Ecology and signal evolution in lizards

Current models of signal evolution explain diversity by invoking a variety of social, perceptual and environmental factors. Social systems and spacing patterns determine the active space of signals and their function. Receiver sensory systems and habitat characteristics interact to constrain signal design. These factors are traditionally implicated in promoting directional evolutionary change, leading to increases or decreases in signal complexity. We examine macro-evolutionary trends in signal design, as reﬂected by display modiﬁer repertoire size, for 124 species of iguanian lizards to identify the importance of ecological factors in display evolution. Possessing a small home range, being arboreal and feeding on moving prey are all correlated with the evolution of large repertoires. However, living in closed habitats is associated with increased evolutionary change in repertoire size, producing greater signal diversity. Ecological factors can thus act either directionally or to promote evolutionary lability.

Lizards of the agamid and iguanid families typically possess a complex display repertoire, which is used in both male-male contests over territories (Carpenter, 1978;Trivers, 1976) and male-female courtship interactions . These visual signals are composed of discrete, and sequentially predictable, movements centred on a head-nod and/or push-up display. Signal complexity varies across species and can be quantified by the number of modifiers used (Jenssen, , 1978Ord et al ., 2001). Modifiers may be either static (e.g. dewlap extensions, crest raising, body compression/inflation, back arching, body raising/tilting and changes in body colouration) or dynamic (e.g. tail-wagging and arm-waving). In a previous study (Ord et al ., 2001), we reported that differences in modifier repertoire size between species are partially explained by differences in male-male competitive intensity and suggested that more complex signals have evolved to improve opponent assessment. In the present study, we report a complementary analysis of the way in which ecological factors have influenced this aspect of signal evolution.
There are at least two possible consequences of natural selection: directional responses generate increases or decreases in the mean value of a trait, while disruptive or diversifying responses promote greater variance. As part of an investigation into the evolution of visual signals in lizards, we examine the pattern of selective response to four ecological factors that are likely to have influenced signal structure: habitat structure, home range size, arboreality and diet.

H ABITAT STRUCTURE
Signals transmitted for any distance inevitably degrade before they reach potential receivers (Dusenbery, 1992;Bradbury & Vehrencamp, 1998). Transmission efficiency is determined by an interaction between signal properties and the physical structure of the environment. Signals designed for long range communication should thus be subjected to minimal or predictable degradation (e.g. ranging in birds: McGregor & Krebs, 1984;Morton, 1986;Naguib & Wiley, 2001;and primates: Whitehead, 1987). Studies of acoustic communication have shown that transmission range in bird song is maximized by tailoring calls to the acoustic environment in which they are typically given (Morton, 1975;Wiley & Richards, 1982). Similarly, the poor visibility characteristics of some habitats has selected for the evolution of conspicuous visual signals (Endler, 1987(Endler, , 1992Fleishman, 1992;Marchetti, 1993;Losos & Chu, 1998). The evolution of communication systems in response to properties of the signalling environment can hence generate and maintain differences in signal design when animals occur in different habitat types.
Habitat choice might influence the complexity of visual signals in at least two ways. In denser habitats, light levels are lower and obstructions potentially reduce signal active space (i.e. the distance over which signals are likely to remain effective). This latter factor will limit the maximum distance over which visual signals can be exchanged. However, as the distance between a signaller and receiver diminishes, the constraint of degradation is relaxed and more complex signals can emerge (Endler, 1992). Lizards in closed habitats may only become aware of conspecifics at close range (Stamps, 1983), and display interactions may consequently be restricted to those conducted over short distances (Jenssen, 1978). This is one mechanism by which closed habitats might enable selection for complex repertoires.
In addition, reductions in ambient light level should interact with properties of the lizard visual system to dictate the maximum rate of display movements. Experimental studies of electroretinographic response in three species of anoles demonstrate that critical flicker fusion frequency, which influences the ability to detect fast-moving objects, declines as the intensity of ambient light falling on a visual stimulus is reduced (Fleishman, Marshall & Hertz, 1995). It follows that the same rapid movements that effectively stimulate the visual system of an opponent in bright sunlight would be ineffective in shade. This sensory constraint should select for slower display movements in closed habi-tats (Fleishman et al ., 1995). If signal structure is unchanged, then the maximum rate of information transmission will be reduced and aggressive interactions involving opponent assessment will become more costly because of the extra time required. Alternatively, animals could add display modifiers to compensate for the constraint on movement rate and enhance information transfer. This latter scenario also predicts that complex signals will evolve with closed habitat preference.
H OME RANGE SIZE Species patrolling smaller home ranges will engage in bouts of territorial defence over shorter distances. In lizards, the intensity and vigour of signal exchanges typically escalates as conspecific distances decrease (Hover & Jenssen, 1976;Scott, 1984;Losos, 1985a;Decourcy & Jenssen, 1994). If opponents are evenly matched in size and condition, signal exchanges may degrade into physical combat (Stamps & Krishnan, 1994), which may also be facilitated by the proximity of participants (Scott, 1984). The increased injury risk under such conditions potentially favours large display repertoires for more efficient and accurate opponent assessment (Jenssen, 1978;Ord et al ., 2001). We therefore test whether species possessing small home ranges are also more likely to evolve complex signals.
A RBOREALITY  examined the evolution of push-up displays in a subset of lizards and found that 'jerky' displays tended to be lost in arboreal species, perhaps because of the physical difficulty of performing signal movements in such environments. However, arboreal habitats could facilitate the evolution of complex sig-nals or elaborate display repertoires as a secondary consequence of the enhanced locomotor flexibility that they require (Moermond, 1979). For example, Losos (1990a) suggested that display rate may coevolve with rapid movements such as running and jumping. Similarly, the ability to negotiate complex habitats is related to visual acuity (Jenssen & Swenson, 1974). Since animal signals are designed to exploit the sensory properties of receivers (Guilford & Dawkins, 1991, 1993Bradbury & Vehrencamp, 1998), species living in complex habitats may be pre-adapted to evolve complex signals. Thus, we assess the relationship between arboreal locomotion and signal structure.

D IET
Most lizards are insectivores, but one subfamily of iguanids is predominantly herbivorous (the iguaninae). Males in this group are typically less aggressive than insectivores (Stamps, 1983). This relationship implies that species feeding on moving prey might have evolved more complex visual signals to mediate aggressive interactions. We therefore investigate whether a diet of typically fast-moving and/or small prey is associated with the evolution of complex signals.

MATERIAL AND METHODS
Published data on modifier repertoire size, habitat preference, home range size, arboreality and diet were compiled from a total of 141 sources for a variety of agamid and iguanid lizard species (see Appendix).
We began by collating together any papers describing display behaviour (total of 215 sources). Studies that did not report display structure in sufficient detail were then excluded. Species ecology was well documented in many of the remaining studies, but additional sources focusing primarily on ecological factors were used to supplement this material when necessary. The final set of sources was used to construct a data set covering 124 species.

D ATA CATEGORIZATION
One inherent problem in comparing communicative systems across a diverse range of species is obtaining an appropriate and reliable index of signal variation that can be standardized. Visual displays in agamid and iguanid lizards almost invariably include the stereotyped head and body movements commonly termed 'head-nod' or 'push-up' displays. Subtle variation does exist within these displays (e.g. differences in display rate, speed, amplitude, etc.) and there are several species for which this variation has been quantified (e.g. Martins & Lamont, 1998). However, we found that this information could not be incorporated into our display complexity index because of methodological differences among published accounts. In addition, these core displays (head-nod and/or push-up) may vary within a species. For example, some anoline lizards are reported to have up to five different core display types, each varying in structure and social context (Hover & Jenssen, 1976;Jenssen & Rothblum, 1977). It was hence impractical to obtain a single value for each species based upon core display characteristics.
We focused instead on 'display modifiers', which are postures or movements that accompany and elaborate core displays . Modifier use varies across species and accounts for a considerable proportion of signal diversity. In addition, information on modifier repertoire size could be readily collated from the literature, using standard criteria. This allowed us to include a large, and taxonomically diverse, range of species in our analyses.
Modifiers scored included: back arching, arm waving, body compression/inflation, body raising, body tilting, eye 'orbing', changes in body colouration, lip smacking, crest raising, tail displays, throat displays (dewlap extensions, gorging of the throat, etc.), and tongue protrusions. To be conservative, and to acknowledge the presence/absence of core displays, four species (see Appendix) that were reported not to possess a core display were scored as having a repertoire size of 0, irrespective of the presence of any modifiers (two species were observed to possess a throat display, while others had a tail display).
Some modifiers are unique to species or species groups. However, our aim was to quantify repertoire size to obtain an index of signal complexity across species. For this reason, we did not compare individual modifiers to assess more subtle differences or similarities in structure with those of other species.
In iguanian lizards, visual displays are typically performed by males. This tendency is reflected in the published literature, with many studies focusing primarily on male signal behaviour. While there are some descriptions of display structure for females and juveniles, repertoire size is invariably either equal to, or smaller than, that reported for males. In addition, although the presence of specific modifiers may be influenced by social context in some species, many are employed across all types of display interaction. Our goal was to obtain an estimate of maximum repertoire size for each species. We therefore used the number of modifiers accompanying male core displays.
We acknowledge the possibility that, in some cases, the original sources from which we obtained repertoire size information may have underestimated the number of modifiers used. This might result when only part of the display repertoire was observed and/or if the focus of the investigator(s) was on other aspects of behaviour. Such errors would have the affect of increasing apparent variation in repertoire size, but there is no reason to expect them to be systematically associated with the other factors of interest. The effect would hence be to reduce the likelihood of detecting significant relationships (see Benton, 1999;Nunn & Barton, 2001).
Habitat preference was classified as 'closed' or 'other'. 'Closed' habitats consisted of environments with large amounts of ground and/or canopy cover and were commonly referred to in the literature as rainforest, forest or closed woodland. 'Other' habitats were typically described as desert, plains, steppe, savannah, prairies, shrubland or open woodland. To be conservative, four species that were reported to occur in various habitats were scored as 'other' (see Appendix). Most data on display structure describe signals produced by males, so data on male home range size was favoured. However, in species that were reported to have no significant difference between the sexes, or when male data were not available, species values were used. When reported, maximum home range size was favoured, otherwise average values were used. All home range values were converted to m 2 . Locomotion was categorized as either 'arboreal' or 'semi-arboreal/ terrestrial' following author interpretations. Diet was categorized as 'moving prey' if a species was exclusively insectivorous or carnivorous. Omnivorous species were those with a diet that included vegetation, but that consisted predominately of moving prey. Species were considered to have a diet of 'non-moving prey' when they were reported to be exclusively herbivorous.

P HYLOGENY
No complete phylogeny of all the species of interest was available, so we compiled a composite tree from several sources. In some cases, several phylogenetic hypotheses were available. Phylogenies were 'ranked' with those based on combined data favoured over purely molecular data, which was, in turn, favoured over purely morphological data. If hypotheses were still equally ranked, we preferred trees based on parsimony methods of tree construction, then those with the fewest number of polytomies (where the precise phylogenetic relationship between species is unknown), and finally those more recent in publication. Species synonyms were checked using the 'EMBL Reptile Database' (http://www.embl-heidelberg.de/ ∼ uetz/livingreptiles.html).

T RAIT EVOLUTION AND CORRELATION
For the continuous variable of home range size, we first log 10 transformed values (Turner, Jennrich & Weintraub, 1969;Christian & Waldschmidt, 1984) and then calculated standardized independent contrasts (Felsenstein, 1985) using the program CAIC v. 2.6.2 (http://evolve.bio.ic.ac.uk/evolve/software/caic/ index.html ; see also Purvis & Rambaut, 1995), which were then used in subsequent regression analyses. Independent contrasts are a common way to control for possible phylogenetic non-independence of species data (Harvey & Pagel, 1991). While the CAIC program can calculate contrasts from trees possessing polytomies, to be consistent with the phylogenetic hypothesis used in the concentrated-changes test (CCT; see below), we conducted analyses by randomly resolving polytomies using MacClade software v. 3.08a (Maddison & Maddison, 1992, 1999. Branch length data were available for only a few species pairs. We therefore set all branch lengths equal to include the maximum number of species in our analyses. We selected the 'Crunch' algorithm for contrast analyses. As required by this method (Purvis & Rambaut, 1995), regressions were forced through the origin. Body size covaries with home range size (Turner et al ., 1969;Christian & Waldschmidt, 1984), so we used additional published data (see Appendix) and regressed contrasts for body size (maximum male snout-vent length) against contrasts for home range size. The residuals from this analysis were then used to control for possible body size effects in subsequent tests (Losos, 1990a).
To investigate patterns of trait evolution, we used the phylogenetic program MacClade v. 3.08a (Maddison & Maddison, 1999) to reconstruct ancestor states for each trait. To use Maddison's (1990) CCT, we first had to randomly resolve polytomies using MacClade. An additional requirement of the CCT is that all data should be dichotomous. We created frequency distribution plots of repertoire and home range size and then used the median value to split the data for each variable into two discrete states. Species with greater than the median number of display modifiers (4; range: 0-8) were scored as having large modifier repertoires, while those with less than or equal to this number were scored as having small modifier repertoires. Similarly, species with smaller than or equal to the median home range size (massfree residual − 0.13; range: − 2.04-1.83) were considered to have small home ranges. All other traits were inherently dichotomous.
We reconstructed five alternative parsimony-based ancestor states and applied the CCT to each. We began by reconstructing ancestor states using Swofford & Maddison's (1987) linear parsimony MINSTATE & MAXSTATE and Maddison's (1991) squared-change parsimony algorithms; the last of these provides identical results to those obtained using generalized least squares and independent contrasts approaches (Martins & Hansen, 1997;Pagel, 1999). MINSTATE and MAXSTATE reconstructions reflect the smallest and largest sets of equally parsimonious values at each node, respectively. Squared-change parsimony minimizes the sum of the squared changes on branches and forces changes to spread out more evenly over the tree. These algorithms are designed for continuously distributed variables; traits that are naturally discrete were therefore reconstructed as if continuous. Areas were then defined manually using the 'fix state' option in MacClade. For continuous variables (i.e. modifier repertoire size and home range size) the median cut-off was used (see above). For discrete variables (i.e. habitat, arboreality and diet), branches with values less than 0.5 were defined as lacking the trait, while those greater than or equal to 0.5 were defined as possessing it.
Finally, we applied the ACCTRAN (which accelerates changes toward the root) and DELTRAN (which delays changes away from the root) linear parsimony algorithms (Swofford & Maddison, 1987). These models require that variables be discrete. Modifier repertoire size and home range size were thus dichotomously transformed prior to ancestor state reconstruction. All other traits were inherently dichotomous.
We used the CCT to examine two hypotheses. First, we examined directional changes in modifier repertoire size. We tested whether evolutionary gains in large repertoire size were more concentrated than expected by chance on branches that also possessed an ecological trait. A significant association would indicate that the trait preceded, or evolved simultaneously with, the evolution of signal complexity. Second, we investigated whether an ecological trait has led to increased variation in repertoire size. We analysed the probability that both gains and losses in large repertoire size were associated with an ecological trait: a significant association would reveal that high levels of evolutionary change in signal complexity were associated with the trait.
The CCT has low statistical power and is therefore likely to be conservative (Lorch & Eadie, 1999). Fol-lowing Ortolani & Caro (1996) and suggestions by Lorch & Eadie (1999), we considered associations with P < 0.05 as highly significant, while P -values falling between 0.05 and 0.10 were considered to be marginally significant. Out of 124 different species for which modifier repertoire data were available (Fig. 1), only those species having complete data on the ecological trait of interest, were used in each analysis (range: 28-104 species). The large number of species being investigated prevented us from calculating an exact probability; we report P -values calculated using a simulation algorithm (Maddison & Maddison, 1992) with 10 000 replicates. This method has been shown to provide results consistent with those of exact P -value calculations (Maddison, 1990).

RESULTS
Large modifier repertoire size evolved between two and 19 times across the agamid and iguanid families and has subsequently been lost between one and 18 times, depending on ancestor reconstruction (MINSTATE, MAXSTATE, ACCTRAN, DELTRAN or squared-change parsimony), tree topology (tree 1 or 2; Fig. 1) and the number of species analysed.
We found no consistent relationship between evolutionary gains in large modifier repertoire size and a pre-existing preference for living in closed habitats. There was a marginally significant association using two sets of evolutionary assumptions (tree 2; Fig. 1, MAXSTATE and DELTRAN), but in most cases, closed habitat preference did not generate increases in repertoire size (Table 1; Fig. 2). However, when both gains and losses in large modifier repertoire size were analysed, there was a significant association between changes in repertoire complexity (regardless of direction) and closed habitat preference (Table 1). This suggests that, while living in closed habitats does not select for increased signal complexity, it does increase the probability of evolutionary change.
Small home range size also resulted in greater evolutionary change in modifier repertoire size in several trait resolutions (Table 1). However, both regression analyses (raw species data corrected for body size: d.f. = 27, R = 0.33, one-tailed P = 0.043; species data corrected for body size and phylogeny (tree 1): d.f. = 27, R = 0.36, one-tailed P = 0.031; species data corrected for body size and phylogeny (tree 2): d.f. = 27, R = 0.28, one-tailed P = 0.075) and most trait evolution analyses (Table 1; Fig. 3) revealed a strong tendency for modifier use to be negatively associated with home range size. Thus, the majority of available evidence supports the hypothesis that having a small home range is associated with increased signal complexity.
Finally, in most evolutionary scenarios, there was a tendency for gains in large modifier repertoire size to occur in regions of the phylogenetic tree reconstructed as being arboreal (Table 1; Fig. 4) or possessing a diet of moving prey (Table 1; Fig. 5). These relationships suggest that an arboreal lifestyle and foraging for moving prey may have enabled, or pre-adapted, species to evolve complex signals.

DISCUSSION
Current theories of signal evolution have identified a variety of selective forces that promote or reduce complexity in communicative systems (e.g. female mate choice decisions: Andersson, 1994;male-male competition: Ord et al., 2001; properties of the signalling environment: Endler, 1992; receiver sensory biases: Ryan & Rand, 1993;predation risk: Stoddard, 1999;and social complexity: Blumstein & Armitage, 1997). Our analyses of macro-evolutionary trends in lizard display modifier use suggest that ecological factors have also played a role in promoting signal diversity (Fig. 6). Specifically, habitat preference can promote evolutionary plasticity in signal repertoires, while possessing a small home range, arboreal locomotion and a diet of moving prey may facilitate (or be coupled with factors that facilitate) the evolution of complex signals.

SIGNAL DIVERSITY
The evolution of visual displays is clearly constrained by an interaction between properties of receiver sensory systems and those of the environment through which signals must be transmitted (Endler & McLellan, 1988;Endler, 1992;Fleishman, 1992). Many studies have identified directional influences of environmental factors on signal structure (Morton, 1975;Wiley & Richards, 1982;Endler, 1987Endler, , 1992Ryan et al., 1990;Fleishman, 1992;Marchetti, 1993;Losos & Chu, 1998). In contrast, we have found that closed habitat choice is significantly associated with high levels of evolutionary lability, suggesting that closed habitats promote a wide range of adaptive solutions for signal design (Fig. 6). Few previous studies have suggested that the environment produces greater levels of signal diversity (although see Alexander, 1962).
At least two non-mutually exclusive hypotheses may explain the evolution of signal diversity in closed habitats. First, because of the conservative nature of the lizard visual system (Fleishman et al., 1995), there is a certain amount of evolutionary inertia in sensory processes. Thus, when distantly related species occur in similar habitats, variation in signal repertoires may The number of evolutionary gains and losses were calculated following alternative ancestor state reconstructions by parsimony using MacClade v. 3.08a (Maddison & Maddison, 1992, 1999. Directional evolutionary change (i.e. increases in large repertoire size) was investigated by determining the number of gains in large repertoire size occurring on branches of the tree also possessing the ecological trait, and whether this distribution could be attributed to chance alone, using Maddison's (1990) concentrated-changes test. Similarly, greater evolutionary variation in repertoire size was investigated by determining the number of both gains and losses in large repertoire size occurring on branches of the tree also possessing the ecological trait and whether this distribution was due to chance alone. Results are given for ancestor reconstructions produced by both tree 1 (Fig. 1a) and tree 2 (Fig. 1b). 'ns' = not significant.   Fig. 1). Data were obtained from a variety of sources (see Appendix). = trait present, ᮀ = trait absent, = equivocal reconstruction. Large repertoire size was assumed to be absent following outgroup analysis at the ancestral node (see Ord et al., 2001). 1  2  3  4  5  6  7  8  9  10  11  12  13  15  14  17  20  21  22  25  27  28  29  30  32  35  37  39  40  41  43  44  48  47  49  50  51  54  55  38  78  77  79  80  81  82  84  83  85  86  87  88  89  90  91  92  93  94  95  96  97  98  99  100  101  102  103  104  105  106  107  108  109  110  111  112  113  114  59  61  60  62  63  64  65  66  67  73  74  75  76  116  117  124  123  122  119  121  118  120  56  58  57 reflect historical constraints on the visual system of each species. Alternatively, the closed habitat species in our analyses are largely anoles (Fig. 2) that occur in tropical environments. Some tropical habitats support a diversity of species, in part because of the opportunity to partition a complex environment (Brooks & McLennan, 1991). There is thus the potential for species groups living in such environments to possess considerable diversity in 'microhabitat' preference. Adaptation to microhabitats is known to produce differences in anoline morphology and behaviour (Losos, 1990a,b). It is hence conceivable that, as overall habitat complexity increases, the proliferation of potential microhabitats leads to a greater range of selective forces acting on signal design, resulting in the observed variation in modifier repertoire size evolution.

SIGNAL COMPLEXITY
Selection has apparently favoured the coevolution of complex signalling behaviour with small home range size (Fig. 6). We have previously found that increased male-male competitive intensity (as reflected by SSD) is also associated with an increasingly elaborate modifier repertoire (Ord et al., 2001). It is therefore tempting to speculate that species possessing small home ranges will consequently experience high levels of male-male competition, at least when habitats are saturated. The maximum distance at which a visual display can be detected is correlated with territory size in at least one species (A. auratus;Fleishman, 1992). Display rate has consistently been found to increase dramatically with decreases in inter-male distance (Hover & Jenssen, 1976;Scott, 1984;Losos, 1985a;Decourcy & Jenssen, 1994). The perceptual constraints on signal design are relaxed at short range, and complex signals may consequently have evolved in species that often interact in this way.
However, while high densities cause some reduction in the home range size of male lizards (Stamps, 1990), the principal effect is of greater home range overlap (Stamps, 1977(Stamps, , 1990Stamps & Krishnan, 1998). We suggest that a direct test of the relationship between modifier use and population density will be necessary to evaluate the probability that complex signals have evolved in response to demographic pressures. The traits of arboreality and feeding on mobile prey both facilitate the evolution of complex signals (Fig. 6). Motion perception (i.e. the ability to detect and localize moving objects) is critical for feeding on small fast-moving prey items (Jenssen & Swenson, 1974;Fleishman, 1992;Fleishman et al., 1995). Similarly, the need quickly to negotiate a complex habitat (Jenssen & Swenson, 1974) is likely to select for enhanced sensory processing. Analyses of retinal structure in anoles have revealed that they possess both a temporal and a central fovea (Fite & Lister, Figure 3. The evolution of large display repertoires and small home range size in iguanian lizards. See Fig. 2 1  2  3  4  5  6  7  8  9  10  11  12  13  15  16  17  18  19  20  21  22  23  24  25  27  28  29  30  31  32  35  37  26  39  40  41  43  46  48  47  49  45  50  42  51  54  55  38  78  77  79  80  81  83  87  88  89  90  91  92  96  98  99  100  101  102  103  105  106  107  108  109  110  111  112  113  114  115  59  62  63  64  65  66  67  68  69  70  71  72  73  74  75  76  117  124  123  122  119  121  118  120  56  57 Large display repertoire Arboreality Figure 5. The evolution of large display repertoires and feeding on moving prey in iguanian lizards. See Fig. 2 legend for details .   1  2  3  4  5  6  7  8  9  10  11  13  15  16  17  18  19  20  21  22  25  28  31  32  34  33  35  36  37  26  39  41  49  47  48  50  51  52  53  54  55  38  78  77  79  81  82  84  83  85  86  87  88  89  90  91  92  96  98  99  100  101  102  103  106  108  109  112  111  114  59  61  60  62  63  64  65  66  74  68  73  72  75  76  116  117  124  123  122  119  121  118  120  56  58  57 Large display repertoire Moving prey 1981). Such bifoveal vision is believed to maximize spatial resolution and visual acuity, allowing capture of small prey, and has otherwise only been reported in raptors (Fite & Lister, 1981). Signals are designed to exploit the perceptual characteristics of receivers (Guilford & Dawkins, 1991;Bradbury & Vehrencamp, 1998). For example, in lizards, the motion pattern of introductory display components is tailored to effectively stimulate the periphery of the visual field and attract the receiver's attention before the more information-rich portion of the display is delivered (Fleishman, 1988a(Fleishman, , b, 1992. Visual system performance will similarly influence the design of other display components by defining the visual cues that can be efficiently detected or discriminated. Improvements in the resolution and/or acuity of the visual system may thus have enabled the evolution of more complex display patterns. Alternatively, the association between repertoire complexity and diet may reflect differences in sociality. Food abundance for insectivores is relatively uniform compared to the typically fluctuating food resources of many herbivores (see Stamps, 1983 and references therein). This patchiness, and the consequent difficulty in defending food resources for herbivores, may have limited the evolution of territoriality, promoting instead non-aggressive aggregations for exploiting clumped resources (Stamps, 1977(Stamps, , 1983. The low frequency of aggressive interactions in herbivores may have relaxed selection for complex signals to mediate territorial disputes. In closing, we note that, while it is important to identify all of the selective forces acting on signal design, the mechanisms through which these act must be explored if we are to develop a truly comprehensive model to explain the evolution of communicative systems. Comparative studies of lizard displays reveal that ecological factors sometimes act directionally, but that they can also promote greater lability in signal evolution (Fig. 6). We anticipate the possibility of similar processes occurring in other taxa.

ACKNOWLEDGEMENTS
We thank Jon Losos, Emília Martins and John Wiens for advice on iguanid phylogeny and comments on our phylogenetic selection criteria; Scott Hocknull for advice on agamid phylogeny and Allen Greer for sharing unpublished phylogenetic hypotheses; Charlie Nunn for advice on concentrated-changes test methodology; and Linda Evans for comments on a previous version of the manuscript. We are also indebted to the staff of the Macquarie University Library for tracking down and obtaining many sources used in this study.