438 THE AMERICAN NATURALIST

COMPARING DIVERSITY INDICES BASED ON COUNTS WEIGHTED BY

BIOMASS OR OTHER IMPORTANCE VALUES
Several indices have been suggested to measure the diversityof a sample of
individualsfrom an ecological community.Two of the more commonly used
indices are Shannon's h = -Y(nJ/N) log (ni/N), and Simpson's A = Y(nJ/N)2(or
itscomplement,d = 1 - A), whereni, i = 1,2, . . . , s, is thenumberof individuals
of species i in a randomsample of N = Ini individuals.A shortcomingof these
indices is that they are statisticallycomparable only if based upon counts, i.e.,
numbersof individuals.In applicationmost ecologists agree thatthe importance
of a species to a communityshould be based on the relativeamountof a continu-
ous quantity,such as biomass or productivity,contributedby membersof that
species (Hurlbert1971; Whittaker1965). In practice these "importance values"
are sometimessubstitutedfortheni's in the diversityindices, withN replaced by
the total of the importancevalues for all species (see e.g., Odum 1970, p. 144;
Pielou 1966). If thisis done the variance of the resultingstatisticis not knownand
it is not appropriateto obtain an estimateof variance by makingthe same typeof
substitution intoexistingexpressionsforthe variance of the index in count form.
In fact such an estimate would depend upon the units of measurementof the
importancevalues, whereas the index itselfis invariant.
Using Shannon's index, Pielou suggests an alternativeestimate of variance
equal to the usual sample variance of adjusted differencesbetween values of h
obtainedfromsuccessively cumulatedsamples. The procedurecould be extended
to otherindices and it has an advantage of not requiringspecies counts. However
it does require a partitionof the sample into independentsubsamples. Pielou
pointsout thattheassumptionthatthese adjusted differencesare independentand
have the same variance as thatof the index itselfis necessary. This assumptionis
not supportedtheoretically.The covariance of successive values of Shannon's h
and Simpson's complementin count formhave been derived(Lyons and Hutche-
son 1977, 1979).
In this paper the asymptoticdistributionand variance of a general diversity
index based on fixed weights is derived. The results may be applied to the
weightedversion of h, d, eh, McIntosh's (1967), Hill's (1973), and Good's (1953)
indices, and various other functionsof the weighted relative frequencies. The
importancevalues are taken intoaccount by assigningthe same fixedweight(i.e.,
not estimatedfromthe sample) to every individualof a species. A large sample
test forcomparingdiversityindices fromtwo independentsamples is suggested.
The test requires that species counts be taken.

RESULTS

Consider a populationof individualsclassifiedinto s species in relative abun-

dances p1, P29 . . . p YPi= 1. Let wi be the fixedweightor importancevalue
,

Am. Nat. 1981. Vol. 118, pp. 438-442.

? 1981 by The Universityof Chicago. 0003-0147/81/1803-0011$02.00

This content downloaded from 129.100.058.076 on November 21, 2016 21:59:20 PM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 NOTES AND COMMENTS 439

assigned to every individualof species i. Then winil/wini representsthe relative

importanceof species i in the sample. The observed frequencies,n1,n2, . . . , ns, in
a randomsample of fixedsize N, have a multinomialdistribution, and each ni can
be writtenas ni =IN' where

1_
= if the kth individualselected is fromspecies i,
Okl, otherwise.

LetXik =
Wi(Xik - Pi), i = 1,2, . . ., s, k = 1,2, . .. , N. By the multivariate central
limittheorem(see e.g., Rao 1965,p. 118),N112Xi= N-1/2W(ni- Npi , i = 192,....
s, are jointly asymptoticallynormallydistributedwith means 0, variances W1p1
(1 - pi), and covariances -ww pjPiP?j ii
Let G(x1,x2, . . . , xs) be any functionwhichis continuousand whose firstand
second partial derivatives exist at the origin, and letg, = aG(OO, ... .., O)/xi, i =
1,2, ... , s, withnot all thegi's equal to zero. Then it can be shown by the delta
method (see e.g., Cramer 1946, p. 366) that N1/2[G(X1,X2, . . ., Xs) - G(0,0, . ..
0)] has limitingnormaldistributionwithmean zero. The variance of the limiting
distributionis
02 = jg?")?p, - (lg)W~p,)2.

All of the diversityindices mentionedabove can be expressed in the generalform

Q[Yfi(w"ini/Ywini)], by proper choice of Q andf. Let G be definedby
G(X1,X2,. . . , Xs) = Q{fil[(Xi + wipi)1Y(Xi + wEp1)]}.

IfJi = nilN, thenT(5,pI2, *. i ) = G(X1,X2, . . iX) =

Q[Yf(vi9i/WN)] =
is the
sample statistic,and T(p1,p2, . p) G(0909 . . ., 0) Q[Yfi(wipi/WP)]is
,

the correspondingpopulation value, where WN = Ywiji, and W, = Ywipi. The

derivativesof G at the originare
1
9= - (Wipi/WY)ipi1W,)19
WP Q-fi(WiPAp?4mpi/WP

i 1,2, . . , s. Upon substitutioninto (1) and some algebraic simplification,

the
variance of N1/2T(O1q52,. . . , i) is
1 -
Q'[Jf("p,/VWP)]2J wsp,[f/(w7^p1/Wp)-
1(Y) pi/Wp)fi(1)p,/WP)]

For Shannon's index Q (x ) -x and fi(x) = x log x for every i, so that

h =T(J1,3P2, . . , ) = -E(WiI/VWN) log (w113i/IWN),and
H = T(p1,p2, * . , PS) = -I(w1ipi/WV) log (wjp1/W1,).

For large N,

Var(h)- NW; Y2p1[log(wipI/W,) + H]2.

If the weightsfor each species are the same then

This content downloaded from 129.100.058.076 on November 21, 2016 21:59:20 PM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
440 THE AMERICAN NATURALIST

Var(h) N-(pilog2pi - H2),

whichis theresultto order 1/Nobtainedby Basharin(1959). The exact expression

for the variance of h is given by Hutcheson (1970).
For Simpson's complement,Q(x) = 1 - x andfi(x) =x2 foreveryi, so thatd
1 - I(wiAI/WN)2, and D = 1 - J(wip1/W,)2, with
4 - W,(l - D )2
Var(d) NW4 Yw2wpi[wipi

This simplifiesto 4/N[YEp3- ('p?)2], the result obtained by Simpson (1949), if

each species is weightedequally.
For McIntosh's complementQ(x) = 1 - x112andfi(x) - x2 foreveryi. ThenI -
1 - [y(wipji/WN)2]1I2, I = 1 - [I(wipl/WP)2]1/2, and

Var() 1W" (Ew p W)-1Yw p,(w1 p, - pWP)

NW2
which reduces to 1/N(YpV!YEp? - Yp?) for equal weights.
For the transformedindex ek, Q(x) = e-x and f(x) = x log x foreveryi, eh -

exp [- Y(Wipi/WN) log(Wipi/WN)], eH = exp [- Y(wipilWp) log(wipilWp)],and

Var(eh) NW2 e2" pwip1[log(WipI/Wp) + H]2 = e2HVar(h).

These variances may be estimatedby replacingpi byp in the above expressions.

Hutcheson (1970) and Lyons and Hutcheson (1978) develop large sample tests
for comparingdiversitiesfromtwo independentsamples using Shannon's and
Simpson's indices. These tests are valid onlyiftheni's are counts. In factthe test
statisticin each case is not invariantto the units of measurementif the ni's are
replaced by importancevalues. Denoting the index of diversityby T and the
estimatorof variance by 6'2, diversitiesfromindependentsamples of sizes N1 and
N2 individualsforany of the weightedindices maybe comparedby the statisticZ
- (T1 - T2)/(ci1+ (r2)1I2. Using an extensionof the methodsabove to two samples,
itcan be shownthatthisstatisticis asymptoticallynormallydistributed.The most
importantfactis thatthe statisticis invariantto the unitsof measurementused to
determinethe wi's.

AN EXAMPLE

Sikora et al. (1972) compare the food habits of inshorejuvenile populationsof

spotted hake, Urophycisregius, and southernhake, U. floridanus, in Georgia
estuaries.Usingbiomass as an importancevalue, the authorswished to determine
if the diversityof the food organismsof these two species differed.They were
unable to make this comparison because the statistical techniques were not
available at that time. Since numbersof individualsas well as biomass weights
were recordedfortheirexperiment,it is possible to test the hypothesisof equal
diversitiesbased on biomass using the methods described above.

This content downloaded from 129.100.058.076 on November 21, 2016 21:59:20 PM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 NOTES AND COMMENTS 441

TABLE 1
ESTIMATEs/(standard deviations) AND VALUES OF COMPARISON STATISTIC Z FOR THREE DIVERSITY
INDICES USING COUNTS WEIGHTED BY BIOMASS AND COUNTS ONLY

Index h d I
Counts weightedby biomass ....... Urophycis 2.474 .864 .631
floridanus (.05205) (.009513) (.01290)
U. regius 2.086 .759 .509
(.06457) (.01889) (.01925)
Z 4.678 4.962 5.271

Index Ih d I

Counts only ...................... Urophycis 2.179 .767 .517

floridanus (.02588) (.005420) (.005615)
U. regius 1.354 .535 .318
(.01350) (.004601) (.003373)
Z 28.239 32.685 30.459

Samples of 342 spottedhake and 192 southernhake were collected. For each of
the two species of hake, the number of individuals of each species of food
organismpresent in the combined contents of the gastrointestinaltracts was
determined.From the southern hake N1 = 3,879 individuals representing50
species were recovered. From the spotted hake N2 = 14,328 individualsrepre-
senting58 species were recovered. Organismswere too decomposed fromdiges-
tive processes to yield reliable weights,thereforeevery individualof a particular
species in the sample was assigned a weightequal to the dryweightin biomass
units(1 biomass unit = .01 mg.) of one organismof the same or similarspecies.
The values of threediversityindices and theirestimatedstandarddeviations,for
each species of hake, along withthe values ofZ forthe comparisontestare given
in table 1 forindices based on counts weightedby biomass units. The resultsfor
indices based on counts only are includedforcomparison.For everyindex based
on biomass units it can be concluded that the southernhake feeds on a greater
diversityof organismsthan the spotted hake. The value of the test statisticZ
would be exactly the same if biomass units were changed to pounds or grams.
The methoddescribed above requiresa randomrepresentativesample of hake
fromeach population. Since the contentsof the stomachs of all individualsare
combinedto arriveat the estimatesof thepi's foreach of thetwo species of hake,
possible differencesbetween feedinghabits of individualhake should not affect
theresults.Pielou's procedurecould be applied to theexample to obtainvariance
estimatesif data were available for the stomach contents of each hake. (The
individual hake could be considered analogous to the quadrats described in
Pielou's example.) However the procedure would be affectedby possible dif-
ferencesin individualfeedinghabits.The greaterthe differencesbetween feeding
habits, the larger the numberof hake required in order to overcome possible
biases in estimatesobtained fromeach individual.

This content downloaded from 129.100.058.076 on November 21, 2016 21:59:20 PM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
442 THE AMERICAN NATURALIST

ACKNOWLEDGMENTS

I thankE. P. Odum and KermitHutcheson fortheirvaluable discussions and

comments,and the refereesforprovidingsuggestionswhich resulted in the im-
provementof this paper.

LITERATURE CITED

Basharin,G. P. 1959. On a statisticalestimateforthe entropyof a sequence of independentrandom

variables. Pages 333-336 in Theoryof probabilityand its applications. Vol. IV. N. Artined.
Society for Industrialand Applied Mathematics. Philadelphia.
Cramer, H. 1946. Mathematicalmethodsof statistics.PrincetonUniversityPress, Princeton,N.J.
Good, I. J. 1953. The populationfrequenciesof species and the estimationof populationparameters.
Biometrika40:237-264.
Hill, M. 0. 1973. Diversityand evenness: a unifyingnotationand its consequences. Ecology 54:427-
432.
Hurlbert,S. H. 1971. The nonconcept of species diversity:a critique and alternativeparameters.
Ecology 52:577-586.
Hutcheson, K. 1970. A test forcomparingdiversitiesbased on the Shannon formula.J. Theor. Biol.
29:151-154.
Lyons, N. I., and K. Hutcheson. 1977. Species diversity:a stoppingrule and FORTRAN program.
Tech. Rep. no. 124. Departmentof Statisticsand ComputerScience, Universityof Georgia,
Athens.
1978. Comparingdiversities:Gini's index. J. Stat. Comput. Simul. 8:75-78.
1979. A stoppingrule based on Simpson's index of diversity.J. Theor. Biol. 77:497-503.
McIntosh,R. P. 1967. An index of diversityand the relationof certainconcepts to diversity.Ecology
48:392-404.
Odum, E. P. 1970. Ecology. Holt, Reinhart& Winston,New York.
Pielou, E. C. 1966. The measurementof diversityin different typesof biologicalcollections. J. Theor.
Biol. 13:131-144.
Rao, C. R. 1965. Linear statisticalinferenceand its applications. Wiley, New York.
Sikora,W. B., R. W. Heard, and M. D. Dahlberg. 1972.The occurrenceand food habitsoftwo species
of hake, Urophycisregius and U. floridanus,in Georgia estuaries. Trans. Am. Fish Soc. no.
3:513-525.
Simpson, E. H. 1949. Measurementof diversity.Nature 163:688.
Whittaker,R. H. 1965. Dominance and diversityin land plant communities.Science 147:250-260.

N. I. LYONS
DEPARTMENT OF STATISTICS AND INSTITUTE OF ECOLOGY
UNIVERSITY OF GEORGIA
ATHENS, GEORGIA 30602
SubmittedMarch 25, 1980; Revised September 9, 1980; Accepted February 6, 1981

This content downloaded from 129.100.058.076 on November 21, 2016 21:59:20 PM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).