You are on page 1of 11

True diversities: A comment on

Lou Josts Entropy and diversity


Sonke Hoffmann
Otto-von-Guericke-University Magdeburg, Faculty of Economics and
Management, Dept. of Economic Policy, P.O. 4120, 39016 Magdeburg, Germany. E-mail: soenke.hoffmann@ww.uni-magdeburg.de

Andreas Hoffmann
University of Applied Sciences Magdeburg-Stendal (FH), Dept. of
Water Resource Management, Breitscheidstrae 2, 39114 Magdeburg, Germany. E-mail: andreas.hoffmann@hs-magdeburg.de

Abstract
Jost (2006) recently discussed Hills (1973) effective number of species and concluded by naming
it the true diversity. Due to the inherent multi-faceted character of diversity we doubt that any
known diversity concept can be called the true one. Instead, we may identify good and bad
concepts, or even best and worst ones, depending on the match-up between properties the
context requires and properties the concept provides. Using this terminology, we agree, that Hills
(1973) effective number is one of the best approaches to quantify community diversity in ecology.
Keywords: Diversity measurement, Entropy, Information

Introduction
In his article Entropy and diversity Jost (2006) gives a very unambiguous statement on what diversity is. According to Jost, there is a class of true diversities,
better known as effective number and usually denoted Na . Jost may aim for the
ultimate guidance through Ricottas (2004) Jungle of biodiversity, but some of
his statements are too absolute in their character. In what follows we comment on
these aspects and argue that the true diversity is not what the name suggests.

What is diversity?
At the very beginning of his article Jost (2006) emphasizes:
Diversity [. . . ] has been confounded with the indices used to measure
it; a diversity index is not necessarily itself a diversity. The radius
of a sphere is an index of its volume but is not itself the volume, and
using the radius in place of the volume in engineering equations will give
dangerously misleading results. This is what biologists have done with
diversity indices.
and justifies this statement as follows:
[. . . ] most common diversity measure, the Shannon-Wiener index, is an
entropy, giving the uncertainty in the outcome of a sampling process.
Saying that something is not itself a diversity requires an unambiguous and commonly accepted understanding of what a diversity is. This prerequisite would
be easily satisfied if diversity was something naturally given a physical quantity,
like volume, mass or energy. But diversity is not. Instead it is something very
multi-faceted and inherently subjective, influenced by at least two sub-questions
(cf. Baumgartner 2006):
(1) Diversity for what purpose?
(2) Diversity of what?
Answers to the first question are manifold but manageable, whereas answers to the
second seem to be bounded only by our imagination. Quite apposite to the matter,
Magurran (1988) compared diversity with a fata-morgana, that takes very different
and blurred shapes from different viewpoints. The often cited plethora of diversity
concepts is nothing but a natural consequence of the plethora of possible answers
to questions (1) and (2). In contrast to physical quantities, diversity is nothing
naturally given but it is implicitly defined by a formal concept that is assumed to
measure the quantity under consideration. This is the common practice in natural
and social sciences. As Peet (1974) for example recognized quite early in ecology:
Diversity, in essence, has always been defined by the indices used to
measure it [. . . ].

From this perspective it is not possible to say what diversity is ultimately. Diversity
in all its dimensions and facets cannot be captured by a single definition or mathematical formalism. Because many different views on diversity exist, many diversites
exist, none of them being per se more a diversity than another. On the other hand it
is equally
is not diversity. Jost argues that the mathematical term
Phard to 1say what
+
S
H c i pi log pi , c R cannot be a diversity because it is an entropy. Generally
seen, H S neither is a diversiy, nor is it an entropy, or any other phenomenological quantity of our real world. Instead, it is nothing but a mathematical expression
with certain inherent properties (see Aczel and Daroczy 1975 for details). And these
properties may or may not fit to a possible being that should be captured by H S .
This being can be entropy, uncertainty, information, inequality, evenness, diversity
or other phenomenons. Entropy is certainly much more objective and physical in its
character than diversity, but nevertheless, it is similarly described by astro physicist
Tim Thompson1 :
The easist answer to the question, What is entropy?, is [. . . ]: Entropy
is what the equations define it to be. You can interpret those equations
to come up with a prosey explanation, but remember that the prose
and the equations have to match up, because the equations give a firm,
mathematical definition [. . . ], that just wont go away.
Jost is clearly right, saying that radius is not volume and that radius is instead a
index variable of volume. However, this does not say anything about the beings
of diversity and entropy. Using the expression H S to measure community diversity,
biologists have certainly not confounded diversity with the indices used to measure
it. Instead they have implicitly defined diversity by using this concept, being very
aware of the fact that H S is not diversity, but one of many ways to make diversity
explicit. The fundamental question arising is not whether H S itself is a diversity or
not, this cannot be answered anyway , but whether this mathematical expression
makes sense in a given context or not. To put it in the words of philosopher Norton
(2003):
there is no correct [...] definition to be found, as one might discover
a gem under a rock. We are looking for a definition that is useful in
deliberative discourse [. . . ]. Proposed definitions will be judged by their
usefulness.
For the sake of ontological and etymological consistency
we have to reformulate
P
Josts statement. We can either say, Expression c i pi log p1i is neither entropy nor
P
diversity but simply a mathematical term, or we say Expression c i pi log p1i can
be used to define and measure entropy as it also can be used to define and measure
one of many facets of diversity. How good H S finally performs as a model of
the quantity under consideration is a fundamentally
different question. Depending
P
on the context, the number that formula c i pi log p1i finally provides may be a
complete nonsense or the most meaningful number of all. In any case, it is an
element in the set of available options.
1

See http://www.tim-thompson.com/entropy1.html

Some proses on H S
The legitimation to use a mathematical expression like H S as a measurement concept
arises from required properties in a given context on the one hand (What should
diversity be?) and the inherent properties of the expression (What is diversity?)
on the other. Such interplay between Thompsons proses and equations is finally
judged with regard to their match-up.
The first, and one of the best matching proses on H S , was given by Ludwig Boltzmann and Josiah Willard Gibbs, who tried to explain classical thermodynamics, and
especially its second law, by statistical means of mechanics. Here, H S quantifies a
macrostate mean value of energy over possible microstates i (e.g. Beck and Schlogl
1993). As a tribute to Boltzman and Gibbs, H S is usually called the BoltzmannGibbs entropy by physicists. Another substantial and well-matching prose was given
by the father of information theory, Claude E. Shannon, who established H S as
a measure of reduced uncertainty (information) in his seminal theory of communication (Shannon 1948). Although the word entropy is, etymologically seen, quite
misleading in the context of information theory, most scientists call H S the Shannon
entropy. In fact, the creator of the word entropy is Rudolf Clausius who proprosed:
[. . . ] to name the quantity S the entropy of the system, after the
Greek word trope, the transformation. I have deliberately chosen the
word entropy to be as similar as possible to the word energy: the two
quantities to be named by these words are so closely related in physical significance that a certain similarity in their names appears to be
appropriate. (Clausius 1850, translation by W.F. Magie)
The name Shannon entropy may be due to the mathematician John von Neumann,
who is quoted as having proposed the word entropy to Shannon for two reasons
(Weinberg 1981):
First, your formula is identical in structure with the entropy of statistical thermodynamics. An second, and more important, no one understands entropy. You will therefore always be at an advantage in an
argument.
Other uses of H S were successively made in economics, political sciences, linguistics
and many more disciplines. The characterizing properties of H S were also found to
be suitable for the ecological measurement of community diversity (Margalef 1958,
Pielou 1974, 1975). The most popular assumption in this context is that
[. . . ] diversity, however defined, is a single statistic in which the number
of species and evenness are confounded (Pielou 1969, see also Magurran
2004, p. 17),
S
S
And
P this is1exactly what H is able to provide. Alternatively, the expression H =
c i pi log pi may also measure average species rarity another possible facet of
community diversity where pi is the relative abundance of species i and log (1/pi )
is the rarity function of that species (Patil and Taillie 1982). The latter proposal

is less popular but not less intuitive: A species with no abundance is, compared to
other species, infinitely rare, while a species with abundance pi = 1 is, compared to
other species, not rare at all.
There are many interpretations of H S . Certainly not all of them have the same
objective and unambiguous status as entropy or information have. Shannon (1956)
himself noticed a bandwagon effect caused by his publication and warned that
information theory
[. . . ] has perhaps ballooned to an importance beyond its actual accomplishments.
Using H S as a diversity measure may or may not be admissible. It depends on
the interplay between the properties required and the properties provided. The
properties required vary with the application contexts and, therefore, H S can neither
be generally accepted nor generally rejected. In any case, rejecting H S as diversity
measure requires much more than simply observing that H S perfectly fits to measure
something else.

Effective numbers: The true diversity?


An important aspect of diversity concepts is the unit of measurement. Many biologists agree, that the quantity of species diversity should be measured in number of
classes (species), corrected by the underlying eveness of the abundance distribution.
An effective (species) number N is maximal and equal to the nominal number N = n
if all species i = 1 . . . n are equally abundant i.e. pi = 1/n i, and if one species
is maximally abundant, i.e. i : pi = 1, then the effective number takes its minimum N = 1. Hill (1973) derived a rather general one-parameter class of diversity
indices Na that meets these requirements. This fact made Na very attractive for a
variety of scientific disciplines: Economists measure the effective number of firms in
an industry (Adelman 1969, Miller 1972, Hannah and Kay 1977), political scientists
measure effective number of parties in a parliament (Laakso and Taagepera 1979,
Dumont and Caulier 2003) and ecologists measure the effective (evenness-corrected)
number of species in a community (Hill 1973, Ricotta 2003b). All three groups may
find the same Na matching to their given context. However, economists measure
industry or market concentration, political scientists measure fragmentation within
the spectrum of political parties, and ecologists measure community evenness or
diversification of species. These examples show, once again, and in incomplete analogy to the Shannon entropy, how little a mathematical formalism determines the
ontological being of what they are able to quantify. Simply observing that H S
or some N may be used to measure something different in other contexts does not
authorize us to sweepingly reject one of the concepts as a diversity measure in ecology. Nevertheless, Jost (2006) wants Na to be exclusively named the class of true
diversities. In the first two sections of his paper he gives two reasons:
(1) Diversity should be proportional to the number of species if all species are
equally abundant. Na is proportional in this sense, H S is not.
4

(2) Let H be a continuous and monotonic


P q function. Any entopy-like formula of
the class H(f (p)), where f (p) = i pi , q R can be transformed to Na .
The first argument may be plausible in many ecologial contexts, but it is not generally sine qua non. If, for instance, diversity is required to incorporate qualitative
differences (disparity) instead of abundance distributions, this assumption becomes
all but reasonable. Imagine a hypothetical community consisting of two rather similar species which may be equally abundant, say, one individual of wasp species and
one of hornet species. Now we double the community richness by adding one rhino
and one albatros. In this context it is hardly plausible to require that diversity has
doubled because the disparity of the community has certainly more than doubled.
And there are biological contexts that require measuring diversity in terms of disparity. Phylogenetic diversity, for example, is based on differences in genes or traits
and not on evenness (Vane-Wright 1991, Faith 1994). Outside biology, similar contexts are even more obvious, e.g. the measurement of economic product diversity
(Dixit and Stiglitz 1977, Gans and Hill 1997) or molecular diversity in pharmaceutical chemistry (Chabala 1998). Josts (2006) normative requirement is appropriate
in large parts of the ecological sciences, but it cannot be used to give Na the status
of the true diversity. Another aspect of Na s fundamentally relative character is
the trade-off between typical diversity properties (cf. Hoffmann 2006): The class
of effective numbers may be proportional to the nominal number if abundances are
equally distributed, but on the other hand this class is neither additive (not even
pseudo additive) nor generally concave two properties that can not be valued as
irrelevant per se. Additivity of a diversity measure is needed if the individuals of a
community are classifiable in more than one way (Pielou 1975). The measure is required to be concave, if diversity is used as portfolio generating function in financial
economics (Fernholz 1999, 2002), if diversity helps finding sparse best-basis selections in applied mathematics (Kreutz-Delgado and Rao 1997, 1998), or, if it needs to
be additively decomposable into within and along community components (Patil
and Taillie 1982, Lande 1996, Ricotta 2003a, Keylock 2005). Moreover, in all optimization contexts concavity is desirable because it causes global maxima while non
concave measures may not do so (Kapur 1994). Due to the general and far reaching
desirability of concavity, this property is sometimes even formulated as an axiom
that characterizes diversity functions (e.g. S. Pavoine et al. 2004). What properties
are finally more valuable is a matter of individual judgements with respect to the
given context and the same holds true for the legitimation to use a mathematical
expression as a measure for phenomenological quantities.
The second one of Josts reasons to call Na the true diversity class is even weaker.
He shows that there exists a transformation for arbitrary functions of a kind H
such that (H(f (p)) = Na (Proof 1 in Jost 2006). The generic algorithm needed to
obtain for any given H is the classical algorithm used to find the equivalent number
of equal sized categories (cf. Hannah and Kay 1977, Laakso and Taagepera 1979).
Applying this algorithm to classical information and entropy measures, Jost observes
that R (x) := exp(x) for entropies of order a and T (x) := expq (x) for entropies of
degree b, where expq (x) is a common q-deformed exponential introduced by Tsallis
(1994). Taking into account the logarithmic nature of entropy-like formalisms, this
observation is not very surprising, and, more importantly, it does not prove anything
5

in favour of Na being a true diversity. To illustrate that point, consider a more


obvious analogue: Let L(p) = ap+b a, b R, a 6= 0 be a sub-class of linear functions.
Now, lets take any two elements in L, say L1 (p) = 3p and L2 (p) = 2p + 1. Both
functions can be transformed into an arbitrary expression of L, for example into
L(p) = p. In this case the transformations would be L1 (x) = 13 x and L2 (x) =
1
(x 1) for L1 and L2 respectively. The generic algorithm to obtain is: (1)
2
subtract b (2) divide by a. It can be easily proved that for arbitrary L there exists
a such that (L(p)) = L(p), but would such proof be a reason to call L(p) = p
the true linear function? Certainly not. Instead we call L one of many linear
functions; one that is characterized by certain properties, such as L(0) = 0, which
is plausible in one context and non-matching to another.

Conclusion
There are good and bad diversity measurement concepts and the distinction between them is determined by properties, contexts and individual judgements. A
major benefit of Hills (1973) one-parameter class of effective numbers Na is its
straightforward unit of measurement and an easy to comprehend proportionality
property. A major drawback is that Na is neither additive nor generally concave.
However, the effective number may be called a good diversity measure, or even
the best in contexts, where a straightforward unit is more valuable than additivity
or concavity. But giving it the absolute status of the true diversity is inadequate
and misleading. Na incorporates some but definitely not all desirable properties of
diversity measures in general.
There are ways to get the plethora of diversity models under better control. Information theorists and physicists successfully generalized Shannon entropy along
different properties (e.g. Renyi 1961, Havrda and Charvat 1967, Borges and Roditi
1998, Kaniadakis et al. 2005) and some economists have already tried to derive a
universal, all-encompassing theory of diversity (e.g. Weitzman 1992, Nehring and
Puppe 2002). Hoffmann (2006) recently discussed a general diversity measure that
includes all classical entropy-like formalisms as well as the effective number class. In
complete analogy to the work of Jost (2006) or Hill (1973) all these models are very
helpful to make the general meaning of diversity successively more comprehensible,
but regarding their true universality they fail.

Bibliography
Aczel, J. and Daroczy, Z. 1975. On measures of information and their characterizations.
- Academic Press.
Adelman, M. 1969. Comment on the H concentration measure as a numbers-equivalent.
- Rev. Econ. Stat. 51: 99101.
Baumgartner, S. 2006. Measuring the diversity of what? And for what purpose? A conceptual comparison of ecological and economic biodiversity indices. Working Paper,
available at SSRN: http://ssrn.com/abstract=894782.
Beck, C. and Schlogl, F. 1993. Thermodynamics of chaotic systems: an introduction. Cambridge nonlinear science series No. 4. Cambridge Univ. Press.
Borges, E. and Roditi, I. 1998. A family of nonextensive entropies. - Phys. Lett. A 246:
399402.
Chabala, J. 1998. Historical overview of the developing field of molecular diversity. - In:
Gordon, E. and Kerwin, J. (eds.), Combinatorial chemistry and molecular diversity
in drug discovery, Wiley-Liss., pp. 316.

Clausius, R. 1850. Uber


die bewegende Kraft der Warme und die Gesetze, welche sich
daraus f
ur die Warmelehre selbst ableiten lassen. - Annalen der Physik und Chemie
155: 368397.
Dixit, A. and Stiglitz, J. 1977. Monopolitstic competition and optimum product diversity.
- Am. Econ. Rev. 67: 297308.
Dumont, P. and Caulier, J. 2003. The effective number of relevant parties: How voting
power improves Laakso-Taageperas index. Working Paper, Facultes Universitaires
Notre Dame de la Paix Namur, Belgium, available at: http://centres.fusl.ac.
be/CEREC/document/people/caulier/enrp.pdf.
Faith, D. 1994. Phylogenetic pattern and the quantification of organismal biodiversity. Philos. T. Roy. Soc. B. 345: 4558.
Fernholz, R. 1999. On the diversity of equity markets. - J. Math. Econ. 31: 393417.
Fernholz, R. 2002. Stochastic portfolio theory. - Springer.
Gans, J. and Hill, R. 1997. Measuring product diversity. - Econ. Lett. 55: 145150.
Hannah, L. and Kay, J. 1977. Concentration in modern industry: theory, measurement
and the U.K. experience. - Macmillan Press.

Havrda, J. and Charvat, F. 1967. Quantification method of classification processes: the


concept of structural -entropy. - Kybernetika 3: 3035.
Hill, M. 1973. Diversity and evenness: a unifying notation and its consequences. - Ecology
54: 427431.
Hoffmann, S. 2006. Concavity and additivity in diversity measurement: re-discovery
of an unknown concept.
Working Paper, FEMM series, Otto-von-GuerickeUniversity, Magdeburg, available at: http://www.ww.uni-magdeburg.de/fwwdeka/
femm/p2006.html.
Jost, L. 2006. Entropy and diversity. - Oikos 113: 363374.
Kaniadakis, G. et al. 2005. Two-parameter deformations of logarithm, exponential, and
entropy: A consistent framework for generalized statistical mechanics. - Phys. Rev.
E 71: 046128104612812.
Kapur, J. 1994. Measures of information and their applications. - Wiley Eastern Limited.
Keylock, C. 2005. Simpson diversity and the Shannon-Wiener index as special cases of a
generalized entropy. - Oikos 109: 203206.
Kreutz-Delgado, K. and Rao, B. 1997. A general approach to sparse basis selection:
majorization, concavity, and affine scaling. Technical Report No. UCSD-CIE-977-1, University of California, San Diego, Department of Electrical and Computer
Engineering.
Kreutz-Delgado, K. and Rao, B. 1998. Measures and algorithms for best basis selection.
- Proc. ICASSP 3: 18811884.
Laakso, M. and Taagepera, R. 1979. Effective number of parties a measure with
application to West Europe. - Comp. Polit. Stud. 12: 327.
Lande, R. 1996. Statistics and partitioning of species diversity, and similarity among
multiple communities. - Oikos 76: 513.
Magurran, A. 1988. Ecological diversity and its measurement. - Princeton University
Press.
Magurran, A. 2004. Measuring biological diversity. - Blackwell Publishing.
Margalef, D. 1958. Information theory in ecology. In: Yearbook of the Society for General
Systems Research, volume 3, pp. 3671. Society for General Systems Research.
Miller, R. 1972. Numbers equivalents, relative entropy, and concentration ratios: a comparison using market performance. - South. Econ. J. 39: 107112.
Nehring, K. and Puppe, C. 2002. A theory of diversity. - Econometrica 70: 11551198.
Norton, B. G. 2003. Defining biodiversity: do we know, what we are trying to save? Faculty
of Forestry Jubilee Lecture Series, University of British Columbia, available at: http:
//www.forestry.ubc.ca/lectures/namkoong/Namkoon series 021604.pdf.
Patil, G. and Taillie, C. 1982. Diversity as a concept and its measurement. - J. Am. Stat.
Assoc. 77: 548561.

Peet, R. 1974. The Measurement of species diversity. - Annu. Rev. Ecol. Syst. 5: 285307.
Pielou, E. 1969. An introduction to mathematical ecology. - Wiley Interscience.
Pielou, E. 1974. Population and community ecology: principles and methods. - Gordon
and Breach.
Pielou, E. 1975. Ecological diversity. - Wiley & Sons.
Renyi, A. 1961. On measures of entropy and information. - In: Proceedings of the Fourth
Berkeley Symposium on Mathematical Statistics and Propability, volume 1, J. Neyman, University of California Press, Berkeley, pp. 547561.
Ricotta, C. 2003a. Additive partition of parametric information and its associated diversity measure. - Acta Biotheor. 51: 91100.
Ricotta, C. 2003b. On parametric evenness measures. - J. Theor. Biol. 222: 189197.
Ricotta, C. 2004. Through the jungle of biodiversity. - Acta Biotheor. 53: 2938.
S. Pavoine, A. D. et al. 2004. From dissimilarities among species to dissimilarities among
communities: a double principal coordinate analysis. - J. Theor. Biol. 228: 523537.
Shannon, C. 1948. A mathematical theory of communication. - Bell Syst. Tech. J. 27:
379423.
Shannon, C. 1956. The bandwagon. - Trans. Inform. Theory IT-2: 3.
Tsallis, C. 1994. What are the numbers that experiments provide?. - Qum. Nova 17:
468471.
Vane-Wright, R. 1991. What to protect? Systematics and the agony of choice. - Biol.
Conserv. 55: 235254.
Weinberg, A. 1981. Book review on: Entropy: a new world view, J. Rifkin and T.
Howard, Viking Press. - Popul. Dev. Rev. 7: 345348.
Weitzman, M. 1992. On diversity. - Q. J. Econ. 107: 363405.

You might also like