Description
The process of classification is central to the daily task of doctors and librarians; and it is
the foundation of study and research in chemistry and biology. Double-entry bookkeeping
and the preparation of financial statements are classification activities of accounting practice.
Classifying national accounting systems has long been an aspect of accounting
research. This paper seeks to extract lessons for accounting researchers from anthropology,
biology, chemistry, cosmology and medicine. In particular, we examine how the classifiers
themselves and the characteristics that they choose can affect classification. We observe
that objectivity is neither possible nor desirable in classification. Despite the arbitrariness,
some classifications can be more reasonable or more useful than others. For previous
accounting classifications, we analyze the classifiers, the scope, the characteristics used,
the data and the classification techniques. We report various problems. We then empirically
investigate the sensitivity of classifications to such issues as the characteristics chosen,
and the countries and sectors included. For this, we hand pick data on the practices of
large listed companies from 12 jurisdictions relating to 14 accounting topics under International
Financial Reporting Standards. We show how different researchers could produce
different classifications,
How arbitrary are international accounting classi?cations?
Lessons from centuries of classifying in many disciplines,
and experiments with IFRS data
Christopher Nobes
a,b,?
, Christian Stadler
c,1
a
School of Management, Royal Holloway, University of London, Egham, Surrey TW20 0EX, UK
b
Discipline of Accounting, University of Sydney, Australia
c
Department of Accounting and Finance, Lancaster University Management School, Lancaster, Lancashire LA1 4YX, UK
Keywords:
Accounting choice
Meta-analysis
Sensitivity
a b s t r a c t
The process of classi?cation is central to the daily task of doctors and librarians; and it is
the foundation of study and research in chemistry and biology. Double-entry bookkeeping
and the preparation of ?nancial statements are classi?cation activities of accounting prac-
tice. Classifying national accounting systems has long been an aspect of accounting
research. This paper seeks to extract lessons for accounting researchers from anthropology,
biology, chemistry, cosmology and medicine. In particular, we examine how the classi?ers
themselves and the characteristics that they choose can affect classi?cation. We observe
that objectivity is neither possible nor desirable in classi?cation. Despite the arbitrariness,
some classi?cations can be more reasonable or more useful than others. For previous
accounting classi?cations, we analyze the classi?ers, the scope, the characteristics used,
the data and the classi?cation techniques. We report various problems. We then empiri-
cally investigate the sensitivity of classi?cations to such issues as the characteristics cho-
sen, and the countries and sectors included. For this, we hand pick data on the practices of
large listed companies from 12 jurisdictions relating to 14 accounting topics under Inter-
national Financial Reporting Standards. We show how different researchers could produce
different classi?cations, particularly depending on which accounting topics are used to rep-
resent the countries.
Ó 2013 Elsevier Ltd. All rights reserved.
Introduction
Classi?cation is a fundamental part of many disciplines.
The classi?cations of diseases and books are vital in the
daily tasks of medical practitioners and librarians, respec-
tively. The Linnaean and Mendeleev classi?cations are cen-
tral to learning and research in biology and chemistry.
Classi?cations have also been made in many other ?elds;
for example, languages (Ruhlen, 1991), economies
(Neuberger & Duffy, 1976), political systems (Shils, 1966),
and legal systems (David & Brierley, 1985). Members of
society are also put into classes, e.g. recently in the UK (Sa-
vage et al., 2013). In all cases, the fundamental purpose of
the classi?cation is to simplify (Rudner, 1966).
The everyday work of accountants involves recording
transactions in the classi?cation system that is double-en-
try bookkeeping. The ?nancial statements which result are
also classi?cations: for example, assets are classed as
non-current or current; the former are then sub-classed
as tangible, intangible or ?nancial (Gröjer, 2001). The
classi?cations are debatable: in the income statement,
should expenses be classi?ed by nature or by function?
0361-3682/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.aos.2013.10.001
?
Corresponding author at: School of Management, Royal Holloway,
University of London, Egham, Surrey TW20 0EX, UK. Tel.: +44 1784
414120.
E-mail addresses: [email protected](C. Nobes), c.stadler@lancaster.
ac.uk (C. Stadler).
1
Tel.: +44 1524 592221.
Accounting, Organizations and Society 38 (2013) 573–595
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
Some classi?cations are metaphysical: the split of equity
?nancial assets into trading or available-for-sale rests
2
not on any observable characteristic, nor even on the real
intentions of managers, but on the declared intentions of
managers.
Classi?cation has also been applied in the ?eld of inter-
national accounting. Just as in other ?elds, classi?cation
has been used to assist understanding of how the many
different objects (in this case, accounting systems) are re-
lated. We explain in more detail below how classi?cation
of accounting systems can be relevant to accounting prac-
tice and research. We use the term ‘accounting system’ to
refer to a set of accounting practices, i.e. policies on recog-
nition, measurement and presentation as used in a com-
pany’s published ?nancial statements. For example, each
individual listed company in the USA has its own account-
ing practices. However, the accounting of all the companies
has many shared characteristics, imposed and enforced by
the Securities and Exchange Commission. The individual
examples of US accounting share so much in common that
they could be said to comprise a ‘system’: the US GAAP
accounting system. Another precise, but quite different,
system is French GAAP as used for unconsolidated ?nancial
statements in France. A country can exhibit more than one
system. For example, although a national GAAP (such as
French GAAP) is still used for unconsolidated ?nancial
statements in most EU countries, the consolidated state-
ments of listed companies are now prepared using Interna-
tional Financial Reporting Standards (IFRS).
The US and French ‘systems’ contain few overt
3
options.
However, partly because of international political negotia-
tions (Camfferman & Zeff, 2007, chap. 5), many options were
included in IFRS; although these are gradually being re-
moved. Therefore, transition from French (or German etc.)
GAAP to IFRS has increased the variation in accounting prac-
tices within a country. Even so, when considering the op-
tions, national factors (including such matters as tax and
legal systems) can still affect a company’s choice. Ball
(2006, p. 15) explains how, even if all entities are complying
with IFRS, the incentives of preparers and enforcers remain
‘primarily local’. As a result, one can discern national pat-
terns of IFRS practice (Kvaal & Nobes, 2010). These could
be seen as different ‘systems’ of generic IFRS. We classify
such systems in the empirical part of this paper.
Many international classi?cations of accounting sys-
tems have been proposed, beginning more than a century
ago. An examination of other ?elds (see below) suggests
that a classi?cation might re?ect its classi?er, and that
the process of classi?cation is awash with judgements of
various kinds. Few accounting classi?ers have even dis-
cussed, let alone investigated, the sensitivity of classi?ca-
tions to changes in the nature of the classi?ers, the
number of objects being classi?ed (countries or accounting
systems), the nature and number of characteristics used to
measure the objects, or the type of companies (e.g. corpo-
rate sectors) included. This paper discusses and empirically
investigates these issues.
Our ?rst objective is to investigate the ways in which
classi?cations in ?elds other than accounting (i.e. anthro-
pology, biology, chemistry, cosmology and medicine) have
been affected by issues analogous to those in the previous
paragraph and how classi?cations have changed dramati-
cally over time. We seek lessons for assessing the robust-
ness of accounting classi?cations.
Our second objective is to apply these lessons. We
examine previous accounting classi?cations, especially to
record the number of countries classi?ed, the number
and type of characteristics used to classify them, and
whether industry sectors were discussed or excluded. We
?nd that the early classi?cations re?ect the classi?ers; in
particular, they vary by the national backgrounds of the
classi?ers. We then ?nd that some classi?ers apparently
used no data, and most of the rest used data collected by
others for other purposes. Few classi?ers speci?ed the date
or the scope of their classi?cations (e.g. was it limited to
listed companies or to non-?nancial companies?), and
few speci?ed a clear purpose.
Our ?nal objective is to investigate empirically whether
accounting classi?cations are anything other than arbi-
trary; whether they can easily be manipulated to back up
particular arguments. To do this, we hand pick data on
the practices in 2011, under International Financial Report-
ing Standards (IFRS), of a large sample of listed companies
from 12 jurisdictions. For these companies, we examine
the observable accounting policy choices on 14 topics,
including presentation issues (e.g. choice of format for
the income statement) and measurement issues (e.g. use
of cost or fair value for investment property). Our analyses
are based on a total of 5689 choices of 514 companies. We
?nd that our classi?cations are highly sensitive to changes
in the set of characteristics measured (i.e. the IFRS policy
topics, in our case), and this is a feature common to classi-
?cations in other ?elds such as biology. However, certain
aspects of the classi?cations are remarkably stable, e.g.
Italy and Spain are always in the same group, and never
with the UK. Furthermore, with minor exceptions, the clas-
si?cations are much more robust to the exclusion of indi-
vidual countries or sectors. We therefore conclude that
our classi?cations based on IFRS choices are not essentially
arbitrary. Nevertheless, our classi?cations could be used to
support or refute the in?uence on accounting of the code/
common legal dichotomy.
The advantage of using the above policy topics is that
IFRS speci?cally allows management to choose among
the options.
4
There is therefore scope for country and indus-
try in?uences to lead to varied practice, unlike the setting of
previous studies of pre-IFRS accounting where management
is constrained by national rules, which can also vary by
industry. Our experiments deal with the accounting policy
choices of actual companies but, for the purposes of other
accounting research, different objects should be classi?ed.
For example, in an analysis of corporate reporting regula-
tion, Leuz (2010) classi?es countries on the basis of facts
and impressions about legal systems and securities laws.
2
As under IAS 39, para. 9.
3
As explained later, we use this term to describe policy options that have
been deliberately inserted into accounting rules.
4
Some caveats will be entered later, but our policy topics are presented
in the IFRS documents as free choices.
574 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
The purpose of a study should guide the choice of character-
istics measured (Roberts, 1995); different accounting classi-
?cations are suitable for different purposes.
The paper contributes by drawing together relevant les-
sons on classi?cation from ?elds other than accounting; by
conducting the ?rst meta-analysis of accounting classi?ca-
tions; by applying the lessons from other ?elds when ana-
lyzing the accounting classi?cation literature with a new
focus on the nature of the classi?ers and of the character-
istics, sectors and countries included; and by assessing the
reliability of previous classi?cations through empirical
investigation of the sensitivity of classi?cation to varia-
tions in such factors, thereby revealing the dramatic effect
of inclusions/exclusions on classi?cations. Our purpose is
not to present a classi?cation but, as a by-product of our
work, we provide data on IFRS practices for the ?rst time
for several jurisdictions (i.e. China, Hong Kong, South Afri-
ca, South Korea and Switzerland), and we provide classi?-
cations which include these countries (and Canada) for the
?rst time.
Our ?ndings on the reliability of classi?cations are
important because hundreds
5
of academic papers refer to
the classi?cations as part of motivating research (Ball,
Kothari, & Robin, 2000; Gray, 1988; O’Donnell & Prather-
Kinsey, 2010; Saudagaran & Biddle, 1995) or to justify an
independent variable (type of accounting system) which is
expected to in?uence issues such as value relevance (e.g.
Ali & Hwang, 2000). Then, there are new uses for classi?ca-
tions, as explanations of which companies volunteer to
adopt IFRS (Tarca, Morris, & Moy, 2013), how jurisdictions
respond to IFRS (Sellhorn & Gornik-Tomaszewski, 2006;
Tyrrall, Woodward, & Rakhimbekova, 2007), how countries
change from one class to another (Xiao, Weetman, & Sun,
2004), how practices on major topics vary over time (Ding,
Richard, & Stolowy, 2008), how companies respond to the
choices available in IFRS (Nobes, 2011), why the amount of
lobbying on IFRS varies by country (Orens, Jorissen, Lybaert,
& van der Tas, 2011), by how much various countries’
domestic accounting requirements vary from IFRS (Ding,
Hope, Jeanjean, & Stolowy, 2007), or how to identify coun-
tries with similar backgrounds when selecting countries
for study (Delvaille, Ebbers, & Saccon, 2005). If the classi?ca-
tions are inappropriate, the research setting or the variables
will be questionable.
For ?nancial analysts, students and policy makers, the
classi?cations are a convenient way of simplifying and
summarizing. So, again, inappropriate classi?cations are
likely to be misleading. For instance, much of the argumen-
tation on the development of new standards is political
(Harrison & McKinnon, 1986), and is now often expressed
in terms of resisting ‘Anglo-American’ accounting. As an
example, German writers have seen the international stan-
dard-setters as a Trojan horse which conceals Anglo-Amer-
ican accounting (Kleekämper, 2000) or as ‘the unknown
enemy from London’
6
(Hennes & Metzger, 2010). Botzem
and Quack (2009) believe that the history of the Interna-
tional Accounting Standards Committee has been wrongly
reported as ‘an Anglo-American success story’ (p. 991). How-
ever, as will be shown, some classi?ers deny the existence of
Anglo-American accounting.
Power (2009) warns researchers not to exaggerate the
international differences (p. 325) and to be wary of resort-
ing to cultural variables to explain them (p. 331). On the
?rst point, Power notes that we can only talk of different
arrangements of balance sheets because all companies
present balance sheets and show very similar things in
them. This paper reinforces those warnings by showing
that some ?ndings about international differences are
unreliable and that classi?cation could be used to prove
or disprove the importance of one commonly used cultural
variable: the legal system. Further, although classi?cations
do identify countries which are different, their main pur-
pose is to group together countries on the basis of their
similarities.
The paper proceeds as follows. The next section shows
how classi?cation in various ?elds can depend on who is
doing the classifying. Then we examine the degree to
which classi?cation depends upon the characteristics cho-
sen to measure the objects being classi?ed, and upon the
de?nition of the characteristics. After that, we perform a
meta-analysis of previous accounting classi?cations, and
we analyze them in order to reveal the apparent effects
on classi?cations of the classi?ers and the potential effects
of various other factors, such as the countries/systems in-
cluded and the characteristics chosen to represent them.
The next four sections report on our empirical investiga-
tions: the data, the ?ndings on policy choice, the analysis
of sensitivity of classi?cation to the manipulation of vari-
ous factors, and the presentation of some new classi?ca-
tions. Then, there are conclusions.
Goldilocks and the forebears
This section examines the degree to which classi?cation
is determined by who is classifying. Bloor (1982, p. 268)
found new support for the claim of Durkheim and Mauss
(1903) that the classi?cation of things reproduces a pattern
of social arrangements more than a pattern of the things. He
argued that even such renowned scientists as Newton and
Boyle were affected by their religious and political ideals
and ‘were arranging the fundamental laws and classi?ca-
tions of their natural knowledge in a way that artfully
aligned them with their social goals’ (p. 290). The ?elds of
cosmology and anthropology are used as examples below.
Cosmology
Throughout most of recorded history, man
7
saw himself
as the unique peak of creation (see below). He lived in a
world which was also in a class of its own, being ?xed and
at the center of the universe. The Copernican revolution,
set in motion in 1543 by the publication of the book
commonly known as the ‘Revolutions’,
8
spread slowly. For
5
As examples from Table 1, Nair and Frank (1980) has 228 citations and
Nobes (1983) has 390 (according to Google Scholar, accessed on 15.4.2013).
6
‘Der unbekannte Feind aus London’.
7
We use the term ‘man’ when discussing authors who did so (i.e. those
until the late twentieth century).
8
De revolutionibus orbium coelestium (On the revolutions of the heavenly
spheres).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 575
espousing it, Galileo was held under house arrest from 1633
to his death in 1642. Even in unorthodox Amsterdam, Joan
Blaeu’s ‘New and very accurate map of the whole world’ of
1662 still gave equal status to Ptolemy’s geocentric beliefs
and heliocentrism (Brotton, 2012, p. 288). However, enlight-
enment eventually reduced anthropocentrism: the earth is
now classi?ed as a planet (i.e. something that moves) orbit-
ing a star which is rather far from the middle of one of many
galaxies. The planet is fairly small, but happens to be in the
Goldilocks zone: at the right distance from its star to be at a
congenial temperature for water-based life forms.
Anthropology
The ‘great chain of being’, derived from Aristotle and
conventional for millennia, is a six-group classi?cation
9
(Lovejoy, 1964). Man is not classi?ed as an animal at all
but as a special creation which is a little lower than the an-
gels. They had spirit only, animals had body only, but man
had both. Man saw himself as unique: not just sui generis
but hors de catégorie. In the eighteenth century, Linnaeus
took homo sapiens down a rung by placing him in the animal
kingdom, though he remained sui generis. The descent of
man continued in the nineteenth century when Darwin out-
rageously suggested evolution from more ‘primitive’ prima-
tes, presumably without spirits; and other types of humans
joined his genus, such as homo neanderthalensis. In the twen-
tieth century, the genus got more crowded, for example with
the arrival of homo ?oresiensis. In the twenty-?rst century,
Wildman, Uddin, Liu, Grossman, and Goodman (2003) went
yet further by proposing
10
that, since modern humans share
99.4% of non-synonymous DNA with chimpanzees, homo
sapiens is a parvenu member of their genus.
Classi?cation and standards
The previous section showed how classi?cation can de-
pend on the mindsets of those doing the classifying, and
how classi?cation can therefore change dramatically over
time without the objects changing. This section examines
the degree to which classi?cation depends upon the char-
acteristics chosen to measure the objects being classi?ed,
and on the de?nition of the characteristics. Foucault
(1970, p. 125) suggested that modernity in science begins
with privileging observation, starting with Roger Bacon.
Sight must replace reliance on ‘self-evident’ axioms. It also
replaces hearsay evidence about sightings, and it is given
greater weight than the less reliable senses of taste, smell
and touch (p. 132). The invention of the telescope and
the microscope helped greatly. It is observation which
guided Copernicus and Galileo, and Linnaeus and Darwin.
However, not even everything visible is relevant and reli-
able: color is not (p. 133). When Linnaeus classi?ed plants,
he used only four observable features: the shape of
elements, the quantity of the elements, their arrangement
related to each other, and their relative magnitudes.
However, there was still much scope in deciding which ele-
ments to observe, as will be explained below. In a book
whose title could be translated
11
as ‘To Think, to Classify’,
Perec (1985) discusses how books can be classi?ed by, inter
alia, alphabetical order of author or title, country of author
or publication, color, date of publication or acquisition, lan-
guage, priority for reading, and so on (p. 39). The ?elds of
cosmology, chemistry, biology and medicine are now used
as examples.
Cosmology
Whether or not a celestial body is classed as a planet
depends, like any classi?cation, on de?nitions (Gröjer,
2001) or standards.
12
The ‘standard’ for a planet was revised
by the International Astronomical Union in 2006 (IAU,
2006), with the result that Pluto (which had only become
a planet, as far as we were concerned, on its discovery in
1930) ceased to be one. The revision was caused by the dis-
covery of bodies larger than Pluto with orbits further from
the sun. The re-de?nition of a planet and the re-classi?ca-
tion of Pluto has both scienti?c and cultural implications
(Basri & Brown, 2006), though not as large as those that
led to the arrest of Galileo for professing the planetary status
of the earth. An important implication for other classi?ca-
tions (e.g. in accounting) is that an object’s place in a classi-
?cation can depend on the range of objects being classi?ed.
Chemistry
Some alchemists had classi?ed elements into solids,
liquids and gases, but this is now seen to produce an
unhelpful classi?cation of such liquids as mercury, molten
lead and liquid nitrogen. So, chemists moved onto observ-
ing various behaviors of elements (e.g. reaction to oxy-
gen), leading to Mendeleev’s periodic table (Aldersley-
Williams, 2011). This approach was later con?rmed by a
more fundamental one (called ‘natural’ in the next para-
graph) when it became possible to count protons, neu-
trons and electrons.
Biology
Linnaeus started his classifying with plants, perhaps be-
cause their characteristics are more easily observable than
such things as the structure of the inner ear of animals
(Foucault, 1970, p. 137). However, he chose to ignore dif-
ferences in leaves, stems and roots, such that the ‘primary
arrangement of the vegetables
13
is to be taken from the
9
God, angels, man, animals, plants and minerals.
10
This proposal has not been generally accepted. For example, Steiper and
Young (2006, p.385) still treat homo and pan as different genera.
11
A translation of ‘Penser/Classer’ was published in 2009 by Godine Press
of Boston under the less literal title of ‘Thoughts of Sorts’.
12
At ?rst sight, the word ‘standard’ has a different meaning in ?nancial
reporting from that used here. It appears to refer to a type of regulation.
Elsewhere in accounting, a ‘standard cost’ ?ts more obviously into the
normal scienti?c meaning. However, the documents issued by the Inter-
national Accounting Standards Board (IASB), for example, are not in
themselves requirements. The IASB is a private sector standard setter. A
regulator such as the European Union can choose to require certain
companies to comply with a standard.
13
That is, plants; Linnaeus classi?ed all things on earth as animal,
vegetable or mineral.
576 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
fruit-body
14
alone’ (Linnaeus, 1751, section 164). In other
words, his system was essentially arbitrary. Whereas the
features of living plants were easy to observe, plants lacked
a fossil record on which to base the evolutionary approach
that was adopted fairly early on for animal classi?cation.
However, analysis of plant DNA has recently solved this
problem and led to a transformation of the botanical classi-
?cation to something ‘natural’, i.e. related to the thing caus-
ing the variation (e.g. Duff & Nickrent, 1999).
Classi?cation of animals has a long history. Socrates
classi?ed man as a ‘featherless biped’, but his pupil Plato
was mocked by Diogenes for repeating it. Ironically, biolo-
gists still include humans and birds in a super-class of tetra-
poda. However, at amoredetailedlevel, humans arenot now
classedwithbirds but withdogs anddolphins (whichosten-
sibly have four feet and no feet, respectively). Looking more
deeply, one can observe ?ve ?ngers not only on a human
handbut alsoona dog’s front pawandinsidea dolphin’s ?ip-
per. Several other mammalian shared characteristics can be
identi?ed, such as giving birth to live young.
As with plants, Linnaeus classi?ed animals by observing
shared characteristics, but the result again depends upon
which characteristics are chosen. As a result, many of Lin-
naeus’ animal classi?cations have alsobeenoverturned. Clas-
si?cation now rests on a search for homologs, which are
shared characteristics inherited from a common ancestor,
suchas seeninthe hand, pawand ?ipper. Ineffect, zoological
classi?cation is nowentirely about descent. For this purpose,
the analysis of DNA became a powerful tool as a supplement
to, andsometimes as a contradictionof, the receivedfossil re-
cord (Stringer, 2011, chap. 1). Again, the zoological classi?ca-
tion is now regarded as ‘natural’ (i.e. less arbitrary, being
based on evolutionary relationships as evidenced by DNA).
However, a caveat should be entered. The biologists’
classi?cations take no account of different possible pur-
poses. For example, if the purpose were to help in planning
the habitats or menus for a newzoological park, it might be
more useful to classify a dolphin with a shark even though
the dolphin is much more closely related to a dog, a human
or even a pterodactyl.
15
Diseases
Medicine is a practical activity, which relies heavily on
the International Classi?cation of Diseases (ICD). This has
been in operation for a century but is revised approximately
everydecade(Bowker &Star, 2000, p. 136). The ICDis a prag-
matic tool with a clear purpose: it helps doctors to identify
diseases and then to record information about patients.
Whereas biologists nowclassify in a monothetic way, using
binary characteristics (e.g. backbone or not), the ICD looks
for a number of shared characteristics (a polythetic system).
Further, whereas chemical elements do not change,
16
and
animal species change very slowly, diseases change rapidly.
Lastly, the ICDsometimes needs to be dramatically expanded.
It beganamong sixEuropeancountries, but hadtobe adjusted
when African and Asian diseases were included (Bowker &
Star, 2000, p. 151). Many of these features remind one of clas-
si?cations of accounting systems: they are polythetic, the sys-
tems change rapidly, and the classi?cations started with
Europe and North America only.
What’s in a name?
A more alarming point must also be made: no classi?ca-
tions are ‘real’. As Buffon pointed out in 1749:
The more we increase the number of divisions in the pro-
duction of nature, the closer we shall approach to the
true, since nothing really exists in nature except individ-
uals, andsincegenera, orders andclasses exist onlyinour
imagination [as cited in Foucault, 1970, p. 146].
We noted earlier that the de?nition of a planet is a mat-
ter of opinion. In biology, it is notable that neither Darwin
nor any follower has set out a de?nition of ‘species’ which
has gained general acceptance. Linnaeus thought that spe-
cies were ?xed in number, immutable in nature and divinely
created. Darwin showed that the ?rst two points were er-
rors, and drew a polite veil over the third. However, we
can now put another interpretation on the origin of species:
they evolved in the brain of homo sapiens. The lack of de?ni-
tions explains why there is debate about whether Neander-
thals and modern humans are part of the same species
(given that they have successfully interbred),
17
and whether
humans are part of the chimpanzee genus. Buffon’s insight
has not yet been taken to its logical conclusion, but the com-
plete abandonment of the apparatus of species, genera, etc. is
being contemplated by biologists (Mishler, 2009, p. 65).
Nevertheless, it seems unlikely that biologists or any
other humans will be able to give up classifying: ‘to classify
is human’ (Bowker & Star, 2000, p. 1). Lévi-Strauss (1958)
suggests that we inevitably perceive the world in terms
of binary opposites, and he encourages a search for under-
lying structures. For example, when discussing the content
of myths, Lévi-Strauss notes that ‘this apparent arbitrari-
ness is belied by the astounding similarity between myths
collected from widely different regions’ (p. 208). Further-
more, the fact that no classi?cations are real does not mean
that classi?cation cannot be useful. For librarians or doc-
tors, various competing classi?cation systems could be al-
most as useful as each other. For example, the Dewey
Decimal system and the Library of Congress system both
work satisfactorily in libraries. Linnaeus’ initial botanical
classi?cation was also of practical use in organizing
information, even though it was arbitrary. However, some
classi?cations might be more useful than others. For exam-
ple, Mendeleev’s classi?cation in chemistry was much
more useful than some earlier classi?cations because it
identi?ed ‘missing’ elements and predicted what they
would be like.
14
That is, the reproductive system.
15
The four types of animal in this sentence other than the shark are all in
the tetrapod clade.
16
Elements cannot be changed by chemical reactions. They can be
changed by (and indeed were formed by) nuclear reactions. Thus, gold is
created from other elements such as base metals (ultimately from
hydrogen) and it could be used to create even heavier elements, but this
does not change the nature, de?nition or ‘standard’ of gold.
17
Modern humans, except for sub-Saharan Africans, contain traces of
Neanderthal DNA; up to 4% in some cases (Green et al., 2010).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 577
Accounting classi?ers can learn from these ?elds. One
relevant lesson from above is the need for detailed per-
sonal observation. Another is that the purposes of a classi-
?cation should be considered. Further, classi?ers should be
deliberate about the characteristics measured; Roberts
(1995, p. 641) shows ‘the incoherence of taxonomies
which rely upon appeals to objectivity’. We apply these
lessons below, while analyzing past accounting
classi?cations.
Analysis of previous accounting classi?cations
There have been many international classi?cations of
accounting, as summarized in Table 1.
18
Several (i.e. items
2, 3, 4, 10, 14 and 15 of Table 1) relate to in?uences on
accounting rather than to accounting itself. Roberts (1995)
calls the former ‘extrinsic’ and the latter ‘intrinsic’; or they
could be called deductive and inductive. This section ?rst
performs a meta-analysis on these classi?cations and then
examines the apparent effects on classi?cation of the classi-
?ers, and the potential effects of various other factors, such
as the countries/systems included and the characteristics
chosen to represent them.
A meta-analysis
Meta-analysis is a procedure which mathematically
integrates the results of previous independent studies. It
can reduce the importance of unbiased errors in the data
or the procedures of particular individual studies. Meta-
analysis is frequently used in medical research, in which
context Egger, Smith, and Phillips (1997) note that atten-
tion must be paid to the selection and weighting of previ-
ous studies. For our meta-analysis of accounting
classi?cations, we include all the studies of Table 1 with
equal weights, in the absence of any objective alternative.
Our analysis covers the 15 countries which host the
world’s largest economies and which have been included
in previous classi?cations. The countries included are
those for which we collect IFRS data for our experiments
below (except China), plus the largest remaining countries:
Brazil, India, Japan and the United States. Russia is ex-
cluded because it was only found in two previous classi?-
cations, and not in terms of published ?nancial reporting.
China is excluded because it was not in any of the former
classi?cations. As will be explained in the next paragraph,
ours is not a traditional meta-analysis which combines
studies by signi?cance levels (see e.g. Christie, 1990), be-
cause the results of classi?cation studies are the groupings
of countries and not signi?cance levels.
Table 2 shows the meta-analysis: the bottom-left trian-
gle relates to all the classi?cations, the top-right triangle to
the intrinsic ones only. For each pair of countries, the ?gure
shown is the percentage of the classi?cations which placed
that pair in the same group. The bracketed number shows
how many classi?cations included the pair. For example,
the bottom-left pairing of Japan and the US shows that
those countries were together for 40% of the ten classi?ca-
tions which included them both. Scores of 0% or 100% re-
veal consensus among the classi?cations. The table also
shows which percentages for country-pairs are signi?-
cantly different from 50%, based on a test of proportion
(two-sided). A signi?cant result indicates a high degree of
con?dence that the relationship of the countries in the pair
(either being or not being grouped together) is not arbi-
trary. Although the various classi?cations consider differ-
ent numbers of countries and result in different numbers
of groups,
19
our method of analyzing country-pairs allows
the combination of these different classi?cations into a
meta-analysis.
The meta-analysis can be summarized as showing two
main features. First, most of the percentages for the coun-
try-pairs are not signi?cantly different from 50%,
20
which
suggests a high degree of arbitrariness in the classi?cations;
however, many relationships are not arbitrary. Similar con-
clusions can be drawn from observing that there are many
scores from 33% to 67%. For example, the results for Italy
in the bottom-left triangle reveal that there is little consen-
sus concerning which countries it should be classi?ed with
(because there are no percentages above 50% which are sta-
tistically signi?cant). On the other hand, there is strong con-
sensus that Italy should not be classi?ed with ‘Anglo’
countries (see the six percentages below 50% which are sta-
tistically signi?cant). Similar remarks apply to France, Spain
and Germany. Secondly, a British group can be identi?ed,
which includes Australia and Hong Kong (see the UK column
and row in the bottom-left triangle). However, North Amer-
ica is not included in that group: the ?rst column of Table 2
shows that only Canada has usually been classi?ed with the
United States. The ?rst row (intrinsic classi?cations only)
shows an even lower tendency for there to be an ‘Anglo-
American’ group.
Several caveats must be entered about this meta-analy-
sis. First, it uses data (i.e. the classi?cations) spanning sev-
eral decades, during which countries might have changed
their relationships. This and other reasons might mean that
the various results should not have been combined. Never-
theless, to the extent that certain pairs of countries retain
their relative positions over many decades (even surviving
a move to IFRS) suggests that the classi?cations are picking
18
d’Arcy (2001) lists some further papers, which we exclude on the
grounds that they replicate or overlap previous papers (e.g. Mueller (1968)
overlaps Mueller (1967), Nair (1982) replicates Nair and Frank (1980), and
Salter and Doupnik (1992) overlaps Doupnik and Salter (1993)); or they are
about the style of rule-making (e.g. Daley and Mueller (1982, for which
d’Arcy references its 1989 re-printing) and AlNajjar (1986)); or their
purpose is not to present a classi?cation (e.g. Previts (1975) provides
criteria, Gray (1988) builds a theory, and Cooke and Wallace (1990) test a
developed/developing country hypothesis).
19
In two of the classi?cations (Nobes (1983) and Doupnik and Salter
(1993)), countries were ?rst divided into two groups, and those were sub-
divided further. For the meta-analysis, we used the two-group classi?ca-
tions (see the footnote of Table 2), which stress similarities rather than
differences. Compared to using the multi-group classi?cations, this
increases the scores in Table 2 for several country-pairs.
20
61% of the scores in the bottom-left triangle and 75% in the top-right
triangle (64 out of 105 country-pairs and 76 out of 101 country-pairs,
respectively; only 101 country-pairs are considered for the top-right
triangle because there are four cases where the country-pair is only
included in one classi?cation but the test requires at least two observa-
tions). The main reason for the lower frequency of signi?cant scores in the
top-right triangle is the reduced power of the tests due to considering fewer
classi?cations.
578 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
Table 1
Features of some classi?cations.
1. Researchers 2. No. of
countries
3. Range of companies (e.g.
sectors, large, listed)
4. Date of data 5. No.
of
topics
6. Type of data 7.Classi?cation
method
8. Classi?cation type
1. Hat?eld (1911) 4 Unspeci?ed Unspeci?ed, c. 1910 0 Impressions of practices Judgement 3 Groups
2. Mueller (1967) 5 Unspeci?ed Unspeci?ed, c. 1965 1 Impressions of purposes Judgement 4 Unconnected groups
3. Seidler (1967) 13 Unspeci?ed Unspeci?ed, c. 1965 1 Impressions of in?uences Judgement 4 Unconnected groups plus
other mentioned countries
4. AAA (1977) 6 Unspeci?ed Unspeci?ed, c. 1975 1 Impressions of in?uences Judgement 5 Unconnected groups
5. da Costa et al.
(1978)
38 Unspeci?ed Unspeci?ed, c. 1973 100 Mixture of rules and impressions of
practices (by Price Waterhouse
partners)
PCA 2 Unconnected groups
6. Frank (1979) 38 Unspeci?ed Unspeci?ed, c. 1973 233 As above PCA, MDS 4 Unconnected groups
7. Nair and Frank
(1980)
38, 46 Unspeci?ed Unspeci?ed, c. 1973 and
c. 1975
233,
264
As above PCA, SSA 4/5 Unconnected groups for
measurement; 7 for disclosure
8. Goodrich (1982) 64 Unspeci?ed Unspeci?ed, c. 1979 26 Impressions of concepts (by Price
Waterhouse partners)
PCA 5 Unconnected groups
9. Nobes (1983) 14 Listed 1980 9 Impressions of practices PCA Hierarchy of 2 groups, leading
to 6 groups
10. Puxty, Willmott,
Cooper, and Lowe
(1987)
4 Unspeci?ed Unspeci?ed, c. 1985 3 Impressions of regulatory style Judgement Positions of the countries with
respect to three regulatory
ideals
11. Shoenthal (1989) 2 Unspeci?ed Unspeci?ed, c. 1987 1 Impressions of competencies of
auditors
Judgement 2 Unconnected groups
12. Doupnik and Salter
(1993)
50 Economically signi?cant
entities
1990 114 Impressions of practices (by
academics and auditors)
Average-
linkage
clustering
Hierarchy of 2 groups, leading
to 9 groups
13. d’Arcy (2001) 14 + IASC Listed; consolidated and
unconsolidated
Unspeci?ed, based on
Ordelheide and Semler
(1995)
129 Rules Clustering,
MDS
4 Groups with MDS
14. Leuz et al. (2003) 31 Listed Based on La Porta et al.
(1998)
9 Facts and impressions relating to
stock markets and investor protection
Clustering by
k-means
3 Groups in order
15. Leuz (2010) 37, 49 Listed ‘2000s’ 13 Facts and impressions on legal
system, securities regulation
Clustering by
k-means
3 Groups, then 5 groups
16. Nobes (2011) 8 Large, listed, consolidated,
excluding ?nancials for some
topics
2008/9 13 Practices PCA, MDS,
clustering
3 Groups by PCA; hierarchy
starting with 2 groups
Key: PCA = principal component analysis. MDS = multi-dimensional scaling. SSA = smallest space analysis.
C
.
N
o
b
e
s
,
C
.
S
t
a
d
l
e
r
/
A
c
c
o
u
n
t
i
n
g
,
O
r
g
a
n
i
z
a
t
i
o
n
s
a
n
d
S
o
c
i
e
t
y
3
8
(
2
0
1
3
)
5
7
3
–
5
9
5
5
7
9
Table 2
Meta-analysis of classi?cations: percentages with which pairs of countries are grouped together.
US % (N) AU % (N) UK % (N) CA % (N) HK % (N) FR % (N) ES % (N) IT % (N) DE % (N) CH % (N) ZA % (N) SK % (N) BR % (N) IN % (N) JP % (N)
US 29 (7) 22 (9) 83 (6) 50 (2) 13
Ã
(8) 14 (7) 17 (6) 25 (8) 20 (5) 20 (5) 50 (2) 20 (5) 25 (4) 43 (7)
AU 40 (10) 75 (8) 33 (6) 50 (2) 0
Ã
(8) 0
Ã
(8) 0
Ã
(7) 0
Ã
(8) 0
Ã
(5) 80 (5) 0 (2) 20 (5) 25 (4) 14 (7)
UK 33 (15) 82
Ã
(11) 50 (6) 100 (2) 22 (9) 0
Ã
(8) 0
Ã
(7) 11
Ã
(9) 20 (5) 100
Ã
(5) 0 (2) 0
Ã
(5) 0
Ã
(4) 0
Ã
(7)
CA 88
Ã
(8) 50 (8) 63 (8) 100 (2) 17 (6) 0
Ã
(6) 0
Ã
(5) 17 (6) 0
Ã
(4) 50 (4) 0 (2) 0
Ã
(4) 0 (3) 33 (6)
HK 75 (4) 75 (4) 100
Ã
(4) 100
Ã
(4) 50 (2) 0 (2) 0 (2) 0 (2) 0 (1) 100 (2) 0 (2) 0 (2) 0 (1) 0 (2)
FR 8
Ã
(13) 0
Ã
(11) 14
Ã
(14) 13
Ã
(8) 25 (4) 63 (8) 71 (7) 78 (9) 80 (5) 20 (5) 50 (2) 40 (5) 25 (4) 43 (7)
ES 10
Ã
(10) 0
Ã
(10) 0
Ã
(11) 0
Ã
(8) 0
Ã
(4) 64 (11) 86 (7) 50 (8) 60 (5) 0
Ã
(5) 50 (2) 60 (5) 50 (4) 57 (7)
IT 13
Ã
(8) 0
Ã
(9) 0
Ã
(9) 0
Ã
(7) 0
Ã
(4) 56 (9) 78 (9) 71 (7) 50 (4) 0
Ã
(5) 50 (2) 60 (5) 50 (4) 50 (6)
DE 15
Ã
(13) 0
Ã
(10) 7
Ã
(14) 13
Ã
(8) 0
Ã
(4) 69 (13) 45 (11) 56 (9) 60 (5) 0
Ã
(5) 50 (2) 40 (5) 25 (4) 57 (7)
CH 14 (7) 0
Ã
(7) 14 (7) 0
Ã
(6) 0 (3) 86 (7) 57 (7) 33 (6) 71 (7) 0
Ã
(4) 0 (1) 25 (4) 25 (4) 20 (5)
ZA 29 (7) 71 (7) 86 (7) 50 (6) 75 (4) 29 (7) 0
Ã
(7) 0
Ã
(7) 14 (7) 17 (6) 0 (2) 0
Ã
(5) 0
Ã
(4) 0
Ã
(5)
SK 25 (4) 0
Ã
(4) 0
Ã
(4) 0
Ã
(4) 0
Ã
(4) 50 (4) 75 (4) 50 (4) 50 (4) 33 (3) 0
Ã
(4) 50 (2) 0 (1) 50 (2)
BR 17 (6) 17 (6) 0
Ã
(6) 0
Ã
(5) 0 (3) 33 (6) 50 (6) 67 (6) 33 (6) 20 (5) 0
Ã
(6) 33 (3) 100
Ã
(4) 60 (5)
IN 14 (7) 29 (7) 14 (7) 0
Ã
(5) 0 (3) 14 (7) 50 (6) 67 (6) 17 (6) 17 (6) 0
Ã
(6) 33 (3) 100
Ã
(5) 50 (4)
JP 40 (10) 10
Ã
(10) 0
Ã
(10) 25 (8) 0
Ã
(4) 50 (10) 56 (9) 38 (8) 67 (9) 43 (7) 14 (7) 50 (4) 50 (6) 29 (7)
This table reports the results of a meta-analysis of the classi?cation studies of Table 1. The bottom-left triangle considers all 16 classi?cations, and the top-right triangle considers only the intrinsic classi?cations
(i.e. excluding studies 2, 3, 4, 10, 14 and 15). For each country-pair, the table shows the frequency (in %) with which the country-pair is classi?ed in the same group. The number in brackets (N) indicates in how
many classi?cations both countries of the country-pair were included.
Ã
indicates that the percentage for the country-pair is signi?cantly different from 50% at the 5% level (two-sided, based on a test of
proportion); the test requires at least two observations, i.e. the country-pair needs to be included in at least two classi?cations; a signi?cant result indicates a high degree of con?dence that the relationship of the
countries in the pair (either being or not being grouped together) is not arbitrary. The countries are: US (United States), Australia (AU), United Kingdom (UK), Canada (CA), Hong Kong (HK), France (FR), Spain (ES),
Italy (IT), Germany (DE), Switzerland (CH), South Africa (ZA), South Korea (SK), Brazil (BR), India (IN) and Japan (JP). For those classi?cation studies of Table 1 which provide more than one classi?cation, we only
use one/the main classi?cation, as follows: for classi?cation study 7, p. 433 (1975 analysis, measurement practices); for 9, Table 8 (we use the two-group classi?cation, not the more detailed one); for 12, Table 1
(again we use the two-group classi?cation not the more detailed one of Table 2); for 13, Fig. 2 (multi-dimensional scaling); for 15, Table 3 Panel C; for 16, Table 4 (principal component analysis).
5
8
0
C
.
N
o
b
e
s
,
C
.
S
t
a
d
l
e
r
/
A
c
c
o
u
n
t
i
n
g
,
O
r
g
a
n
i
z
a
t
i
o
n
s
a
n
d
S
o
c
i
e
t
y
3
8
(
2
0
1
3
)
5
7
3
–
5
9
5
up something fundamental. However, whether the insights
from this analysis can be relied upon at all depends greatly
on whether there are biased errors in the data or the meth-
ods used by the previous classi?ers. This is a central issue
of this paper, so we return to the worth of Table 2 after we
have examined that issue.
The classi?ers
The above discussions of cosmology and anthropology
showed how susceptible classi?cation can be to the nature
of the classi?ers. For accounting classi?cations, most of the
early writers had US or UK origins, so they were most
familiar with US and UK accounting, and had noticed the
differences. They then ?tted the rest of the world around
that starting point, often leading to a three-way classi?ca-
tion: US, UK and other. This explanation is consistent with
classi?cations 1, 3 and 4 of Table 1: Hat?eld (1911) and the
similar
21
ones of Seidler (1967) and of the American
Accounting Association (AAA, 1977, p. 105). These classi?ca-
tions were all drawn up by Americans. However, classi?ca-
tion 2 was produced by Gerhard Mueller, whose initial
education
22
was in Germany. Thus, Mueller had a different
Weltanschauung, in which the US and the UK are together
in one class, and the other three classes are each typi?ed
by a different continental European country. This suggests
that the nature of the classi?ers affected the classi?cations.
An extreme version of the above approach, of starting
with the US and the UK, can be found in paper 11 of Table 1
and in Alexander and Archer (2000). In these, the writers
(all from North America or the UK) identify some differ-
ences between the US and the UK (though these relate to
the context of accounting rather than to accounting prac-
tices) and then conclude that the US and the UK cannot
be classi?ed together. This would be like observing that
two cousins exhibit many differences, and therefore can-
not be closely related.
The range of countries
The objects being classi?ed (i.e. ‘accounting systems’ or
countries that use particular accounting systems) vary in
number from two to 50 (column 2 of Table 1). Communist
countries were generally excluded, because they had no
published ?nancial reporting. Later in this paper, we in-
clude Chinese companies using IFRS. Roberts (1995) points
out that it should be accounting systems rather than coun-
tries that are classi?ed. This became particularly relevant
when the widespread use of IFRS began in 2005 because,
in many countries, IFRS is only used for certain types of
reporting, such that one country now uses two or more
systems. This point was not adopted by any of the classi?-
cations in Table 1, though it was discussed at length in
Nobes (1998). The last classi?cation of Table 1 still appears
to classify countries, but it is the set of IFRS practices used
by companies in a country (i.e. the accounting system) that
is the object of classi?cation. The same applies later in this
paper.
The range of companies
The scope of the data (e.g. restrictions by sector or list-
ing status) is recorded in column (3). As may be seen, most
classi?cations did not specify a scope. This reduces their
usefulness, because the practices of listed companies vary
from those of unlisted companies; and, even among listed
companies, size has a major effect.
23
The last classi?cation
in Table 1 was the only one to mention sectors. It included
companies in all sectors, but displayed the sectoral mix
and excluded data on ?nancial companies for topics for
which sector-speci?c practices were anticipated. On such
grounds, the exclusion of ?nancial companies is common
in much research involving accounting data. However, this
creates a different problem: in all countries, the ?nancial
sector is signi?cant and, in some (e.g. Australia, Italy, Spain
and the UK), it is the most important sector among large
listed companies, as shown later. Therefore, exclusion of
the sector presents a misleading picture of a whole account-
ing system.
The importance of sector in in?uencing accounting pol-
icy was ?rst systematically investigated by Jaafar and
McLeay (2007), who examined three policy issues for com-
panies from 13 EU countries using national accounting
rules, in a pre-IFRS world. Consequently, Jaafar and McLeay
were not investigating policy choices only but a mixture of
different requirements and different choices. They found
that country was a much stronger explanatory variable
than sector, but that sector had some in?uence. Apart from
the ?nancial sector, which is excluded from many studies
on policy choice, a sector which might make idiosyncratic
choices is extractives, given the degree to which US prac-
tices dominate.
24
Jaafar and McLeay found some evidence
of this; and it might be important for countries in which
extractive companies constitute a large industry sector
(e.g. Canada).
The period measured
The users of classi?cations should also be aware that
countries can change their positions over time.
25
Table 1
(column 3) gives information on the dates of the data used
for the classi?cations, noting that most classi?ers have not
speci?ed a date.
The characteristics chosen
The discussions about chemistry, biology and cosmol-
ogy above showed that the nature and de?nitions of the
characteristics chosen as the basis for classi?cation is vital.
It is therefore inevitable that classi?ers must use judge-
ment in selecting and de?ning the characteristics used to
21
Seidler discusses the US and UK groups in some detail, suggests a
French group without naming any members of it, and proposes a
Communist group, mentioning only the Soviet Union. The AAA has British
and US groups, plus two continental European groups and ‘Communistic’.
22
Until moving to California at age 22, and then taking various degrees.
23
For example, see Nobes and Perramon (2013).
24
For example, under IFRS, there are no detailed rules on accounting
issues associated with extraction.
25
For example, see Nobes (1998).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 581
represent the objects to be classi?ed. We now examine this
aspect for the ‘intrinsic’ classi?cations of Table 1, i.e. those
that classify countries by their accounting rules/practices
rather than by in?uences on the accounting.
The number of speci?ed characteristics per system/
country (column 5 of Table 1) varies from zero to 264.
Classi?cations 5 to 8 were all
26
based on the surveys of Price
Waterhouse (PW) (1973, 1976, 1979) which had begun as a
list of detailed differences between US and UK accounting.
They therefore did not ask (about a country) such important
questions as: (i) are depreciation expenses determined by tax
rules? or (ii) is deferred tax accounted for? They asked in-
stead such peripheral questions as whether or not self-insur-
ance provisions were maintained in an internal account by
systematic charges to income (PW, 1973, Question 124),
which was known to be a topic of US/UK difference.
Not surprisingly, the PW data showed that the US and
the UK were the most different
27
of any pairing of the 36
countries examined for 1973. Consequently, by using these
data, da Costa, Bourgeois and Lawson (1978) found again
that the world had three types of country: US-led (contain-
ing most of continental Europe), UK-led, and unclassi?able
(i.e. Canada and the Netherlands). Frank (1979) and Nair
and Frank (1980) identi?ed four groups from the same data,
two of which were those dominated by the US and UK,
respectively. Goodrich (1982) used the 1979 PW data and
identi?ed ?ve groups, two of which were headed by the
US and the UK, though there is another headed by Jersey
which (remarkably) also contains Guatemala, Germany,
Italy, the Netherlands and Zaire.
Doupnik and Salter (1993) started with PW’s list of
characteristics but report an attempt to eliminate those
not representative of the fundamental features of the
accounting systems being classi?ed. d’Arcy (2001) used a
large number of characteristics from a KPMG database
which was not speci?cally prepared for her purpose. In
Nobes (2011), the characteristics measured are the policy
choices made by companies on topics such as those listed
in Table 3. The issue of which characteristics to measure
was discussed, and certain presentation topics were de-
leted on the grounds of less importance.
It was notedabovethat thepurposes of a classi?cationcan
be relevant for choosing the characteristics to be measured.
Someaccountingresearchers haveoutlinedthevarious possi-
ble purposes of classifying, including those of classifying
accounting systems, but few
28
have speci?ed the purpose of
their own attempts, apart from organizing knowledge.
The type of data used to measure the characteristics
As may be seen from Table 1 (column 6), the type of
data used to measure the characteristics also varies. Only
the last classi?cation was based on the collection of data
relating to the accounting practices of actual companies.
Clearly, the early classi?cations, which either used no data
(classi?cations 1–4) or relied upon impressions of practices
or of in?uences on practices (classi?cations 8–11), are less
satisfactory than detailed observation of practices. In the
non-accounting ?elds reviewed earlier, the quality of data
for classi?cation improved over time in various ways. For
accounting research, the annual reports of hundreds of
listed companies can now be collected quickly, and many
are available in English.
29
By contrast, in the 1960s and
1970s, if the US researchers had wanted to collect the ?nan-
Table 3
IFRS policy topics.
Topic IFRS policy options Standard
b
1.
a
– Income statement by nature – By function or neither IAS 1.99
2.
a
– No inclusion of a line for EBIT or operating pro?t – Line included IAS 1.82
3. – Equity accounting results included in ‘operating’ – Immediately after, or after ‘?nance’ IAS 1.82
c
4. – Balance sheet showing net assets – Showing assets = credits IAS 1.54
c
5. – Balance sheet with liquidity decreasing (cash at top) – Liquidity increasing IAS 1.54
c
6. – Indirect operating cash ?ows – Direct IAS 7.18
7. – Dividends received shown as operating cash ?ow – Not IAS 7.31
8.
a
– Interest paid shown as operating cash ?ow – Not IAS 7.31
9. – Some property at fair value – Only cost IAS 16.29
10. – Investment property at fair value – At cost IAS 40.30
11.
a
– Some designation of ?nancial instruments at fair value – None IAS 39.9
12.
a
– FIFO only for inventory cost – Weighted average used IAS 2.25
13. – Actuarial gains and losses to OCI – Corridor method or to income in full IAS 19.92/3
14. – Proportionate consolidation of joint ventures – Equity method IAS 31.30
This table shows 14 IFRS policy topics on which choices were observable in 2011. Topics 1–8 are presentation issues and topics 9–14 are measurement
issues. The topics are as in Kvaal and Nobes (2010). Most topics are binary choices but topics 1, 3 and 13 allow a choice between three options. For these, we
de?ne binary choices: for topic 1, we distinguish whether or not the income statement is by nature because the ‘neither’ cases are usually more similar to
‘by function’ than ‘by nature’; for topic 3, we consider the key issue to be whether or not the item is included in operating pro?t; for topic 13, we combine
the options ‘corridor method’ and ‘to income in full’ because we consider the key issue to be whether or not actuarial gains and losses are ever charged in
the income statement, which is not the case under ‘actuarial gains and losses to OCI’.
a
Not appropriate, and therefore not collected for ?nancial companies.
b
Versions of the standards ruling in 2011.
c
IAS 1 speci?es lists of items to be shown in ?nancial statements, but does not specify their order.
26
Classi?cation 8 (Goodrich, 1982) is based on the concepts part of the
survey, but still excludes the two examples of important questions given
later in this paragraph.
27
See Exhibit 1 of da Costa, Bourgeois, and Lawson (1978).
28
An exception is that Leuz, Nanda, and Wysocki (2003) make a
classi?cation in order to better understand international variations in
earnings management.
29
Nobes and Perramon (2013) ?nd that English language reports contain
the same information as the originals.
582 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
cial statements of all the members of the main French stock
market index (for example), that would have proved very
dif?cult, and many of the reports would not have been in
English. This might be what held back researchers from de-
tailed observation, but a more likely explanation is that
accounting researchers were not yet accustomed to an
empirical approach (Watts & Zimmerman, 1979).
Data which mix rules and impressions of practices
(classi?cations 5–7 and 12) are incoherent. The usefulness
of data relating to rules alone (e.g. classi?cation 13) can
also be questioned. For example, IFRS (IAS 38, para. 72) al-
lows certain intangible assets to be measured at fair value
rather than on a cost basis, whereas German and US GAAPs
require a cost basis. However, this difference in rules is of
doubtful signi?cance if no IFRS companies choose fair va-
lue to measure intangibles, which is the case in the sample
of German and UK companies of Christensen and Nikolaev
(2013). It is surely more signi?cant, for example, that the
majority of IFRS-using UK companies choose fair value
for investment properties whereas it is very rare for IFRS-
using German companies to do so,
30
even though all the
companies from both countries are using identical rules.
The quality of data
Any of the methods of measuring the characteristics cho-
sen for classi?cation can involve error. The PWdata used for
classi?cations 5 to 7 of Table 1 certainly contain errors. Sim-
ple examples
31
are that the UK is scored as not allowing
weighted average cost for inventory valuation which was
not (and never has been) the case; and that, on several topics,
the UK has different scores from Ireland, even though they
shared the same accounting rules. The data based on KPMG
information (used for classi?cation 13) also produces errone-
ous scores. For example,
32
the KPMG topics related to ‘provi-
sions’ were interpreted for most countries as referring to
provisions as de?nedin IAS 37 (i.e. liabilities of uncertaintim-
ing or amount), but the scoring for Australia was based on the
rules for items such as ‘bad debt provisions’ (i.e. impairments,
in current IFRS terminology). The last classi?cation (Nobes,
2011), in common with most others, did not provide speci?c
information about how the data were collected and coded.
This makes it dif?cult to replicate the studies.
The techniques of classi?cation
Table 1 (column 7) also shows the techniques used for
classi?cation. These range from qualitative assessments
to several different statistical methods. The resulting clas-
si?cations range (see column 8) from apparently unrelated
groups of countries to hierarchies (family trees or dendro-
grams) of related countries. Roberts (1995, p. 649) warns
against pushing the evolutionary analogies of the family
trees too far, and points out that dendrograms can summa-
rize similarities and differences without invoking evolu-
tion. Following from this, Roberts (1995, p. 656) also
criticizes the use of the term ‘species’ in an accounting
classi?cation, as in Nobes (1983). We use ‘system’ for a
set of objects with important characteristics in common,
although the apparently scienti?c word ‘species’ might
not be entirely out of place, given the discussion above
about the vagueness of the term in biology.
Roberts (p. 652) convincingly suggests that analogies
with the classi?cation of languages might be more appro-
priate, given that languages both converge (they interbreed)
and diverge, whereas species only diverge. Even so, as dis-
cussed earlier, the Linnaean system did not begin as evolu-
tionary but was based on assessing shared characteristics.
When evolution was added in (greatly aided later by the
analysis of DNA), the broad outline of the animal classi?ca-
tion survived although many details were revised. The same
might apply to the hierarchical accounting classi?cations.
That is, although they were prepared by assessing common
characteristics, the inclusion of evolution might not upset
the results. For example, the common ancestor of UK and
US accounting could perhaps be traced to nineteenth cen-
tury UK practice. By contrast, the common ancestor of
French and UK accounting lies much further in the past, per-
haps in the middle of the 16th century.
33
The statistical methods of classi?cation employed in
our subsequent empirical analyses are outlined in Appen-
dix A. In principle, they are sensitive to which countries
are included. For example, a clustering program starts by
?nding the two nearest countries, showing them together
and then treating the average of them as a ‘country’ for
the next stage of clustering. So, exclusion of one country
can affect the ‘seeding’ of the ?rst cluster, which can then
affect the position of many other countries.
Summary of factors affecting classi?cation
In sum, early accounting classi?cations seem to
have been affected by the national backgrounds of the
classi?ers or of the data gatherers. We have suggested,
above, ways in which classi?cation has also been affected
by the data used. First, the choice of which characteristics
to measure has a profound effect on the results. Once cho-
sen, the way of measuring the characteristics has var-
ied: several classi?ers apparently used no data, some
used incoherent data (mixtures of rules and impressions
of practices); and others used data which are arguably of
limited practical relevance (differences in rules).
Many classi?ers did not specify the scope of the objects
being classi?ed (e.g. large or listed companies) or the date.
However, some classi?ers have entered caveats. Frank
(1979, p. 596) notes that the topics included in his PW data
vary in importance. Frank does not make a selection or
comment on the mixture of rules and impressions of
practices in the data, but warns that the coding scheme
which turns that mixture into data for classi?cation might
introduce errors. Nair and Frank (1980) note that the clas-
30
In the companies comprising the main German and UK stock market
indices (Kvaal & Nobes, 2010).
31
Several examples are given in Nobes (1981).
32
These examples are discussed in Nobes (2004).
33
The earliest double-entry bookkeeping records in France and England
date from 1299 and 1305, respectively. However, both were isolated
examples kept by Italian ?rms of merchants in versions of Italian. Domestic
practice might instead be traced to translations of Pacioli’s tractatus on
bookkeeping, which were produced in France and England in the middle of
the 16th century (Coomber, 1956; Yamey, 1997).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 583
si?cations differ if based on presentation topics rather than
on measurement topics. d’Arcy (2001, p. 333) points out
that different topics would lead to different classi?cations,
and notes the inherent problem of using data on rules in-
stead of practices. Nobes (2011) mentions the need for
judgement in identifying important characteristics, and ex-
cludes some characteristics on these grounds.
We can now return to the meta-analysis of Table 2. We
have noted above that the early classi?ers might have been
particularly aware of US and UK differences, and that many
subsequent classi?cations were based on PW data which
had been originally designed to reveal such differences. This
couldexplainwhyno ‘Anglo-American’ groupwas generally
found.
The lessons of this section for accounting researchers
are that (i) a classi?cation should be based on detailed
observation of characteristics, (ii) the characteristics cho-
sen should ideally be informed by the purpose of the clas-
si?cation, and at least be deliberately chosen and overt,
(iii) related to this, any claims of objectivity are incoherent,
(iv) accounting practices are a better representation of an
‘accounting system’ than rules are, and (v) the set of com-
panies included in the accounting ‘system’ and the period
of the data should be speci?ed. It is further clear from this
survey that the effect of inclusions or exclusions of coun-
tries, sectors (especially ?nancial and extractive) and char-
acteristics needs to be empirically investigated in order to
see whether classi?cation is robust to manipulation of
these issues or whether it is instead essentially arbitrary.
We now proceed with that.
Data
Our sample
34
includes companies from the world’s
largest economies which use IFRS, as follows: (i) the
countries with the six largest stock markets where IFRS
was required from 2005 (i.e. Australia, France, Germany,
Italy, Spain and the UK), (ii) the two other countries with
large stock markets with a longer history of IFRS usage:
South Africa and Switzerland, (iii) Canada and South Kor-
ea, where IFRS has been recently adopted, and (iv) Hong
Kong and companies from China which use IFRS.
35
In or-
der to include Canada and South Korea, we use ?nancial
statements from 2011 onwards. In particular, we use
company reports for years ended 31 December 2011 (or
latest before) for 12 countries.
36
Our sample includes
the largest
37
listed companies in each of these jurisdic-
tions, which comprise 65%
38
of the total market capitali-
zation of these countries. Companies with foreign
in?uence or which are subsidiaries are excluded. In total,
we examine the IFRS practices of 514 companies. Details
of the sample are provided in Appendix B.
Table 4 shows the sample by country and sector. As ex-
plained, we wish to investigate whether exclusion of cer-
tain sectors might affect classi?cation. Prior literature
indicates that the ?nancial and extractive sectors have idi-
osyncratic policies.
39
Given the topic of this paper, we
should admit that the classi?cation of companies by sector
exhibits the dif?culties typical of classi?cation. There are
several accepted versions. We have chosen the ‘Industry
Classi?cation Benchmark’ (ICB) of the index company FTSE.
The corresponding data are from Worldscope (data code
WC07040). Consistently with our recommended approach,
we have used judgement to adjust it for our purposes, in
particular to calculate country totals for extractive
companies.
40
We concluded above that practices are the best
representation of an accounting system. We record the
IFRS practices of companies and use them as the character-
istics to be measured in order to classify a country. Even for
companies which are fully complying with IFRS, there is
considerable scope for varied practice because, for
example: (i) the recognition of expenses (e.g. impairments)
or assets (e.g. development projects) relies on the exercise
of judgement against somewhat vague criteria, (ii) the
measurement of liabilities (e.g. provisions) or assets (e.g.
the fair value of investment properties) involves estima-
tion, and (iii) many standards offer choices to companies.
The ?rst two of these are hard to assess (although, see an
attempt by Wehrfritz, Haller, & Walton, 2012), but data
on the third can be hand-picked from the annual reports
of companies. These data provide a good indication of the
in?uence of factors such as country and sector because
the differences in practices are caused by management
choices and not by regulations. There is some constraint
34
Our results show that the inclusion or exclusion of particular countries
generally does not affect how the remaining countries are classi?ed.
35
Although China has not fully adopted IFRS, the majority of the largest
listed Chinese companies prepares IFRS ?nancial statements, because they
are listed on the Hong Kong Stock Exchange (HKEx), which required IFRS
from 2005. Consequently, Chinese companies with a listing in Hong Kong
and Mainland China prepared two sets of ?nancial statements (IFRS and
Chinese GAAP). However, from 2010, HKEx accepts Chinese GAAP ?nancial
statements, and six companies in our sample have stopped preparing IFRS
?nancial statements.
36
The one exception to this is that we include the ?rst available IFRS
?nancial statements for ten Canadian and four South Korean companies
which have year-ends other than 31 December 2011. This enables the
inclusion of six Canadian and four South Korean ?nancial companies; in
particular, our Canadian sample would otherwise not include any bank
because all Canadian banks in our sample have 31 October year-ends.
37
Findings of country in?uence would probably be even stronger for
smaller companies because of less international in?uence (Nobes &
Perramon, 2013).
38
According to Worldscope data for 2011 (Worldscope code: WC07210).
39
Christensen and Nikolaev (2013) show that real-estate ?rms (which are
part of the ?nancial sector) choose to use fair value for investment property
more frequently than other ?rms. Jaafar and McLeay (2007, p. 180) refer to
special practices in the extractive industries.
40
We de?ne extractive companies as those in sector 0530 (oil and gas
producers), sector 1770 (mining), sub-sector 1753 (aluminum) and sub-
sector 1755 (non-ferrous metals); additionally, we classify Fortescue
Metals Group of sub-sector 1757 (iron and steel) as an extractive company.
We believe that using ICB codes results in a better industry classi?cation
than using primary SIC codes (Worldscope data code WC07021). If using
SIC codes, we would identify extractive companies as ‘mining’ (SIC codes
starting with the digits 10, 12, 13 or 14) and ?nancial companies as
‘?nance, insurance and real estate’ (SIC codes starting with the digit 6).
Using ICB codes, we classify 57 (129) companies as extractive (?nancial),
and using SIC codes we would have classi?ed 79% (94%) in the same way.
The main difference is that many integrated oil and gas companies (e.g. BP)
are not classi?ed as extractive but as ‘manufacturing’ using SIC codes due to
their petroleum re?ning businesses. Additionally, the classi?cation of some
companies using SIC codes is unsuitable for our purposes: e.g. China Oil?eld
Services is classi?ed as an extractive company even though it does no
extraction.
584 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
on changing policy,
41
but IFRS speci?cally removes re-
straints on choices made on ?rst-time use of IFRS (IFRS 1,
para. 11). As explained below, there are some rare examples
of national regulators, in the ?nancial sector, adding to IFRS
requirements.
The list of policy topics used by Kvaal and Nobes (2010)
is shown here as Table 3, after deleting the topics on which
choice was removed from IFRS by 2011. There are eight
presentation topics and six measurement topics. For the
last two topics in the table, changes to IFRS had already
been made by 2011 but were not compulsory for any of
our companies.
42
Kvaal and Nobes (2012) found that there
was little early adoption of IFRS changes, but we report on
this below. For ?nancial companies, Kvaal and Nobes
(2010) omitted topics on which there were sector-speci?c
presentation practices in?uenced by pre-IFRS laws. We do
not do that, because part of our purpose is to investigate
the effects of including or excluding certain sectors. Still,
we omit ?ve topics for ?nancial companies because they
are not appropriate for the sector, as explained in Appendix
C.
When the purpose of research is to investigate whether
IFRS policy choices are associated with country, then it is
appropriate to examine as many policy topics as are obser-
vable. However, as discussed earlier, for assessing a coun-
try’s accounting ‘system’ or a country’s pro?le of IFRS
practices, judgement is needed to exclude (or give lower
weight to) topics likely to be of little importance to users
of ?nancial statements (e.g. the liquidity order of assets
in a balance sheet). We investigate the sensitivity of classi-
?cations to such exclusion of topics.
Appendix C provides details about our data collection
and our coding procedures used to generate binary choice
data from the 14 IFRS policy topics. Our empirical analyses
below are based on a total of 5,689 hand-picked IFRS policy
choices of the 514 companies from 12 countries.
Findings on policy choice
Policy choices
Table 5 reports, by country, the percentages of compa-
nies in our sample which chose particular options. For sev-
eral topics, the policy choice was observable for all 514
companies. However, we only count companies for which
the policy is observable, which explains why the ‘N’ in
Table 5 is smaller for certain topics, notably investment
property measurement (topic 10).
Table 5 includes several countries for which data on
IFRS practices have not previously
43
been presented: China,
Hong Kong, South Africa, South Korea and Switzerland. Some
features of these countries stand out. First, the practices in
South Korea are unusually uniform: for most topics, over
90% of companies make the same choice, and for no topic
do fewer than 80% of companies make the same choice. Sec-
ondly, the majority of companies in China, Hong Kong, South
Africa and Switzerland chose to take actuarial gains/losses to
income, even though that practice was to be outlawed by a
change to IFRS which had already been issued. Similarly, a
majority of companies in South Africa and a large minority
in Switzerland chose proportional consolidation, even
though that practice was to be outlawed. This is further evi-
dence of no widespread early adoption of changes to IFRS,
and of the strong in?uence of pre-IFRS practices and there-
fore national patterns of IFRS practice, as documented in
the papers mentioned in the previous section.
Examples of sectoral effect
As explained earlier, some previous researchers ob-
served an association between pre-IFRS practices and sec-
Table 4
Sample by country and sector.
Sector AU UK CA CN HK FR ES IT DE CH ZA SK
P
0/1 Extractives 6 9 21 7 0 2 1 1 0 0 7 3 57
0/1 Other oil and gas, basic materials 4 3 2 3 0 2 3 0 7 3 1 4 32
2 Industrials 7 15 2 12 4 9 8 7 6 2 6 13 91
3 Consumer goods 2 10 2 3 3 7 1 5 7 3 2 5 50
4 Health care 2 3 0 1 0 2 1 1 2 4 2 0 18
5 Consumer services 8 20 8 4 3 7 3 8 6 0 6 4 77
6 Telecommunications 1 4 3 1 1 1 1 1 1 1 2 3 20
7 Utilities 1 5 1 3 3 2 4 3 2 0 0 2 26
8 Financials 20 22 10 13 8 6 9 12 6 6 6 11 129
9 Technology 0 2 0 2 0 2 1 0 2 1 0 4 14
P
51 93 49 49 22 40 32 38 39 20 32 49 514
This table reports descriptive statistics of the sample companies. The countries are Australia (AU), United Kingdom (UK), Canada (CA), China (CN), Hong
Kong (HK), France (FR), Spain (ES), Italy (IT), Germany (DE), Switzerland (CH), South Africa (ZA) and South Korea (SK). Sector is according to the ?rst digit of
the Industry Classi?cation Benchmark (ICB), except that we show all the extractive companies (sectors 0530 and 1770; sub-sectors 1753 and 1755; and
Fortescue Metals Group) together in the ?rst row, and all the remaining companies of sector 0 (oil and gas) and sector 1 (basic materials) together in the
second row.
41
IAS 8 imposes certain conditions and disclosure requirements (paras.
14 and 29).
42
IAS 19 was amended in 2011 to remove the option on the treatment of
actuarial gains and losses. IAS 31 was replaced in 2011, thus removing the
option of proportional consolidation. Both changes were only compulsory
for 2013 onwards.
43
Data on the other countries is included elsewhere; for example in
Nobes (2011).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 585
tor. In some cases, this was driven by sectors having differ-
ent accounting rules.
44
By contrast, apart from a few iso-
lated examples of additional jurisdiction-based rules,
45
there are no sector-speci?c accounting requirements in IFRS.
Even so, signi?cant differences in IFRS choices between sec-
tors are evident when we split our entire sample into three
sectors: ?nancial, extractive and other. Table 6 shows partic-
ularly clear examples from our data. More importantly for
this paper, there are also signi?cant differences between
the sectors within countries. As may be seen, compared to
other companies in their countries, Australian ?nancial com-
panies prefer fair value for investment properties, Canadian
?nancial companies prefer not to recognize actuarial gains/
losses as OCI (they prefer the corridor method), Canadian
extractive companies prefer to proportionally consolidate
joint ventures, and British ?nancial companies are less likely
to show net assets but more likely to start the balance sheet
with cash. In all these cases, a v
2
test of independence shows
that a null hypothesis of no association with sector can be
rejected at the 1% level. These are examples of how a coun-
try’s sectoral mix might affect its mean scores on topics,
which might then affect classi?cation, as examined in the
next section.
Sensitivity of classi?cations
Introduction
We have suggested above how the accounting classi?-
cations, particularly the early ones, appear to have been
sensitive to the nature of the classi?ers. In this section,
we empirically investigate how sensitive classi?cations
can be to various aspects of the data used.
As recorded in Table 1, several different statistical
methods of classi?cation have been employed by previous
researchers. We begin with principal component analysis.
Table 7 shows the principal components for one version
of the data: all 14 topics for all sectors of all 12 countries.
This analysis leads to a three-group initial classi?cation,
summarized as ‘run’ 1 in Table 8. In this table, we report
the results for 11 different versions of the data. For ease
of comparison, Germany is always shown in Group 1.
The Kaiser–Meyer–Olkin (KMO) measure of sampling
adequacy should be higher than 0.6 (or possibly 0.5) for
the data to be considered suitable for factor analysis (Kai-
Table 5
Percentages of policy choice by country and topic.
IFRS policy choice N AU UK CA CN HK FR ES IT DE CH ZA SK
1. Income statement by nature 385 35 11 5 44 36 29 96 81 24 29 15 3
2. Operating pro?t not shown 385 42 1 31 31 29 3 0 0 12 0 0 0
3. Equity pro?ts in operating 423 59 35 48 4 0 8 23 14 35 39 7 4
4. Balance sheet showing net assets 514 100 76 0 39 82 0 0 0 0 5 0 0
5. Balance sheet with liquidity decreasing 514 100 10 100 24 14 10 22 29 26 50 9 98
6. Indirect cash ?ows 514 4 98 100 98 100 100 91 95 100 95 66 100
7. Dividends received as operating 348 87 37 85 5 30 79 39 20 71 43 86 91
8. Interest paid as operating 381 86 61 74 44 43 79 52 69 61 64 96 89
9. Some property at fair value 504 10 10 2 0 5 0 0 0 0 0 0 0
10. Investment property at fair value 216 93 68 36 21 94 20 5 0 5 80 40 3
11. Some fair value designation 383 10 3 13 0 7 24 4 4 6 7 23 19
12. FIFO only 329 21 42 23 6 15 11 22 19 0 36 23 6
13. Actuarial gains/losses to OCI 414 85 89 72 8 36 60 68 30 59 35 28 83
14. Proportionate consolidation of JVs 379 6 25 55 9 0 71 70 38 17 43 59 17
This table reports the percentages of companies per country and topic which make the respective IFRS policy choice in 2011. The countries are as in Table 4.
N is the number of observations/companies. See Table 3 and Appendix C for details of the topics.
Table 6
Examples of sectoral differences in policy choice.
Country IFRS policy choice N % Financials % Extractives % Others p-value
All 10. Investment property at fair value 216 61 0 9
The process of classification is central to the daily task of doctors and librarians; and it is
the foundation of study and research in chemistry and biology. Double-entry bookkeeping
and the preparation of financial statements are classification activities of accounting practice.
Classifying national accounting systems has long been an aspect of accounting
research. This paper seeks to extract lessons for accounting researchers from anthropology,
biology, chemistry, cosmology and medicine. In particular, we examine how the classifiers
themselves and the characteristics that they choose can affect classification. We observe
that objectivity is neither possible nor desirable in classification. Despite the arbitrariness,
some classifications can be more reasonable or more useful than others. For previous
accounting classifications, we analyze the classifiers, the scope, the characteristics used,
the data and the classification techniques. We report various problems. We then empirically
investigate the sensitivity of classifications to such issues as the characteristics chosen,
and the countries and sectors included. For this, we hand pick data on the practices of
large listed companies from 12 jurisdictions relating to 14 accounting topics under International
Financial Reporting Standards. We show how different researchers could produce
different classifications,
How arbitrary are international accounting classi?cations?
Lessons from centuries of classifying in many disciplines,
and experiments with IFRS data
Christopher Nobes
a,b,?
, Christian Stadler
c,1
a
School of Management, Royal Holloway, University of London, Egham, Surrey TW20 0EX, UK
b
Discipline of Accounting, University of Sydney, Australia
c
Department of Accounting and Finance, Lancaster University Management School, Lancaster, Lancashire LA1 4YX, UK
Keywords:
Accounting choice
Meta-analysis
Sensitivity
a b s t r a c t
The process of classi?cation is central to the daily task of doctors and librarians; and it is
the foundation of study and research in chemistry and biology. Double-entry bookkeeping
and the preparation of ?nancial statements are classi?cation activities of accounting prac-
tice. Classifying national accounting systems has long been an aspect of accounting
research. This paper seeks to extract lessons for accounting researchers from anthropology,
biology, chemistry, cosmology and medicine. In particular, we examine how the classi?ers
themselves and the characteristics that they choose can affect classi?cation. We observe
that objectivity is neither possible nor desirable in classi?cation. Despite the arbitrariness,
some classi?cations can be more reasonable or more useful than others. For previous
accounting classi?cations, we analyze the classi?ers, the scope, the characteristics used,
the data and the classi?cation techniques. We report various problems. We then empiri-
cally investigate the sensitivity of classi?cations to such issues as the characteristics cho-
sen, and the countries and sectors included. For this, we hand pick data on the practices of
large listed companies from 12 jurisdictions relating to 14 accounting topics under Inter-
national Financial Reporting Standards. We show how different researchers could produce
different classi?cations, particularly depending on which accounting topics are used to rep-
resent the countries.
Ó 2013 Elsevier Ltd. All rights reserved.
Introduction
Classi?cation is a fundamental part of many disciplines.
The classi?cations of diseases and books are vital in the
daily tasks of medical practitioners and librarians, respec-
tively. The Linnaean and Mendeleev classi?cations are cen-
tral to learning and research in biology and chemistry.
Classi?cations have also been made in many other ?elds;
for example, languages (Ruhlen, 1991), economies
(Neuberger & Duffy, 1976), political systems (Shils, 1966),
and legal systems (David & Brierley, 1985). Members of
society are also put into classes, e.g. recently in the UK (Sa-
vage et al., 2013). In all cases, the fundamental purpose of
the classi?cation is to simplify (Rudner, 1966).
The everyday work of accountants involves recording
transactions in the classi?cation system that is double-en-
try bookkeeping. The ?nancial statements which result are
also classi?cations: for example, assets are classed as
non-current or current; the former are then sub-classed
as tangible, intangible or ?nancial (Gröjer, 2001). The
classi?cations are debatable: in the income statement,
should expenses be classi?ed by nature or by function?
0361-3682/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.aos.2013.10.001
?
Corresponding author at: School of Management, Royal Holloway,
University of London, Egham, Surrey TW20 0EX, UK. Tel.: +44 1784
414120.
E-mail addresses: [email protected](C. Nobes), c.stadler@lancaster.
ac.uk (C. Stadler).
1
Tel.: +44 1524 592221.
Accounting, Organizations and Society 38 (2013) 573–595
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
Some classi?cations are metaphysical: the split of equity
?nancial assets into trading or available-for-sale rests
2
not on any observable characteristic, nor even on the real
intentions of managers, but on the declared intentions of
managers.
Classi?cation has also been applied in the ?eld of inter-
national accounting. Just as in other ?elds, classi?cation
has been used to assist understanding of how the many
different objects (in this case, accounting systems) are re-
lated. We explain in more detail below how classi?cation
of accounting systems can be relevant to accounting prac-
tice and research. We use the term ‘accounting system’ to
refer to a set of accounting practices, i.e. policies on recog-
nition, measurement and presentation as used in a com-
pany’s published ?nancial statements. For example, each
individual listed company in the USA has its own account-
ing practices. However, the accounting of all the companies
has many shared characteristics, imposed and enforced by
the Securities and Exchange Commission. The individual
examples of US accounting share so much in common that
they could be said to comprise a ‘system’: the US GAAP
accounting system. Another precise, but quite different,
system is French GAAP as used for unconsolidated ?nancial
statements in France. A country can exhibit more than one
system. For example, although a national GAAP (such as
French GAAP) is still used for unconsolidated ?nancial
statements in most EU countries, the consolidated state-
ments of listed companies are now prepared using Interna-
tional Financial Reporting Standards (IFRS).
The US and French ‘systems’ contain few overt
3
options.
However, partly because of international political negotia-
tions (Camfferman & Zeff, 2007, chap. 5), many options were
included in IFRS; although these are gradually being re-
moved. Therefore, transition from French (or German etc.)
GAAP to IFRS has increased the variation in accounting prac-
tices within a country. Even so, when considering the op-
tions, national factors (including such matters as tax and
legal systems) can still affect a company’s choice. Ball
(2006, p. 15) explains how, even if all entities are complying
with IFRS, the incentives of preparers and enforcers remain
‘primarily local’. As a result, one can discern national pat-
terns of IFRS practice (Kvaal & Nobes, 2010). These could
be seen as different ‘systems’ of generic IFRS. We classify
such systems in the empirical part of this paper.
Many international classi?cations of accounting sys-
tems have been proposed, beginning more than a century
ago. An examination of other ?elds (see below) suggests
that a classi?cation might re?ect its classi?er, and that
the process of classi?cation is awash with judgements of
various kinds. Few accounting classi?ers have even dis-
cussed, let alone investigated, the sensitivity of classi?ca-
tions to changes in the nature of the classi?ers, the
number of objects being classi?ed (countries or accounting
systems), the nature and number of characteristics used to
measure the objects, or the type of companies (e.g. corpo-
rate sectors) included. This paper discusses and empirically
investigates these issues.
Our ?rst objective is to investigate the ways in which
classi?cations in ?elds other than accounting (i.e. anthro-
pology, biology, chemistry, cosmology and medicine) have
been affected by issues analogous to those in the previous
paragraph and how classi?cations have changed dramati-
cally over time. We seek lessons for assessing the robust-
ness of accounting classi?cations.
Our second objective is to apply these lessons. We
examine previous accounting classi?cations, especially to
record the number of countries classi?ed, the number
and type of characteristics used to classify them, and
whether industry sectors were discussed or excluded. We
?nd that the early classi?cations re?ect the classi?ers; in
particular, they vary by the national backgrounds of the
classi?ers. We then ?nd that some classi?ers apparently
used no data, and most of the rest used data collected by
others for other purposes. Few classi?ers speci?ed the date
or the scope of their classi?cations (e.g. was it limited to
listed companies or to non-?nancial companies?), and
few speci?ed a clear purpose.
Our ?nal objective is to investigate empirically whether
accounting classi?cations are anything other than arbi-
trary; whether they can easily be manipulated to back up
particular arguments. To do this, we hand pick data on
the practices in 2011, under International Financial Report-
ing Standards (IFRS), of a large sample of listed companies
from 12 jurisdictions. For these companies, we examine
the observable accounting policy choices on 14 topics,
including presentation issues (e.g. choice of format for
the income statement) and measurement issues (e.g. use
of cost or fair value for investment property). Our analyses
are based on a total of 5689 choices of 514 companies. We
?nd that our classi?cations are highly sensitive to changes
in the set of characteristics measured (i.e. the IFRS policy
topics, in our case), and this is a feature common to classi-
?cations in other ?elds such as biology. However, certain
aspects of the classi?cations are remarkably stable, e.g.
Italy and Spain are always in the same group, and never
with the UK. Furthermore, with minor exceptions, the clas-
si?cations are much more robust to the exclusion of indi-
vidual countries or sectors. We therefore conclude that
our classi?cations based on IFRS choices are not essentially
arbitrary. Nevertheless, our classi?cations could be used to
support or refute the in?uence on accounting of the code/
common legal dichotomy.
The advantage of using the above policy topics is that
IFRS speci?cally allows management to choose among
the options.
4
There is therefore scope for country and indus-
try in?uences to lead to varied practice, unlike the setting of
previous studies of pre-IFRS accounting where management
is constrained by national rules, which can also vary by
industry. Our experiments deal with the accounting policy
choices of actual companies but, for the purposes of other
accounting research, different objects should be classi?ed.
For example, in an analysis of corporate reporting regula-
tion, Leuz (2010) classi?es countries on the basis of facts
and impressions about legal systems and securities laws.
2
As under IAS 39, para. 9.
3
As explained later, we use this term to describe policy options that have
been deliberately inserted into accounting rules.
4
Some caveats will be entered later, but our policy topics are presented
in the IFRS documents as free choices.
574 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
The purpose of a study should guide the choice of character-
istics measured (Roberts, 1995); different accounting classi-
?cations are suitable for different purposes.
The paper contributes by drawing together relevant les-
sons on classi?cation from ?elds other than accounting; by
conducting the ?rst meta-analysis of accounting classi?ca-
tions; by applying the lessons from other ?elds when ana-
lyzing the accounting classi?cation literature with a new
focus on the nature of the classi?ers and of the character-
istics, sectors and countries included; and by assessing the
reliability of previous classi?cations through empirical
investigation of the sensitivity of classi?cation to varia-
tions in such factors, thereby revealing the dramatic effect
of inclusions/exclusions on classi?cations. Our purpose is
not to present a classi?cation but, as a by-product of our
work, we provide data on IFRS practices for the ?rst time
for several jurisdictions (i.e. China, Hong Kong, South Afri-
ca, South Korea and Switzerland), and we provide classi?-
cations which include these countries (and Canada) for the
?rst time.
Our ?ndings on the reliability of classi?cations are
important because hundreds
5
of academic papers refer to
the classi?cations as part of motivating research (Ball,
Kothari, & Robin, 2000; Gray, 1988; O’Donnell & Prather-
Kinsey, 2010; Saudagaran & Biddle, 1995) or to justify an
independent variable (type of accounting system) which is
expected to in?uence issues such as value relevance (e.g.
Ali & Hwang, 2000). Then, there are new uses for classi?ca-
tions, as explanations of which companies volunteer to
adopt IFRS (Tarca, Morris, & Moy, 2013), how jurisdictions
respond to IFRS (Sellhorn & Gornik-Tomaszewski, 2006;
Tyrrall, Woodward, & Rakhimbekova, 2007), how countries
change from one class to another (Xiao, Weetman, & Sun,
2004), how practices on major topics vary over time (Ding,
Richard, & Stolowy, 2008), how companies respond to the
choices available in IFRS (Nobes, 2011), why the amount of
lobbying on IFRS varies by country (Orens, Jorissen, Lybaert,
& van der Tas, 2011), by how much various countries’
domestic accounting requirements vary from IFRS (Ding,
Hope, Jeanjean, & Stolowy, 2007), or how to identify coun-
tries with similar backgrounds when selecting countries
for study (Delvaille, Ebbers, & Saccon, 2005). If the classi?ca-
tions are inappropriate, the research setting or the variables
will be questionable.
For ?nancial analysts, students and policy makers, the
classi?cations are a convenient way of simplifying and
summarizing. So, again, inappropriate classi?cations are
likely to be misleading. For instance, much of the argumen-
tation on the development of new standards is political
(Harrison & McKinnon, 1986), and is now often expressed
in terms of resisting ‘Anglo-American’ accounting. As an
example, German writers have seen the international stan-
dard-setters as a Trojan horse which conceals Anglo-Amer-
ican accounting (Kleekämper, 2000) or as ‘the unknown
enemy from London’
6
(Hennes & Metzger, 2010). Botzem
and Quack (2009) believe that the history of the Interna-
tional Accounting Standards Committee has been wrongly
reported as ‘an Anglo-American success story’ (p. 991). How-
ever, as will be shown, some classi?ers deny the existence of
Anglo-American accounting.
Power (2009) warns researchers not to exaggerate the
international differences (p. 325) and to be wary of resort-
ing to cultural variables to explain them (p. 331). On the
?rst point, Power notes that we can only talk of different
arrangements of balance sheets because all companies
present balance sheets and show very similar things in
them. This paper reinforces those warnings by showing
that some ?ndings about international differences are
unreliable and that classi?cation could be used to prove
or disprove the importance of one commonly used cultural
variable: the legal system. Further, although classi?cations
do identify countries which are different, their main pur-
pose is to group together countries on the basis of their
similarities.
The paper proceeds as follows. The next section shows
how classi?cation in various ?elds can depend on who is
doing the classifying. Then we examine the degree to
which classi?cation depends upon the characteristics cho-
sen to measure the objects being classi?ed, and upon the
de?nition of the characteristics. After that, we perform a
meta-analysis of previous accounting classi?cations, and
we analyze them in order to reveal the apparent effects
on classi?cations of the classi?ers and the potential effects
of various other factors, such as the countries/systems in-
cluded and the characteristics chosen to represent them.
The next four sections report on our empirical investiga-
tions: the data, the ?ndings on policy choice, the analysis
of sensitivity of classi?cation to the manipulation of vari-
ous factors, and the presentation of some new classi?ca-
tions. Then, there are conclusions.
Goldilocks and the forebears
This section examines the degree to which classi?cation
is determined by who is classifying. Bloor (1982, p. 268)
found new support for the claim of Durkheim and Mauss
(1903) that the classi?cation of things reproduces a pattern
of social arrangements more than a pattern of the things. He
argued that even such renowned scientists as Newton and
Boyle were affected by their religious and political ideals
and ‘were arranging the fundamental laws and classi?ca-
tions of their natural knowledge in a way that artfully
aligned them with their social goals’ (p. 290). The ?elds of
cosmology and anthropology are used as examples below.
Cosmology
Throughout most of recorded history, man
7
saw himself
as the unique peak of creation (see below). He lived in a
world which was also in a class of its own, being ?xed and
at the center of the universe. The Copernican revolution,
set in motion in 1543 by the publication of the book
commonly known as the ‘Revolutions’,
8
spread slowly. For
5
As examples from Table 1, Nair and Frank (1980) has 228 citations and
Nobes (1983) has 390 (according to Google Scholar, accessed on 15.4.2013).
6
‘Der unbekannte Feind aus London’.
7
We use the term ‘man’ when discussing authors who did so (i.e. those
until the late twentieth century).
8
De revolutionibus orbium coelestium (On the revolutions of the heavenly
spheres).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 575
espousing it, Galileo was held under house arrest from 1633
to his death in 1642. Even in unorthodox Amsterdam, Joan
Blaeu’s ‘New and very accurate map of the whole world’ of
1662 still gave equal status to Ptolemy’s geocentric beliefs
and heliocentrism (Brotton, 2012, p. 288). However, enlight-
enment eventually reduced anthropocentrism: the earth is
now classi?ed as a planet (i.e. something that moves) orbit-
ing a star which is rather far from the middle of one of many
galaxies. The planet is fairly small, but happens to be in the
Goldilocks zone: at the right distance from its star to be at a
congenial temperature for water-based life forms.
Anthropology
The ‘great chain of being’, derived from Aristotle and
conventional for millennia, is a six-group classi?cation
9
(Lovejoy, 1964). Man is not classi?ed as an animal at all
but as a special creation which is a little lower than the an-
gels. They had spirit only, animals had body only, but man
had both. Man saw himself as unique: not just sui generis
but hors de catégorie. In the eighteenth century, Linnaeus
took homo sapiens down a rung by placing him in the animal
kingdom, though he remained sui generis. The descent of
man continued in the nineteenth century when Darwin out-
rageously suggested evolution from more ‘primitive’ prima-
tes, presumably without spirits; and other types of humans
joined his genus, such as homo neanderthalensis. In the twen-
tieth century, the genus got more crowded, for example with
the arrival of homo ?oresiensis. In the twenty-?rst century,
Wildman, Uddin, Liu, Grossman, and Goodman (2003) went
yet further by proposing
10
that, since modern humans share
99.4% of non-synonymous DNA with chimpanzees, homo
sapiens is a parvenu member of their genus.
Classi?cation and standards
The previous section showed how classi?cation can de-
pend on the mindsets of those doing the classifying, and
how classi?cation can therefore change dramatically over
time without the objects changing. This section examines
the degree to which classi?cation depends upon the char-
acteristics chosen to measure the objects being classi?ed,
and on the de?nition of the characteristics. Foucault
(1970, p. 125) suggested that modernity in science begins
with privileging observation, starting with Roger Bacon.
Sight must replace reliance on ‘self-evident’ axioms. It also
replaces hearsay evidence about sightings, and it is given
greater weight than the less reliable senses of taste, smell
and touch (p. 132). The invention of the telescope and
the microscope helped greatly. It is observation which
guided Copernicus and Galileo, and Linnaeus and Darwin.
However, not even everything visible is relevant and reli-
able: color is not (p. 133). When Linnaeus classi?ed plants,
he used only four observable features: the shape of
elements, the quantity of the elements, their arrangement
related to each other, and their relative magnitudes.
However, there was still much scope in deciding which ele-
ments to observe, as will be explained below. In a book
whose title could be translated
11
as ‘To Think, to Classify’,
Perec (1985) discusses how books can be classi?ed by, inter
alia, alphabetical order of author or title, country of author
or publication, color, date of publication or acquisition, lan-
guage, priority for reading, and so on (p. 39). The ?elds of
cosmology, chemistry, biology and medicine are now used
as examples.
Cosmology
Whether or not a celestial body is classed as a planet
depends, like any classi?cation, on de?nitions (Gröjer,
2001) or standards.
12
The ‘standard’ for a planet was revised
by the International Astronomical Union in 2006 (IAU,
2006), with the result that Pluto (which had only become
a planet, as far as we were concerned, on its discovery in
1930) ceased to be one. The revision was caused by the dis-
covery of bodies larger than Pluto with orbits further from
the sun. The re-de?nition of a planet and the re-classi?ca-
tion of Pluto has both scienti?c and cultural implications
(Basri & Brown, 2006), though not as large as those that
led to the arrest of Galileo for professing the planetary status
of the earth. An important implication for other classi?ca-
tions (e.g. in accounting) is that an object’s place in a classi-
?cation can depend on the range of objects being classi?ed.
Chemistry
Some alchemists had classi?ed elements into solids,
liquids and gases, but this is now seen to produce an
unhelpful classi?cation of such liquids as mercury, molten
lead and liquid nitrogen. So, chemists moved onto observ-
ing various behaviors of elements (e.g. reaction to oxy-
gen), leading to Mendeleev’s periodic table (Aldersley-
Williams, 2011). This approach was later con?rmed by a
more fundamental one (called ‘natural’ in the next para-
graph) when it became possible to count protons, neu-
trons and electrons.
Biology
Linnaeus started his classifying with plants, perhaps be-
cause their characteristics are more easily observable than
such things as the structure of the inner ear of animals
(Foucault, 1970, p. 137). However, he chose to ignore dif-
ferences in leaves, stems and roots, such that the ‘primary
arrangement of the vegetables
13
is to be taken from the
9
God, angels, man, animals, plants and minerals.
10
This proposal has not been generally accepted. For example, Steiper and
Young (2006, p.385) still treat homo and pan as different genera.
11
A translation of ‘Penser/Classer’ was published in 2009 by Godine Press
of Boston under the less literal title of ‘Thoughts of Sorts’.
12
At ?rst sight, the word ‘standard’ has a different meaning in ?nancial
reporting from that used here. It appears to refer to a type of regulation.
Elsewhere in accounting, a ‘standard cost’ ?ts more obviously into the
normal scienti?c meaning. However, the documents issued by the Inter-
national Accounting Standards Board (IASB), for example, are not in
themselves requirements. The IASB is a private sector standard setter. A
regulator such as the European Union can choose to require certain
companies to comply with a standard.
13
That is, plants; Linnaeus classi?ed all things on earth as animal,
vegetable or mineral.
576 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
fruit-body
14
alone’ (Linnaeus, 1751, section 164). In other
words, his system was essentially arbitrary. Whereas the
features of living plants were easy to observe, plants lacked
a fossil record on which to base the evolutionary approach
that was adopted fairly early on for animal classi?cation.
However, analysis of plant DNA has recently solved this
problem and led to a transformation of the botanical classi-
?cation to something ‘natural’, i.e. related to the thing caus-
ing the variation (e.g. Duff & Nickrent, 1999).
Classi?cation of animals has a long history. Socrates
classi?ed man as a ‘featherless biped’, but his pupil Plato
was mocked by Diogenes for repeating it. Ironically, biolo-
gists still include humans and birds in a super-class of tetra-
poda. However, at amoredetailedlevel, humans arenot now
classedwithbirds but withdogs anddolphins (whichosten-
sibly have four feet and no feet, respectively). Looking more
deeply, one can observe ?ve ?ngers not only on a human
handbut alsoona dog’s front pawandinsidea dolphin’s ?ip-
per. Several other mammalian shared characteristics can be
identi?ed, such as giving birth to live young.
As with plants, Linnaeus classi?ed animals by observing
shared characteristics, but the result again depends upon
which characteristics are chosen. As a result, many of Lin-
naeus’ animal classi?cations have alsobeenoverturned. Clas-
si?cation now rests on a search for homologs, which are
shared characteristics inherited from a common ancestor,
suchas seeninthe hand, pawand ?ipper. Ineffect, zoological
classi?cation is nowentirely about descent. For this purpose,
the analysis of DNA became a powerful tool as a supplement
to, andsometimes as a contradictionof, the receivedfossil re-
cord (Stringer, 2011, chap. 1). Again, the zoological classi?ca-
tion is now regarded as ‘natural’ (i.e. less arbitrary, being
based on evolutionary relationships as evidenced by DNA).
However, a caveat should be entered. The biologists’
classi?cations take no account of different possible pur-
poses. For example, if the purpose were to help in planning
the habitats or menus for a newzoological park, it might be
more useful to classify a dolphin with a shark even though
the dolphin is much more closely related to a dog, a human
or even a pterodactyl.
15
Diseases
Medicine is a practical activity, which relies heavily on
the International Classi?cation of Diseases (ICD). This has
been in operation for a century but is revised approximately
everydecade(Bowker &Star, 2000, p. 136). The ICDis a prag-
matic tool with a clear purpose: it helps doctors to identify
diseases and then to record information about patients.
Whereas biologists nowclassify in a monothetic way, using
binary characteristics (e.g. backbone or not), the ICD looks
for a number of shared characteristics (a polythetic system).
Further, whereas chemical elements do not change,
16
and
animal species change very slowly, diseases change rapidly.
Lastly, the ICDsometimes needs to be dramatically expanded.
It beganamong sixEuropeancountries, but hadtobe adjusted
when African and Asian diseases were included (Bowker &
Star, 2000, p. 151). Many of these features remind one of clas-
si?cations of accounting systems: they are polythetic, the sys-
tems change rapidly, and the classi?cations started with
Europe and North America only.
What’s in a name?
A more alarming point must also be made: no classi?ca-
tions are ‘real’. As Buffon pointed out in 1749:
The more we increase the number of divisions in the pro-
duction of nature, the closer we shall approach to the
true, since nothing really exists in nature except individ-
uals, andsincegenera, orders andclasses exist onlyinour
imagination [as cited in Foucault, 1970, p. 146].
We noted earlier that the de?nition of a planet is a mat-
ter of opinion. In biology, it is notable that neither Darwin
nor any follower has set out a de?nition of ‘species’ which
has gained general acceptance. Linnaeus thought that spe-
cies were ?xed in number, immutable in nature and divinely
created. Darwin showed that the ?rst two points were er-
rors, and drew a polite veil over the third. However, we
can now put another interpretation on the origin of species:
they evolved in the brain of homo sapiens. The lack of de?ni-
tions explains why there is debate about whether Neander-
thals and modern humans are part of the same species
(given that they have successfully interbred),
17
and whether
humans are part of the chimpanzee genus. Buffon’s insight
has not yet been taken to its logical conclusion, but the com-
plete abandonment of the apparatus of species, genera, etc. is
being contemplated by biologists (Mishler, 2009, p. 65).
Nevertheless, it seems unlikely that biologists or any
other humans will be able to give up classifying: ‘to classify
is human’ (Bowker & Star, 2000, p. 1). Lévi-Strauss (1958)
suggests that we inevitably perceive the world in terms
of binary opposites, and he encourages a search for under-
lying structures. For example, when discussing the content
of myths, Lévi-Strauss notes that ‘this apparent arbitrari-
ness is belied by the astounding similarity between myths
collected from widely different regions’ (p. 208). Further-
more, the fact that no classi?cations are real does not mean
that classi?cation cannot be useful. For librarians or doc-
tors, various competing classi?cation systems could be al-
most as useful as each other. For example, the Dewey
Decimal system and the Library of Congress system both
work satisfactorily in libraries. Linnaeus’ initial botanical
classi?cation was also of practical use in organizing
information, even though it was arbitrary. However, some
classi?cations might be more useful than others. For exam-
ple, Mendeleev’s classi?cation in chemistry was much
more useful than some earlier classi?cations because it
identi?ed ‘missing’ elements and predicted what they
would be like.
14
That is, the reproductive system.
15
The four types of animal in this sentence other than the shark are all in
the tetrapod clade.
16
Elements cannot be changed by chemical reactions. They can be
changed by (and indeed were formed by) nuclear reactions. Thus, gold is
created from other elements such as base metals (ultimately from
hydrogen) and it could be used to create even heavier elements, but this
does not change the nature, de?nition or ‘standard’ of gold.
17
Modern humans, except for sub-Saharan Africans, contain traces of
Neanderthal DNA; up to 4% in some cases (Green et al., 2010).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 577
Accounting classi?ers can learn from these ?elds. One
relevant lesson from above is the need for detailed per-
sonal observation. Another is that the purposes of a classi-
?cation should be considered. Further, classi?ers should be
deliberate about the characteristics measured; Roberts
(1995, p. 641) shows ‘the incoherence of taxonomies
which rely upon appeals to objectivity’. We apply these
lessons below, while analyzing past accounting
classi?cations.
Analysis of previous accounting classi?cations
There have been many international classi?cations of
accounting, as summarized in Table 1.
18
Several (i.e. items
2, 3, 4, 10, 14 and 15 of Table 1) relate to in?uences on
accounting rather than to accounting itself. Roberts (1995)
calls the former ‘extrinsic’ and the latter ‘intrinsic’; or they
could be called deductive and inductive. This section ?rst
performs a meta-analysis on these classi?cations and then
examines the apparent effects on classi?cation of the classi-
?ers, and the potential effects of various other factors, such
as the countries/systems included and the characteristics
chosen to represent them.
A meta-analysis
Meta-analysis is a procedure which mathematically
integrates the results of previous independent studies. It
can reduce the importance of unbiased errors in the data
or the procedures of particular individual studies. Meta-
analysis is frequently used in medical research, in which
context Egger, Smith, and Phillips (1997) note that atten-
tion must be paid to the selection and weighting of previ-
ous studies. For our meta-analysis of accounting
classi?cations, we include all the studies of Table 1 with
equal weights, in the absence of any objective alternative.
Our analysis covers the 15 countries which host the
world’s largest economies and which have been included
in previous classi?cations. The countries included are
those for which we collect IFRS data for our experiments
below (except China), plus the largest remaining countries:
Brazil, India, Japan and the United States. Russia is ex-
cluded because it was only found in two previous classi?-
cations, and not in terms of published ?nancial reporting.
China is excluded because it was not in any of the former
classi?cations. As will be explained in the next paragraph,
ours is not a traditional meta-analysis which combines
studies by signi?cance levels (see e.g. Christie, 1990), be-
cause the results of classi?cation studies are the groupings
of countries and not signi?cance levels.
Table 2 shows the meta-analysis: the bottom-left trian-
gle relates to all the classi?cations, the top-right triangle to
the intrinsic ones only. For each pair of countries, the ?gure
shown is the percentage of the classi?cations which placed
that pair in the same group. The bracketed number shows
how many classi?cations included the pair. For example,
the bottom-left pairing of Japan and the US shows that
those countries were together for 40% of the ten classi?ca-
tions which included them both. Scores of 0% or 100% re-
veal consensus among the classi?cations. The table also
shows which percentages for country-pairs are signi?-
cantly different from 50%, based on a test of proportion
(two-sided). A signi?cant result indicates a high degree of
con?dence that the relationship of the countries in the pair
(either being or not being grouped together) is not arbi-
trary. Although the various classi?cations consider differ-
ent numbers of countries and result in different numbers
of groups,
19
our method of analyzing country-pairs allows
the combination of these different classi?cations into a
meta-analysis.
The meta-analysis can be summarized as showing two
main features. First, most of the percentages for the coun-
try-pairs are not signi?cantly different from 50%,
20
which
suggests a high degree of arbitrariness in the classi?cations;
however, many relationships are not arbitrary. Similar con-
clusions can be drawn from observing that there are many
scores from 33% to 67%. For example, the results for Italy
in the bottom-left triangle reveal that there is little consen-
sus concerning which countries it should be classi?ed with
(because there are no percentages above 50% which are sta-
tistically signi?cant). On the other hand, there is strong con-
sensus that Italy should not be classi?ed with ‘Anglo’
countries (see the six percentages below 50% which are sta-
tistically signi?cant). Similar remarks apply to France, Spain
and Germany. Secondly, a British group can be identi?ed,
which includes Australia and Hong Kong (see the UK column
and row in the bottom-left triangle). However, North Amer-
ica is not included in that group: the ?rst column of Table 2
shows that only Canada has usually been classi?ed with the
United States. The ?rst row (intrinsic classi?cations only)
shows an even lower tendency for there to be an ‘Anglo-
American’ group.
Several caveats must be entered about this meta-analy-
sis. First, it uses data (i.e. the classi?cations) spanning sev-
eral decades, during which countries might have changed
their relationships. This and other reasons might mean that
the various results should not have been combined. Never-
theless, to the extent that certain pairs of countries retain
their relative positions over many decades (even surviving
a move to IFRS) suggests that the classi?cations are picking
18
d’Arcy (2001) lists some further papers, which we exclude on the
grounds that they replicate or overlap previous papers (e.g. Mueller (1968)
overlaps Mueller (1967), Nair (1982) replicates Nair and Frank (1980), and
Salter and Doupnik (1992) overlaps Doupnik and Salter (1993)); or they are
about the style of rule-making (e.g. Daley and Mueller (1982, for which
d’Arcy references its 1989 re-printing) and AlNajjar (1986)); or their
purpose is not to present a classi?cation (e.g. Previts (1975) provides
criteria, Gray (1988) builds a theory, and Cooke and Wallace (1990) test a
developed/developing country hypothesis).
19
In two of the classi?cations (Nobes (1983) and Doupnik and Salter
(1993)), countries were ?rst divided into two groups, and those were sub-
divided further. For the meta-analysis, we used the two-group classi?ca-
tions (see the footnote of Table 2), which stress similarities rather than
differences. Compared to using the multi-group classi?cations, this
increases the scores in Table 2 for several country-pairs.
20
61% of the scores in the bottom-left triangle and 75% in the top-right
triangle (64 out of 105 country-pairs and 76 out of 101 country-pairs,
respectively; only 101 country-pairs are considered for the top-right
triangle because there are four cases where the country-pair is only
included in one classi?cation but the test requires at least two observa-
tions). The main reason for the lower frequency of signi?cant scores in the
top-right triangle is the reduced power of the tests due to considering fewer
classi?cations.
578 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
Table 1
Features of some classi?cations.
1. Researchers 2. No. of
countries
3. Range of companies (e.g.
sectors, large, listed)
4. Date of data 5. No.
of
topics
6. Type of data 7.Classi?cation
method
8. Classi?cation type
1. Hat?eld (1911) 4 Unspeci?ed Unspeci?ed, c. 1910 0 Impressions of practices Judgement 3 Groups
2. Mueller (1967) 5 Unspeci?ed Unspeci?ed, c. 1965 1 Impressions of purposes Judgement 4 Unconnected groups
3. Seidler (1967) 13 Unspeci?ed Unspeci?ed, c. 1965 1 Impressions of in?uences Judgement 4 Unconnected groups plus
other mentioned countries
4. AAA (1977) 6 Unspeci?ed Unspeci?ed, c. 1975 1 Impressions of in?uences Judgement 5 Unconnected groups
5. da Costa et al.
(1978)
38 Unspeci?ed Unspeci?ed, c. 1973 100 Mixture of rules and impressions of
practices (by Price Waterhouse
partners)
PCA 2 Unconnected groups
6. Frank (1979) 38 Unspeci?ed Unspeci?ed, c. 1973 233 As above PCA, MDS 4 Unconnected groups
7. Nair and Frank
(1980)
38, 46 Unspeci?ed Unspeci?ed, c. 1973 and
c. 1975
233,
264
As above PCA, SSA 4/5 Unconnected groups for
measurement; 7 for disclosure
8. Goodrich (1982) 64 Unspeci?ed Unspeci?ed, c. 1979 26 Impressions of concepts (by Price
Waterhouse partners)
PCA 5 Unconnected groups
9. Nobes (1983) 14 Listed 1980 9 Impressions of practices PCA Hierarchy of 2 groups, leading
to 6 groups
10. Puxty, Willmott,
Cooper, and Lowe
(1987)
4 Unspeci?ed Unspeci?ed, c. 1985 3 Impressions of regulatory style Judgement Positions of the countries with
respect to three regulatory
ideals
11. Shoenthal (1989) 2 Unspeci?ed Unspeci?ed, c. 1987 1 Impressions of competencies of
auditors
Judgement 2 Unconnected groups
12. Doupnik and Salter
(1993)
50 Economically signi?cant
entities
1990 114 Impressions of practices (by
academics and auditors)
Average-
linkage
clustering
Hierarchy of 2 groups, leading
to 9 groups
13. d’Arcy (2001) 14 + IASC Listed; consolidated and
unconsolidated
Unspeci?ed, based on
Ordelheide and Semler
(1995)
129 Rules Clustering,
MDS
4 Groups with MDS
14. Leuz et al. (2003) 31 Listed Based on La Porta et al.
(1998)
9 Facts and impressions relating to
stock markets and investor protection
Clustering by
k-means
3 Groups in order
15. Leuz (2010) 37, 49 Listed ‘2000s’ 13 Facts and impressions on legal
system, securities regulation
Clustering by
k-means
3 Groups, then 5 groups
16. Nobes (2011) 8 Large, listed, consolidated,
excluding ?nancials for some
topics
2008/9 13 Practices PCA, MDS,
clustering
3 Groups by PCA; hierarchy
starting with 2 groups
Key: PCA = principal component analysis. MDS = multi-dimensional scaling. SSA = smallest space analysis.
C
.
N
o
b
e
s
,
C
.
S
t
a
d
l
e
r
/
A
c
c
o
u
n
t
i
n
g
,
O
r
g
a
n
i
z
a
t
i
o
n
s
a
n
d
S
o
c
i
e
t
y
3
8
(
2
0
1
3
)
5
7
3
–
5
9
5
5
7
9
Table 2
Meta-analysis of classi?cations: percentages with which pairs of countries are grouped together.
US % (N) AU % (N) UK % (N) CA % (N) HK % (N) FR % (N) ES % (N) IT % (N) DE % (N) CH % (N) ZA % (N) SK % (N) BR % (N) IN % (N) JP % (N)
US 29 (7) 22 (9) 83 (6) 50 (2) 13
Ã
(8) 14 (7) 17 (6) 25 (8) 20 (5) 20 (5) 50 (2) 20 (5) 25 (4) 43 (7)
AU 40 (10) 75 (8) 33 (6) 50 (2) 0
Ã
(8) 0
Ã
(8) 0
Ã
(7) 0
Ã
(8) 0
Ã
(5) 80 (5) 0 (2) 20 (5) 25 (4) 14 (7)
UK 33 (15) 82
Ã
(11) 50 (6) 100 (2) 22 (9) 0
Ã
(8) 0
Ã
(7) 11
Ã
(9) 20 (5) 100
Ã
(5) 0 (2) 0
Ã
(5) 0
Ã
(4) 0
Ã
(7)
CA 88
Ã
(8) 50 (8) 63 (8) 100 (2) 17 (6) 0
Ã
(6) 0
Ã
(5) 17 (6) 0
Ã
(4) 50 (4) 0 (2) 0
Ã
(4) 0 (3) 33 (6)
HK 75 (4) 75 (4) 100
Ã
(4) 100
Ã
(4) 50 (2) 0 (2) 0 (2) 0 (2) 0 (1) 100 (2) 0 (2) 0 (2) 0 (1) 0 (2)
FR 8
Ã
(13) 0
Ã
(11) 14
Ã
(14) 13
Ã
(8) 25 (4) 63 (8) 71 (7) 78 (9) 80 (5) 20 (5) 50 (2) 40 (5) 25 (4) 43 (7)
ES 10
Ã
(10) 0
Ã
(10) 0
Ã
(11) 0
Ã
(8) 0
Ã
(4) 64 (11) 86 (7) 50 (8) 60 (5) 0
Ã
(5) 50 (2) 60 (5) 50 (4) 57 (7)
IT 13
Ã
(8) 0
Ã
(9) 0
Ã
(9) 0
Ã
(7) 0
Ã
(4) 56 (9) 78 (9) 71 (7) 50 (4) 0
Ã
(5) 50 (2) 60 (5) 50 (4) 50 (6)
DE 15
Ã
(13) 0
Ã
(10) 7
Ã
(14) 13
Ã
(8) 0
Ã
(4) 69 (13) 45 (11) 56 (9) 60 (5) 0
Ã
(5) 50 (2) 40 (5) 25 (4) 57 (7)
CH 14 (7) 0
Ã
(7) 14 (7) 0
Ã
(6) 0 (3) 86 (7) 57 (7) 33 (6) 71 (7) 0
Ã
(4) 0 (1) 25 (4) 25 (4) 20 (5)
ZA 29 (7) 71 (7) 86 (7) 50 (6) 75 (4) 29 (7) 0
Ã
(7) 0
Ã
(7) 14 (7) 17 (6) 0 (2) 0
Ã
(5) 0
Ã
(4) 0
Ã
(5)
SK 25 (4) 0
Ã
(4) 0
Ã
(4) 0
Ã
(4) 0
Ã
(4) 50 (4) 75 (4) 50 (4) 50 (4) 33 (3) 0
Ã
(4) 50 (2) 0 (1) 50 (2)
BR 17 (6) 17 (6) 0
Ã
(6) 0
Ã
(5) 0 (3) 33 (6) 50 (6) 67 (6) 33 (6) 20 (5) 0
Ã
(6) 33 (3) 100
Ã
(4) 60 (5)
IN 14 (7) 29 (7) 14 (7) 0
Ã
(5) 0 (3) 14 (7) 50 (6) 67 (6) 17 (6) 17 (6) 0
Ã
(6) 33 (3) 100
Ã
(5) 50 (4)
JP 40 (10) 10
Ã
(10) 0
Ã
(10) 25 (8) 0
Ã
(4) 50 (10) 56 (9) 38 (8) 67 (9) 43 (7) 14 (7) 50 (4) 50 (6) 29 (7)
This table reports the results of a meta-analysis of the classi?cation studies of Table 1. The bottom-left triangle considers all 16 classi?cations, and the top-right triangle considers only the intrinsic classi?cations
(i.e. excluding studies 2, 3, 4, 10, 14 and 15). For each country-pair, the table shows the frequency (in %) with which the country-pair is classi?ed in the same group. The number in brackets (N) indicates in how
many classi?cations both countries of the country-pair were included.
Ã
indicates that the percentage for the country-pair is signi?cantly different from 50% at the 5% level (two-sided, based on a test of
proportion); the test requires at least two observations, i.e. the country-pair needs to be included in at least two classi?cations; a signi?cant result indicates a high degree of con?dence that the relationship of the
countries in the pair (either being or not being grouped together) is not arbitrary. The countries are: US (United States), Australia (AU), United Kingdom (UK), Canada (CA), Hong Kong (HK), France (FR), Spain (ES),
Italy (IT), Germany (DE), Switzerland (CH), South Africa (ZA), South Korea (SK), Brazil (BR), India (IN) and Japan (JP). For those classi?cation studies of Table 1 which provide more than one classi?cation, we only
use one/the main classi?cation, as follows: for classi?cation study 7, p. 433 (1975 analysis, measurement practices); for 9, Table 8 (we use the two-group classi?cation, not the more detailed one); for 12, Table 1
(again we use the two-group classi?cation not the more detailed one of Table 2); for 13, Fig. 2 (multi-dimensional scaling); for 15, Table 3 Panel C; for 16, Table 4 (principal component analysis).
5
8
0
C
.
N
o
b
e
s
,
C
.
S
t
a
d
l
e
r
/
A
c
c
o
u
n
t
i
n
g
,
O
r
g
a
n
i
z
a
t
i
o
n
s
a
n
d
S
o
c
i
e
t
y
3
8
(
2
0
1
3
)
5
7
3
–
5
9
5
up something fundamental. However, whether the insights
from this analysis can be relied upon at all depends greatly
on whether there are biased errors in the data or the meth-
ods used by the previous classi?ers. This is a central issue
of this paper, so we return to the worth of Table 2 after we
have examined that issue.
The classi?ers
The above discussions of cosmology and anthropology
showed how susceptible classi?cation can be to the nature
of the classi?ers. For accounting classi?cations, most of the
early writers had US or UK origins, so they were most
familiar with US and UK accounting, and had noticed the
differences. They then ?tted the rest of the world around
that starting point, often leading to a three-way classi?ca-
tion: US, UK and other. This explanation is consistent with
classi?cations 1, 3 and 4 of Table 1: Hat?eld (1911) and the
similar
21
ones of Seidler (1967) and of the American
Accounting Association (AAA, 1977, p. 105). These classi?ca-
tions were all drawn up by Americans. However, classi?ca-
tion 2 was produced by Gerhard Mueller, whose initial
education
22
was in Germany. Thus, Mueller had a different
Weltanschauung, in which the US and the UK are together
in one class, and the other three classes are each typi?ed
by a different continental European country. This suggests
that the nature of the classi?ers affected the classi?cations.
An extreme version of the above approach, of starting
with the US and the UK, can be found in paper 11 of Table 1
and in Alexander and Archer (2000). In these, the writers
(all from North America or the UK) identify some differ-
ences between the US and the UK (though these relate to
the context of accounting rather than to accounting prac-
tices) and then conclude that the US and the UK cannot
be classi?ed together. This would be like observing that
two cousins exhibit many differences, and therefore can-
not be closely related.
The range of countries
The objects being classi?ed (i.e. ‘accounting systems’ or
countries that use particular accounting systems) vary in
number from two to 50 (column 2 of Table 1). Communist
countries were generally excluded, because they had no
published ?nancial reporting. Later in this paper, we in-
clude Chinese companies using IFRS. Roberts (1995) points
out that it should be accounting systems rather than coun-
tries that are classi?ed. This became particularly relevant
when the widespread use of IFRS began in 2005 because,
in many countries, IFRS is only used for certain types of
reporting, such that one country now uses two or more
systems. This point was not adopted by any of the classi?-
cations in Table 1, though it was discussed at length in
Nobes (1998). The last classi?cation of Table 1 still appears
to classify countries, but it is the set of IFRS practices used
by companies in a country (i.e. the accounting system) that
is the object of classi?cation. The same applies later in this
paper.
The range of companies
The scope of the data (e.g. restrictions by sector or list-
ing status) is recorded in column (3). As may be seen, most
classi?cations did not specify a scope. This reduces their
usefulness, because the practices of listed companies vary
from those of unlisted companies; and, even among listed
companies, size has a major effect.
23
The last classi?cation
in Table 1 was the only one to mention sectors. It included
companies in all sectors, but displayed the sectoral mix
and excluded data on ?nancial companies for topics for
which sector-speci?c practices were anticipated. On such
grounds, the exclusion of ?nancial companies is common
in much research involving accounting data. However, this
creates a different problem: in all countries, the ?nancial
sector is signi?cant and, in some (e.g. Australia, Italy, Spain
and the UK), it is the most important sector among large
listed companies, as shown later. Therefore, exclusion of
the sector presents a misleading picture of a whole account-
ing system.
The importance of sector in in?uencing accounting pol-
icy was ?rst systematically investigated by Jaafar and
McLeay (2007), who examined three policy issues for com-
panies from 13 EU countries using national accounting
rules, in a pre-IFRS world. Consequently, Jaafar and McLeay
were not investigating policy choices only but a mixture of
different requirements and different choices. They found
that country was a much stronger explanatory variable
than sector, but that sector had some in?uence. Apart from
the ?nancial sector, which is excluded from many studies
on policy choice, a sector which might make idiosyncratic
choices is extractives, given the degree to which US prac-
tices dominate.
24
Jaafar and McLeay found some evidence
of this; and it might be important for countries in which
extractive companies constitute a large industry sector
(e.g. Canada).
The period measured
The users of classi?cations should also be aware that
countries can change their positions over time.
25
Table 1
(column 3) gives information on the dates of the data used
for the classi?cations, noting that most classi?ers have not
speci?ed a date.
The characteristics chosen
The discussions about chemistry, biology and cosmol-
ogy above showed that the nature and de?nitions of the
characteristics chosen as the basis for classi?cation is vital.
It is therefore inevitable that classi?ers must use judge-
ment in selecting and de?ning the characteristics used to
21
Seidler discusses the US and UK groups in some detail, suggests a
French group without naming any members of it, and proposes a
Communist group, mentioning only the Soviet Union. The AAA has British
and US groups, plus two continental European groups and ‘Communistic’.
22
Until moving to California at age 22, and then taking various degrees.
23
For example, see Nobes and Perramon (2013).
24
For example, under IFRS, there are no detailed rules on accounting
issues associated with extraction.
25
For example, see Nobes (1998).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 581
represent the objects to be classi?ed. We now examine this
aspect for the ‘intrinsic’ classi?cations of Table 1, i.e. those
that classify countries by their accounting rules/practices
rather than by in?uences on the accounting.
The number of speci?ed characteristics per system/
country (column 5 of Table 1) varies from zero to 264.
Classi?cations 5 to 8 were all
26
based on the surveys of Price
Waterhouse (PW) (1973, 1976, 1979) which had begun as a
list of detailed differences between US and UK accounting.
They therefore did not ask (about a country) such important
questions as: (i) are depreciation expenses determined by tax
rules? or (ii) is deferred tax accounted for? They asked in-
stead such peripheral questions as whether or not self-insur-
ance provisions were maintained in an internal account by
systematic charges to income (PW, 1973, Question 124),
which was known to be a topic of US/UK difference.
Not surprisingly, the PW data showed that the US and
the UK were the most different
27
of any pairing of the 36
countries examined for 1973. Consequently, by using these
data, da Costa, Bourgeois and Lawson (1978) found again
that the world had three types of country: US-led (contain-
ing most of continental Europe), UK-led, and unclassi?able
(i.e. Canada and the Netherlands). Frank (1979) and Nair
and Frank (1980) identi?ed four groups from the same data,
two of which were those dominated by the US and UK,
respectively. Goodrich (1982) used the 1979 PW data and
identi?ed ?ve groups, two of which were headed by the
US and the UK, though there is another headed by Jersey
which (remarkably) also contains Guatemala, Germany,
Italy, the Netherlands and Zaire.
Doupnik and Salter (1993) started with PW’s list of
characteristics but report an attempt to eliminate those
not representative of the fundamental features of the
accounting systems being classi?ed. d’Arcy (2001) used a
large number of characteristics from a KPMG database
which was not speci?cally prepared for her purpose. In
Nobes (2011), the characteristics measured are the policy
choices made by companies on topics such as those listed
in Table 3. The issue of which characteristics to measure
was discussed, and certain presentation topics were de-
leted on the grounds of less importance.
It was notedabovethat thepurposes of a classi?cationcan
be relevant for choosing the characteristics to be measured.
Someaccountingresearchers haveoutlinedthevarious possi-
ble purposes of classifying, including those of classifying
accounting systems, but few
28
have speci?ed the purpose of
their own attempts, apart from organizing knowledge.
The type of data used to measure the characteristics
As may be seen from Table 1 (column 6), the type of
data used to measure the characteristics also varies. Only
the last classi?cation was based on the collection of data
relating to the accounting practices of actual companies.
Clearly, the early classi?cations, which either used no data
(classi?cations 1–4) or relied upon impressions of practices
or of in?uences on practices (classi?cations 8–11), are less
satisfactory than detailed observation of practices. In the
non-accounting ?elds reviewed earlier, the quality of data
for classi?cation improved over time in various ways. For
accounting research, the annual reports of hundreds of
listed companies can now be collected quickly, and many
are available in English.
29
By contrast, in the 1960s and
1970s, if the US researchers had wanted to collect the ?nan-
Table 3
IFRS policy topics.
Topic IFRS policy options Standard
b
1.
a
– Income statement by nature – By function or neither IAS 1.99
2.
a
– No inclusion of a line for EBIT or operating pro?t – Line included IAS 1.82
3. – Equity accounting results included in ‘operating’ – Immediately after, or after ‘?nance’ IAS 1.82
c
4. – Balance sheet showing net assets – Showing assets = credits IAS 1.54
c
5. – Balance sheet with liquidity decreasing (cash at top) – Liquidity increasing IAS 1.54
c
6. – Indirect operating cash ?ows – Direct IAS 7.18
7. – Dividends received shown as operating cash ?ow – Not IAS 7.31
8.
a
– Interest paid shown as operating cash ?ow – Not IAS 7.31
9. – Some property at fair value – Only cost IAS 16.29
10. – Investment property at fair value – At cost IAS 40.30
11.
a
– Some designation of ?nancial instruments at fair value – None IAS 39.9
12.
a
– FIFO only for inventory cost – Weighted average used IAS 2.25
13. – Actuarial gains and losses to OCI – Corridor method or to income in full IAS 19.92/3
14. – Proportionate consolidation of joint ventures – Equity method IAS 31.30
This table shows 14 IFRS policy topics on which choices were observable in 2011. Topics 1–8 are presentation issues and topics 9–14 are measurement
issues. The topics are as in Kvaal and Nobes (2010). Most topics are binary choices but topics 1, 3 and 13 allow a choice between three options. For these, we
de?ne binary choices: for topic 1, we distinguish whether or not the income statement is by nature because the ‘neither’ cases are usually more similar to
‘by function’ than ‘by nature’; for topic 3, we consider the key issue to be whether or not the item is included in operating pro?t; for topic 13, we combine
the options ‘corridor method’ and ‘to income in full’ because we consider the key issue to be whether or not actuarial gains and losses are ever charged in
the income statement, which is not the case under ‘actuarial gains and losses to OCI’.
a
Not appropriate, and therefore not collected for ?nancial companies.
b
Versions of the standards ruling in 2011.
c
IAS 1 speci?es lists of items to be shown in ?nancial statements, but does not specify their order.
26
Classi?cation 8 (Goodrich, 1982) is based on the concepts part of the
survey, but still excludes the two examples of important questions given
later in this paragraph.
27
See Exhibit 1 of da Costa, Bourgeois, and Lawson (1978).
28
An exception is that Leuz, Nanda, and Wysocki (2003) make a
classi?cation in order to better understand international variations in
earnings management.
29
Nobes and Perramon (2013) ?nd that English language reports contain
the same information as the originals.
582 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
cial statements of all the members of the main French stock
market index (for example), that would have proved very
dif?cult, and many of the reports would not have been in
English. This might be what held back researchers from de-
tailed observation, but a more likely explanation is that
accounting researchers were not yet accustomed to an
empirical approach (Watts & Zimmerman, 1979).
Data which mix rules and impressions of practices
(classi?cations 5–7 and 12) are incoherent. The usefulness
of data relating to rules alone (e.g. classi?cation 13) can
also be questioned. For example, IFRS (IAS 38, para. 72) al-
lows certain intangible assets to be measured at fair value
rather than on a cost basis, whereas German and US GAAPs
require a cost basis. However, this difference in rules is of
doubtful signi?cance if no IFRS companies choose fair va-
lue to measure intangibles, which is the case in the sample
of German and UK companies of Christensen and Nikolaev
(2013). It is surely more signi?cant, for example, that the
majority of IFRS-using UK companies choose fair value
for investment properties whereas it is very rare for IFRS-
using German companies to do so,
30
even though all the
companies from both countries are using identical rules.
The quality of data
Any of the methods of measuring the characteristics cho-
sen for classi?cation can involve error. The PWdata used for
classi?cations 5 to 7 of Table 1 certainly contain errors. Sim-
ple examples
31
are that the UK is scored as not allowing
weighted average cost for inventory valuation which was
not (and never has been) the case; and that, on several topics,
the UK has different scores from Ireland, even though they
shared the same accounting rules. The data based on KPMG
information (used for classi?cation 13) also produces errone-
ous scores. For example,
32
the KPMG topics related to ‘provi-
sions’ were interpreted for most countries as referring to
provisions as de?nedin IAS 37 (i.e. liabilities of uncertaintim-
ing or amount), but the scoring for Australia was based on the
rules for items such as ‘bad debt provisions’ (i.e. impairments,
in current IFRS terminology). The last classi?cation (Nobes,
2011), in common with most others, did not provide speci?c
information about how the data were collected and coded.
This makes it dif?cult to replicate the studies.
The techniques of classi?cation
Table 1 (column 7) also shows the techniques used for
classi?cation. These range from qualitative assessments
to several different statistical methods. The resulting clas-
si?cations range (see column 8) from apparently unrelated
groups of countries to hierarchies (family trees or dendro-
grams) of related countries. Roberts (1995, p. 649) warns
against pushing the evolutionary analogies of the family
trees too far, and points out that dendrograms can summa-
rize similarities and differences without invoking evolu-
tion. Following from this, Roberts (1995, p. 656) also
criticizes the use of the term ‘species’ in an accounting
classi?cation, as in Nobes (1983). We use ‘system’ for a
set of objects with important characteristics in common,
although the apparently scienti?c word ‘species’ might
not be entirely out of place, given the discussion above
about the vagueness of the term in biology.
Roberts (p. 652) convincingly suggests that analogies
with the classi?cation of languages might be more appro-
priate, given that languages both converge (they interbreed)
and diverge, whereas species only diverge. Even so, as dis-
cussed earlier, the Linnaean system did not begin as evolu-
tionary but was based on assessing shared characteristics.
When evolution was added in (greatly aided later by the
analysis of DNA), the broad outline of the animal classi?ca-
tion survived although many details were revised. The same
might apply to the hierarchical accounting classi?cations.
That is, although they were prepared by assessing common
characteristics, the inclusion of evolution might not upset
the results. For example, the common ancestor of UK and
US accounting could perhaps be traced to nineteenth cen-
tury UK practice. By contrast, the common ancestor of
French and UK accounting lies much further in the past, per-
haps in the middle of the 16th century.
33
The statistical methods of classi?cation employed in
our subsequent empirical analyses are outlined in Appen-
dix A. In principle, they are sensitive to which countries
are included. For example, a clustering program starts by
?nding the two nearest countries, showing them together
and then treating the average of them as a ‘country’ for
the next stage of clustering. So, exclusion of one country
can affect the ‘seeding’ of the ?rst cluster, which can then
affect the position of many other countries.
Summary of factors affecting classi?cation
In sum, early accounting classi?cations seem to
have been affected by the national backgrounds of the
classi?ers or of the data gatherers. We have suggested,
above, ways in which classi?cation has also been affected
by the data used. First, the choice of which characteristics
to measure has a profound effect on the results. Once cho-
sen, the way of measuring the characteristics has var-
ied: several classi?ers apparently used no data, some
used incoherent data (mixtures of rules and impressions
of practices); and others used data which are arguably of
limited practical relevance (differences in rules).
Many classi?ers did not specify the scope of the objects
being classi?ed (e.g. large or listed companies) or the date.
However, some classi?ers have entered caveats. Frank
(1979, p. 596) notes that the topics included in his PW data
vary in importance. Frank does not make a selection or
comment on the mixture of rules and impressions of
practices in the data, but warns that the coding scheme
which turns that mixture into data for classi?cation might
introduce errors. Nair and Frank (1980) note that the clas-
30
In the companies comprising the main German and UK stock market
indices (Kvaal & Nobes, 2010).
31
Several examples are given in Nobes (1981).
32
These examples are discussed in Nobes (2004).
33
The earliest double-entry bookkeeping records in France and England
date from 1299 and 1305, respectively. However, both were isolated
examples kept by Italian ?rms of merchants in versions of Italian. Domestic
practice might instead be traced to translations of Pacioli’s tractatus on
bookkeeping, which were produced in France and England in the middle of
the 16th century (Coomber, 1956; Yamey, 1997).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 583
si?cations differ if based on presentation topics rather than
on measurement topics. d’Arcy (2001, p. 333) points out
that different topics would lead to different classi?cations,
and notes the inherent problem of using data on rules in-
stead of practices. Nobes (2011) mentions the need for
judgement in identifying important characteristics, and ex-
cludes some characteristics on these grounds.
We can now return to the meta-analysis of Table 2. We
have noted above that the early classi?ers might have been
particularly aware of US and UK differences, and that many
subsequent classi?cations were based on PW data which
had been originally designed to reveal such differences. This
couldexplainwhyno ‘Anglo-American’ groupwas generally
found.
The lessons of this section for accounting researchers
are that (i) a classi?cation should be based on detailed
observation of characteristics, (ii) the characteristics cho-
sen should ideally be informed by the purpose of the clas-
si?cation, and at least be deliberately chosen and overt,
(iii) related to this, any claims of objectivity are incoherent,
(iv) accounting practices are a better representation of an
‘accounting system’ than rules are, and (v) the set of com-
panies included in the accounting ‘system’ and the period
of the data should be speci?ed. It is further clear from this
survey that the effect of inclusions or exclusions of coun-
tries, sectors (especially ?nancial and extractive) and char-
acteristics needs to be empirically investigated in order to
see whether classi?cation is robust to manipulation of
these issues or whether it is instead essentially arbitrary.
We now proceed with that.
Data
Our sample
34
includes companies from the world’s
largest economies which use IFRS, as follows: (i) the
countries with the six largest stock markets where IFRS
was required from 2005 (i.e. Australia, France, Germany,
Italy, Spain and the UK), (ii) the two other countries with
large stock markets with a longer history of IFRS usage:
South Africa and Switzerland, (iii) Canada and South Kor-
ea, where IFRS has been recently adopted, and (iv) Hong
Kong and companies from China which use IFRS.
35
In or-
der to include Canada and South Korea, we use ?nancial
statements from 2011 onwards. In particular, we use
company reports for years ended 31 December 2011 (or
latest before) for 12 countries.
36
Our sample includes
the largest
37
listed companies in each of these jurisdic-
tions, which comprise 65%
38
of the total market capitali-
zation of these countries. Companies with foreign
in?uence or which are subsidiaries are excluded. In total,
we examine the IFRS practices of 514 companies. Details
of the sample are provided in Appendix B.
Table 4 shows the sample by country and sector. As ex-
plained, we wish to investigate whether exclusion of cer-
tain sectors might affect classi?cation. Prior literature
indicates that the ?nancial and extractive sectors have idi-
osyncratic policies.
39
Given the topic of this paper, we
should admit that the classi?cation of companies by sector
exhibits the dif?culties typical of classi?cation. There are
several accepted versions. We have chosen the ‘Industry
Classi?cation Benchmark’ (ICB) of the index company FTSE.
The corresponding data are from Worldscope (data code
WC07040). Consistently with our recommended approach,
we have used judgement to adjust it for our purposes, in
particular to calculate country totals for extractive
companies.
40
We concluded above that practices are the best
representation of an accounting system. We record the
IFRS practices of companies and use them as the character-
istics to be measured in order to classify a country. Even for
companies which are fully complying with IFRS, there is
considerable scope for varied practice because, for
example: (i) the recognition of expenses (e.g. impairments)
or assets (e.g. development projects) relies on the exercise
of judgement against somewhat vague criteria, (ii) the
measurement of liabilities (e.g. provisions) or assets (e.g.
the fair value of investment properties) involves estima-
tion, and (iii) many standards offer choices to companies.
The ?rst two of these are hard to assess (although, see an
attempt by Wehrfritz, Haller, & Walton, 2012), but data
on the third can be hand-picked from the annual reports
of companies. These data provide a good indication of the
in?uence of factors such as country and sector because
the differences in practices are caused by management
choices and not by regulations. There is some constraint
34
Our results show that the inclusion or exclusion of particular countries
generally does not affect how the remaining countries are classi?ed.
35
Although China has not fully adopted IFRS, the majority of the largest
listed Chinese companies prepares IFRS ?nancial statements, because they
are listed on the Hong Kong Stock Exchange (HKEx), which required IFRS
from 2005. Consequently, Chinese companies with a listing in Hong Kong
and Mainland China prepared two sets of ?nancial statements (IFRS and
Chinese GAAP). However, from 2010, HKEx accepts Chinese GAAP ?nancial
statements, and six companies in our sample have stopped preparing IFRS
?nancial statements.
36
The one exception to this is that we include the ?rst available IFRS
?nancial statements for ten Canadian and four South Korean companies
which have year-ends other than 31 December 2011. This enables the
inclusion of six Canadian and four South Korean ?nancial companies; in
particular, our Canadian sample would otherwise not include any bank
because all Canadian banks in our sample have 31 October year-ends.
37
Findings of country in?uence would probably be even stronger for
smaller companies because of less international in?uence (Nobes &
Perramon, 2013).
38
According to Worldscope data for 2011 (Worldscope code: WC07210).
39
Christensen and Nikolaev (2013) show that real-estate ?rms (which are
part of the ?nancial sector) choose to use fair value for investment property
more frequently than other ?rms. Jaafar and McLeay (2007, p. 180) refer to
special practices in the extractive industries.
40
We de?ne extractive companies as those in sector 0530 (oil and gas
producers), sector 1770 (mining), sub-sector 1753 (aluminum) and sub-
sector 1755 (non-ferrous metals); additionally, we classify Fortescue
Metals Group of sub-sector 1757 (iron and steel) as an extractive company.
We believe that using ICB codes results in a better industry classi?cation
than using primary SIC codes (Worldscope data code WC07021). If using
SIC codes, we would identify extractive companies as ‘mining’ (SIC codes
starting with the digits 10, 12, 13 or 14) and ?nancial companies as
‘?nance, insurance and real estate’ (SIC codes starting with the digit 6).
Using ICB codes, we classify 57 (129) companies as extractive (?nancial),
and using SIC codes we would have classi?ed 79% (94%) in the same way.
The main difference is that many integrated oil and gas companies (e.g. BP)
are not classi?ed as extractive but as ‘manufacturing’ using SIC codes due to
their petroleum re?ning businesses. Additionally, the classi?cation of some
companies using SIC codes is unsuitable for our purposes: e.g. China Oil?eld
Services is classi?ed as an extractive company even though it does no
extraction.
584 C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595
on changing policy,
41
but IFRS speci?cally removes re-
straints on choices made on ?rst-time use of IFRS (IFRS 1,
para. 11). As explained below, there are some rare examples
of national regulators, in the ?nancial sector, adding to IFRS
requirements.
The list of policy topics used by Kvaal and Nobes (2010)
is shown here as Table 3, after deleting the topics on which
choice was removed from IFRS by 2011. There are eight
presentation topics and six measurement topics. For the
last two topics in the table, changes to IFRS had already
been made by 2011 but were not compulsory for any of
our companies.
42
Kvaal and Nobes (2012) found that there
was little early adoption of IFRS changes, but we report on
this below. For ?nancial companies, Kvaal and Nobes
(2010) omitted topics on which there were sector-speci?c
presentation practices in?uenced by pre-IFRS laws. We do
not do that, because part of our purpose is to investigate
the effects of including or excluding certain sectors. Still,
we omit ?ve topics for ?nancial companies because they
are not appropriate for the sector, as explained in Appendix
C.
When the purpose of research is to investigate whether
IFRS policy choices are associated with country, then it is
appropriate to examine as many policy topics as are obser-
vable. However, as discussed earlier, for assessing a coun-
try’s accounting ‘system’ or a country’s pro?le of IFRS
practices, judgement is needed to exclude (or give lower
weight to) topics likely to be of little importance to users
of ?nancial statements (e.g. the liquidity order of assets
in a balance sheet). We investigate the sensitivity of classi-
?cations to such exclusion of topics.
Appendix C provides details about our data collection
and our coding procedures used to generate binary choice
data from the 14 IFRS policy topics. Our empirical analyses
below are based on a total of 5,689 hand-picked IFRS policy
choices of the 514 companies from 12 countries.
Findings on policy choice
Policy choices
Table 5 reports, by country, the percentages of compa-
nies in our sample which chose particular options. For sev-
eral topics, the policy choice was observable for all 514
companies. However, we only count companies for which
the policy is observable, which explains why the ‘N’ in
Table 5 is smaller for certain topics, notably investment
property measurement (topic 10).
Table 5 includes several countries for which data on
IFRS practices have not previously
43
been presented: China,
Hong Kong, South Africa, South Korea and Switzerland. Some
features of these countries stand out. First, the practices in
South Korea are unusually uniform: for most topics, over
90% of companies make the same choice, and for no topic
do fewer than 80% of companies make the same choice. Sec-
ondly, the majority of companies in China, Hong Kong, South
Africa and Switzerland chose to take actuarial gains/losses to
income, even though that practice was to be outlawed by a
change to IFRS which had already been issued. Similarly, a
majority of companies in South Africa and a large minority
in Switzerland chose proportional consolidation, even
though that practice was to be outlawed. This is further evi-
dence of no widespread early adoption of changes to IFRS,
and of the strong in?uence of pre-IFRS practices and there-
fore national patterns of IFRS practice, as documented in
the papers mentioned in the previous section.
Examples of sectoral effect
As explained earlier, some previous researchers ob-
served an association between pre-IFRS practices and sec-
Table 4
Sample by country and sector.
Sector AU UK CA CN HK FR ES IT DE CH ZA SK
P
0/1 Extractives 6 9 21 7 0 2 1 1 0 0 7 3 57
0/1 Other oil and gas, basic materials 4 3 2 3 0 2 3 0 7 3 1 4 32
2 Industrials 7 15 2 12 4 9 8 7 6 2 6 13 91
3 Consumer goods 2 10 2 3 3 7 1 5 7 3 2 5 50
4 Health care 2 3 0 1 0 2 1 1 2 4 2 0 18
5 Consumer services 8 20 8 4 3 7 3 8 6 0 6 4 77
6 Telecommunications 1 4 3 1 1 1 1 1 1 1 2 3 20
7 Utilities 1 5 1 3 3 2 4 3 2 0 0 2 26
8 Financials 20 22 10 13 8 6 9 12 6 6 6 11 129
9 Technology 0 2 0 2 0 2 1 0 2 1 0 4 14
P
51 93 49 49 22 40 32 38 39 20 32 49 514
This table reports descriptive statistics of the sample companies. The countries are Australia (AU), United Kingdom (UK), Canada (CA), China (CN), Hong
Kong (HK), France (FR), Spain (ES), Italy (IT), Germany (DE), Switzerland (CH), South Africa (ZA) and South Korea (SK). Sector is according to the ?rst digit of
the Industry Classi?cation Benchmark (ICB), except that we show all the extractive companies (sectors 0530 and 1770; sub-sectors 1753 and 1755; and
Fortescue Metals Group) together in the ?rst row, and all the remaining companies of sector 0 (oil and gas) and sector 1 (basic materials) together in the
second row.
41
IAS 8 imposes certain conditions and disclosure requirements (paras.
14 and 29).
42
IAS 19 was amended in 2011 to remove the option on the treatment of
actuarial gains and losses. IAS 31 was replaced in 2011, thus removing the
option of proportional consolidation. Both changes were only compulsory
for 2013 onwards.
43
Data on the other countries is included elsewhere; for example in
Nobes (2011).
C. Nobes, C. Stadler / Accounting, Organizations and Society 38 (2013) 573–595 585
tor. In some cases, this was driven by sectors having differ-
ent accounting rules.
44
By contrast, apart from a few iso-
lated examples of additional jurisdiction-based rules,
45
there are no sector-speci?c accounting requirements in IFRS.
Even so, signi?cant differences in IFRS choices between sec-
tors are evident when we split our entire sample into three
sectors: ?nancial, extractive and other. Table 6 shows partic-
ularly clear examples from our data. More importantly for
this paper, there are also signi?cant differences between
the sectors within countries. As may be seen, compared to
other companies in their countries, Australian ?nancial com-
panies prefer fair value for investment properties, Canadian
?nancial companies prefer not to recognize actuarial gains/
losses as OCI (they prefer the corridor method), Canadian
extractive companies prefer to proportionally consolidate
joint ventures, and British ?nancial companies are less likely
to show net assets but more likely to start the balance sheet
with cash. In all these cases, a v
2
test of independence shows
that a null hypothesis of no association with sector can be
rejected at the 1% level. These are examples of how a coun-
try’s sectoral mix might affect its mean scores on topics,
which might then affect classi?cation, as examined in the
next section.
Sensitivity of classi?cations
Introduction
We have suggested above how the accounting classi?-
cations, particularly the early ones, appear to have been
sensitive to the nature of the classi?ers. In this section,
we empirically investigate how sensitive classi?cations
can be to various aspects of the data used.
As recorded in Table 1, several different statistical
methods of classi?cation have been employed by previous
researchers. We begin with principal component analysis.
Table 7 shows the principal components for one version
of the data: all 14 topics for all sectors of all 12 countries.
This analysis leads to a three-group initial classi?cation,
summarized as ‘run’ 1 in Table 8. In this table, we report
the results for 11 different versions of the data. For ease
of comparison, Germany is always shown in Group 1.
The Kaiser–Meyer–Olkin (KMO) measure of sampling
adequacy should be higher than 0.6 (or possibly 0.5) for
the data to be considered suitable for factor analysis (Kai-
Table 5
Percentages of policy choice by country and topic.
IFRS policy choice N AU UK CA CN HK FR ES IT DE CH ZA SK
1. Income statement by nature 385 35 11 5 44 36 29 96 81 24 29 15 3
2. Operating pro?t not shown 385 42 1 31 31 29 3 0 0 12 0 0 0
3. Equity pro?ts in operating 423 59 35 48 4 0 8 23 14 35 39 7 4
4. Balance sheet showing net assets 514 100 76 0 39 82 0 0 0 0 5 0 0
5. Balance sheet with liquidity decreasing 514 100 10 100 24 14 10 22 29 26 50 9 98
6. Indirect cash ?ows 514 4 98 100 98 100 100 91 95 100 95 66 100
7. Dividends received as operating 348 87 37 85 5 30 79 39 20 71 43 86 91
8. Interest paid as operating 381 86 61 74 44 43 79 52 69 61 64 96 89
9. Some property at fair value 504 10 10 2 0 5 0 0 0 0 0 0 0
10. Investment property at fair value 216 93 68 36 21 94 20 5 0 5 80 40 3
11. Some fair value designation 383 10 3 13 0 7 24 4 4 6 7 23 19
12. FIFO only 329 21 42 23 6 15 11 22 19 0 36 23 6
13. Actuarial gains/losses to OCI 414 85 89 72 8 36 60 68 30 59 35 28 83
14. Proportionate consolidation of JVs 379 6 25 55 9 0 71 70 38 17 43 59 17
This table reports the percentages of companies per country and topic which make the respective IFRS policy choice in 2011. The countries are as in Table 4.
N is the number of observations/companies. See Table 3 and Appendix C for details of the topics.
Table 6
Examples of sectoral differences in policy choice.
Country IFRS policy choice N % Financials % Extractives % Others p-value
All 10. Investment property at fair value 216 61 0 9