You are on page 1of 6

Journal of Ethnopharmacology 142 (2012) 259–264

Contents lists available at SciVerse ScienceDirect

Journal of Ethnopharmacology
journal homepage: www.elsevier.com/locate/jep

An imprecise probability approach for the detection of over and underused


taxonomic groups with the Campania (Italy) and the Sierra Popoluca
(Mexico) medicinal flora
Caroline S. Weckerle a, Stefano Cabras b, Maria Eugenia Castellanos c, Marco Leonti d,n
a
Institute of Systematic Botany, University of Zürich, Zollikerstrasse 107, CH-8008 Zürich, Switzerland
b
Department of Mathematics and Informatics, Universita di Cagliari, Via Ospedale 72, 09124 Cagliari (CA), Italy
c
Department of Statistics and O.R., Rey Juan Carlos University, C/Tulipan, SN-28938 Madrid, Spain
d
Dipartimento Farmaco Chimico Tecnologico, Universita di Cagliari, Facolta di Farmacia, Via Ospedale 72, 09124 Cagliari (CA), Italy

a r t i c l e i n f o abstract

Article history: Aim of the study: We use the IDM model to test for over- and underuse of plant taxa as source for
Received 5 February 2012 medicine. In contrast to the Bayes approach, which only considers the uncertainty around the data of
Received in revised form medicinal plant surveys, the IDM model also takes the uncertainty around the inventory of the flora
2 April 2012
into account, which is used for the comparison between medicinal and local floras.
Accepted 1 May 2012
Available online 10 May 2012
Materials and methods: Statistical analysis of the medicinal flora of Campania (Italy) and of the
medicinal flora used by the Sierra Popoluca (Mexico) was performed with the IDM model and the
Keywords: Bayes approach. For Campania 423 medicinal plants and 2237 vascular plant species and for the Sierra
Bayesian approach Popoluca 605 medicinal plants and 2317 vascular plant species were considered.
Imprecise Dirichlet Model (IDM)
Results: The IDM model (s ¼4) indicates for Campania the Lamiaceae and Rosaceae as overused, and the
Medicinal plants
Caryophyllaceae, Poaceae, and Orchidaceae as underused. Among the Popoluca the Asteraceae and
Pharmacopoeia
Quantitative ethnobotany Piperaceae turn out to be overused, while Cyperaceae, Poaceae, and Orchidaceae are underused. In
comparison with the Bayes approach, the IDM approach indicates fewer families as over- or underused.
Conclusions: The IDM model leads to more conservative results compared to the Bayes approach. Only
relatively few taxa are indicated as over- or underused. The larger the families (nj’s) are, the more
similar do the results of the two approaches turn out. In contrast to the Bayes approach, small taxa with
most or all species used as medicine (e.g., nj ¼2, xj ¼2) tend not to be indicated as overused with the
IDM model.
& 2012 Elsevier Ireland Ltd. All rights reserved.

1. Introduction ethnopharmacological data in a wider context with different bio-


informatic tools.
Ethnobotanical field studies frequently report on the most Since biological diversity is evidence for chemical diversity
important plant families based on a simple count of species used (Mc Chesney, 1996) the crucial question is: How is biological
as medicine. Such ranks of families are, however, not very in- diversity exploited for medicinal purposes? Beside empirically
structive for the understanding of local plant use. In particular are learned pharmacologic properties, plant selection is traditionally
species counts per taxonomic groups (mostly plant family) not conditioned by ecological factors, cultural history and organo-
representative because larger families tend to be overemphasized leptic features. However, the influence of media such as books,
with respect to small ones (Moerman et al., 1999). By introducing newspapers, TV and the World Wide Web progressively influence
a quantitative comparison of the medicinal flora with the total medicinal plant selection also in places without historical written
available flora and testing for under- and overused taxonomic records of plant use (Leonti, 2011).
groups Moerman (1979) directed the epistemological considera- Recently we introduced the Bayesian approach as a new method
tion about human plant selection towards a more systematic to test for under- and overused taxonomic groups, which, in con-
approach. Also, Heinrich and Verpoorte (2012) point out the trast to regression analysis, allows for hypothesis testing and takes
importance to look at data in multiple ways and to analyze the uncertainty over the sampled data into account (Weckerle et al.,
2011). The application of these methods depends on the availability
of floral inventories, i.e., flora checklists. However, for many in-
n
Corresponding author. Tel.: þ39 706758712; fax: þ 39 706758553. habited regions of the world such checklists do not exist, which
E-mail addresses: marcoleonti@netscape.net, mleonti@unica.it (M. Leonti). limits the application of quantitative and comparative approaches.

0378-8741/$ - see front matter & 2012 Elsevier Ireland Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jep.2012.05.002
260 C.S. Weckerle et al. / Journal of Ethnopharmacology 142 (2012) 259–264

Flora checklists may also be incomplete and often do not comprise post-Mixe-Zoquean (post-Olmec) culture with the application
cultivated and semi-cultivated species and/or comprise larger areas of a statistical method and the concept of cognates (Leonti
than the region accessed by humans for the extraction of medicinal et al., 2003b).
plants. Moreover, surveys of medicinal plants are almost never
complete (Leonti et al., 2012). This means that in most cases there is 2.2. Total flora
an overall uncertainty over the total flora (n) as well as over the
medicinal flora (t). Although the census of the Mexican flora is incomplete, it is
The observed proportion, t/n, may thus differ from the ‘‘true’’ estimated that between 29,000 and 34,000 species do occur,
proportion, y, because there is uncertainty over n and t. The idea placing the vascular flora of Mexico the fourth highest worldwide
of the existence of such a true proportion y has been used in (Bye, 1993; Challenger, 1998, p. 35). Homeland of the Sierra
regression analysis (Moerman et al., 1999) and in the binomial Popoluca is the ‘‘Sierra de Los Tuxtlas’’, the northernmost tropical
model described in Bennett and Husby (2008). In Weckerle et al. rainforest of the continent (Dirzo and Garcia, 1992). The flora
(2011) this idea has been replaced by an uncertainty around y checklist by Ramirez-Ramirez (1999) records around 2400
represented by a specific prior distribution, p(y). Such precision in species, but there may be up to 3000 species growing in the
defining a specific prior distribution (i.e., the uniform prior) is Sierra (Chevalier and Buckles, 1995, p. 182). This high biological
given by the assumption that the overall flora, n, is exactly known diversity is due to the overlap of the Central and South American
(as also assumed in Moerman et al. (1999) and Bennett and Husby (neotropic) and the North American (holarctic) floral kingdoms
(2008)). In this paper we also remove the latter assumption and and conditioned by the climatic and topographic diversity of the
consider therefore a set of priors p(y) where the width of the set region. The flora checklist (Ramirez-Ramirez, 1999)) was updated
is directly related to the uncertainty around n. Such an un- according to Stevens (2001 onwards) and THE ANGIOSPERM
certainty, however, must be small enough in order to learn PHYLOGENY GROUP (2009).
something about y based on the observed values of t and n. We
therefore introduce Imprecise Probability Calculation using the
Imprecise Dirichlet Model (IDM) described in Walley (1996), 2.3. Statistics
which allows for the estimation of under- and overuse of
taxonomic groups without assuming a specific prior distribution This statistical approach addresses the same questions posed
based on y. Instead we assume that this prior is imprecise. We by Moerman et al. (1999), Bennett and Husby (2008), and
exemplify the testing for under- and overused taxonomic groups Weckerle et al. (2011). We want to test for over- and underused
with IDM with the Campania (cf. Weckerle et al., 2011) and the taxonomic groups but without considering a specific prior dis-
Popoluca dataset (cf. Leonti et al., 2001,2003a) and compare the tribution of y. This means we want to evaluate the probability
results with the Bayesian approach. yjA[0, 1] for a certain taxonomic group, j, j ¼1,y, J of J groups,
that a species belonging to this group j is used as medicine.
Therefore, we collected the number of species, nj and the number
2. Methods of them used in medicine, namely xj for all J groups. We assume
that yj is random and that it can assume any value in [0, 1]. In
2.1. Medicinal flora contrast to the Bayesian method presented in Weckerle et al.
(2011), however, we do not apply a specific prior distribution. For
Altogether, the data used for Campania (Italy) includes 423 example, we do not assume that just because we have not suf-
different native medicinal plant species belonging to 83 families ficient reason to believe that some value has a higher probability
and are the same as published before in Weckerle et al. (2011). than another; yj is uniformly distributed between 0 and 1. The
Also the taxonomical classification into J¼7 groups has remained assumption that nj is unknown is especially relevant when the
the same. We have shown that the selection of medicinal plants in number of n’s is relatively small (i.e., small taxa—high uncertainty).
Campania is causally influenced by the herbal written by Mat- If we would assume another prior such as for instance the Jeffreys
thioli in the 16th century but very probably also by other books prior, yj  Beta(1/2, 1/2), we may reach a different conclusion. Since
and means (Leonti et al., 2010). we avoid assuming a specific prior distribution on the common
The medicinal flora of the Popoluca was gathered during field- proportion y or yj as done in Weckerle et al. (2011) we suppose that
research from 1999–2000 in the Sierra Santa Marta, Veracruz, this prior is imprecise. The set of priors in the imprecise setting
Mexico. Voucher specimens (Leonti 1-599) were deposited at the comprises quite a large set of different priors also containing the
National Mexican Herbarium (MEXU) UNAM (Mexico, D.F.), Her- Jeffreys and the Uniform prior used in Weckerle et al. (2011).
barium-Hortorium of the Colegio de Postgraduados Chapingo Therefore, results obtained with the approach of imprecise prob-
(CHAPA) Texcoco (Mexico), Instituto Mexicano del Seguro Social ability are much more robust with respect to those obtained with
(IMSS-M; Mexico, D.F.), Instituto de Ecologı́a (XAL) Xalapa (Ver- the Bayesian/Uniform prior used in Weckerle et al. (2011).
acruz, Mexico), Center for Pharmacognosy and Phytotherapy The model we use is the Imprecise Dirichlet Model (IDM) fully
ULSOP (London, UK) and the herbarium of the University of Zurich described in Walley (1991,1996), and Walley et al. (1996). The
Z/ZT (Switzerland; cf. Leonti et al., 2001,2003a). Medicinal species whole procedure in IDM is regulated by just one parameter, s,
that were obtained from local markets are excluded from the which regulates how wide the set of priors is. For instance, s ¼4
analysis but introduced species, by the Nahuas or Europeans, are leads to quite a large set that includes many of the priors usually
included as they are now part of the local flora. The larger dataset adopted for Bayesian inference on proportions such as the Uni-
of the Popoluca flora prompted us to use a lower hierarchical form and Jeffreys prior. One may be tempted to consider the set of
classification level and divide the flora into J ¼15 major taxo- all priors, which may be obtained with s¼N. However, with such
nomic groups. The sampling of the Sierra Popoluca medicinal a set of priors we are unable to learn anything about y whatever
plant wealth revealed over 600 different species. While the in- sample we observe. Taking the above considerations into account
digenous plant selection shows an association with the doctrine we would like to obtain some summaries about the posterior
of signatures and organoleptic properties such as taste, smell, distribution of y and yj. For instance, the lower and upper
color and appearance (Leonti et al., 2002) it was also possible to intervals for the expected posterior value of yj, conditionally only
trace down the ethnopharmacopoeia of the Sierra Popoluca to a on the observed xj, nj is ðxj =nj þ s, xj þ s=nj þ sÞ, while for y they are
C.S. Weckerle et al. / Journal of Ethnopharmacology 142 (2012) 259–264 261

  Pj Pj Table 1
t tþs
nj þ s , nj þ s where t ¼ x and n ¼
j¼1 j
n . The most
j¼1 j Example for excel calculation. The 95% posterior credible interval of yj and y
(n¼10, t¼ 2) are shown.
informative quantities that consider all the uncertainty around yj
and y are the quantiles of their corresponding posteriors. Namely Group (J) A (nj) B (xj) IDM inf. IDM sup.
the upper and lower limit for the lower quantile, i.e. 2.5% are the
ð2:5%Þ
1 2 2 0.053 1.000
ð2:5%Þ
solutions y j and y j of the following two equations: 2 8 0 0.000 0.610

ð2:5%Þ 
Prðyj r y j xj ,nj Þ ¼ 0:025 Total/Common 10 2 0.019 0.684

ð2:5%Þ 
Prðyj r y j xj ,nj Þ ¼ 0:025:
n¼10 and t¼2 and for a certain class j, nj ¼2 and xj ¼2 (Table 1).
Note, that because the upper and the lower probabilities must
ð2:5%Þ
Using the Bayes/uniform prior we have that the interval for yj with
ð2:5%Þ
match the level of 0.025 we have that y j Z yj . The most 95% probability is (0.29, 0.99), while for y the corresponding
conservative lower bound for yj is therefore y j
ð2:5%Þ
. Analogously, 95% interval is (0.06, 0.52) and we conclude that there are no differ-
ences between y and yj. The same conclusion, but with different
the lower and upper limit for the higher quantile, i.e. 97.5% are
ð97:5%Þ
intervals, is obtained if we assume the Jeffreys prior, that is
ð97:5%Þ
the solutions, y j and y j of the following two equations: Beta(1/2, 1/2), in fact we obtain the corresponding intervals
 ð97:5%Þ   (0.33,0.99) for yj and (0.04,0.50) for y. Using the imprecise approach
Pr yj r y j xj ,nj ¼ 0:975
provides even a stronger evidence for the same conclusion being
 yIj ¼ ð0:02,0:68Þ \ yIj ¼ ð0:05,1:00Þ a |: The comparison of the above
ð97:5%Þ  
Pr yj r y j xj ,nj ¼ 0:975: intervals is illustrated in Fig. 1.

Note, that because the upper and the lower probabilities must
ð97:5%Þ ð97:5%Þ
match the level of 0.025 we have that y j Z yj The lower
and the upper probabilities are calculated with respect to the two
beta distributions, Beta(xj þs, nj  xj) provided that xj onj and
Beta(xj, nj  xj þs), for xj 40. Therefore the required solutions can
be obtained as the quantiles of the following Beta distributions:
ð2:5%Þ
yj ¼qBeta(0.025, xj þ s, nj xj), for xj onj and 1 otherwise,
y ð2:5%Þ
j
¼qBeta(0.025, xj, nj  xj þ s), for xj 40 and 0 otherwise,
ð97:5%Þ
yj ¼ qBeta(0.975, xj þs, nj  xj), for xj onj and 1 otherwise,
y ð97:5%Þ
j
¼ qBeta(0.975, xj, nj  xj þs), for xj 40 and 0 otherwise,

which can be simply obtained using the INV.BETA() function in


Excel. For the parameter y the formulas are pretty much similar,
just replace xj with t and nj with n for 0ot on:
ð2:5%Þ
y ¼qBeta(0.025, t þs, n  t)
y ð2:5%Þ ¼qBeta(0.025, t, n t þs)
ð97:5%Þ
y ¼ qBeta(0.975, t þs, n  t)
y ð97:5%Þ ¼ qBeta(0.975, t, n t þs),

Fig. 1. Analysis of n¼ 10 species (Comparison of 95% probability intervals).


Based on the above quantities, we state that a class j is over- or
Comparison of the 95% posterior credible intervals (probability intervals) of y and
underused if the interval for yj has nothing in common with the yj obtained with the Uniform (Bayes), Jeffreys and the Imprecise model (IDM) with
corresponding interval of y when considering the imprecise bounds. n¼10, t¼ 2, nj ¼ 2 and xj ¼2.
   
ð2:5%Þ ð97:5%Þ ð2:5%Þ ð97:5%Þ
Essentially we require that y j ,yj \ y ,y ¼ |:
We provide the minimum and maximum values of the probabilities The different approaches, however, tend to provide similar
yj and y that we are willing to assume considering all possible results when the sample size increases. For instance, suppose
uncertainties around such parameters. In fact, the uncertainty that n¼ 100, t ¼20, nj ¼20, xj ¼20. With the Bayes/uniform prior
we are willing to assume is regulated by the parameter s. Let us the 95% probability interval of yj is (0.84, 0.99), while for y it is
consider the expressions for the expected values of yj and y: If we (0.13, 0.29). With the Jeffreys prior we have (0.88, 0.99) for yj and
allow to increase s indefinitely (s¼ N), we end up with both (0.13, 0.29) for y. Using the imprecise approach we obtain an
posterior expectations to be between 0 and 1, so that we have interval for yj of (0.88, 1.0) and (0.12, 0.32) for y (see Fig. 2). The
learnt nothing from the sample. The same happens of course for the two examples above show that the different approaches may lead
    to the same conclusions. The imprecise approach, however, leads
I ð2:5%Þ ð97:5%Þ I ð2:5%Þ ð97:5%Þ
intervals yj ¼ y j ,yj and y ¼ y j ,yj because to more conservative intervals because a larger amount of
both will be between 0 and 1 and therefore we will not be able uncertainty is considered for the unknown probabilities.
to make any decision about the over- or underuse of a taxonomic
group. In this case we say that the inference is vacuous. For
this reason, from now on we consider s¼4 (also suggested in 2.4. Excel calculation
Walley (1996) because with s¼4 we include many of the widely
used priors). In order to exemplify the different approach we For the calculation of the most conservative lower bound of
reconsider the example made in Weckerle et al. (2011). Suppose yj ðy ð2:5%Þ
j
Þ and y ðy
ð2:5%Þ
Þ as well as for the two upper bounds of
262 C.S. Weckerle et al. / Journal of Ethnopharmacology 142 (2012) 259–264

Table 2 shows the posterior distribution of the common pro-


portion y and the probability for each taxonomic group yj. The
Monocots and Early Diverging Eudicots tend to differ from the
common mean which is around 19%, i.e., Monocots are underused,
while Early Diverging Eudicots are marginally overused with the
Bayes approach, but marginally not overused with the IDM model.
Thus, the results obtained with Bayes are very similar to those with
the IDM model, albeit more conservative (Fig. 3).

Fig. 2. Analysis of n ¼100 species (Comparison of 95% probability intervals).


Comparison of the 95% posterior credible intervals (probability intervals) of y and
yj obtained with the Uniform (Bayes), Jeffreys and the Imprecise model (IDM) with
n¼ 100, t ¼20, nj ¼ 20 and xj ¼20.

Table 2
Bayes and IDM intervals for the Campanian dataset divided into J ¼7 taxonomic
groups, showing over- and underuse of taxonomic groups.

Fig. 3. Comparison of the 95% probability intervals obtained with the Uniform
(Bayes) and Imprecise model (IDM) for the Campanian data.

Table 3
Bayes and IDM intervals for the Campanian dataset, showing over- and underused
families (J ¼128).

Family (J) nj xj Bayes inf. Bayes sup. IDM inf. IDM sup.

n, number of species; x, number of medicinal species. Over- and underused taxa Overused
are underlined. Dashed line indicates underuse.
Lamiaceae 92 33 0.268 0.461 0.253 0.484
Rosaceae 74 25 0.240 0.452 0.222 0.481
Salicaceae 11 6 0.277 0.789 0.177 0.873
ð97:5%Þ ð97:5%Þ
yj (y j ) and y (y ) we can use the excel function Inverse Malvaceae 20 8 0.218 0.616 0.164 0.694
Primulaceae 14 6 0.213 0.677 0.142 0.770
beta (INV.BETA or BETAINV). Suppose that the total number of Caprifoliaceae 10 5 0.234 0.766 0.139 0.861
species is 10 (n), while the total number of used species is 2 (t) Violaceae 10 5 0.234 0.766 0.139 0.861
distributed in 2 (taxonomic) groups. Group 1 consists of 2 species Urticaceae 7 4 0.245 0.843 0.122 0.933
and both are used (n1 ¼2, x1 ¼2), while none of the 8 species Solanaceae 7 4 0.245 0.843 0.122 0.933
Moraceae 3 3 0.398 0.994 0.118 1.000
of group 2 are used (n2 ¼8, x2 ¼ 0). In the formula we use A and B Apocynaceae 4 3 0.284 0.947 0.099 0.996
instead of nj and xj. The formula for the lower bound is Cucurbitaceae 5 3 0.223 0.882 0.085 0.968
‘‘ ¼INV.BETA(0,025;B;A  B þ4)’’, while formula for the upper Loranthaceae 2 2 0.292 0.992 0.053 1.000
bound is ‘‘ ¼ INV.BETA(0,975;B þ4;A B)’’. Melanthaceae 2 2 0.292 0.992 0.053 1.000

Underused
Geraniaceae 24 0 0.001 0.137 0.000 0.292
Amaranthaceae 31 1 0.008 0.162 0.001 0.275
3. Results and discussion Cyperaceae 45 1 0.005 0.115 0.001 0.200
Fabaceae 226 28 0.087 0.173 0.083 0.187
Caryophyllaceae 106 4 0.015 0.093 0.010 0.128
3.1. Analysis with Bayes and IDM for Campania (Italy)
Poaceae 206 14 0.041 0.111 0.037 0.127
Orchidaceae 65 0 0.000 0.054 0.000 0.124
Here we present the results comparing the Bayesian approach
(uniform) as discussed in Weckerle et al. (2011) with the IDM
Common 0.173 0.206 0.173 0.207
model. We consider an application to species divided into J¼7
taxonomic groups. Underlined taxa are indicated as over- or underused with both approaches.
C.S. Weckerle et al. / Journal of Ethnopharmacology 142 (2012) 259–264 263

Comparison of the 95% posterior credible intervals obtained 3.2. Analysis with Bayes and IDM of the Popoluca dataset (Mexico)
from lower and upper posterior probabilities (probability intervals)
of y and yj obtained with the Uniform (Bayes) and the Imprecise Here we present an application to the species of the Popoluca
model (IDM) for the 7 higher taxonomic groups of Campania (Italy). flora divided into J¼15 taxonomic groups. Table 4 shows the
Table 3 shows the Bayes and IDM calculations for the over- posterior distribution of the common proportion y and the prob-
and underuse of the J ¼128 families of the Campanian vascular abilities of the taxonomic groups yj calculated with both, the
flora. The upper part shows the families whose 95% posterior Bayesian approach and the IDM model. Again we find very similar
credible interval lies above the 95% posterior credible interval of results for both approaches with the IDM model being more
the overall proportion for the flora y (0.173, 0.207) and which are conservative. With respect to the results obtained with the Bayes
thus overused, while the lower part of the table shows the plant model only the taxonomic group 12 (Berberidopsidales–Caryo-
families whose 95% posterior credible interval lies below the 95% phyllales) are not indicated as overused with the IDM model.
posterior credible interval of the overall proportion for the flora, Comparison of the 95% posterior credible intervals obtained
and which are thus underused. With IDM only the Lamiaceae and from lower and upper posterior probabilities (probability inter-
Rosaceae are pointed out as overused, while underused families vals) of y and yj obtained with the Uniform (Bayes) and the
with IDM are Orchidaceae, Poaceae and Caryophyllaceae (Fig. 4). Imprecise model (IDM) for the 15 higher taxonomic groups of the
‘‘Sierra Popoluca’’ flora (Veracruz, Mexico).
Table 5 shows the Bayes and IDM values of the over- and
underused plant families of the Popoluca flora (J ¼199). The
complete dataset showing nj and xj together with the 95% poster-
ior probability for each of the 199 plant families is available as
supplementary data (Appendix A). The fact that the IDM model
assumes that n and nj are not exactly known, taxonomic groups
(e.g., families) with relatively low n’s are not pointed out as
incompatible with H0. In Weckerle et al. (2011) with the Bayes
approach several small families have been indicated as over-
used such as Loranthaceae and Melanthaceae (both nj ¼2, xj ¼2),
Moraceae (nj ¼3, xj ¼3) or Apocynaceae (nj ¼4, xj ¼3) and it is of
course questionable how relevant and stringent such data based
on small nj’s is. In fact, when approached with IDM, all these
smaller families are not pointed out as incompatible with H0.
Thus, the results with IDM and s¼ 4 are very conservative. A less
conservative estimate can be achieved by defining s¼ 2, which
still comprises both, Uniform and Jeffreys prior (Table 5). Thereby
one obtains two sets of results with different measures for
stringency, one measure being more conservative than the other.
The obtained amplitude of stringency allows for a more rationale
contextualization and discussion of the data.

4. Conclusions

Fig. 4. Comparison of the 95% probability intervals obtained with the Uniform Considering uncertainty in ethnobotanical data is important,
(Bayes) and Imprecise model (IDM) for the Popoluca data. as medicinal plant inventories (t) tend not to be complete, while

Table 4
Bayes and IDM intervals for the Popoluca flora divided into J¼ 15 taxonomic groups, showing over- and underuse of the groups.

Over- and underused taxa are underlined. Dashed line indicates underuse.
264 C.S. Weckerle et al. / Journal of Ethnopharmacology 142 (2012) 259–264

Table 5 Bye, R., 1993. The role of humans in the diversification of plants in Mexico. In:
Bayes and IDM intervals for the Popoluca flora, showing over- and underused Ramamoorthy, T.P., Bye, R., Lot, A., Fa, J. (Eds.), Biological Diversity of Mexico.
families (J ¼ 199). Oxford University Press, New York, Oxford. (pp: 707–731).
Challenger, A., 1998. Utilización y Conservación de los Ecosistemas Terrestres
Family (J) nj xj Bayes Bayes IDM IDM IDM IDM de México, Pasado, Presente y Futuro. Comisión Nacional para el Conocimiento
inf. sup. inf. sup. inf. sup. y Uso de la Biodiversidad, México D.F.
s¼4 s ¼4 s¼2 s ¼2 Chevalier, J.M., Buckles, D., 1995. A land without Gods: process theory, malde-
velopment and the Mexican Nahuas. Zed Books, London & New Jersey.
Overused Dirzo, R., Garcia, M.C., 1992. Rates of deforestation in Los Tuxtlas, a neotropical
Asteraceae 116 57 0.402 0.581 0.387 0.597 0.394 0.590 area in Southeast Mexico. Conservation Biology 6, 84–90.
Piperaceae 46 22 0.341 0.619 0.307 0.656 0.321 0.639 Heinrich, M., Verpoorte, R., 2012. Statistical tools in ethnopharmacology. Journal of
Acanthaceae 23 12 0.328 0.709 0.266 0.766 0.291 0.745 Ethnopharmacology 139, 691–692.
Leonti, M., Vibrans, H., Sticher, O., Heinrich, M., 2001. Ethnopharmacology of the
Amaranthaceae 15 9 0.354 0.802 0.260 0.867 0.299 0.848
Popoluca, México: an evaluation. Journal of Pharmacy and Pharmacology 53,
Lamiaceae 26 12 0.287 0.647 0.235 0.706 0.255 0.681
1653–1669.
Commelinaceae 12 7 0.316 0.808 0.213 0.882 0.251 0.861
Leonti, M., Sticher, O., Heinrich, M., 2002. Medicinal plants of the Popoluca,
Anacardiaceae 9 6 0.348 0.878 0.211 0.945 0.262 0.933
México: organoleptic properties as indigenous selection criteria. Journal of
Aristolochiaceae 9 6 0.348 0.878 0.211 0.945 0.262 0.933
Ethnopharmacology 81, 307–315.
Oxalidaceae 5 4 0.359 0.957 0.157 0.997 0.223 0.996
Leonti, M., Ramirez-R., F., Sticher, O., Heinrich, M., 2003a. Medicinal flora of the
Crassulaceae 3 3 0.398 0.994 0.118 1.000 0.194 1.000
Popoluca, México: a botanico-systematical perspective. Economic Botany 57,
Chrysobalanaceae 4 3 0.284 0.947 0.099 0.996 0.147 0.995
218–230.
Dilleniaceae 4 3 0.284 0.947 0.099 0.996 0.147 0.995 Leonti, M., Sticher, O., Heinrich, M., 2003b. Antiquity of medicinal plant usage in two
Picramniaceae 4 3 0.284 0.947 0.099 0.996 0.147 0.995 Macro-Mayan ethnic groups (México). Journal of Ethnopharmacology 88, 119–124.
Bixaceae 2 2 0.292 0.992 0.053 1.000 0.094 1.000 Leonti, M., Cabras, S., Weckerle, C.S., Solinas, M.N., Casu, L., 2010. The causal
Underused dependence of present plant knowledge on herbals—contemporary medicinal
Thelypteridaceae 19 0 0.001 0.168 0.000 0.349 0.000 0.249 plant use in Campania (Italy) compared to Matthioli (1568). Journal of
Cyperaceae 35 2 0.018 0.187 0.006 0.281 0.007 0.225 Ethnopharmacology 130, 379–391.
Leonti, M., 2011. The Future is written: impact of scripts on the cognition,
Poaceae 93 5 0.024 0.120 0.017 0.158 0.018 0.134
selection, knowledge and transmission of medicinal plant use and its implica-
Orchidaceae 99 2 0.006 0.070 0.002 0.111 0.002 0.085
tions for ethnobotany and ethnopharmacology. Journal of Ethnopharmacology
134, 542–555.
Common 0.244 0.279 0.243 0.280 0.243 0.280 Leonti, M., Cabras, S., Castellanos, M.E., Weckerle, C.S., 2012. Reply to the
commentary: Regression residual vs. Bayesian analysis of medicinal floras.
Underlined taxa are indicated as over- or underused with both approaches. Journal of Ethnopharmacology 139, 695–697.
Mc Chesney, J.D., 1996. Biological diversity, chemical diversity, and the search for
new pharmaceuticals. In: Balick, M.J., Elisabetsky, E., Laird, S.A. (Eds.),
Medicinal Resources of the Tropical Forest: Biodiversity and its Importance
to Human Health. Columbia University Press, New York. (pp. 11–18).
flora checklists (n) might approximate completeness. The present Moerman, D.E., 1979. Symbols and selectivity: a statistical analysis of native
method, i.e., the IDM model, considers both, the uncertainty of American medical ethnobotany. Journal of Ethnopharmacology 1, 111–119.
n and t for the calculation of over- and underused taxa and plant Moerman, D.E., Pemberton, R.W., Kiefer, D., Berlin, B., 1999. A comparative analysis
of five medicinal floras. Journal of Ethnobiology 19, 49–67.
groups. Compared to the Bayesian approach (cf. Weckerle et al., Ramirez-Ramirez, F., 1999. Flora y Vegetación de la Sierra de Santa Marta,
2011), which assumes complete inventories of species of a local Veracruz. Tesis, Facultad de Ciencias. Universidad Nacional Autónoma de
flora, the IDM model leads to more conservative results. In fact, México, México, DF, pp. 217–350.
Stevens, P.F., 2001 onwards. Angiosperm Phylogeny Website. Version 9, June 2008
the parameter s40 assumes s more plants, with respect to those
[and more or less continuously updated since]. /http://www.mobot.org/
in a floral inventory, for which we do not know if they are used for MOBOT/research/APweb/S (September 19, 2009).
medicine or not. With IDM, taxa with relatively high n’s tend to THE ANGIOSPERM PHYLOGENY GROUP, 2009. An update of the Angiosperm
produce similar results as with Bayes. Small taxa with all or most Phylogeny Group classification for the orders and families of flowering plants:
APG III. Botanical Journal of the Linnean Society 161, 105–121.
species used as medicine (e.g., nj ¼2, xj ¼2) are not indicated as
Walley, P., 1991. Statistical reasoning with imprecise probabilities. Chapman and
overused with IDM, which, however, tends to be the case with the Hall, London, UK.
Bayesian approach. Less conservative estimates can be achieved Walley, P., 1996. Inferences from multinomial data: learning about a bag of
by defining s¼1, 2, or 3, which leads to different measures and marbles. Journal of the Royal Statistical Society Series B 58, 3–57.
Walley, P., Gurrin, L., Burton, P., 1996. Analysis of clinical data using imprecise
amplitudes of stringency allowing for a more rationale contextua-
prior probabilities. The Statistician 45, 457–485.
lization and discussion of the data. More rigorous application of Weckerle, C.S., Cabras, S., Castellanos, M.E., Leonti, M., 2011. Quantitative methods
statistical methods and hypothesis testing in ethnobotany may lead in ethnobotany and ethnopharmacology: considering the overall flora-hypothesis
to progresses in the understanding of human-plant relationships. testing for over- and underused plant families with the Bayesian approach.
Journal of Ethnopharmacology 137, 837–843.

Appendix A. Supporting information

Supplementary data associated with this article can be found in


the online version at http://dx.doi.org/10.1016/j.jep.2012.05.002.

References

Bennett, B.C., Husby, C.E., 2008. Patterns of medicinal plant use: an examination of
the Ecuadorian Shuar medicinal flora using contingency table and binomial
analysis. Journal of Ethnopharmacology 116, 422–430.

You might also like