## Introduction

Of all the Crustaceans, the Oniscidea is undoubtedly the group that has been most successful in colonizing terrestrial environments. Although these isopods are found in a variety of different habitats, they are characterized by low dispersal ability and a high degree of stenoecy. A combination of these characteristics is often a determinant of the high degree of morphological and/or genetic variation exhibited by species of this suborder over time and space, at both micro- and macro-scales (Gentile & Sbordoni, 1998; Sarbu *et al.*, 2000). Oniscidea are also very sensitive to habitat heterogeneity. Recent studies of Oniscidea from Mediterranean islands have shown that the number of species is directly proportional to habitat heterogeneity, which may also influence community structure (Sfenthourakis, 1996a; G. Gentile and R. Argano, unpubl. data). As a result of these characteristics, Oniscidea are a valuable tool when investigating the evolutionary dynamics of insular biota, and represent a good biological model for the study of colonization processes.

Many studies of the Oniscidea of Mediterranean islands have been carried out, primarily focusing on local faunas (Arcangeli, 1953; Ferrara & Taiti, 1978; Taiti & Ferrara, 1980, 1989; Caruso *et al.*, 1987; Argano & Manicastri, 1996; Sfenthourakis, 1996a,b; G. Gentile & R. Argano, unpubl. data). Up until now, with the exception of the work by Sfenthourakis (1996a,b), there has been little use made of the available data, despite their biogeographical relevance. In fact, these data could prove very useful, not only to investigate general biogeographical patterns of the Mediterranean area, but also to address the species–area issue, a topic of renewed interest among biogeographers.

In this study, numerical taxonomic techniques were applied to the data in order to perform an analysis of the Oniscidea fauna from a large sample of islands within the Mediterranean Sea. We use the species–area relationship to compare five continental archipelagos of the Mediterranean Sea and the oceanic archipelago of the Canary Islands. A range of linear and nonlinear models were used to verify the findings of Sfenthourakis (1996a), who investigated which model (linear, logarithmic or semi-logarithmic) should be applied to Oniscidea of the Canary, Aegean and Tuscanian islands. We also considered the biological relevance of the slope and intercept of the species–area relationship for Oniscidea of the islands of the Mediterranean Sea.

In general, estimates of slopes and intercepts can be affected by bias introduced by the area-range effect. Martin (1981) has shown that slope estimates may vary when the smallest and largest island ranges of some archipelagos are examined separately or cumulatively. In particular, he observed that slope estimates would be higher if based on ranges of small islands, whereas they would be lower when considering ranges of larger islands. Additionally, if the areas of two archipelagos overlap, but the range of one is extended to include larger or smaller islands, then the slope of the log *S*/log *A* curve would be lower or higher respectively. The influence of spatial and temporal scale on the nature of the species–area relationship, in relation to speciation, has been discussed by Lomolino (2000).

We also consider the ‘small island effect’ (SIE). The SIE refers to the existence of two different patterns in the species–area curve, whereas traditional models, such as the log *S*/log *A* model, can usually describe only one pattern. The SIE shows that below a certain threshold value, the number of species can vary independently of area and that in this case, a sigmoid curve describes better the species–area relationship (Lomolino, 2000). To address this issue, Lomolino & Weiser (2001) used simple linear regression with a breakpoint transformation (McGee & Carleton, 1970; Besier & Sugihara, 1997). For the estimation of the breakpoint, which is the upper limit of the SIE, they used the equation:

where *S* and *A* are species richness and area respectively. *T* is the upper limit of the SIE and (log *A* ≥ *T*) is a logical variable that returns 1 or 0 if true or false respectively. Parameters of the equations were estimated by iteration, with *T* (in units of log *A*) being incremented at each iteration. In this equation, *x*-values of islands smaller than *T* are reduced to 0, whereas *x*-values of islands larger or equal to *T* are decreased by the amount *T*.

With respect to the traditional models, the equation proposed by Lomolino & Weiser (2001) has the advantage that it describes in more detail the species–area relationship when a SIE exists. However, the equation is not very appropriate to assess whether or not a SIE exists in a certain data set because it *a priori* assumes a SIE and imposes it on the model. In fact, *x*-values are reduced to 0 when islands are smaller than *T* so that log *S* is estimated as a constant (*b*_{0}).

We used our data to compare slopes and intercepts for six archipelagos that differ in island size. We also investigated the possible occurrence of the SIE on the shape of the species–area curve by using both the model proposed by Lomolino & Weiser (2001) and a more general model of piecewise regression that does not assume *a priori* the existence of a SIE. Lastly, we investigated the possible occurrence of an area effect, as reported by Martin (1981), by estimating determination coefficients (*R*^{2}), slopes (*z*) and intercepts (*k*) by both adding islands of increasing/decreasing size, and using sliding-windows that encompassed islands of increasing size. Although the SIE effect exists in nature, it is still debated how common it is (Lomolino, 2000, 2002; Williamson *et al.*, 2001; Barrett *et al.*, 2003). In this regard, the inclusion of a disproportionately high number of large islands in biogeographical surveys could be one of the reasons why many studies failed to detect the effect (Lomolino, 2000; Lomolino & Weiser, 2001). In the present study, the island size distributions for each archipelago always showed a leptokurtic, right-skewed pattern, thus removing this bias.

Additionally, to gather more information from the data, we looked for possible covariation patterns between regression parameters and area distribution skewness, when the area-range effect was investigated. Thus, at each step in the cumulative and sliding-window analyses the skewness of the area distribution was also calculated.