Ecological Indicators 67 (2016) 703–713

Contents lists available at ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Biased richness and evenness relationships within Shannon–Wiener

index values
W.L Strong ∗
Yukon Research Centre, Yukon College, 500 College Drive, P.O. Box 2799, Whitehorse, Yukon Y1A 5K4, Canada

a r t i c l e i n f o a b s t r a c t

Article history: The purpose of this analysis was to empirically model and graphically illustrate the numerical rela-
Received 15 October 2015 tionships between richness (S, 4–35 species) and evenness (E) with respect to Shannon–Wiener index
Received in revised form 21 February 2016 (H , loge -based) values. Thirty-two richness-based third-order polynomial regression models (R > 0.99,
Accepted 21 March 2016
P < 0.001, n = 28–71) were constructed to characterize these relationships. A composite diagram showed
Available online 25 April 2016
richness varied curvilinearly, with steepness increasing and the spacing between curves decreasing
with greater evenness and H . Maximum H values for each richness curve were equal to loge S (when
Keywords:
E = 1), whereas minima were approximated by evenness values of ∼1/S (when H = 0). It was concluded
Biodiversity
Entropy from multiple and polynomial regression analyses that: (i) evenness contributed more than richness
Evenness (E:S ≥3:1) to determining H , based on standardized partial beta-coefficients; (ii) the differential in
Plant community E:S ratios increased with greater richness; (iii) the patterns of H sample variation between maximum
Richness unevenness and perfect evenness was convexo-concave shaped; and (iv) richness as an explanatory
Shannon–Wiener variable of H was likely an artifact of evenness (0–1 scale) being rescaled according to individual
H maxima. H was redefined as a logarithm-weighted measure of evenness at a given level of rich-
ness, which means H is either an imperfect index of diversity or a biased measure of evenness. It
was also found that the fundamental components of the Shannon–Wiener index measure dominance
concentration rather than evenness, with the reversal in emphasis due to multiplication of the H equa-
tion by −1. H -derived effective species numbers (exp H , D) increasingly deviated from those of the
diversity model D = S × E in response to increasing richness (up to 69% for 35 species), particularly
when evenness was between 0.15 and 0.40. Of two cross-validated H prediction methods (P < 0.001,
n = 325), the collective use of individual richness-based polynomial regression equations (r = 0.954) was
better than a single multiple regression model that incorporated a broad spectrum of richness levels
(r = 0.882). A simple graphic model was constructed to illustrate patterns of evenness variation as a func-
tion of changing richness and H values. Based on the identified biases, particularly E:S ratios, it was
recommended that use of H be discontinued as a basis for assessing diversity in ecological research or,
at the very least, accompanied by independent analyzes of richness and evenness.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction diversity, Shannon (species) diversity, Shannon entropy, Shannon

The Shannon–Wiener index (Shannon and Weaver, 1949), also
known as the Shannon–Weiner [sic] diversity, Shannon–Weaver
Abbreviations: D, diversity; E, evenness; EDw , Lorenz curve derived evenness
index; E , Camargo evenness index; H , Shannon–Wiener index; −H , an H value
prior to multiplication by −1; IRPR, independent richness-based polynomial regres-
sion; J , Pielou evenness index; loge , natural logarithm; K–S, Kolmogorov–Smirnov

one-sample test; Hmax , maximum value of H ; pi , proportional abundance values;
P, probability level; r, Pearson product-moment correlation coefficient; R, correla-
2
tion coefficient with >1 independent-variable; Radj , sample size adjusted explained
variance; S, species richness; SEE, standard error of estimate.
∗ Tel.: +1 867 667 2924; fax: +1 867 667 2924.
E-mail address: wstrong@yukoncollege.yk.ca
information index, and H , has been used in biological studies as a
measure of diversity since the mid-1950s (e.g., MacArthur, 1955;
Patten, 1959). As a measure of entropy (Hill, 1973; Jost, 2006;
Shannon and Weaver, 1949), or the amount of variation among
abundance values, H is currently defined as the degree of “uncer-
tainty in the . . . identity of an individual that is randomly chosen
from . . . [a] dataset” (Tuomisto, 2012, p. 1206; and is similarly
defined in ecological reference texts by authors such as Barbour
et al., 1999; Kent and Coker, 1992; Krebs, 2009; Magurran, 2003),
which is similar to the characterization given by Shannon (1948,
pp. 392–393). Others, however, have emphasized the interaction
between richness (number of unique taxa or species) and even-
ness (distribution of abundance, Smith and Wilson, 1996) as the key
attribute of H (e.g., Buzas and Hayek, 1996; Hurlbert, 1971; Lloyd

http://dx.doi.org/10.1016/j.ecolind.2016.03.043
1470-160X/© 2016 Elsevier Ltd. All rights reserved.
704 W.L Strong / Ecological Indicators 67 (2016) 703–713
and Ghelandi, 1964; Molles, 2002; Risser and Rice, 1971; Whittaker,
1972).
Despite a long history of use, uncertainty appears to still
exist regarding both the understanding and interpretation of H ,
although transformation to numbers of equally common species
or “effective species” (exp H ) has been advocated as a standard
for comparing diversity among samples and studies (e.g., Hill,
1973; Jost, 2006; MacArthur, 1965). This numerical transforma-
tion, however, does little to advance the ecological understanding
of the relationship between richness and evenness. Some studies
evade the interaction issue by merely indicating that a statisti-
cally significant difference occurred between compared samples
without explanation. Opinions on the merits of H , as a measure
of diversity, range from being a “dubious index” with “no direct
biological interpretation” (Goodman, 1975, p. 244; also Camargo,
2008, p. 282) and an index with “serious conceptual and statistical
problems which make comparisons of species richness or species
abundance across communities nearly impossible” (Barrantes and
Sandoval, 2009, p. 451) to a reluctance to recommend its use
(Magurran, 2003, p. 101; Whittaker, 1972, p. 224) to being “the
most profound and useful of all diversity indices” (Jost, 2006,
p. 364).
Yue et al. (2007) and Barrantes and Sandoval (2009) consid-
ered H to be the best known and most widely used of all diversity
indices. A word search of the Thomson Reuters WEB OF SCIENCETM
database found that H was used in at least 1527 research studies
in 2012 through 2014. This number does not include academic the-
ses and dissertations (PROQUEST® Dissertation and Theses Global
database), which represented at least an additional 1717 docu-
ments, nor technical reports. The number of annually published
studies that referenced H grew from >119 in 1994 to >1073 in 2014,
or a nine-fold increase.
A review of 200 arbitrarily selected biological publications from
the 2012 to 2014 period (W.L Strong, unpublished data) indicated
that ecosystem diversity was the focus of 65% of the studies, with
genetics representing an additional 21%. Inferential statistical test-
ing was used to identify differences among H values in two-thirds
of these studies. The remaining studies primarily used regression
and correlation to describe relationships between H , and various
abiotic (e.g., Cağlar and Albayrak, 2012) and biotic factors (e.g.,
Con et al., 2013), or made only qualitatively interpretations of
the data (e.g., Akhoundi et al., 2013). About 58% of the biological
studies used natural logarithms (loge ) as a basis for calculating
H , with ∼21% not clearly identifying the applied logarithm base,
although it affects the numerical outcome of H and can limit the
comparison of results among studies (Magurran, 2003). The use
of H -derived effective species numbers as a basis for diversity
comparison was not a widely applied practice among biologists,
based on their use in only four of the 200 reviewed studies (Adie
et al., 2013; Külköylüoğlu et al., 2012, 2013; Skácelová and Lepš,
2014).
Three issues complicate the interpretation of H from an ecolog-
ical perspective: (1) the currently advocated definition emphasizes
uncertainty (e.g., Jost, 2006, p. 363–364; Tuomisto, 2012, p. 1206),
as if diversity were merely a statistical probability issue rather than
an ecological phenomenon; (2) the differential logarithm weighting
of proportional abundance values in the equation affects the out-
come of H calculations (Odum, 1969; Peet, 1974, p. 295; Ricotta,
2003, p. 184; Whittaker, 1972, p. 224) and, therefore, its interpre-
tation; and (3) the numerical relationship between richness and
evenness, and H is vague. For example, is or how is richness repre-
sented in H ? This question arises because H is calculated from
dimensionless proportional abundance values, without richness
being a parameter in the Shannon–Wiener equation. In addition,
do richness and evenness have equally weighted roles in the cal-
culation of H ? Based on the conceptual model advocated by Jost
(2010, p. 212) and Tuomisto (2012, p. 1205) and indirectly by other
such as Hurlbert (1971, p. 577), diversity (D) should be proportional
to the relationship between richness (S) and evenness (E) (i.e., D (as
represented by exp H ) = S × E).
Strong quantitative associations have been reported between
H and richness (DeJong, 1975; Monk, 1967; Risser and Rice, 1971;
Tramer, 1969) and H and (un)evenness (DeJong, 1975; Risser and
Rice, 1971), so presumably multivariate regression equations could
be constructed to quantify their numerical relationships as pre-
dictors of H (Buzas and Hayek, 1996; Stirling and Wilsey, 2001);
but no comprehensive models appear to exist that clearly demon-
strate such interactions. With regard to the Stirling and Wilsey
(2001) analysis, their multivariate models are of questionable valid-
ity, because the Pielou (1966) J evenness index (J = H /loge S, loge S
is also known as Hmax ) was used as an explanatory variable of H ,
i.e., there is a lack of independence between H and J . Their use
of J as an evenness measure probably stemmed from its popular-
ity relative to other indices at the time. Among the 200 reviewed
biological publications, J occurred in 73 (36.5%). The large and
growing number of studies that use H suggest it is becoming
an expected standard for assessing biological diversity, and the
above indicated issues and questions need to be addressed if H
values are to be understood from a numerical perspective rather
than just calculated. From a practical perspective, the use of H
as a resource management tool (e.g., evaluation of herbicide and
mechanical treatment effects as forest site management practices
– Seiwa et al., 2012; Wu et al., 2013) could have substantial and
long-term adverse environmental implications, if it is prone to mis-
interpretation.
Tuomisto (2010) suggested that the understanding of diversity
lay not in the form of the numbers (e.g., entropies, probabilities,
effective species), but in the ecological meaning of the variation
in the abundance values that are the basis for calculating such
indices. Based on this philosophical perspective, the objectives of
this analysis were to: (i) graphically model the numerical relation-
ships between evenness and loge -based H values, and richness
(5–35 species); (ii) estimate the amount of potential variation that
occurs among H values, when calculated for different species abun-
dance combinations at individual levels of richness; (iii) construct
multiple regression models to predict H as a basis for determining
if richness and evenness contribute equally to its calculation; (iv)
determine if H -derived “effective species” numbers (D) conform
to the diversity model D = S × E; (v) propose a more ecologically-
oriented definition of H than is currently advocated; and (vi)
create an easily understood two-dimensional graph to summarize
the numerical relationships of evenness to richness and H with
respect to the analyzed data. Although H is used in various fields of
research, this assessment is primarily oriented toward terrestrial
vegetation and plant community analyses.
2. Materials and methods
2.1. Synthetic numerical sequences (DATASET A)
Data for modeling the relationship between richness and even-
ness, and H were created by compiling a series of synthetic
numerical sequences for each richness level for four through 35
species (DATASET A). This range of richness appears to encompass
the number of species that typically occur in terrestrial plant com-
munity composition samples or relevés (e.g., De Grandpré et al.,
2011; Hart and Chen, 2008; Strong, 2015; Wu et al., 2013). To create
each numerical series, an initial sequence of abundance values with
an indisputable H and evenness value was established. The initial
sequence for each series consisted of equally abundant species that
totaled 100 (e.g., first sequence for five species – 20, 20, 20, 20,
 W.L Strong / Ecological Indicators 67 (2016) 703–713
and 20; H = 1.609, evenness = 1 or perfect). Additional numerical
sequences for each series were created by simultaneously deduct-
ing from the last and adding to the first species in the sequence, an
amount sufficient to reduce H by ∼0.05 (i.e., second sequence for
five species – 29.8, 20, 20, 20, and 10.2; H = 1.559, see Table A1).
When the abundance of an individual species was reduced to 0.01,
the deduction process shifted to the next species in the sequence.
This procedure was repeated using the newly created sequence,
until H was reduced to a value of about 0.05 or less (Table A1).
The number of sequences for each level of richness was approxi-
mately equal to loge S/0.05. The incremental reduction of H values
created numerical sequences, which began with perfect evenness
and ended with near maximum unevenness (i.e., maximum dom-
inance concentration) at each level of richness. This allowed the
comparison of numerical sequences based on a single rate of change
within and among different richness levels, i.e., a form of standard-
ization. However, no indication of overall H variance was possible
for individual richness levels from these data due the sequence
development process. For each numerical sequence, an index of
evenness was calculated.
2.2. Assessment of H variation (DATASET B)
To determine how much variation potentially occurs in H num-
bers among different combinations of species abundance values,
2500 random numerical sequences were compiled for each of four
richness levels: 5, 15, 25, and 35 species (DATASET B). Because
sequences created from completely random values tended to bunch
in the middle of E · H scatter diagrams, it was necessary to increase
the diversity among sequences. To expand the range of diversity,
the abundance of individual species were constrained to different
upper-limits. The upper-limits for the four richness level were:
level 5 – 100, 50, 25, 10, and 5; level 15 – 100, 60, 35, 20, 10[n = 2],
5[3], 3[3], and 1[3]; level 25 – 100, 60, 35, 20, 10[n = 3], 5[5], 3[6],
and 1[7]; and level 35 – 100, 75, 50, 20, 10[n = 3], 7.5[3], 5[6], 3[8],
1[7], and 0.1[4]. The assigned upper-limits were arbitrary, but an
attempt was made to mimic the general compositional abundance
of species in natural plant communities with a mid-level of
diversity. The number of species with lower values increased with
richness. The final sequences for DATASET B were regenerated by
species using the random number generation function available in
Microsoft® Office Excel 2003. H and evenness were calculated for
each sequence of abundance values. Standard errors of estimate
(SEE) were used to assess variability among H values by richness
level.
2.3. Index calculations
H values were calculated using the equation (Shannon and
S
Weaver, 1949): H  = − i=1 (pi × loge pi ); where pi is ai /ai , ai
is the abundance of the ith species, and i is one through S. For
this analysis, abundance values were equated with plant cover
percentages. H values were calculated using loge because of
its more common usage among the 200 reviewed studies (see
Introduction).
Evenness (EDw = 1 − Dw ) was calculated based on the dominance
concentration equation (Strong, 2002): Dw = maxi [bi /pi − i/S];
where pi is the proportional abundance of the ith species, bi is the
sequential cumulative totaling of pi values arrayed from largest
to smallest, i is one through S, and maxi is the largest calculated
ith value. EDw was used as a measure of evenness because (i) the
index is based on the Lorenz curve method of assessment, which
is the preferred approach for estimating (un)evenness (Gosselin,
2001); (ii) species abundance values are treated equally; (iii) EDw
values are strongly correlated with other commonly used even-
ness measures such as the Camargo index and Gini coefficient
705
(Strong, 2002, p. 242); (iv) EDw fulfills the four basic requirements
of an evenness index as defined by Smith and Wilson (1996); and
(v) the computation of EDw is independent of H , unlike the often
referenced Pielou (1966) J evenness index. For comparative pur-
poses, the more commonly cited Camargo (1993) E evenness index
(Tuomisto, 2012, p. 1208) was calculated for individual numer-
ical sequences in DATASET A. The Camargo Index was defined
S S
as: E  = 1 − ( i1=1 i2=1+1 |pi1 − pi2 |/S). EDw and E values ranged
from 0 to 1, with zero representing maximum unevenness and one
representing perfect evenness.
The Shannon–Wiener index was reverse calculated using the
equation −H = exp ( (pi × loge pi )), or exp (H × −1) to determine
the fundamental diversity attribute measured by H .
2.4. Regression modeling
Third-order polynomial regression was used to model nonlinear
trends in evenness and H variation relative to richness in DATASET
A and DATASET B. Second-order or quadratic regression models
explained nearly as much variance as third-order models, but the
latter better fit the data especially the tails. Typically, second-order
models over-estimated Hmax  or perfect evenness, which was the
only indisputably value in each datasets, i.e., an anchor point. Hav-
ing the regression models intersect or at least closely approximate

Hmax was important to maximize consistencies among richness
trend lines.
Forward stepwise, conventional multiple linear regression with
richness and evenness as independent-variables was used to con-
struct equations to predict H (dependent-variable). These analyses
were based on subsets of numerical sequences from DATASET A
(n = 350 of 1811), because richness values within the total dataset
were not normally distributed based on Kolmogorov–Smirnov one-
sample tests (K–S, P > 0.05). The occurrence of both normally and
linearly (relative to H ) distributed independent-variables at the
˛ < 0.05 level was an assumed requirement to maximize the poten-
tial of obtaining viable models (e.g., Tabachnick and Fidell, 2013,
p. 78; Zar, 2010, p. 430). The lack of normality was due to the
continuously increased frequency of numerical sequences with
increasing richness (Table A2). To mitigate this problem, numer-
ical sequences were randomly selected from DATASET A based
on richness. The number of selected sequences per richness level
corresponded with the distribution pattern of a normal curve.
Associated with each randomly selected numerical sequence were
corresponding H and evenness values that were used for regres-
sion modeling. Multiple correlation (R) was used to assess the
strength of the relationship between independent (n = 2) and the
dependent multiple regression variables. Coefficients of deter-
mination (R2 ) were adjusted according to available degree of
freedom to avoid over-estimating the proportion of explained
2 ).
variance (Radj
Multiple regression residuals (observed minus predicted val-
ues) were also expected to be linearly distributed, homoscedastic
(equal variance) based on Fmax -tests (Sokal and Rohlf, 1995, p. 397),
and normally distributed (Tabachnick and Fidell, 2013, p. 127).
Residuals with high peakedness or leptokurtic (kurtosis >3) dis-
tributions were also considered acceptable, so long as the lower
limit of the deviant frequency class was zero and the class width
was small relative to the range of variation. Comparison and
characterization of residuals was based on histograms, scatter dia-
grams, linear regression, and Pearson product-moment correlation
(r).
As an alternative to multiple regression, individual richness-
based polynomial regression equations (Table A2) were used as
the basis for predicting H . It was expected that individual poly-
nomial equations when used collectively would more accurately
706 W.L Strong / Ecological Indicators 67 (2016) 703–713

predict H than the best available multiple regression. The rationale Table 1
Average richness (S), evenness (EDw ), and Shannon–Wiener index (H ) values in
was that the multiple regression models would be less sensitive to
DATASET A–C and their associated data subsets.
differences in richness.
STATISTICA (release 5.0) was used for multiple regression and Richness range n S EDw H
variable characterization, whereas Microsoft® Office Excel 2003 DATASET A Average (standard deviation)
was used for polynomial regression modeling. 4–35 1811 21.3 (8.8) 0.47 (0.26) 1.49 (0.91)

DATASET A subsets
4–35 (Eq. 1a) 350 18.3 (5.8) 0.47 (0.26) 1.46 (0.87)
2.5. Regression model validation (DATASET C) 4–35 (Eq. 1b) 350 18.3 (5.8) 0.47 (0.26) 1.42 (0.84)
4–19 200 11.4 (2.9) 0.51 (0.25) 1.28 (0.70)
To cross-validate potentially viable regression models, H values 20–35 200 27.5 (3.0) 0.43 (0.26) 1.55 (0.95)
were calculated for 325 real plant community relevés (DATASET DATASET B subsets
C, sourced from Strong, 1999, 2014, 2015). These relevés con- 5 2500 5 (0) 0.57 (0.09) 1.09 (0.19)
tained four to 30 species. Relevés were sampled by the same field 15 2500 15 (0) 0.43 (0.06) 1.75 (0.20)
25 2500 25 (0) 0.41 (0.06) 2.10 (0.22)
methods, but differed with respect to vegetation physiognomy and
35 2500 35 (0) 0.40 (0.06) 2.33 (0.22)
species composition. The botanical composition of these mid- and
high-latitude boreal vegetation relevés primarily included conifer DATASET C
4–30 325 14.4 (5.5) 0.46 (0.09) 1.71 (0.53)
and deciduous tree-dominated, and graminoid-dominated stands.
Validation was performed by comparing observed (DATASET C)
and regression equation predicted H residuals using the same
techniques as employed to identify viable conventional multiple richness level was equal to loge S. The resulting evenness values
regression models (i.e., histograms, regression, correlation, etc.). were contoured using polynomial regression.
Separate validation testing was done using (i) the best available
multiple regression and (ii) the collective use of multiple richness- 3. Results
based polynomial regressions. Models accepted as valid had both
linearly and normally distributed residuals, but the better proce- 3.1. Richness and evenness relationships in H
dure for predicting H was considered the method that explained
the most variance. DATASET A contained 1811 numerical sequences, with an aver-
age richness of 20.5 (Table 1). Data subsets with 350 samples had
2.6. Richness and evenness ratios average richness values that were three species less than the full
dataset. In contrast, the average richness of these subsets was 3.9
Standardized partial beta- or b -coefficients were used to esti- species greater than DATASET C. The latter dataset also had a nar-
mate the relative contribution made by richness and evenness in rower range of richness (maximum 30 species). Among datasets
the determination of H (McGrew and Monroe, 2000, p. 223; Sokal and subsets, average evenness (EDw ) was typically in the middle of
and Rohlf, 1995, pp. 613–614; Zar, 2010, p. 433), using both lin- the scale (range 0.40–0.57). DATASET A and its 350 sample subsets,
ear conventional and correlation matrix-based (Tabachnick and and DATASET C had essentially identical mean EDw values, although
Fidell, 2013, p. 131) multiple regression. Correlation matrix-based the latter had a smaller standard deviation. H values were also sim-
regression modeling was used to increase the number of avail- ilar for DATASET A and its 350 sample subsets (Table 1), whereas
able E:S estimates, when independent-variables were not or could DATASET C had a slight greater average (Table 1). Among other
not be fit by common transformation functions to a normally data subsets, EDw and H values varied according to the richness
distribution. range, with average EDw values greatest when richness was only
Both conventional and correlation matrix-based models were five species.
based on subsets of numerical sequences extracted from DATASET Systematically different curvilinear relationships occurred
A. These data subsets either spanned the range of richness between EDw and H values when stratified by richness. Each had
(4–35 species, n = 350, i.e., same data subset as used for con- an inclined trend line relative to increasing evenness (Fig. 1). The
ventional multiple regression modeling), or included only the steepness of the trend lines increased and the spacing decreased
lower (4–19, n = 200) or upper (20–35, n = 200) half of the rich- with increasing richness. The lower end of each richness curve
ness range. The two latter subsets were used to assess the relative occurred along the evenness or x-axis of Fig. 1, where EDw val-
importance of the independent-variables for predicting H at dif- ues were equal to ∼1/S, when H was zero. The upper end of
ference levels of richness. Selection of numerical sequences for each richness level was equal to loge S. Third-order polynomial
inclusion in these two latter analyses followed the same proce- regression models closely approximated the distribution of EDw
dures as used for conventional multiple regression modeling (see · H values when stratified by richness, with multiple correla-
Section 2.4). tion coefficients (R) of >0.994 (Table A2). The fit of individual
polynomial regression curves to evenness was poorest for 4–6
species, although their individual standard errors of estimate were
2.7. Evenness response to H <0.03.
Scatter diagrams based on 2500 sets of EDw and H values
To illustrate the general response pattern of evenness to chang- per richness level (5, 15, 25, and 35) had curvilinear distribu-
ing H values, a two-dimensional, trivariate graph was constructed tion patterns (Fig. 2). Samples for 15, 25, and 35 species had
based on selected richness (5, 15, 25, 30, and 35) and evenness (0, relative compact distribution patterns that occurred in the cen-
0.2, 0.4, 0.6, 0.8, and 1.0) values. The H numbers for this graph tral to slightly lower portion of the evenness scale (0.25–0.6).
were based on the average of values generated by the Table A2 Data sequences with five species had a narrower and more elon-
and the DATASET B equations, working on the assumption that the gate sample distribution pattern than the other richness levels.
true model at each level of richness occurred somewhere between Standard errors of estimate for H among the four richness level
these two trend lines. Averaged H values were prorated (i.e., a ranged from 0.045 to 0.093. H regression trend lines generated
1–2% downward adjustment) so that perfect evenness for each by polynomial Eq. (A.2), (A.12), (A.22), and (A.32) consistently
 W.L Strong / Ecological Indicators 67 (2016) 703–713 707

Fig. 1. Systematic patterns of variation in richness (4–35 species, n = 28–71, respec-

tively) relative to evenness (EDw ) and incrementally-reduced Shannon–Wiener
index (H , loge ) values. Values derived from DATASET A. Lines are third-order poly-
nomial regression trend lines. See Table A2 for individual equations and their
associated statistical attributes.

occurred below randomly generated data sequences (Fig. 2). The

H values in these two datasets deviated (maximum range/2) up
to ±0.14 (S = 5), ±0.32 (S = 15), ±0.40 (S = 25), and ±0.44 (S = 35)
from their mid-line position in the 0.4–0.6 portion of the evenness
scale.
When plotted by richness level according to EDw and H val-
ues, numerical sequences in DATASET C had curvilinear distribution
patterns that overlapped those of DATASET B in Fig. 2. The
DATASET C distribution patterns, however, were less consistent
with respect to regression trend lines, because of the small sam-
ple sizes (i.e., 5 species, n = 6; 15 species, n = 10; and 25 species,
n = 3).

3.2. H multiple regression models

Two conventional multiple regression models (Eq. (1a) and Eq.

(1b), Table 2) that incorporated the full range of richness values
conformed to the model acceptance criteria, except for the lack
of normally distributed residuals (Fig. 3, K–S <0.01). The form and
characteristics of Eq. (1a) were replicated by Eq. (1b) using an inde-
pendent set of numerical sequences (Table 2). For this reason, only
Eq. (1a) is summarized in the following. The lack of normally dis-
tributed residuals for Eq. (1a) was due to 212 or 60% of the 350
possible values occurring in the 0–0.1 class of the frequency distri-
bution (Fig. 3). Overall, 80% of residual values occurred within ±0.1
of the observed H value. Because the deviations between observed
and predicted values were small, this lack of normally distributed
residuals was not considered a limitation to the reliable estimation
of H . All alpha and beta-coefficients of regression were significant Fig. 2. Distribution of 2500 randomly generated sets of numerical sequences per
at the P < 0.001 level. richness level (DATASET B) relative to evenness (EDw ) and Shannon–Wiener H val-
Eq. (1a) explained nearly all of the variance in observed H values ues, with polynomial regression curves and associated attributes. The lower line
of each richness level is the polynomial regression curve from Fig. 1. Dashed lines
(Table 2). Residuals in the model were considered homoscedastic represent trajectory extrapolations to the known end-points.
based on an Fmax -test probability value >0.05. All of the deviant
data points beyond two standard deviations from the regression
trend line (n = 16) represented over-estimates (Fig. 3). These over-
estimates formed no particular pattern based on their distribution
across the richness spectrum.
708 W.L Strong / Ecological Indicators 67 (2016) 703–713

Table 2
Richness and evenness (EDw and E ) based multiple regression equations for predicting Shannon–Wiener index (Ĥ ) values, and their associated evenness and richness (E:S)
ratios as derived from standardized partial beta-coefficients. Analysis based on DATASET A. See Table 1 for the mean and standard deviations of the independent-variables
by richness and sample size.
2
Eq. Regression model type and equation Richness range n R Radj SEE P E:Sc

Conventional multiple regression models

0a Ĥ = −0.8031 + 0.0348S + 3.2800EDw 4–35 1811 0.982 0.964 0.17 <0.001 –
1ab Ĥ = −0.6859 + 0.0336S + 3.2198EDw 4–35 350 0.992 0.984 0.11 <0.001 –
1b Ĥ = −0.6985 + 0.0346S + 3.1845EDw 4–35 350 0.990 0.979 0.12 <0.001 –

Conventional multiple regression models based on standardized partial beta-coefficients

2a Ĥ = 0 + 0.2265S + 0.9556EDw 4–35 350 0.992 0.984 0.11 <0.001 4.2:1
2b Ĥ = 0 + 0.2396S + 0.9780EDw 4–35 350 0.990 0.979 0.12 <0.001 4.1:1

EDw correlation matrix-based multiple regression models based on standardized partial beta-coefficients
3 Ĥ = 0 + 0.2265S + 0.9556EDw 4–35 350 0.992 0.984 0.11 <0.001 4.3:1
√
4 Ĥ = 0 + 0.2493S + 0.9547 EDw 4–35 350 0.992 0.984 0.11 <0.001 3.8:1

5 Ĥ = 0 + 0.2593S + 0.9827EDw 4–19 200 0.992 0.983 0.09 <0.001 3.8:1
√
6 Ĥ = 0 + 0.2569S + 0.9837 EDw 4–19 200 0.993 0.986 0.08 <0.001 3.8:1
7 Ĥ = 0 + 0.0658S + 0.9945EDw 20–35 200 0.997 0.994 0.07 <0.001 15.1:1
√
8 Ĥ = 0 + 0.0589S + 0.9911 EDw 20–35 200 0.993 0.987 0.11 <0.001 16.8:1

E correlation matrix-based multiple regression models based on standardized partial beta-coefficientsd

9 Ĥ = 0 + 0.2547S + 0.9013E 4–35 350 0.941 0.884 0.29 <0.001 3.6:1
√
10 Ĥ = 0 + 0.2946S + 0.9435 E 4–35 350 0.981 0.962 0.17 <0.001 3.2:1
11 Ĥ = 0 + 0.3084S + 0.9433E 4–19 200 0.946 0.895 0.23 <0.001 3.1:1
√
12 Ĥ = 0 + 0.3345S + 0.9826 E 4–19 200 0.981 0.963 0.14 <0.001 2.9:1
13 Ĥ = 0 + 0.0826S + 0.9413E 20–35 200 0.944 0.972 0.31 <0.001 11.4:1
√
14 Ĥ = 0 + 0.0708S + 0.9835 E 20–35 200 0.986 0.972 0.16 <0.001 13.9:1
a
This was not a valid model because the residuals were not normally (Kolmogorov–Smirnov one-sample test, P < 0.01) nor linear distributed. It was included only for
comparative purposes.
b
Equations (1a and 2a) and (1b and 2b) were derived from the same datasets. The “a” and “b” denote replicate analyses based on different sets of numerical sequences.
Equations (3) and (4) were derived from the Eq. (1a) dataset.
c
Values represent the size of the EDw beta-coefficient relative to the S beta-coefficient.
d
No valid conventional regression models were available due to a lack of normally distributed E values. Construction of these models paralleled those of EDw to allow
direct comparison.

DATASET C validation testing of Eq. (1a) and the combined use

of individual richness-based polynomial regression (IRPR) mod-
els yielded residuals that were normally distributed, linear, and
homoscedastic (Fig. 4). The correlation between observed and pre-
dicted H values was strongest for the IRPR model, although both
correlation coefficients were very strong relative to a critical test
value of ∼0.1 at the ˛ 0.05 level (degrees of freedom = 323). The
standard error of estimate was ∼18% smaller for the IRPR than
the Eq. (1a) model, with each under-estimating observed H by
∼20 and ∼37%, respectively, based on the deviation of their beta-
coefficients from a value of one. Eq. (1a) had a y-intercept that was
notably above zero (Fig. 4A). Fewer anomalous residuals occurred
in the IRPR (Fig. 4B) than the Eq. (1a) scatter diagram. In Fig. 4A,
the samples that deviated most from the regression line usually
contained only four species. Removal of this richness level (n = 9 of
2 )
325) from the regression increased total variance explanation (Radj
from 0.779 to 0.830.

3.3. Richness versus evenness

Correlation matrix-based (Eq. (3–8)) and conventional (Eq. (2a)

Fig. 3. Distribution and characterization of multiple regression Eq. (1a) residuals
based on simple linear regression and Pearson product-moment correlation. The
curved line within the Distribution of Residuals inset is a normal distribution curve,
with normality assessed using a Kolmogorov–Smirnov one-sample test (K–S). The
Fmax -test was used to assess homoscedasticity.
and (2b)) multiple regressions based on standardized partial beta-
coefficients explained similar amounts of variance and had similar
standard errors of estimate (Table 2). All of these equations were
significant at the P < 0.001 level and explained ∼90% or more of
the variance in observed H values. EDw :S ratios ranged from 3.8
to 4.2:1, respectively, for 4–35 species. These ratios were similar to
equation based on 4–19 species, but the differential jumped ∼4-fold
for 20–35 species (Table 2). Correlation matrix-based regression
models that used E as the measured of evenness had E:S ratios
that were slightly but consistently less than those based on EDw
(Table 2).
 W.L Strong / Ecological Indicators 67 (2016) 703–713
Fig. 4. Validation results for multiple regression Eq. (1a) (A) and individual
richness-based polynomial regression (IRPR) equations (B) based on simple
linear regression and the distribution of residuals. Dashed lines indicate a
1:1 ratio between the independent- and dependent-variables. Observed H
values were based DATASET C. Residuals were check for normality based
on Kolmogorov–Smirnov one-sample tests (K–S) and homoscedasticity by the
Fmax -tests.
3.4. H and effective species
Effective species numbers calculated by the diversity model
D = S × E, using DATASET A derived values, produced linear
sequences of values relative to evenness, regardless of the richness
level (Fig. 5). H -derived effective species numbers (exp H ) based
on the same data yielded curvilinear patterns relative to evenness
(Fig. 5). Deviation from the S × E lines increased with richness and
was greatest in the center of evenness scale. Exp H values were
consistently less than those of the S × E models, when richness was
greater than seven species. For five species, the maximum differen-
tial was 0.4 species. In contrast, it was as great as 10 effective species
for the 35 species richness level. Maximum absolute deviations
for 15, 25, and 35 species occurred when evenness was 0.50–0.55,
but the greatest proportional deviations occurred at low to mod-
erate levels of evenness (0.15–0.40). For the latter, the maximum
709
Fig. 5. A comparison of effective species numbers (D) generated by the model
D = S × E and exp H for selected levels of richness. H values were from DATASET
A. Each dot represents an individual exp H value. S – richness, and H –
Shannon–Wiener index.
deviations from the expected S × E values were 48% (S = 15), 61%
(S = 25), and 69% (S = 35).
3.5. H equation deconstruction
Reverse calculation of the Shannon–Wiener equation yielded
−H values that ranged from 0 to 1. Values of −H were
strongly and inversely correlated with EDw (r = −0.867, P < 0.001,
n = 1811).
3.6. Graphic model of H relationships
Fig. 6 illustrates the numerical relationships among even-
ness (EDw ), richness, and H . H values were curvilinear functions
of richness and EDw , with the most rapid changes in even-
ness occurring when richness was <15 species. The vertical
710 W.L Strong / Ecological Indicators 67 (2016) 703–713
Fig. 6. A graphic model that illustrates the response of evenness to differences in
richness (5–35 species) and Shannon–Wiener H values. H values based on the aver-
age of the Eq. (1a) and individual richness-based polynomial regression trend line
values.
spacing between EDw increments decreased with increasing H
(Fig. 6).
4. Discussion
4.1. Procedural considerations
Individual richness-based polynomial regressions (IRPR) were
strong predictors of H , based on a >98% explanation of variance
by each equations. It might be validly argued that these very high
degrees of explanation were a function of the method used to
compile DATASET A. However, only slightly less explanation was
obtained for each of the four sets of 2500 random regenerated
2 = 84 − 95%). Even if the DATASET A com-
numerical sequences (Radj
pilation method optimized the amount of explained variance, this
does not invalidate the unequal numerical relationships identified
in E:S ratios with respect to the determination of H (Table 2). As
well, the IRPR models offer a perspective of the numerical rela-
tionship between evenness and richness, when the amount of
change in H is standardized within and across different richness
levels.
Another potential procedural concern might be the force-fitting
of richness values into normal distributions. Whereas this proce-
dure was not ideal, few options existed to correct the problem,
except to calculate H from individual IRPR equations. For exam-
ple, direct random sampling of DATASET A was not a viable option,
because any properly formulated and implemented sampling strat-
egy would largely replicate the original distribution pattern. As
well, conventional data transformation functions (e.g., square root,
logs, reciprocal, powers) cannot convert a systematically increas-
ing set of frequencies into a normal distribution. Even if possible,
interpretation of the results might be problematic (Tabachnick and
Fidell, 2013, p. 86). Because regression models based on DATASET A
(Eq. 0, n = 1811) and the 350 sample subsets produced very similar
equations (Table 2), force-fitting of the richness into normal dis-
tributions (Eqs. 1a and 1b) had essentially no effect on the size of
individual regression beta-coefficients or the amount of variance
explanation.
4.2. H evenness and richness relationships
A measure of diversity should assess the relationship between
richness and evenness in an unbiased manner. However, mul-
tiple regression E:S ratios indicate that richness plays a much
smaller role than evenness in determining H , and its contribu-
tion decreased with increasing richness. Conceptually, richness
establishes the starting-point, end-point, and general trajectory of
values for a specific level of richness, but evenness dictates the
actual H value (cf. Fig. 1). Therefore, an H value can be defined
as a logarithm-weighted measure (i.e., pi × loge pi ) of evenness
at a specific level of richness. This makes H a quasi-evenness
index, which is contrary to the conclusion by Yue et al. (2007,
p. 1611) that H “cannot express the evenness component” of
Shannon–Wiener index. In studies that identified richness as more
important than evenness for explaining differences in diversity
(e.g., Kricher, 1972; Loya, 1972; Tramer, 1969), they typically had
diverse richness with similar evenness, which neutralized the influ-
ence of evenness as an explanatory variable. This response pattern
is consistent with one of the desirable properties of a diversity
index (No. 2, Pielou, 1975, p. 7). It, however, should be appreci-
ated that the recognition of richness as the more important variable
in these studies was based on the comparison of difficult bio-
logical groups (i.e., an ecological consideration) rather than the
relationship between richness and evenness within the calcula-
tion of H values. This is a subtle but important interpretational
distinction.
Although variation in H for a given level of richness is driven
by evenness, the strong inverse correlation between EDw and −H
indicates the Shannon–Wiener index actually measures dominance
concentration. This is not a contradiction of the previously stated
definition because the two measures are inversely related (i.e.,
Dominance Concentration = 1 − E + error). The transformation from
a measure of dominance concentration to evenness occurs due to
multiplication of the principal components of the Shannon–Wiener
equation ( (pi × loge pi )) by −1. The inclusion of −l in the equation
appears to have been purely an interpretation convenience used to
remove the negative sign from the −H portion of the calculation
(Shannon and Weaver, 1949, p. 15), which occurs when propor-
tional abundances are converted to loge values. If the negative sign
of −H had not been removed, the interpretation of H might have
evolved much differently and been far less contentious, because
anti-logged −H values would have range from 0 to 1 regardless of
richness, rather than individual richness levels having their own
scaling.
Because the Shannon–Wiener index is calculated exclusively
from abundance values, how richness accounts for a portion of the
variance in H had not been explained. One possible explanation
is that richness is indirectly represented through the stretching
of the 0–1 EDw scale to accommodate the increased lengthening
of the H scale with greater richness. For example, maximum H
or loge S for five species (1.61) is less than half the value for 35
species (3.55), although both richness levels have EDw values of
1.0 (cf. Figs. 1 and 6). When richness exceeds >25–30 species, this
bias is less due to reduced absolute differences between loge S val-
ues (Fig. 1). This rescaling issue may limit the legitimate statistical
comparison of H samples, except those with the same richness
level.
4.3. Variation in H
DATASET A values were consistently located beneath those of
DATASET B in Fig. 2, but both were part of the same sample
distribute pattern when they had the same level of richness. This
suggests the distribution patterns of richness-based E · H values
 W.L Strong / Ecological Indicators 67 (2016) 703–713
are lens-shaped, with a convex upper and a concave lower bound-
ary (Fig. 2). The degree of upward arching increases with richness.
The potential range of E · H values varies according to the degree
of evenness. For example, only one combination of proportional
abundance values can satisfy the requirements of perfect even-
ness (i.e., pi = 1/S) and absolute dominance concentration (pi = 1,
S = 1) at any level of richness. However, the number of possible E ·
H combinations that represent either lesser or greater evenness
progressively increases toward the center of the evenness scale
along with variability. These different ranges of variation along the
evenness scale are a function of how variable any combination of
proportional abundance value can be and still yield a similar H
value.
If DATASET A and DATASET B indicate the potential range of E ·
H values at a given richness level, then a regression line located
midway between the upper and lower limits of the data along
the evenness scale would represent a more accurate H model
than those associated with Fig. 2. Based on this scenario, H val-
ues would deviate about ±20% from this new trend line within
the 0.4–0.6 range of evenness. Because Eq. (1a) and the IRPR equa-
tions were both derived from DATASET A, a 20% downward shift
in H values would explain most of its under-prediction in Fig. 4A
and nearly all of it Fig. 4B. The poorer ability of Eq. (1a) to pre-
dict H values relative to the IRPR method occurs because multiple
regression modeling primarily reflects the central tendency of the
data. Therefore, more extreme values are less reliably predicted.
Although both Eq. (1a) and the IRPR method of H prediction were
statistically significant, neither is advocated for use because each
represents a subset of the overall E · H variance at a given level of
richness.
4.4. H interpretation issues
Fig. 6 can be used to visualize the expected nonlinear and sys-
tematically changing relationships between richness and evenness,
with respect to H . A previous analysis by Buzas and Hayek (1996,
p. 41) hints at a similar presentation, but Fig. 6 is more comprehen-
sive. The primary advantage of Fig. 6 is the trivariate perspective
it offers of evenness, which is not directly possible from multiple
regression equations. Because Fig. 6 was constructed from a limited
sample (DATASET A and B), it is only an approximation of the
true relationships between richness and evenness relative to H . A
much larger (e.g., ×1000) and a much more compositionally diverse
dataset than was used would be required to create definitively
models.
Even if H could be reliably predicted, how to interpret the values
is an open question. Loge weighting exaggerates the importance
of species proportional abundance values ≤0.15, which are often
the most common components of an assemblage (e.g., Akhoundi
et al., 2013; La Roi and Hnatiuk, 1980; Strong, 2014), while
systematically reducing those >0.55 relative to the intervening
values (cf. Peet, 1974, p. 295). As a result, the actual relation-
ships among species proportional abundances are distorted. This
could explains why H -derived effective species numbers (exp H )
deviate from the expected numbers as per the D = S × E diver-
sity model, particularly in the lower portion of the evenness
scale. Such distortion could easily lead to a misinterpretation of
diversity.
The statistical comparison of H values as a basis for testing
diversity hypotheses may not produce insightful or logical results
that reflect the true interactions between richness and evenness,
if a null hypothesis is accepted (cf. Yue et al., 2007, p. 1612).
For example, if samples with five and 35 species had H values
of ∼1.5, the associated levels of evenness would be ∼0.86 (high
711
evenness) and ∼0.28 (low evenness), respectively (cf. Fig. 6). No
form of statistical analysis or even the most liberal qualitative
assessment could justify recognition of a difference between the
two H values, despite a substantial difference in richness and even-
ness. Superficially, it could be argued that the two samples were
similar from a diversity perspective. Yet, proportional abundance
values (pi ) taken from DATASET A seem incongruent. The second
sample (pi = 0.659, ∼0.029[n = 11], 0.025, and 0.0001[n = 23]) com-
pared to the first (pi = 0.335, 0.200[n = 3] and 0.065) had seven times
more species, the dominant was twice as large, and 34 of 35 species
had pi < 0.03 compared to none in the first sample. If the two sam-
ples were forest communities, with the largest pi in each sample
representing the tree canopy, it is unlikely they would be consid-
ered similar from either an ecologically or a diversity perspective,
and would likely be entirely different plant communities that exist
under distinctly different ecological regimes. Unfortunately, such
anomalies in richness and evenness are difficult to recognize based
on H values alone. Similar concerns were expressed by Harper
and Hawksworth (1996) who thought the Shannon–Wiener index
produced counter-intuitive diversity results, when species rich
communities with low evenness could have lower H values than
those with low richness and high evenness. This inconsistency may
stem from the limited role that richness plays in the determination
of H .
Most diversity studies statistically test and report richness
differences in conjunction with H analyses. However, fewer like-
wise assess (un)evenness or use an index that is independent
of H . As a result, it can be difficult to determine whether rich-
ness or (un)evenness, or both were responsible for differences
between tested groups (cf. Zhang et al., 2014, p. 316); or if
they differed when no significant H difference was reported. For
this reason, it would be prudent to test simultaneously grouped
samples for differences in H , richness, and evenness. Evenness
measures such the Pielou J and the Heip (1974) index should
be avoided due to the use of H in their equations (Alatalo,
1981).
5. Conclusions
The Shannon–Wiener index is a biased measure of diversity
based on the assessed data and applied methods of analysis due
to the unequal roles of richness and evenness. The logical solu-
tion to this problem would be to abandon its use in favor of a
more justifiable diversity index, possibly one based on absolute
rather than proportional abundance values (Ricotta, 2003), or to
solely rely on independent analyses of richness and (un)evenness.
The latter option would likely fulfill the basic needs of most com-
parative ecological studies. It, however, seems unlikely that use
of H will be abandoned in the foreseeable future due its deep
entrenchment in biodiversity assessment methods, even though
separate analyses of richness and (un)evenness would require min-
imal, if any, additional effort. As an alternative to abandonment,
summary statistics and the results of inferential testing of rich-
ness and (un)evenness should be provided in conjunction with
H to ensure the proper portrayal of abundance data in ecology
research.
Acknowledgements
Yukon Energy, Mines and Resources Library acquired
and provided documents that were locally unavailable. M.
Jacqueline Redburn provided comments on a draft of the
manuscript.
712 W.L Strong / Ecological Indicators 67 (2016) 703–713

Appendix A.

Table A1
An example of how series of synthetic numerical sequences were compiled for five species (cf. Section 2.1) with associated H and evenness (EDw ) values.

No. Species H H reduction EDw

1 2 3 4 5

Abundance value
1 20.00 20.00 20.00 20.00 20.00 1.6094 – 0.000
2 29.80 20.00 20.00 20.00 10.20 1.5593 ∼0.050 0.098
3 33.50 20.00 20.00 20.00 6.50 1.5097 ∼0.050 0.135
4 36.10 20.00 20.00 20.00 3.90 1.4600 ∼0.050 0.161
5 38.05 20.00 20.00 20.00 1.95 1.4101 ∼0.050 0.180
6 39.47 20.00 20.00 20.00 0.53 1.3604 ∼0.050 0.195
7 42.79 20.00 20.00 17.20 0.01 1.3107 ∼0.050 0.228
8 47.29 20.00 20.00 12.70 0.01 1.2609 ∼0.050 0.273
9 50.64 20.00 20.00 9.35 0.01 1.2108 ∼0.050 0.306
10 53.29 20.00 20.00 6.70 0.01 1.1612 ∼0.050 0.333
11 55.49 20.00 20.00 4.50 0.01 1.1111 ∼0.050 0.355
12 57.29 20.00 20.00 2.70 0.01 1.0613 ∼0.050 0.373
13 58.76 20.00 20.00 1.23 0.01 1.0112 ∼0.050 0.388
14 59.84 20.00 20.00 0.15 0.01 0.9617 ∼0.050 0.398
15 63.28 20.00 16.70 0.01 0.01 0.9122 ∼0.050 0.433
16 66.68 20.00 13.30 0.01 0.01 0.8623 ∼0.050 0.467
17 69.53 20.00 10.45 0.01 0.01 0.8124 ∼0.050 0.495
18 71.98 20.00 8.00 0.01 0.01 0.7624 ∼0.050 0.520
19 74.11 20.00 5.87 0.01 0.01 0.7122 ∼0.050 0.541
20 75.93 20.00 4.05 0.01 0.01 0.6627 ∼0.050 0.559
21 77.50 20.00 2.48 0.01 0.01 0.6130 ∼0.050 0.575
22 78.83 20.00 1.15 0.01 0.01 0.5626 ∼0.050 0.588
23 79.84 20.00 0.14 0.01 0.01 0.5127 ∼0.050 0.598
24 82.70 17.27 0.01 0.01 0.01 0.4631 ∼0.050 0.627
25 85.67 14.30 0.01 0.01 0.01 0.4134 ∼0.050 0.657
26 88.32 11.65 0.01 0.01 0.01 0.3629 ∼0.050 0.683
27 90.67 9.30 0.01 0.01 0.01 0.3125 ∼0.050 0.707
28 92.74 7.23 0.01 0.01 0.01 0.2626 ∼0.050 0.727
29 94.60 5.37 0.01 0.01 0.01 0.2123 ∼0.050 0.746
30 96.24 3.73 0.01 0.01 0.01 0.1623 ∼0.050 0.762
31 97.67 2.30 0.01 0.01 0.01 0.1126 ∼0.050 0.777
32 98.89 1.08 0.01 0.01 0.01 0.0627 ∼0.050 0.789
33 99.84 0.13 0.01 0.01 0.01 0.0130 ∼0.050 0.798

Table A2
Individual richness-based polynomial regression (IRPR) equations that characterize the distribution of incrementally-reduced H values (DATASET A) for 32 synthetic datasets
based on differences in richness (cf. Fig. 1).

Eq. Richness Equation SEE R R2 n P-value

adj

A.1 4 Ĥ 2
= −0.4946 + 2.1253EDw + 0.6885EDw − 0.9062EDw 3 0.028 0.995 0.984 28 <0.001
A.2 5 Ĥ = −0.4529 + 2.4768EDw + 0.3938EDw 2 − 0.7807EDw 3 0.023 0.999 0.988 33 <0.001
A.3 6 Ĥ = −0.3835 + 2.5448EDw + 0.5221EDw 2 − 0.8678EDw 3 0.019 0.999 0.989 36 <0.001
A.4 7 Ĥ = −0.3475 + 2.7086EDw + 0.4742EDw 2 − 0.8673EDw 3 0.016 0.999 0.990 39 <0.001
A.5 8 Ĥ = −0.3273 + 2.9025EDw + 0.3411EDw 2 − 0.8130EDw 3 0.015 0.999 0.991 42 <0.001
A.6 9 Ĥ = −0.2941 + 2.9875EDw + 0.3824EDw 2 − 0.8564EDw 3 0.013 0.999 0.991 45 <0.001
A.7 10 Ĥ = −0.2728 + 3.0958EDw + 0.3558EDw 2 − 0.8523EDw 3 0.012 0.999 0.991 46 <0.001
A.8 11 Ĥ = −0.2618 + 3.2440EDw + 0.2430EDw 2 − 0.8018EDw 3 0.011 0.999 0.991 48 <0.001
A.9 12 Ĥ = −0.2413 + 3.2995EDw + 0.3016EDw 2 − 0.8495EDw 3 0.010 0.999 0.991 50 <0.001
A.10 13 Ĥ = −0.2322 + 3.4150EDw + 0.2241EDw 2 − 0.8140EDw 3 0.010 0.999 0.991 52 <0.001
A.11 14 Ĥ = −0.2114 + 3.4364EDw + 0.3191EDw 2 − 0.8807EDw 3 0.009 0.999 0.991 53 <0.001
A.12 15 Ĥ = −0.1989 + 3.4941EDw + 0.3295EDw 2 − 0.8896EDw 3 0.009 0.999 0.992 53 <0.001
A.13 16 Ĥ = −0.1912 + 3.5791EDw + 0.2902EDw 2 − 0.8784EDw 3 0.008 0.999 0.992 55 <0.001
A.14 17 Ĥ = −0.1847 + 3.6549EDw + 0.2605EDw 2 − 0.8681EDw 3 0.008 0.999 0.992 57 <0.001
A.15 18 Ĥ = −0.1745 + 3.7028EDw + 0.2675EDw 2 − 0.8755EDw 3 0.008 0.999 0.993 58 <0.001
A.16 19 Ĥ = −0.1629 + 3.7308EDw + 0.3144EDw 2 − 0.9087EDw 3 0.008 0.999 0.993 58 <0.001
A.17 20 Ĥ = −0.1541 + 3.7687EDw + 0.3432EDw 2 − 0.9329EDw 3 0.008 0.999 0.993 60 <0.001
A.18 21 Ĥ = −0.1530 + 3.8678EDw + 0.2611EDw 2 − 0.8907EDw 3 0.008 0.999 0.993 61 <0.001
A.19 22 Ĥ = −0.1462 + 3.9371EDw + 0.2028EDw 2 − 0.8719EDw 3 0.008 0.999 0.993 62 <0.001
A.20 23 Ĥ = −0.1402 + 3.9414EDw + 0.2651EDw 2 − 0.8993EDw 3 0.008 0.999 0.993 64 <0.001
A.21 24 Ĥ = −0.1348 + 3.9836EDw + 0.2710EDw 2 − 0.9056EDw 3 0.008 0.999 0.993 64 <0.001
A.22 25 Ĥ = −0.1233 + 3.9782EDw + 0.3611EDw 2 − 0.9663EDw 3 0.007 0.999 0.993 64 <0.001
A.23 26 Ĥ = −0.1240 + 4.0591EDw + 0.2765EDw 2 − 0.9231EDw 3 0.008 0.999 0.993 65 <0.001
A.24 27 Ĥ = −0.1179 + 4.0755EDw + 0.3248EDw 2 − 0.9559EDw 3 0.007 0.999 0.993 66 <0.001
A.25 28 Ĥ = −0.1111 + 4.0990EDw + 0.3421EDw 2 − 0.9672EDw 3 0.007 0.999 0.993 67 <0.001
A.26 29 Ĥ = −0.1086 + 4.1419EDw + 0.3308EDw 2 − 0.9645EDw 3 0.008 0.999 0.994 68 <0.001
A.27 30 Ĥ = −0.1056 + 4.2110EDw + 0.2411EDw 2 − 0.9109EDw 3 0.008 0.999 0.994 68 <0.001
A.28 31 Ĥ = −0.1005 + 4.2051EDw + 0.3294EDw 2 − 0.9675EDw 3 0.007 0.999 0.994 69 <0.001
A.29 32 Ĥ = −0.0965 + 4.2328EDw + 0.3382EDw 2 − 0.9756EDw 3 0.007 0.999 0.994 69 <0.001
A.30 33 Ĥ = −0.0934 + 4.2671EDw + 0.3250EDw 2 − 0.9681EDw 3 0.007 0.999 0.994 70 <0.001
A.31 34 Ĥ = −0.0887 + 4.2828EDw + 0.3558EDw 2 − 0.9885EDw 3 0.007 0.999 0.994 70 <0.001
A.32 35 Ĥ = −0.0867 + 4.3305EDw + 0.3026EDw 2 − 0.9536EDw 3 0.008 0.999 0.994 71 <0.001

Total 1811
 W.L Strong / Ecological Indicators 67 (2016) 703–713
References
Adie, H., Rushworth, I., Lawes, M.J., 2013. Pervasive, long-lasting impact of his-
torical logging on composition, diversity and above ground carbon stocks in
Afrotemperate forest. For. Ecol. Manage. 310, 887–895, http://dx.doi.org/10.
1016/j.foreco.2013.09.037.
Akhoundi, M., Mirzaei, A., Beghaei, A., Alten, B., Depaquit, J., 2013. Sand fly (Diptera:
Pyschodidae) distribution in the endemic and non-endemic foci of visceral leish-
maniasis in northwestern Iran. J. Vector Ecol. 38, 97–104, http://dx.doi.org/10.
1111/j.1948-7134.2013.12014.x.
Alatalo, R.V., 1981. Problems in the measurement of evenness in ecology. Oikos 37,
199–204, http://dx.doi.org/10.2307/3544465.
Barbour, M.G., Burk, J.H., Pitts, W.D., Gilliam, F.S., Schwartz, M.W., 1999. Terrestrial
Plant Ecology, 3rd ed. Benjamin/Cummings, Menlo Park.
Barrantes, G., Sandoval, L., 2009. Conceptual and statistical problems associated with
the use of diversity indices in ecology. Rev. Biol. Trop. 57, 451–461.
Buzas, M.A., Hayek, L.-A.C., 1996. Biodiversity resolution: an integrated approach.
Biodivers. Lett. 3, 40–43, http://dx.doi.org/10.2307/2999767.
Cağlar, S., Albayrak, S., 2012. Assessment of ecological quality status of
Küçükçekmece Bay (Marmara Sea) by applying BENTIX, AMBI, BOPA and BO2A
biotic indexes. Medit. Mar. Sci. 13, 198–207.
Camargo, J.A., 1993. Must dominance increase with the number of subordinate
species in competitive interactions? J. Theor. Biol. 161, 537–542, http://dx.doi.
org/10.1006/jtbi.1993.1072.
Camargo, J.A., 2008. Revisiting the relation between species diversity and informa-
tion theory. Acta Biotheor. 56, 275–283, http://dx.doi.org/10.1007/s10441-008-
9053-x.
Con, T.V., Thang, N.T., Ha, D.T.T., Khiem, C.C., Quy, T.H., Lam, V.T., Do, T.V., Sato,
T., 2013. Relationship between aboveground biomass and measures of struc-
ture and species diversity in tropical forests of Vietnam. For. Ecol. Manage. 310,
213–218, http://dx.doi.org/10.1016/j.foreco.2013.08.034.
De Grandpré, L., Boucher, D., Bergeron, Y., Gagnon, D., 2011. Effects of small canopy
gaps on boreal mixedwood understory vegetation dynamics. Community Ecol.
12, 67–77, http://dx.doi.org/10.1556/ComEc.12.2011.1.9.
DeJong, T.M., 1975. A comparison of three diversity indices based on their compo-
nents of richness and evenness. Oikos 26, 222–227, http://dx.doi.org/10.2307/
3543712.
Goodman, D., 1975. The theory of diversity-stability relationships in ecology. Quat.
Rev. Biol. 50, 237–266, http://dx.doi.org/10.1086/408563.
Gosselin, F., 2001. Lorenz partial order: the best known logical framework to
define evenness indices. Community Ecol. 2, 197–207, http://dx.doi.org/10.
1556/ComEc.2.2001.2.7.
Harper, J.L., Hawksworth, D.L., 1996. Preface. In: Hawksworth, D.L. (Ed.), Biodiver-
sity: Measurement and Estimation. Chapman & Hall, London, pp. 5–12.
Hart, S.A., Chen, H.Y.H., 2008. Fire, logging, and overstory affect understory abun-
dance, diversity, and composition in boreal forest. Ecol. Monogr. 78, 123–140,
http://dx.doi.org/10.1890/06-2140.1.
Heip, C., 1974. A new index measuring evenness. J. Mar. Bio. Assoc. U.K. 54, 555–557.
Hill, M.O., 1973. Diversity and evenness: a unifying notation and its consequences.
Ecology 54, 427–432, http://dx.doi.org/10.2307/1934352.
Hurlbert, S.H., 1971. The nonconcept of species diversity: a critique and alternative
parameters. Ecology 52, 577–586, http://dx.doi.org/10.2307/1934145.
Jost, L., 2006. Entropy and diversity. Oikos 113, 363–375, http://dx.doi.org/10.1111/
j.2006.0030-1299.14714.x.
Jost, L., 2010. The relation between evenness and diversity. Diversity 2, 207–232,
http://dx.doi.org/10.3390/d2020207.
Kent, M., Coker, P., 1992. Vegetation Description and Analysis: A Practical Approach.
CRC Press, Boca Raton.
Krebs, C.J., 2009. Ecology: The Experimental Analysis of Distribution and Abundance,
6th ed. Benjamin Cummings, San Francisco.
Kricher, J.C., 1972. Bird species diversity: the effect of species richness and equi-
tability on the diversity index. Ecology 53, 278–282, http://dx.doi.org/10.2307/
1934082.
Külköylüoğlu, O., Yavuzatmaca, M., Akmemir, D., Sari, N., 2012. Distribution and
local species diversity of freshwater Ostracoda in relation to habitat in the
Kahramanmaraş Province of Turkey. Int. Rev. Hydrobiol. 97, 247–261, http://
dx.doi.org/10.1002/iroh.201290002.
Külköylüoğlu, O., Akdemir, D., Sari, N., Yavuzatmaca, M., Oral, C., Başak, E., 2013.
Distribution and ecology of Ostracoda (Crustacea) from troughs in Turkey. Turk.
J. Zool. 37, 277–287.
La Roi, G.H., Hnatiuk, R.J., 1980. The Pinus contorta forests of Banff and Jasper National
Parks: a study in comparative synecology and syntaxonomy. Ecol. Monogr. 50,
1–29, http://dx.doi.org/10.2307/2937244.
Lloyd, M., Ghelandi, R.J., 1964. A table for calculating the ‘equitability’ component of
species diversity. J. Anim. Ecol. 33, 217–225, http://dx.doi.org/10.2307/2628.
Loya, Y., 1972. Community structure and species diversity of hermatypic
corals at Eilat, Red Sea. Mar. Biol. 13, 100–123, http://dx.doi.org/10.1007/
BF00366561.
Magurran, A.E., 2003. Measuring Biological Diversity. Blackwell Publishing Com-
pany, Oxford.
713
MacArthur, R., 1955. Fluctuations of animal populations and a measure of commu-
nity stability. Ecology 36, 533–536, http://dx.doi.org/10.2307/1929601.
MacArthur, R.H., 1965. Patterns of species diversity. Biol. Rev. 40, 510–533, http://
dx.doi.org/10.1111/j.1469-185X.1965.tb00815.x.
McGrew Jr., J.C., Monroe, C.B., 2000. An Introduction to Statistical Problem Solving
in Geography, 2nd ed. McGraw-Hill Companies Ltd., Boston.
Molles Jr., M.C., 2002. Ecology: Concepts and Applications, 2nd ed. McGraw-Hill,
New York.
Monk, C.D., 1967. Tree species diversity in the eastern deciduous forest with par-
ticular reference to north central Florida. Am. Nat. 101, 173–187, http://dx.doi.
org/10.1086/282482.
Odum, E.P., 1969. The strategy of ecosystem development. Science 164, 262–270,
http://dx.doi.org/10.1126/science.164.3877.262.
Patten, B.C., 1959. An introduction to the cybernetics of the ecosystem: the tropic-
dynamic aspects. Ecology 40, 221–231, http://dx.doi.org/10.2307/1930032.
Peet, R.K., 1974. The measurement of species diversity. Annu. Rev. Ecol. Syst. 15,
285–307, http://dx.doi.org/10.1146/annurev.es.05.110174.001441.
Pielou, E.C., 1966. The measurement of diversity in different types of biolog-
ical collections. J. Theor. Biol. 13, 131–144, http://dx.doi.org/10.1016/0022-
5193(66)90013-0.
Pielou, E.C., 1975. Ecological Diversity. John Wiley & Sons Inc., New York.
Ricotta, C., 2003. Parametric scaling from species relative abundance to abso-
lute abundances in the computation of biological diversity: a first proposal
using Shannon’s entropy. Acta Biotheor. 51, 181–188, http://dx.doi.org/10.1016/
S0022-5193(03)00026-2.
Risser, P.G., Rice, E.L., 1971. Diversity in tree species in Oklahoma upland forests.
Ecology 52, 876–880, http://dx.doi.org/10.2307/1936036.
Seiwa, K., Eto, Y., Hishita, M., Masaka, K., 2012. Effects of thinning intensity on species
diversity and timber production in a conifer (Cryptomeria japonica) plantation
in Japan. J. For. Res. 17, 468–478, http://dx.doi.org/10.1007/s10310-011-0316-
z.
Shannon, C.E., 1948. A mathematical theory of communication. Bell Syst. Tech. J. 27,
379–423, http://dx.doi.org/10.1002/j.1538-7305.1948.tb01338.x.
Shannon, C.E., Weaver, W., 1949. The Mathematical Theory of Communication. Uni-
versity of Illinois Press, Urbana.
Skácelová, O., Lepš, J., 2014. The relationship of diversity and biomass in phy-
toplankton communities weakens when accounting for species proportions.
Hydrobiologia 724, 67–77, http://dx.doi.org/10.1007/s10750-013-1723-2.
Smith, B., Wilson, J.B., 1996. A consumer’s guide to evenness indices. Oikos 76, 70–82,
http://dx.doi.org/10.2307/3545749.
Sokal, R.R., Rohlf, F.J., 1995. Biometry, 2nd ed. W.H. Freeman and Company, New
York.
Stirling, G., Wilsey, G., 2001. Empirical relationships between species richness, even-
ness, and proportional diversity. Am. Nat. 158, 286–299, http://dx.doi.org/10.
1086/321317.
Strong, W.L, 1999. Botanical, vegetation and ecological resource survey of the pro-
posed Coal Valley II Mine extension and subregional ecosection classification. In:
Coal Valley Mine Extension Project. Luscar Ltd., Edson, Alberta, Canada, Consult.
Rep. 9.
Strong, W.L, 2002. Assessing species abundance unevenness within and between
plant communities. Community Ecol. 3, 237–246, http://dx.doi.org/10.1556/
ComEc.3.2002.2.9.
Strong, W.L, 2014. Northernmost North American Pinus contorta var. latifolia (lodge-
pole pine) sociations and vegetation diversity relative to its central range east
of the Rocky Mountains. Nord. J. Bot. 32, 222–232, http://dx.doi.org/10.1111/j.
1756-1051.2013.00126.x.
Strong, W.L, 2015. High-latitude Yukon boreal-cordilleran grassland plant commu-
nities. Arctic 68, 69–78, http://dx.doi.org/10.14430/arctic4453.
Tabachnick, B.G., Fidell, L.S., 2013. Using Multivariate Statistics, 6th ed. Pearson
Education Inc., Boston.
Tramer, E.J., 1969. Bird species diversity: components of Shannon’s formula. Ecology
50, 927–929, http://dx.doi.org/10.2307/1933715.
Tuomisto, H., 2010. A consistent terminology for quantifying species diversity? Yes,
it does exist. Oecologia 164, 853–860, http://dx.doi.org/10.1007/s00442-010-
1812-0.
Tuomisto, H., 2012. An updated consumer’s guide to evenness and related indices.
Oikos 121, 1203–1218, http://dx.doi.org/10.1111/j.1600-0706.2011.19897.x.
Whittaker, R.H., 1972. Evolution and measurement of species diversity. Taxon 21,
213–251, http://dx.doi.org/10.2307/1218190.
Wu, L., Liu, J., Takashima, A., 2013. Effects of selective logging on stand structure and
tree species diversity in a subtropical evergreen broad-leaved forest. Ann. For.
Sci. 70, 535–543, http://dx.doi.org/10.1007/s13595-013-0292-x.
Yue, T.X., Ma, S.N., Wu, S.X., Zhan, J.Y., 2007. Comparative analyses of the scaling
diversity index and its applicability. Int. J. Rem. Sens. 28, 1611–1623, http://dx.
doi.org/10.1080/01431160600887714.
Zar, J.H., 2010. Biostatistical Analysis, 5th ed. Prentice Hall, Upper Saddle River.
Zhang, D., Zhang, J., Yang, W., Wu, F., Huang, Y., 2014. Plant and soil seed bank
diversity across a range of ages of Eucalyptus grandis plantations afforested on
arable lands. Plant Soil 376, 307–325, http://dx.doi.org/10.1007/s11104-013-
1954-z.