .17 i19 804 .32 i19 519 .67
i20 677 .18 i20 804 .32 i20 519 .62
Valid N (listwise) 677 Valid N (listwise) 804 Valid N (listwise) 519

Figure 2.38 Proportion of correct responses for items, organized by group.

 Another Random Document on
Scribd Without Any Related Topics
of burnt terra Siena, but some places take a dingier tint, perhaps
from the effect of the explosion, while the north side is partially
varied with the sulphur yellow of the Lichen candelarius. The
western end, which, though the part opposite to the entrance of the
Acropolis, is in fact the back of the temple, represented the contest
of Neptune and Minerva; and the relative position of the deities, and
of the benefits they offered to the Athenians, is well explained by
Wilkins by reference to a medal. Behind the statues the tympanum is
faced by upright slabs of marble, some of which are hollowed out at
the back, I do not know why, and you might, perhaps, creep behind
the two figures which remain, which are those said to be of Hadrian
and Sabina; but though all these statues were finished all round, as
you know, by having seen them in London, there was certainly no
provision made for a close inspection after they were up in their
places.
These two remaining statues have lost their heads, and
consequently all that entitled them to their appellation, and the
remaining fragment of cornice has fallen from its first position, and
rests upon the figures. The metopes of this front are almost
obliterated, but the sculpture of the inner frieze is exquisite, perhaps
the finest piece of the whole circuit. Sometimes when this is lighted
by the declining sun, the effect is inconceivably beautiful; but this
effect never could have had place in the perfect state of the building,
because it is only owing to the want of ceiling and roof in the
peristyle, that it is ever illuminated by the direct light of the sky.
You would imagine from the way in which Wilkins endeavours to
correct an error which Wheeler never made, that the columns of this
temple were in five pieces, but this is not true; the number of drums
or frusta is eleven or twelve, but the junction in those which have
suffered no violence is so beautifully fine, that it is frequently
impossible to trace it, even on the smooth surface of white marble.
The inner range of columns is more shattered than the outer, and
here we begin to be sensible of the immediate effects of the
explosion occasioned by the bombs of the Venetians in 1687. The
space between the internal columns, forming the pronaos, appears
to have been filled up with a metal grating resting on a sill of
marble. The holes for fastening it still remain, and the situation of
the plinth, on which this grating rested, is indicated by the parts cut
away to receive it at the foot of the columns. The doorway is 16 feet
8 inches wide in the original work, but it has been contracted by
blocks of marble, at some more recent period; probably when the
edifice was converted into a Greek church, and this has misled
Stuart. Besides the evidence of this being a posterior contraction,
arising from the inferior nature of the work, and its total want of
connexion with the wall of the cell, we may observe also that some
of the stones used are covered with inscriptions. No ornamental
architrave remains to the door, but there are vestiges whence we
may conclude that this part was of metal, as in the Propylæa.
When you have once entered the building you see the manner in
which it has suffered. The powder must have been near the middle;
and on each side, both the wall of the cell, and the columns of the
peristyle have been thrown down. At the eastern end, the walls and
the inner range of columns are destroyed. Perhaps towards the
west, the walls still existed between the ancient temple and the
opisthodomus, and protected in some degree the western end; but
that wall is now entirely destroyed, and the surface of the wall of the
cell everywhere shattered, exposing a raw white fracture.
But we are not yet fairly within the temple; we must first consider
the Opisthodomus. It is paved with slabs of white marble 8 inches
thick, 5 feet 8 inches long, and 3 feet 11 inches wide, all bedded in
rough masonry, or on the native rock. There are four square blocks
of a larger size, being 5 feet 10 inches square, and 14 inches thick;
placed in corresponding situations, so as to divide the area of the
room into nine equal parts, and appearing to have received the
columns which supported the roof. Stuart has imagined six such
columns, but without sufficient reason.
We may distinguish on the pavement marks of the openings of the
doors, consisting of grooves forming portions of circles, which have
probably received plates of metal. The centre from which these
segments are described, is, as has been already mentioned, clear of
the thickness of the walls. It is doubtful if there were originally any
opening between this chamber and the cell of the temple. If such
existed, it was probably a small side door, but a large central
doorway doubtless existed here while the building was occupied as a
church. Stuart has remarked a sinking round the larger space which
formed the hypæthral cell of the temple, at the distance of 15 feet
from the walls; but he has not noticed some peculiarities in the
pavement, which are of considerable importance in understanding
the construction and arrangement of the building. There is first a
pavement of marble slabs, 8 inches thick, as in the opisthodomus,
10 feet 3 inches and a half wide at the end, and not quite so much
at the side: next to this is a course 4 feet wide, and 14 inches thick,
doubtless intended to receive the columns of the internal peristyle.
At one angle of the parallelogram, formed by this series of slabs,
there are faint traces of the position of a column about 3 feet 9
inches in diameter, and as these blocks are regularly spaced, we may
ascertain that the distance from centre to centre of the columns was
about 11 feet 6 inches, and consequently the space between them
was 7 feet 9 inches, while in the other intervals, the space from
centre to centre was 12 feet 11 inches, and the clear
intercolumniation about 9 feet 2 inches. The ancients always placed
the internal columns, and those of their courts, proportionally farther
apart than the external ones. At Pæstum, where the outer columns
are only one and a tenth diameter apart, the inner are nearly one
and two fifths; at Egina, those of the external peristyle are about
one diameter and two thirds asunder, those of the internal, two and
one third. In the Propylæa, the spaces between the external
columns seem to be about one diameter and a half, if we exclude
the large middle entrance, the internal about two and a half. In the
latter case, the internal columns were Ionic, and it is very possible
that this was also the case in the Parthenon, but we have not the
slightest fragment remaining; a circumstance which seems to me
very remarkable. Was no use made of them in the church? and if so,
why were they rejected?
 The lines of the small columns traced by Stuart, are sufficiently
evident, but they were probably parts of the church, not of the
temple; for besides that their diminutive size renders them
unsuitable for the internal structure of such a building, and that their
remaining fragments are of very indifferent workmanship, their
situation does not correspond with the division of the marble slabs,
which in an edifice so regularly and systematically constructed, is
alone a sufficient reason for rejecting them. It may seem remarkable
that the original columns should not have left on the pavement more
distinct traces of their existence; but while the Greeks united with
wooden blocks, and sometimes by other means, the different
portions of the column, they placed the lowest part, without
preparation, on the smooth pavement. In some instances, in the
temple of Theseus, less than half the column remains, and not the
slightest indication is visible on the pavement to shew that the other
half ever existed. Within the sinking of the pavement already
mentioned, we find again the eight-inch marble paving, but in one
part slabs of limestone occupy its place. These are supposed to have
supported the statue of the goddess. Mr. B. very justly observed,
that where every thing is thus of white marble, a statue of that
material would have wanted its just consequence. It was necessary
to employ a more expensive substance to give it sufficient relief in
the imagination.
A considerable portion of the paving remains in its place, but some
has been removed, and in the opisthodomus a large slab of marble
moulded on the edge, appeared underneath the pavement, which
must have been buried at the time of the erection of the temple.
Before leaving the inside, I must conduct you up a staircase, which
the Turks have made in a little tower at the south-west angle of the
building. We may by its means observe many particulars, which are
not so easily discernible from below. The examination of the
Athenian edifices leads to the conclusion, that the white Pentelic
marble was obtainable in slabs of considerable length and breadth,
but that the thickness was more limited. In the walls of the cell of
the Parthenon, the courses are of 1 foot 8 inches and two thirds. In
the Erectheum and the Propylæa, they are somewhat less. When a
greater width is required, either to form a plinth course, or to cover
the opening of a doorway, the slab is uniformly set on edge.
In the frusta of the columns, the stones are thicker; one in the
Parthenon being 3 feet 7.8 inches in thickness; but probably that
part of the quarry which would furnish blocks of such a thickness,
did not afford them of considerable length. The architrave of the
Parthenon is 5 feet 9 inches wide, and is got out in three
thicknesses, probably on account of the difficulty of procuring pieces
14 feet long, and 2 feet 10 inches and a half thick, which would have
been required, had it been constructed, as is more usually the case,
in two blocks. In the frieze, there was not the same necessity that
the work should be perfectly solid, and accordingly only two beds
were used, leaving a vacancy of about 10 inches in the middle. Thus
this peculiarity which Clarke discovered to be owing to the
dishonesty of the workmen, and Wilkins considers a proof of great
science and foresight, may with greater probability be attributed to
the nature of the quarry.
The ceiling of the pronaos and posticum was formed by marble
joists supporting slabs of the same material, which contained the
lacunariæ; but in the lateral peristyle, these slabs were laid from the
wall to the external epistylium, without the intervention of joists.
You may see here in several places marks of ancient ornamental
painting, and in some instances of painting of two different styles
and dates, one of which has been over the other. In parts sheltered
from the weather, there are even indications of painting on the
outside cornice. There are some very curious contrivances in the
construction of the cornice, at the south-west angle, with which I
will not at present trouble you, as I do not think them very good.
I have already explained to you the system of covering in these
temples; the ornamental tiles are disposed in the Theseum, one over
each triglyph; in the Parthenon there was also one over each
metope; in the first building therefore, they were a little more than 4
feet apart, in the latter about 3 feet 6 inches. For some unknown
reason, the tiles themselves do not correspond with this disposition
of the ornaments; the latter are therefore entirely detached from the
former; and the real stops of the convex tiles are small triangular
projections, placed back, so as not to be seen from below; three of
them occupying the dimensions of two antefixæ and the intervening
space. Vast heaps of ruins lie around, as you may suppose, but on
the south side, part of a continued basement of coarse limestone
may be seen under the three steps. Three courses of this limestone
are exposed, and part of the fourth; each about 1 foot 9 inches in
height. The uppermost projects 4 feet from the step, and the upper
surface is sunk about 5 inches, probably to receive a covering of
marble: the next course projects 18 inches, the two lowest each
about half an inch. I cannot tell how far this basement extends, or
what its whole height may be; but in front there is an appearance of
vaults, which puzzle me much more, as they seem to imply a very
considerable elevation of the front of the temple above the ground
on which it stood. I only perceived them by some holes broken in at
the summit, and they are too much filled up with rubbish to enable
us to trace their extent; perhaps they were only partial hollows, filled
up in ancient times as well as in modern.
On the north side, the heaps are perhaps still greater than on the
south, and Fauvel told us that the earth had not been opened, and
consequently, as none of the friezes on this side remained on the
building, it was probable that a great quantity of sculpture would be
found. On speaking subsequently to Lusieri about this circumstance,
he assured me that he had begun to dig in that part, but the Turks
informed him that limekilns had been erected there, in which all the
sculpture that could be found had been burnt by preference. He
continued however his excavations, till finding the limekilns, without
the occurrence of any bas-reliefs, he at length gave up the search. It
is amusing to hear how uniformly Fauvel and Lusieri disagree in their
statements: the latter strenuously denies that he has done any thing
to injure any building, except in the one instance of removing the
Cariatic columns; while the former accuses him of having occasioned
a great deal of ruin in his operations to remove the sculpture, and
even of wantonly destroying objects where no advantage was to be
derived. The last charge is improbable, and in some degree
invalidates the former. Indeed it is said on the other hand, that
Fauvel endeavoured to obtain these spoils for his own government,
and that he now exaggerates the evil from disappointment; but I am
afraid considerable mischief has been done. With regard to the
sculptures taken away, it must be observed that they were most of
them very much exposed, that the young Turks are eager enough to
break off whatever they can reach, in hopes of disposing of them to
the Franks, and that the petty officers and sailors, who either in
merchant ships, or vessels of war, visit these shores, have a great
propensity to break off fragments as memorials. Thousands of
broken pieces, evidently from the building, and which they might
almost fit on to it, lie about, but nothing will satisfy them, unless
they perform their share of mutilation; and repeatedly, on my visits
to the Acropolis, I have beheld with sorrow new fractures on the
drapery of the Caryatides, the only objects now within their reach.
Less remains of the eastern end, or front of the temple, than of
the west. We see there holes which are said to have afforded the
means of support to certain shields which were suspended on the
architrave, whose history is not very clear; and there is occasionally
a slight green circular stain, produced by the edges of the shields
themselves. They were about 3 feet 6 inches in diameter. There was
one under each metope, and under the drops of each triglyph are
the holes by which the bronze letters of an inscription were fixed. I
made a copy of these, but I am afraid it will be impossible to restore
the letters.
Having despatched the Propylæa and the Parthenon, I have now
to give you some account of the triple temple of Erectheus, Minerva
Polias, and the nymph Pandrosus. And first of the hexastyle temple,
which is that of Erectheus, and which is sometimes also called the
temple of Neptune, because it contained an altar to that deity, and
because here was said to be the well of salt water which sprung up
at the contest between Neptune and Minerva; but Pausanias, who
calls the whole edifice a double building, διπλοῦν ὄικημα, does not
specify in which part of it this well was found.
 Stuart in his view represents six columns as erect in their places,
with the continued architrave and frieze; there are now only five.
The architrave over these remains, but the frieze is gone, except two
pieces of the black marble, one of which is misplaced; the other
column was taken away by lord Elgin, and is now in the British
Museum. These columns are very much damaged in the lower part,
so that it is difficult to determine their exact diameter. The base is
not well given in Stuart. They have had a very extended apophysis,
like that of Jupiter Stator at Rome, and I am persuaded that Stuart
has not made them too small. They are therefore very nearly, or
perhaps quite ten diameters in height, and they do not appear too
slender; but you will recollect that they have a very large and
ornamental capital, which certainly takes off from the appearance of
height, and communicates a character, which like that of the
Corinthian order, is in harmony with more delicate proportions. The
contrast between the plainness of the frieze, and the richness of
decoration obtained in the flank, by the continuation of the
ornaments of the capitals, must have struck you as an inconsistency,
but in the original, the frieze is of black marble, and the cramp holes
on the surface shew it to have been enriched, probably with figures
of gilt bronze; so that the whole together possessed in the highest
degree that air of gaiety and splendour, which seems so congenial
both to the character and taste of the Athenians. This black marble,
[29]
having lost its polish, and being stained by the weather, and
perhaps by lichens, has changed its hue into a dull gray, and has
been passed over, without notice of the material. I believe Mr.
Bedford was the first to remark it; but in mentioning the name of the
discoverer in this instance, I do not mean to claim merit in other
circumstances, when I do not name the source of my information.
The fact is, that I rarely know to whom the honour ought to be
attributed: some things have been pointed out to me by Fauvel,
others by Lusieri, or by other persons, and I have combined them
unconsciously with Sharp’s observations and my own; but I believe
the man who has done most towards the complete elucidation of
these antiquities is Baron Haller, a German, who unfortunately for
me, and for all lovers of the fine arts, died last year.
 Passing to the inside of the Erectheum, we find all the pavement
of the cell removed. The walls for the most part are formed of single
blocks, occupying its whole thickness; cramped together, both
horizontally and vertically, but the sort of dado-course which you
may observe above the base, in the flank elevation given by Stuart,
is, or was, in two thicknesses set on edge. The inner slabs have
been taken away, so that it is wonderful the wall stands. The parts
intended to be exposed both inside and outside, are of Pentelic
marble, but those which were entirely hid are of Magnesian
limestone, and this peculiarity enables us to determine that the
pavement of the cell was lower than that of the portico, for not only
the course, which rising above the pavement as high as the top of
the base moulding, might be expected to be so, is of marble; but
there is another course of marble below it, and again, as you
approach the cross wall, marked (a) on Stuart’s plan, there is a third
course of marble under the dado. This cross wall does not rise to the
height of the outside pavement, and there are no traces of its
junction with the side walls above, while on the contrary, though the
cross wall (b) is entirely destroyed, and the lower parts of the side
walls where this would have joined, are smooth and even up to the
height of four courses above what I have called the dado-course, yet
above that, the interrupted surface shows the insertion of the bond-
stones, and attests the existence of such a wall. It is probable that
the lower part was open, and perhaps there were pilasters of metal,
and six columns, either of metal or marble in the opening. This back
part is so filled up with rubbish, that I could only determine one
course lower than those in the first cell, to be of marble. In the
north wall there is a row of small holes, very neatly bored; there are
several such ranges both here, and in the Propylæa, but I do not
know their use.
The greatest peculiarities of this building exist in the wall at the
western end. I will say nothing about the windows and pilasters, of
which we have no other Greek examples, nor of a recess in the
upper part at the south-west angle, taken out of the thickness of the
walls, which is quite unaccountable, but direct your attention to the
basement. In this we find a doorway under one of the internal
pilasters, and consequently also under one of the external
semicolumns. M. Fauvel contends, that this opening is not coeval
with the building, but has been broken through for some purpose of
later convenience; and Stuart, by altogether omitting it, seems to
have been of the same opinion. It is however of the original
construction, for the face of the work within the opening
corresponds with that of the face of the building, without any marks
of the stones having been cut, or in any way altered, and there are
no marks of cramps, which since the walls are cramped, as above-
mentioned, in both directions, must inevitably have been visible had
there originally been no opening. These seem to me pretty strong
proofs, but there is another perhaps still more convincing. I have
already mentioned the method adopted in these edifices to obtain
strength over the openings, by setting the blocks on edge, and
making the course double the height of the adjoining courses; this is
the case here; a large block, equal to two courses, occurs above the
opening. A door in such a situation appears remarkable, but an
opening under the angle of the building at the very point of its
junction with the Pandroseum, is still more extraordinary: this
requires greater strength, and accordingly is covered by a great
stone occupying three courses; 14 feet 6 inches and three quarters
long, 4 feet 9 inches and three quarters high, and 2 feet 2 inches
and three quarters thick. The top of the door marked by Stuart, as
giving an entrance to the Pandroseum, is immediately under the
surbase moulding of the gallery, 2 feet 3 inches and a quarter above
the pavement of the portico of the Erectheum; that of the door
under the gallery is two courses lower, or 3 feet 2 inches and a half,
while the door under the angle is yet one course lower: all this
contrivance may perhaps have had some reference to the access to
the spring of salt water; but without considerable excavations the
problem cannot be solved.
I have put together what I had to say of this cell, whether, in fact,
belonging to one, two, or three temples; and I will now conduct you
to the tetrastyle portico, supposed to be that of Minerva Polias,
which, to say the truth, is rather awkwardly joined to the first; at
least the junction at the posterior angle seems to be ill managed.
This is unfortunately a powder magazine, and the intercolumniations
are consequently filled up with walls of rubble, so that the
proportions of the columns cannot be observed, but for some reason
every body at once prefers this to the hexastyle portico; and the
capitals of the columns, though almost buried in the coarse masonry,
are universally admired. As the difference between these, and those
of the Erectheum, is so small as to be passed over as a trifle of no
consequence in an examination of the prints, it would be desirable to
find out from what the difference of effect arises, but I have not
been able to determine the question. The bases and lower parts of
the columns are buried, but one of them is accessible by digging,
and it appears to have been ornamented with inserted pieces of
coloured glass. I apprehend that the plain fillet under the necking of
the capital in this temple, and the flat eye of the capital in both,
have been finished either with ornaments of gilt bronze, or of
coloured glass, or stones. The angular volute, in this example, is
made so thin, as to be very sensibly translucent; we did not gain
admittance to the inside, for in order to avoid the danger of
explosions, the Turks have walled up the opening, and are obliged to
make a hole in the wall when they want any powder. I have very
little to say about the Pandroseum; the opening I have noticed at
the angle of its junction with the gallery, indicates a considerable
depth to the internal pavement. Lusieri says, that he has dug, and
that there are certainly no traces of a wall or spring; but it is not at
all clear that it is to be sought for in this part, nor am I satisfied that
he dug deep enough. Externally, a wall comes against the pedestal
of the caryatides, and the parts behind it are less finished than those
in front, but no conclusion can be drawn from this, since the building
was not entirely completed; some of the pilaster capitals are only
partially carved, and in the back front of the gallery the last finish of
the walls is not carried down to the ground. The little pateræ in the
architrave of the Pandroseum were probably intended to be carved,
if they were not rather enriched with flowers of gilt metal.
 All the three fine buildings of the Acropolis have been used in
succession as powder magazines, and have suffered from
explosions.
 LETTER XLVIII.

OTHER ANTIQUITIES OF ATHENS.

Athens, March, 1818.

It is fortunate for us that we have more remains of the two cities
where architecture was carried to the greatest perfection, than of
any other. The immensity of Rome, and the vast multitude of public
buildings which adorned it, might lead one to expect that we should
meet there with more remains than elsewhere, but Athens never
was a very large city, nor do the public buildings in it appear to have
been constructed on a larger scale than in many others. In each of
these cities we probably see the remains of some of the finest
examples. Judging from the fragments found at Rome, we may
pronounce that there were many other buildings of great beauty, but
none which we could wish to exchange for the temple of Mars Ultor,
of Jupiter Stator, of Jupiter Tonans, of Antoninus and Faustina, or for
the portico of the Pantheon. The temple of Jupiter Capitolinus was
indeed larger than any of these, but we may doubt if its architecture
were better, or even so good. Here there are fewer objects to
distract the attention, and we may be assured that the Parthenon,
the Erectheum, the Propylæa, and the temple of Theseus, were the
principal objects of beauty in the time of Athenian splendour, and if
there were others which rivalled, there were none which pretended
to surpass them. That we have remains of the best edifices of
Hadrian and Herodes Atticus, is not quite so certain, but we know
from several examples that architecture had fallen, at that period,
from the dignity and purity which it possessed in the time of Pericles.
 I have, on a former occasion, attempted to explain what were the
peculiarities of situation which gave effect to the Roman buildings.
They occupied in many instances points of land advancing into the
general line of the valley of the Tiber. The spectator was in the
centre, the objects were round him. In Athens it was exactly the
reverse; the objects were grouped together on a hill in the centre,
which displayed its magnificence on every side. At Rome, the
beholder was dazzled by the multiplicity of objects. At Athens he was
impressed by their simplicity and unity, for, from every point of view,
the public edifices which crowded the summit, and were disposed on
the slopes of the Acropolis, would combine to form a single whole.
In both nature and art seemed to have united to produce an
harmonious effect, and even in the style of architecture, the richness
and grandeur of the Roman, and the grace and elegance of the
Athenian, seem alike suited to the disposition of the buildings, and
the stations they occupied. It is remarkable, that both these cities
should have been so admirably placed. Paris hardly offers a single
marked situation; Naples would have been better, had some of its
principal edifices occupied the Chiatamone; Milan is on a flat;
Florence merely in a fine valley. London would be preferable to any
of these for the display of architecture, but we have taken no
advantage of the steep bank rising from the Thames, which, though
rather too low, would nevertheless afford admirable situations for
public buildings. The present circumstances of Athens and Rome are
no less strikingly opposed to each other than the situation and style
of architecture. Rome is adorned, and frequently incumbered with
modern magnificence; the Athenian ruins are either insulated or
surrounded by mere huts. At Rome the buildings are numerous, and
very much decayed; at Athens they are few, and much more perfect.
Indeed, the mere lapse of time seems to have had very little effect
on those of the latter city. Earthquakes have shaken, explosions
have shattered, and avarice has despoiled them, but a great deal of
what remains, remains absolutely perfect, except in the more
delicate and exposed sculpture, and even of the sculpture a
considerable portion is as fresh as if it were only just finished. The
modern manners of Greece and Italy have introduced a style of
domestic architecture in the two cities completely different. In one
country each family inhabits a story, or perhaps only a suite of
apartments, and many families live consequently under the same
roof. In the other, such a system would be profanation: each family
occupies its own little house, shut out from the rest of the world by
being placed in a court, and instead of six or seven stories, as in
Italy, we here rarely find more than two. Nature seems to delight in
adding to these contrasts. At Rome the atmosphere is remarkably
quiet; at Athens the winds are frequent and impetuous. At Rome, on
the contrary, the recurrence of thunder-storms is frequent, and they
are extremely violent; at Athens, thunder is rare: but this remark is
so very wide of my subject, that it reminds me to return to it.
You will think I have said enough of the Acropolis, but I must still
take you round the outside of the walls, and we will notice a few
adjoining antiquities as we proceed. There is a sitting statue of early
Greek workmanship on the ascent to the Acropolis, and a beautiful,
but much injured capital of a peculiar style near the entrance; there
is also in the way up a fountain of brackish water, which is said to
run only in summer. A well within the Acropolis, the water of which is
not good, is perhaps connected with this. Beginning below the
Propylæa, we observe between that edifice and the lower modern
battery, some ancient piers built up in the present walls, which seem
to announce the most ancient access to the Acropolis, the materials
and style of masonry are like that which I have mentioned as
existing under the north-west angle of the Propylæa, and as being in
all likelihood, prior to the age of Pericles. Taking our course along
the northern side of the hill, we meet with the grotto of Pan; a
hollow in the rock of no great extent, with numerous square and
circular recesses cut apparently for the reception of votive tablets. At
a little distance there is a vault descending from the Acropolis, and
some steps passing over part of it; these steps have been thought to
belong to the entrance of the original Acropolis, but this is at best
very doubtful, and I have no theory as to their probable object or
date. Proceeding farther, the rock is extremely uneven, both in
height and direction, and there are many hollows in it, but of no
considerable depth. Some of them appear to have had votive
tablets, and perhaps architectural ornaments. In the wall above, we
see parts of an enormous entablature; it is twice interrupted, but
recurs again at the same level, as if it were placed by design, and
not accidentally. The architrave, triglyphs, and cornice, are of
limestone; but the metopes are slabs of marble. Two theories have
been made for this; one is supported by M. Fauvel, who thinks it a
portion of the regular finish of the work. If so its size was
proportioned, not to the mere height of the wall, for in some places
the space below this crowning is hardly more than the height of the
entablature itself, but to that of the rock and wall together. The
other theory takes into consideration what we are told of the mode
of erecting these walls after the Persian invasion, when to avoid the
effects of Lacedæmonian jealousy, they were reconstructed in the
greatest haste, and of any materials which first came to hand,
however previously employed. These fragments, therefore, may
either have belonged to the old temple of Minerva, or to that of
Jupiter Olympus, which was also a large edifice of the Doric order. I
should incline to Fauvel’s opinion, were it not that several marble
frusta of columns, a little farther on, which seem to correspond in
size with this entablature, give some additional weight to the latter
theory. These frusta are not finished; the circular shape is
determined, and a smoothed ring on the edge of each frustum
marks the intended size of the column, but the rest of the surface is
rough, and the projections left as means to lift them into their places
still remain. Farther on there are vestiges of very ancient walls below
the foot of the rock, which may be traced at intervals more than half
round the Acropolis; at the east end is a large cavern, which
penetrates the rock to a considerable extent: it seems to be formed
by the destruction of a loose breccia, which in some parts becomes a
mere gravel. Turning round to the south side we meet with several
ancient foundations composed of large blocks of Magnesian
limestone. Just above the choragic monument of Thrasyllus, there
are two detached columns, with triangular capitals, ornamented with
leaves and volutes, but of little beauty; these are also testimonials of
the same sort with that monument, and have supported tripods, as
is shewn by the cramp-holes remaining at the top. In one of them a
statue has probably at some period taken the place of the tripod,
and the square pedestal placed on the abacus to receive it still
remains. Below this monument is a large hollow, which is supposed
to have contained the theatre of Bacchus, for it is now generally
acknowledged, that what Stuart has published under that name, is
the theatre of Herodes Atticus. The range of arches extending from
the hollow to the latter building seems a very mixed production, with
no very clear intimations of genuine antiquity. The Odeum, or
theatre of Herodes, is partly of brick, and partly of Magnesian
limestone.
I shall not attempt to follow the order of place in the few
remaining remarks I have to make on Athenian antiquities.
I mentioned the Temple of Jupiter Olympus on our arrival, but so
important an edifice must not be passed over with so slight a notice.
A building under this name was begun by Pisistratus, or perhaps still
earlier. This we may suppose to have been destroyed by the
Persians. Pericles seems to have done nothing towards its
completion or re-erection; perhaps Jupiter, in his time, was not a
popular deity. Livy mentions it as built by Antiochus, and as the only
temple worthy the majesty of the god; but the passage is defective.
Vitruvius also says that it was built by Antiochus, and that Cossutius,
a Roman, was the architect. According to Pliny, Sylla carried away its
columns to Rome, but in the immense multitude of fragments
remaining in that city, there is not a single example of a column, or a
portion of a column, of large diameter, of Pentelic marble. Several
Asiatic sovereigns are said to have paid their court to Augustus by
contributing to its restoration (an odd way of gaining favour), and to
this epoch I would attribute all the existing columns, for it contains
several particulars which render it probable that it was prior to
Hadrian’s time. To judge of the date of a building, either by its
design or execution, it is necessary to compare many different
works, for every edifice will have something peculiar to itself, and
without several examples it is impossible to distinguish these
individual peculiarities from those which are characteristic of the
age; and we have few examples of architecture in Greece, whose
date we can determine, between the time of Alexander and that of
Hadrian. My principal guide in this instance is in the foliage of the
capitals. I have already mentioned to you the difference of forms
adopted by the artists of Greece and Rome in this respect, and that
the latter usually made the lower divisions of the leaves to lap over
the other, (fig. 4) while the former only made them touch; but there
also appears to have been a pretty regular progress among the
Greeks themselves in the arrangement of these divisions. In the
earlier examples, the upper point of the lower division just touches
the lowest point of the division above it, (fig. 1). This is the case in
the monument of Lysicrates, (I am sorry to say that Stuart is not
good authority in this respect, he has nowhere sufficiently attended
to the character of the foliage), and such also is the case in this
temple of Jupiter Olympus. Afterwards the upper point of the lower
division touched not the point, but the side of the division above it:
(fig. 2) of this we have only fragments. In the latest specimens two
points of the lower divisions touch, or nearly touch, the sides of the
upper: (fig. 3) of this the fragments are very numerous, and often
executed in a very dry and tasteless manner. What I suppose Clarke
to mean by his early Corinthian capital, is often thus formed, and is
the work of the lower empire. In the arch of Hadrian this practice is
begun, but not fully established, and the leaves are gracefully
drawn. Now if you ask me why I refer these columns to the time of
Augustus rather than that of Antiochus, to which these observations
seem to apply at least equally well, I must refer you to the authority
of Pliny, and the spoliation of Sylla; for I have no internal evidence.
 Fig 1. Fig 2.

Fig 3. Fig 4.

In all ruins the mind speculates on what the building has been,
and where the remains are magnificent, on the power, the riches,
the zeal, and intelligence requisite to produce it; and finds no small
degree of pleasure in that employment. In this temple there is ample
field for such speculations, enough still remains to indicate both the
disposition of the building, and that of the court in which it stood,
which was probably surrounded with columns. The columns
remaining belonged to the temple itself; the smaller ones were more
easily taken away; the size of the large ones has been their
protection. One of them, which was standing in Stuart’s time, has
however been destroyed by a governor of Athens, and the materials
employed in the erection of a mosque, but the experiment did not
succeed; for the pashaw of Negropont hearing of it, made it a
pretence to extort money from him, on the ground that he had
appropriated to himself the property of the grand signor. These
columns are above 6 feet in diameter, and nearly 60 feet high, that
is, they are somewhat thicker than those of the portico of Covent
Garden theatre, and almost twice as high. They are of the Corinthian
order, and their sculptured capitals still remain. There must have
been originally, at least one hundred and sixteen of them: they are
of Pentelic marble, but many of the blocks are much veined with
mica slate, and resemble cipollino, but with a purer ground. It
appears probable that thicker blocks might be obtained of this, than
of the pure white marble. The workmanship is excellent, though
perhaps not equal to that of the Phidian architecture. Their physical
beauty is enhanced by the various effects of their grouping, as seen
in different positions, and by the stains of a yellowish, or rather of
an orange hue, which time has produced in all the edifices of this
marble. It is probably owing to the action of the air on a small
quantity of iron, contained in the mica which the Pentelic marble is
never without. As perfect buildings, perhaps the original colour was
the best, but as ruins, their beauty is certainly increased by the
present tints. These remains are unincumbered by any modern
building, except a little sort of hut, erected on a piece of the
architrave, the traditional residence of a Stylite; and they are placed
on an artificial platform, on a bank rising from the Ilissus, supported
by a buttressed wall, part of which still remains. The height of the
bases is unequal, and the plinths of the inner columns rest on blocks
of hard limestone, but there is a sinking of about two inches below
them, as if to receive a marble pavement. One of the outer range of
plinths also rests on a similar limestone, except in front, where there
is a block of marble, and the top of this would have been exactly
level with the surface of the marble pavement. The three first
columns of the south range rest on marble, and the paving between
them is of the same material.
The gateway known by the name of the Arch of Hadrian is near
these columns. It has perhaps rather a foolish and unmeaning look,
and the more so, from the comparison of its little columns with their
gigantic neighbours. Yet still it is an interesting monument; the
beauty of the material, the excellence of the workmanship, the
almost perfect state in which it exists, and a certain lightness and
even elegance in the disposition of its upper part, demand a
considerable degree of admiration. Wilkins has proposed a new
reading of the inscription over the gateway, by which he makes the
city of Theseus to lie on the outside of this archway, and the city of
Hadrian between it and the Acropolis. I could not resist the
temptation of making a view of it, standing directly in front, and
looking north-westward. In this position,[30] the writer tells us that
the Acropolis is out of the field of view, while according to my
notions it occupies half the picture.
I will spare you any details of the Gymnasium of Ptolemy, or of the
Stoa, since I have no additions to make to your knowledge of either.
The narrow ridge of the Areopagus would hardly afford room for a
court of justice, but the rocks are cut in all parts, apparently for the
reception of buildings, and this is also in a considerable degree the
case with the hill called Lycabettus; indeed all the three hills,
Lycabettus, Pnyx, and Musæum, are covered with traces of human
labour. Among them are subterranean cones, probably receptacles
for corn, and from one of these cones we observed a passage into a
cylindrical pit, not bigger than a well. Fragments of terra cotta also
abound in some places, and now and then a small piece of marble,
but this material was probably little, if at all used, in private
buildings.
The area on the Pnyx, supposed to be the place of assembly, is in
part sustained by a wall of vast stones, forming a line convex
outwards; the largest of them is about 10 feet by 8 on the face, but
I do not know its thickness. The space above is nearly a large sector
of a circle, not much less than a semicircle, with the beema (βημα)
in the centre of the circle. This remains as a raised platform with
steps up to it, but the area slopes in all directions from the beema, a
circumstance very inconvenient for public speaking. Just above this
is another area, somewhat similar, but on a smaller scale, and the
rocks above it are cut in a manner which might make one imagine a
sort of pulpit in this part also. From this upper area the sea may be
seen, which from the lower would be quite invisible. In both
instances the speaker must have turned his back upon it. Can these
circumstances have anything to do with the change in the situation
of the beema attributed to the thirty tyrants? By the side of the
lower beema, which is much more distinct than the other, there are
a number of little, square recesses, which are supposed to have
been intended to receive tablets of notices, or of decrees.
After descending a little from this area, we begin the ascent of the
third and highest of these three hills; which abounds like the others
with vestiges of ancient habitancy, and here and there exhibits a
small portion of the foundation of the city wall. The monument of
Philopappus crowns the summit. I have nothing particular to say of
this edifice, but the view from it is very fine, comprising the whole of
the plain of Athens, and the Saronic Gulf, and their surrounding
mountains, beyond which other mountains appear in distant
perspective. Indeed when lighted up, as I have seen it under a
brilliant sunset, it presents a scene of matchless splendour, as
enchanting to the senses as to the imagination.
The Stadium of Herodes Atticus is a little way out of Athens,
among (if I may so express myself) the roots of Hymettus. The
eminences which immediately bound it seem to have approached
each other, but their absolute union at the circular end is artificial.
The short valley thus obtained was lengthened by two great piers of
rubble-work; the stone facing which once completed them is gone,
and exposes the rubble, evidently laid in successive courses with a
coat of stucco or mortar between each. On the right hand are vast
foundations of the same nature, forming, as is supposed, the
substruction of the temple of Victory, and of the immense flight of
steps continued to it from the Stadium: on the left are traces of
other edifices, which are said to have constituted parts of the temple
of Fortune. The channel of an occasional torrent passes under the
mound at the upper end. The whole effect must have been very
splendid in its original state, but it is now almost reduced again to its
natural condition. All its marbles and squared stones have been
taken away, and the fragments of rubble which remain are hardly
distinguishable from the native rock, which bursts frequently from
the soil all around. The course was perhaps further lengthened by a
magnificent bridge, certainly not wanted for crossing the Ilissus, and
as the width of the way at the top must have been more than 60
feet, it could hardly have been intended for a mere passage, even if
the Ilissus were really a river. Stuart figures the three arches as
remaining, but they have all now disappeared, and some peasants
were at work when I was there in detaching the squared masonry of
the piers, so that a few shapeless masses of rubble will probably, in
a few years, be all that remains.
The neighbourhood of Athens is everywhere scattered over with
ruins, a large portion of which appear to have been tombs. With so
many unappropriated fragments, it is almost impossible that we
should not find some which might be conceived to be those of the
celebrated men of Athens, and accordingly we find many of their
names attached to these fragments. They are however, nothing
more than foundations, rarely rising more than a few inches above
the soil. The evidence of their appropriation is very unsatisfactory,
and there is nothing in the remains themselves which would enable
us to determine the nature of the edifice. Of those mentioned by
Pausanias, it is probable that the greater part were little more than
those of our burying-grounds, a small mound of earth, with a short
column, or steelee, generally thickest upwards, placed instead of a
headstone. Three men in ten days could have performed but little,
and there was a law to limit to this degree of exertion the expense
of a sepulchre. The place of the Academy is guessed at, rather than
known. There can hardly be any great error, but there are no
remains, for the few fragments of capitals, and other mouldings of
buildings in that direction, cannot be traced to it. The two hills of
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com