Island biogeography is not a single-variable discipline: the small island effect debate


  • Kostas A. Triantis,

    Corresponding author
    1. Azorean Biodiversity Group, Departamento de Ciências Agrárias–CITAA, Universidade dos Açores, Pico da Urze, 9700-042, Angra do Heroísmo, Terceira, Açores, Portugal
    2. Biodiversity Research Group, Oxford University Centre for the Environment, South Parks Road, Oxford OX1 3QY, UK
    3. Department of Ecology and Taxonomy, Faculty of Biology, National and Kapodistrian University, Athens GR-15784, Greece
    Search for more papers by this author
  • Spyros Sfenthourakis

    1. Section of Animal Biology, Department of Biology, University of Patras, GR-26504 Patra, Greece
    Search for more papers by this author

Kostas Triantis, Azorean Biodiversity Group, Departamento de Ciências Agrárias– CITAA, Universidade dos Açores, Pico da Urze, 9700-042 Angra do Heroísmo, Terceira, Açores, Portugal.


In some island systems, an ‘anomalous’ feature of species richness on smaller islands, in comparison with larger ones, has been observed. This has been described as the small island effect (SIE). The precise meaning of the term remains unresolved, as does the explanation for the phenomenon and even whether it exists. Dengler (2010; Diversity Distrib, 16, 256–266.) addresses a number of conceptual and methodological issues concerning the nature and the detection of the SIE but fails to settle conclusively most of the issues he raises. We contend that his approach is theoretically flawed, especially in its treatment of habitat diversity. We offer a few suggestions of what is needed to advance understanding of the SIE.

Our ultimate theory of species diversity may not mention area, because area seldom exerts a direct effect on a species’ presence. More often area allows a large enough sample of habitats, which in turn control species occurrence’.

MacArthur & Wilson (1967; p. 8)

In some island systems, an ‘anomalous’ feature of species richness on smaller islands, in comparison with larger ones, has been identified; a phenomenon that has been described as the small island effect. Dengler (2010) addresses a number of conceptual and methodological issues concerning the nature and especially the methods used for the detection of the small island effect, introducing a number of criteria for the study of the phenomenon. He does so with direct reference to the data set presented in a recent paper of ours (Sfenthourakis & Triantis, 2009), criticizing both our approach and results. He extends his criticism to the method introduced by Triantis et al. (2006) for the study of the SIE. We contend that he fails to settle conclusively most of the issues he raises and that his overview of the theoretical background and methods for the detection of small island effect is itself flawed.

The small island effect

The term small island effect has been applied to several subtly different phenomena and contexts. Perhaps, the first use of the term was by Rand & Rabor (1960) who studied differences in body size and other attributes between bird species on small islands and the mainland. They defined small island effects as ‘the apparent ecological differences between faunas and populations of small islands and those of near-by larger land masses’. Similarly, Mees (1969) applied the term to size differences among birds in the family Zosteropidae living on large and small islands. A little later, Soulé (1980) used the term to refer to the effects of a chronically small population on genetic variation. Shortly, after the paper by Rand & Rabor (1960), the observation was made that in some systems the general positive relationship of species richness with area is absent below a certain threshold of island size. This was also described as the small island effect [e.g. Preston, 1962; Wiens, 1962; Niering, 1963; MacArthur & Wilson, 1967; Whitehead & Jones, 1969; Heatwole & Levins, 1973; Heatwole, 1975; see also Lack (1971) and Carlquist (1974) for similar, but not identical, considerations of the phenomenon]. This usage has recently become dominant within the literature, mainly as a result of an influential paper by Lomolino & Weiser (2001).

Still, even with the incorporation of the small island effect (hereafter SIE) within the current paradigm of biogeography, as an ‘anomalous’ feature of species richness on smaller islands compared with larger ones, it nonetheless remains unclear what exactly the term stands for, what mechanisms underlie the SIE and whether it even exists (see Lomolino, 2000; Lomolino & Weiser, 2001; Gentile & Argano, 2005; Triantis et al., 2006; Burns et al., 2009; Dengler, 2010). Here lies the first major conceptual flaw of Dengler’s approach; as he himself recognizes, there are various concepts attributed to the term SIE, leading to different methodological approaches. For example, the method applied by Lomolino & Weiser (2001) assumed that species richness will vary independently of area below a threshold of island size when a SIE is present, while the method applied by Gentile & Argano (2005) is checking for the existence of a threshold, below which the relationship between species richness and area is different than the same relationship above the threshold. On the other hand, Triantis et al. (2006) suggested that we need to consider additional variables for the study of the SIE, in addition to area, and introduced a new method for the detection of the upper limit of the SIE. However, Dengler decides to assign the approach of Lomolino & Weiser (2001), i.e. species richness varying independently of area below a certain threshold, as the ‘classic’ one (‘SIE sensu stricto’) and the rest as ‘deviating SIE concepts’ (‘SIE sensu lato’). This is simply wrong; there is currently no strong theoretical or empirical evidence for accepting one particular approach as ‘correct’ (not to mention ‘classic’). Although a traditional link is certainly identified between Niering’s (1963; fig. 7) first graphical representation of the SIE and Lomolino & Weiser’s method for the detection of the phenomenon, this does not definitively identify their method as being the most appropriate one. Moreover, the use of terms such as ‘SIE sensu stricto’ and ‘SIE sensu lato’ could be seen as further complicating the overall discussion; we can understand the use of these terms in taxonomy and phylogeny, but we see no necessity for using them for theoretical constructs such as the SIE.

The discussion over the nature of the SIE, even its very existence, is still inconclusive, and it is obvious that further theoretical and empirical studies are needed. Elevating the first to come and/or most popular approach to the level of the norm, without a robust theoretical documentation and strong factual evidence, is certainly premature sensu stricto.

Methodological issues

Dengler presents nine different methods applied so far for the detection of the SIE. This comparative presentation is a valuable addition to the discussion. Furthermore, we agree with his approach to apply nonlinear regressions for the various regressions involved in the methods, an innovation that may overcome some of the major drawbacks of traditional data transformation. Additionally, we agree that the inclusion of islands with no species in the study of SIE could provide new and interesting insights (e.g. Morrison, 2011).

Nonetheless, several statistical and conceptual issues remain to be clarified and resolved:

  • 1 In the title of his paper, Dengler claims robustness for the methodological steps he introduces for the study of the SIE, but as his method has only been tested on a single data set it appears too early to judge how robust and appropriate the method may be. It would be helpful if Dengler’s method had been applied to more data sets and/or simulated data sets, for which he can control for a SIE or not. Note also that the regressions he has used should be further evaluated by statistical examination for normality and homoscedasticity. Dengler, in his review of our manuscript, elucidated that ‘robustness’ is referring to the internal logic of the method. Then, it is this ‘internal logic’ that we consider to be flawed.
  • 2 It is not clear that all the studies have been correctly assigned in terms of the methods used for the detection of the SIE. For example, the study by Panitsa et al. (2006) was assigned to the ‘Visual inspection’ category, when in fact these authors applied the Lomolino & Weiser (2001) breakpoint regression method (Panitsa et al., 2006, p. 1225, last paragraph). Similarly, for the study of Morrison (1997), Dengler reports that the ‘Criterion for SIE’ was the ‘higher R2 of the SIE model’; subsequently, the ‘Methodological shortcoming was: No penalization for additional parameter(s’)’. Although Morrison did report R2 values, he used a partial F-test to determine whether the additional parameter in the curvilinear model was statistically significant, as a test for an SIE (p. 453).
  • 3Triantis et al. (2006; hereafter KAT) suggested that we need to consider additional variables for the study of the SIE, apart from area, and introduced a new method for the detection of the upper limit of the SIE. The method is based on an a priori theoretical model according to which island area affects habitat diversity directly and area and habitat diversity directly affect species number per island. Following this, each data set (among the 16 used) was tested for the existence of a certain island size under which the direct effects of area were eliminated. This detection was carried out through the sequential exclusion of islands, from largest to smallest, and the simultaneous estimation of the standardized partial regression coefficient of area; when this coefficient was found to be equal or smaller than zero, the respective area was assigned as the upper limit of SIE. Note that an SIE was detected in six of the 16 cases considered (37.5%).

Dengler claims that the KAT method suffers from five major shortcomings: (1) It incorporates a de facto comparison of log S = z1 log A + z2 log H vs. log S = (log A < T) z3 log H + (log A > T) (z1 log A + z2 log H), and thus, we should penalize the two additional parameters. This is a misunderstanding; no comparison is made with any simpler model incorporating habitat diversity or with any other method, owing to the distinctive nature of the method, i.e. inclusion of habitat diversity measures. The upper limit of the SIE is estimated based on the value of the standardized partial regression coefficient of area and not on any measure of fit. Thus, no penalization of parameters is needed in the particular context of the method. (2) High correlation between H and A causes imprecise parameter estimates. We concur with the suggestion that high collinearity between area and habitat diversity can have possible confounding effects and that this is an issue to be addressed when the method is applied, but the need for this is discussed in KAT. The use of path analysis can partially amend this problem (e.g. Grace et al., 2010 for a discussion on the structural equation modelling). (3) ‘Forcing’ the regression through the origin. The equation used for the estimation of path coefficients includes a constant, so the regression is not forced through the origin of the axes. Unfortunately, this point was omitted in error by Triantis et al. (2006), and we therefore take the opportunity of correcting that omission here. (4) Deviating concept of SIE (‘cryptic SIE’). As noted earlier, we reject the notion that there is a single correct formulation of the SIE; labelling approaches different from Lomolino & Weiser (2001) as ‘deviating’ is unhelpful. The SIE agenda should not be constrained by imposing a particular concept and dismissing all others as deviations. (5) Problematic habitat definition (see below).

Dengler compared the upper limit estimations for the SIE we presented in Sfenthourakis & Triantis (2009) following KAT with what he considers as implementations of the SIE sensu lato, i.e. all methods apart from the one suggested by Lomolino & Weiser (2001). He argues that the lack of similar values between the two sets of values indicates a lack of support for the method we have applied. Nevertheless, as he also recognizes, the estimated values from the KAT method derive from a totally independent approach. We have assessed the possible mechanisms establishing the SIE through studying the composition of island faunas in the Aegean Sea in terms of specialist and generalist species representation, below and above the SIE threshold. In this case, the SIE threshold was estimated as the island size where the cumulative ratio of specialists to generalists diverged from that of the overall data set, and it was almost identical with the threshold estimated by the KAT method. We concluded that the relative representation of specialist and generalist species on islands of different sizes plays an important role in shaping SIE-related patterns. According to Dengler, ‘this coincidence does not hold for the thresholds of the SIEs sensu lato determined in this article, which removes any support for the priority setting proposed by Sfenthourakis & Triantis (2009)’. Again, we suggest that it is unhelpful to attempt to elevate one usage to a position of primacy as though we were dealing with taxonomic naming conventions for species and genera.

Habitat diversity is important

As indicated in the quotation at the beginning of this paper, MacArthur & Wilson (1967), among others, highlighted the necessity for incorporating environmental heterogeneity into theories and models for explaining species diversity patterns (e.g. Williams, 1964; Kohn & Walsh, 1994; Ricklefs & Lovette, 1999; Triantis et al., 2003; Hortal et al., 2009). This is reinforced at small spatial scales, where environmental heterogeneity and area can become increasingly decoupled, and area alone cannot efficiently express the total effects of habitat diversity and island size on species richness (e.g. Triantis et al., 2005). One of the first to recognize this was Fosberg (1948), who studied the flora of atolls in the Pacific Ocean and concluded that ‘The flora is larger (species more numerous) in more or less direct proportion to the amount of rainfall and the area of the islet’. In this context, Triantis et al. (2006) highlighted the need to consider habitat diversity explicitly in the study of the SIE. This is a crucial conceptual difference with the other methods that consider only area. Moreover, even if the method proves to be wrong, the challenge for incorporating measures of environmental heterogeneity remains.

Dengler follows a rather ambiguous line of reasoning on this issue. Towards the end of his paper, he admits that habitat diversity should not be ignored, suggesting that it could be incorporated in the study of the SIE as an explanatory variable for the residuals of the best species–area model. He suggests that plotting the residuals of the habitat–area relationship against the residuals of the species–area relationship could provide insights for the relative importance of habitat diversity. This could certainly act as supplemental to other available approaches. However, there are two points to consider. First, it has to be determined which species–area model to apply, because there are at least 20 different functions proposed so far, without strong evidence supporting the a priori selection of one or even a few of them; hence, a multimodel approach could be helpful (e.g. Guilhaumon et al., 2008; Tjørve, 2009). Second, Dengler’s approach considers area to be more meaningful and/or effective in describing species richness than habitat diversity. Although area has proved to be the best single explanatory variable for species richness, when habitat diversity is also considered it is often found to be a better predictor of species richness than area (e.g. Lack, 1973; Kohn & Walsh, 1994; Ricklefs & Lovette, 1999; Sfenthourakis & Triantis, 2009). Thus, an equally plausible approach in such cases might be to use area to explain the residuals of the species–habitat relationship.

The core of our criticism of Dengler’s approach relates to the way he treats habitat diversity as used in the KAT method and in ecological studies overall. He comments ‘These habitat definitions [the ones used in Sfenthourakis & Triantis, 2009] are idiosyncratic, generally not transferable among island datasets and constructed from an anthropogenic view [our emphasis]. Moreover, this approach does not account for the proportional areas of the habitats on the islands. Nor does it reflect the varying degrees of similarity between the habitat types […]. Thus, it is questionable as to the meaning of an SIE and of its upper limit L if habitats are defined and used in such a way.’ In another part of his paper, he adds: ‘apart from solving the other methodological shortcomings of the method applied byTriantis et al. (2006)(see Table 1), one should apply a uniform habitat definition for all island data sets to be compared. To find such a general, globally applicable and, at the same time, ecologically meaningful measure of habitat diversity will certainly not be an easy task. And finally ‘To disentangle these two processes by including a measure of habitat diversity into the SAR models […] seems hardly plausible since there are infinite possibilities of how to quantify habitat diversity – and each will result in different outcomes’.

Hence, although Dengler suggested possible ways of incorporating habitat diversity in the SIE search programme, he discourages any effort towards quantifying it. It might be ideal, but probably impossible, or even misguided according to some conceptual approaches to the term ‘habitat’, if in the future we could define an International Prototype of Habitat Unit similar to the International Prototype of Kilogram (see Meanwhile, we cannot just ignore environmental heterogeneity and its possible effects on species diversity. Although well-defined obvious and discrete variables or quantities, such as area, elevation, etc., will remain easy to understand and explain, as in other fields of science, we are usually left with rather abstract, open and generalized concepts; we just hope to increase the degree of abstraction producing concepts with higher degrees of generality (see Pickett et al., 2007). Hence, although we are aware of the many pitfalls in habitat diversity/heterogeneity conceptualization and quantification, this is not a reason to abandon it. Similar to SIE, niche theory has many pitfalls and problems of definition (e.g. Colwell & Rangel, 2009); should we abandon niche theory too?

A few pointers for future research

The current uncertainty over the nature, the driving mechanisms and the methods for the study of the SIE are understandable, given the relatively short time since the patterns’ recognition and the limited number of studies addressing it. Even for better established and extensively explored patterns of biogeography, such as nestedness, no consensus has yet been reached regarding the mechanisms generating the pattern and the appropriate metric to quantify it (e.g. Ulrich et al., 2009). Hence, we stress the lack of strong consensus along with the growing number of applied and theoretical studies of SIE and the need for a critical review of the state of art and for perspectives guiding future research.

Such a review should incorporate at least three major tasks as follows: (1) An evaluation of the various views on what constitutes a SIE in respect to their ability to effectively describe actual patterns, hence increasing the utility of the term; (2) A classification of the available explanations/mechanisms and the hypotheses generated by these explanations. According to Sfenthourakis & Triantis (2009), there are two main categories of explanation as follows: (i) demographic stochasticity (i.e. extinction rates independent of island area) and (ii) idiosyncratic habitat diversity at small spatial scales [see discussion about the Small island habitat effect in Whittaker & Fernández-Palacios (2007); see also Heatwole et al. (1981)]. To date, explanations have mainly been promulgated as hypotheses related to species richness, but a broader theoretical framework should also introduce hypotheses on community composition, species abundances, species co-occurrence and generalist-specialist species representation (e.g. Burns et al., 2009; Sfenthourakis & Triantis, 2009; Morrison, 2011); and (3) An evaluation of the available methods applied to date for the detection of the SIE, possibly leading to the introduction of new methods. Dengler undertook the latter task but failed to place his evaluation within the general framework set by the two previous tasks.

No single cause is sufficient to explain the vast majority of ecological phenomena, and in many cases, a single variable is not even enough to effectively describe them. Habitat diversity has a well-established effect on species diversity, at least at certain spatial scales. Therefore, before we restrict our approaches to a single variable (area), we should at least try to incorporate other variables known to have a significant influence. The ‘debate’ over the existence and the nature of the small island effect is certainly not resolved yet. Although the pattern is gaining space in the ecological literature, further theoretical and empirical studies are needed to fully resolve these issues. For example, it is possible that the SIE is manifested through some critical value range, rather than a strict threshold point (e.g. Ficetola & Denoël, 2009).


We are grateful to Richard Ladle, Mike Weiser, Kevin Burns, Joaquin Hortal, Sam Scheiner, François Guilhaumon, Aris Parmakelis, Even Tjørve and especially Rob Whittaker for comments and suggestions on previous versions of the manuscript. We thank L. Morrison and an anonymous reviewer for comments on a draft of this manuscript. We have to note that we did not have the opportunity to express our opinion on Dengler’s paper before or during the evaluation process; however, we thank J. Dengler for comments on a previous version of the manuscript. KAT is supported by a FCT Fellowship (SFRH/BPD/44306/2008).


Kostas Triantis was recently appointed Assistant Professor in the Biology Department of the University of Athens. He is mainly interested in island biogeography of oceanic and mainland islands and macroecology. His other research interests include the following: scale in ecological and biogeographical analysis, species–area relationship, environmental heterogeneity and conservation biogeography.

Spyros Sfenthourakis is an Assistant Professor of Ecology and Biogeography in the University of Patras. He is interested in processes shaping species diversity, mainly in insular communities. He has published in the fields of island biogeography, community assembly, phylogeography and systematics of terrestrial isopods, as well as biodiversity patterns in Greece. He is also interested in the reproductive and migratory behaviour of bird species.

Editor: Marcel Rejmanek