Introduction

Habitat structural complexity is considered an important factor influencing the population dynamics and trophic organization of terrestrial arthropods (Price et al. 1980; Kareiva 1987; McCoy and Bell 1991; Hunter and Price 1992; Döbel and Denno 1994; Siemann 1998; Rypstra et al. 1999). Specifically for natural enemies, habitat structure is thought to affect their accumulation and conservation in both natural and managed ecosystems (for reviews, see Cromartie 1981; Altieri and Letourneau 1982; Sheehan 1986; Andow 1991; Landis et al. 2000; Sunderland and Samu 2000; Symondson et al. 2002). By incorporating the effects of habitat structure on higher trophic levels, it is possible to more thoroughly understand the extent to which top-down effects from natural enemies cascade throughout a food web (McCoy and Bell 1991; Hunter and Price 1992; Denno et al. 2002). However, information on the effect of habitat structure on the accumulation of natural enemies and reasons underlying such a response has never been synthesized in any quantitative manner.

The relative complexity of a habitat is determined largely by the number of different structural elements per unit volume (McCoy and Bell 1991). Moreover, how habitat complexity is defined, as well as its ensuing effects on natural enemies, depends in large part on the spatial scale at which effects are examined (McCoy and Bell 1991; Landis et al. 2000). From an arthropod’s perspective, habitat structure can vary from simple to complex at several relevant spatial scales. From large to small, these spatial scales are: landscape (e.g. across a matrix or gradient of different habitats), habitat (e.g. within a specific habitat type), and small-scale (e.g. within individual plants or plant parts).

To assess landscape-level complexity and its effects on natural enemies, it is important to consider how complexity varies across several different habitats (e.g. agricultural plantings, old fields, woodlots, etc.) (Landis et al. 2000). Although studies on the effects of landscape complexity on natural enemies are sparse at present (but see Marino and Landis 1996; Menalled et al. 1999), researchers generally recognize the potential importance of landscape-level complexity for the density and diversity of natural enemy complexes (Thies and Tscharntke 1999; Landis et al. 2000). At present, more is known about how variation in structural complexity impacts natural enemy abundance at smaller spatial scales within a habitat or plant. Within a single habitat type, structure can vary from simple to complex in terms of its plant species diversity (e.g. monoculture versus polyculture), plant structural diversity (low versus high profile vegetation), and detrital diversity (leaf litter absent or present). In general, researchers acknowledge that increases in structural complexity at the habitat level may encourage invertebrate predators and parasitoids (Uetz 1991; Rypstra et al. 1999; Landis et al. 2000; Sunderland and Samu 2000).

At a smaller spatial scale yet, structure can vary at the level of a single plant or among individual plants. For example, different crop varieties have leaves that vary dramatically in the density of hair tufts (domatia) associated with the major leaf veins. Predatory mites find refuge in domatia (Grostal and O’Dowd 1994; Agrawal 1997; Roda et al. 2000) suggesting that structural variation at the small spatial scale of a leaf can impact natural enemy abundance.

Historically, ecologists have recognized that the abundance and diversity of invertebrate natural enemies is promoted in complex-structured habitats (Lowrie 1948; Barnes 1954; Kajak 1960; Luczak 1963; Duffey 1966). However, many of these earlier studies were correlative, anecdotal, and lacked an experimental approach. Recently, a growing number of studies have employed an experimental approach in the field to elucidate the effects of habitat complexity on natural enemy abundance. Although field experiments yield important insights regarding ecological patterns and processes, they are logistically formidable to manage and thus often suffer from low replication. As a result, the power of such experiments to detect a statistically significant effect is often low (Arnqvist and Wooster 1995; Van Zandt and Mopper 1998).

Meta-analysis offers a solution to the dilemma of low statistical power in individual experiments because it allows for a synthesis of results from multiple independent studies that test the same hypothesis (Hedges and Olkin 1985; Gurevitch et al. 1992; Gurevitch and Hedges 1993; Cooper and Hedges 1994; Arnqvist and Wooster 1995). Meta-analytical techniques quantify the magnitude of a treatment effect (e.g. a habitat structural manipulation) from individual studies as well as calculate the overall magnitude and significance of the cumulative effect across all studies examined (Hedges and Olkin 1985; Rosenberg et al. 2000).

The objective of this analysis is to provide insight into the effects of habitat structure on the aggregation and retention of invertebrate natural enemies in both natural and managed ecosystems; an important goal from both theoretical and applied perspectives. Specifically, this assessment addresses four issues related to the response of natural enemies to habitat complexity and seeks to: (1) establish whether natural enemies exhibit a consistent pattern of accumulation in complex compared to simple-structured habitats, (2) assess whether habitat manipulations at two spatial scales (within-habitat and within-plant) similarly influence the effects of complexity on natural enemies, (3) test if the aggregation of natural enemies in complex habitats results from a numerical response to abundant prey (e.g. herbivores), and (4) assess if different natural enemy guilds respond similarly to manipulations of habitat structure. This meta-analysis aims to synthesize the results of published experimental studies and present a clear consensus of the effect of habitat structural complexity on the abundance of arthropod predators and parasitoids. In addition, potential mechanisms underlying the response of natural enemies to complex habitats will be discussed.

Materials and methods

Survey of studies and criteria for inclusion

We conducted an exhaustive and systematic survey of the literature and ultimately assembled a set of 43 published studies where habitat structure was manipulated and the density responses of 62 independent predator taxa were assessed. Although the literature search yielded many potential studies to be included in the analysis, only studies that satisfied three criteria were selected. First, we included only experimental studies in which one or more elements of habitat complexity (e.g. leaf litter volume or diversity, leaf density, or vegetation density or diversity) were manipulated or where study sites were specifically chosen to represent structurally complex and structurally simple habitats (Hanks 1991; Leddy 1996; Marino and Landis 1996; Tooker and Hanks 2000). Studies in which the complexity of a single habitat was manipulated over time (i.e. pre- versus post-tillage assessment of predator densities in an agricultural field) were excluded from this review. Second, studies must have reported treatment means, variance estimates and sample sizes for natural enemy abundance. These measures are necessary in order to calculate an effect size (treatment compared to control) for each individual study in the meta-analysis (Rosenberg et al. 2000). In studies where it was especially difficult to estimate natural enemy abundance and where prey were sessile, we accepted measures of sessile prey mortality as an estimate of natural enemy density when those data were available (Hanks 1991; Marino and Landis 1996; Tooker and Hanks 2000). Third, studies must have included an appropriate control. For studies that compared monocultural plantings with bi- or polycultural plantings, the monoculture was treated as the “simple-structured” control.

One of the assumptions of meta-analysis is that each observation is independent of all others (Arnqvist and Wooster 1995; Rosenberg et al. 2000). To minimize the effects of non-independence, results for only one taxon per study were used for our meta-analysis. If the responses of multiple predator taxa were published in the same report, we selected the focal predator (identified by the author) or one of the most abundant predators for meta-analysis. When data on enemy responses were presented across several sampling dates, we selected the mid-season date when the greatest difference between treatment and control densities occurred to calculate effect size.

The data set

Included in our assessment were studies that examined natural enemy responses to manipulated habitat structure at two spatial scales: habitat and within-plant. Because there was only one landscape-level study in our search that met the above criteria (Marino and Landis 1996), it was pooled with studies where vegetation complexity was manipulated at the habitat level.

At the habitat spatial scale, studies were pooled into one of two categories for analysis, either “detritus structure” or “vegetation structure.” Included in the detritus category were studies in which leaf litter, mulch, or groundcover was manipulated. For these studies, habitat structure was altered by either supplementing or reducing the amount of detritus in a natural system, placing straw or mulch within a managed garden or crop, or using organic fertilizers to promote the growth of ground cover.

Also at the habitat scale, predator responses were measured in response to manipulations of vegetation structure. Many studies in this category compared predator densities between simple-structured monocultures and complex-structured bicultures or polycultures. Predator responses were also compared between agricultural fields with different intensities of weed management or in natural systems where plant species density was manipulated. Natural enemy abundance was also assessed in urban plantings where their abundance was compared on the same plant species between monocultures and more diverse plantings.

At the within-plant spatial scale, natural-enemy responses were compared between simple and complex-structured treatments achieved by plant structural manipulations. Plant architecture was manipulated by altering the amount of pubescence associated with leaf domatia, by changing leaf density, or by varying branch density.

To determine if natural enemy taxa responded similarly to structural manipulations, enemies were sorted into ten guilds for analysis. In some instances, natural enemies were not separated into specific taxa, but were pooled for analysis into groupings such as the “natural enemy assemblage”. For spiders, studies were sorted into one of three categories for analysis: hunters (species that do not build webs), web-builders, or the spider assemblage. All 62 taxa were ultimately sorted into one of the following ten natural enemy guilds for analysis: web-building spiders, hunting spiders, spider assemblage, hemipteran predators, coccinellid beetles, carabid beetles, predaceous mites, ants, parasitoids, and the natural enemy assemblage at large. Of the 62 independent cases, there were 43 instances in which habitat structure was experimentally increased and 19 instances where structure was decreased. Enemy responses to increased and decreased habitat complexity were analyzed separately for each spatial-scale category and guild.

To test if natural enemies aggregated in complex habitats because prey were more abundant there, prey abundance was compared between treatments and controls. In all, there were 18 studies that assessed how prey and their natural enemies responded to increases in habitat complexity and nine studies that determined their responses to decreased habitat complexity. Of these studies, all but one focused on herbivorous prey. The remaining study (Bultman and Uetz 1984) measured the abundance of Collembola (a detritivore) in manipulated treatments (detrital complexity increased and decreased) relative to an unmanipulated control.

Meta-analyses

Means (X̄), standard deviations (SD), and sample sizes (n) for controls and treatments were extracted from the text, tables or graphs of published reports. If data were available only in graphs, they were first scanned before using the imaging program Scion (http://www.scioncorp.com/) to extract means and variance estimates. For each study, the magnitude of the effect of structural manipulations on natural enemy abundance (d i) was estimated as the standardized mean difference between the control and experimental mean (see Electronic Supplementary Material):

$$d_{{\text{i}}} = {{{\bar{X}^{{\text{E}}} - \bar{X}^{{\text{C}}} }} \over{{{\text{SD}}}}}J$$
(1)

where X̄ E is the mean abundance of natural enemies in manipulated habitats, X̄ C is their mean abundance in the control habitat, SD is the pooled standard deviation, and J is a correction term for bias associated with sample size.

Separate meta-analyses (Gurevitch and Hedges 1993; Rosenberg et al. 2000) were conducted to determine the effect of increasing or decreasing habitat/plant structural complexity on the abundance of (1) all natural enemies (pooled total) independent of spatial scale, (2) natural enemies at the habitat spatial scale (detritus and vegetation analyzed separately) and within-plant scale, (3) natural enemies in each of the ten guilds, and (4) prey at the habitat spatial scale. The effect size (d * ++ or Hedge’s d) (Hedges and Olkin 1985; Rosenberg et al. 2000) for each meta-analysis was calculated as:

$$d^{*}_{{ + + }} = {{{{\sum\limits_{i = 1}^n {w_{{\text{i}}} d_{{\text{i}}} } }}} \over{{{\sum\nolimits_{{\text{i}} = 1}^n {w_{{\text{i}}} } }}}}$$
(2)

where d i is the effect size for the i th study, and w is the weight (reciprocal of the sampling variance) for the i th study.

Effect size measures the magnitude of the structural manipulation on natural-enemy density. A positive value of d suggests that increasing habitat complexity resulted in an increase in natural enemy density, whereas a negative value of d indicates that the manipulation reduced natural enemy density. An effect of habitat/plant structure was considered significant if the 95% bias-corrected bootstrap confidence interval (CI) of the effect size (d * ++) did not overlap zero (Rosenberg et al. 2000). Bias-corrected bootstrap CIs were calculated for each d * ++ from re-sampling tests generated from 999 iterations (Adams et al. 1997; Rosenberg et al. 2000). All calculations were performed using MetaWin statistical software (Rosenberg et al. 2000).

An inherent problem when analyzing published literature is that studies with non-significant results often go unpublished. To address the issue of publication bias (the “file-drawer problem” sensu Arnqvist and Wooster 1995), we generated a fail-safe number for each effect size (d * ++) (Rosenthal 1979; Rosenberg et al. 2000). A fail-safe number represents the number of non-significant studies needed in an analysis to change a significant result into a non-significant one. Thus, the higher the fail-safe number, the more credence given to a significant result.

Results

Overall response of natural enemies to alterations in habitat structural complexity

There was a significant overall effect of altered habitat complexity (vegetation and detritus manipulations considered collectively) on the abundance of arthropod natural enemies. In general, enhancing the structural complexity of the habitat resulted in a significant increase in natural enemies, both predators and parasitoids (d=0.87, CI=0.35 to 1.36, df=42, fail-safe number=407; Fig. 1A). Recall that when CIs do not bracket zero, there is a significant treatment effect (P<0.05). Note too that effect sizes exceeding 0.7 are considered large for studies involving insect populations (Tonhasca and Byrne 1994). Thus, increasing the structural complexity of the habitat resulted in a large effect on natural enemy abundance. The large fail-safe number indicated little sample bias and that an additional 407 non-significant studies would be necessary to dismiss the significant effect and that sample bias towards studies that report significant results is an implausible explanation for this result. Similarly, decreasing habitat complexity resulted in a significant negative effect on natural enemy abundance (d=−0.64, CI=−1.04 to −0.30, df=18, fail-safe=111; Fig. 1A).

Fig. 1
figure 1

Effect size of experimental increases and decreases in habitat structural complexity (relative to an unmanipulated control) on A the overall abundance of natural enemies independent of spatial scale and habitat, and on B the abundance of natural enemies at the spatial scales of habitat (detritus and above-ground living vegetation) and plant (architecture). Mean effect size ±95% bootstrap confidence intervals (CI) from meta-analyses are shown. The effect of the structural treatment/manipulation was considered statistically significant if the 95% CI did not overlap zero (dashed line). The asterisk represents significant treatment effect (P<0.05), ns=non-significant treatment effect (P>0.05). Sample sizes are given to the right of each effect size

Full size image

Response of natural enemies to altered structural complexity at two spatial scales

At the spatial scale of a single habitat, increases in structural complexity resulted in a significant increase in natural enemy abundance (Fig. 1B). Natural enemy increases were large when structural complexity was increased by manipulating the amount of detritus (e.g. thatch, leaf litter or mulch) (d=2.37, CI=0.65 to 5.60, df=6, fail-safe=33) as well as when the structural complexity of the living vegetation was enhanced via no-till/mowing practices, intercropping, or polyculture (d=0.69, CI=0.21 to 1.28, df=31, fail-safe=100). Increasing architectural complexity at the within-plant spatial scale (e.g. altering leaf density, branch structure, or the structural complexity of leaf domatia) resulted in a smaller, yet still significant, increase in natural enemy abundance (d=0.43, CI=0.01 to 0.83, df=3, fail-safe=0; Fig. 1B).

Decreasing structural complexity resulted in a significant reduction in natural enemy abundance at both the habitat and within-plant spatial scales. For example, when the structural complexity of detritus was reduced at the habitat spatial scale natural enemy abundance was drastically reduced (d=−1.05, CI=−8.09 to −0.76, df=2, fail-safe=6; Fig. 1B). Decreases in habitat structure at the within-plant spatial scale (e.g. leaf density reduced or leaf domatia excised) also promoted reductions in natural enemy abundance (d=−0.70, CI=−1.46 to −0.27, df=6, fail-safe=16; Fig. 1B). However, decreases in vegetation complexity at the habitat level produced a marginally non-significant decrease in natural enemy abundance (d =−0.44, CI=−1.29 to 0.01, df=8; Fig. 1B).

Responses of natural-enemy guilds to altered habitat complexity

Of the nine natural enemy guilds analyzed (ants were excluded due to no sample size), seven exhibited significant positive density responses to increased structural complexity of vegetation (polycultures, weedy fields, interplantings) or detritus at the habitat spatial scale (Fig. 2). For these seven guilds, hunting spiders showed the strongest response (d =2.34, CI=0.48 to 4.62, df=3, fail-safe=4), followed by web building spiders (d=2.09, CI=0.55 to 5.09, df=4, fail-safe=17), the natural enemy assemblage at large (d=1.63, CI=0.04 to 4.15, df=3, fail-safe=3), the pooled spider assemblage (d=1.30, CI=0.12 to 3.89, df=5, fail-safe=5), hemipteran predators (d=1.21, CI=0.37 to 2.27, df=7 fail-safe=19), parasitoids (d=0.69, CI=0.05 to 1.36, df=1, fail-safe=0), and mites (d=0.62, CI=0.19 to 1.04, df=1, fail-safe=0). Only the predaceous beetles, coccinellids (d=−0.42, CI=−1.26 to 0.39, df=6) and carabids (d=−0.48, CI=−2.65 to 0.62, df=3), failed to show a significant response to increased habitat complexity.

Fig. 2
figure 2

Effect size of experimental increases and decreases in habitat structural complexity on the abundance of natural enemies in ten different guilds. Mean effect size ±95% bootstrap CIs from meta-analyses are shown. The effect of the structural treatment/manipulation was considered statistically significant if the 95% CI did not overlap zero (dashed line). The asterisk represents significant treatment effect (P<0.05), ns non-significant treatment effect (P>0.05). Sample sizes are given to the right of each effect size

Full size image

For studies in which habitat structural complexity was decreased, four of six natural-enemy guilds were adversely affected (Fig. 2). Web-building spiders showed the strongest negative density response to decreased habitat complexity (d=−1.69, CI =−5.71 to −0.97, df=2, fail-safe=2), followed by ants (d=−1.04, CI=−1.61 to −0.58, df=1, fail-safe=1), hunting spiders (d=−0.86, CI=−1.99 to −0.28, df=2, fail-safe=4), and the spider assemblage at large (d =−0.83, CI=−0.88 to −0.79, df=2, fail-safe=14). Mites failed to respond significantly to a decrease in habitat complexity (d=−0.54, CI=−2.60 to 0.38, df=3). Hemipteran predators showed a slight positive response to decreased habitat structure (d=0.18, CI=0.14 to 0.20, df=1, fail-safe = 0), but the low fail-safe number instills little confidence in this result. Carabid and coccinellid beetles, parasitoids, and the natural enemy assemblage at large were excluded due to low or no sample size.

Prey response to altered habitat complexity

Potential prey (mostly herbivores) did not exhibit a significant density response when habitat complexity was increased (d=−0.60, df=16, CI=−1.52 to 0.34). Similarly, when habitat complexity was simplified, herbivores failed to show a significant density response (d=−0.14, df=8, CI=−0.38 to 0.49). Thus, there is little evidence that the strong patterns of predator accumulation in complex-structured habitats results from a numerical response to enhanced prey density.

Discussion

General patterns of enemy accumulation in complex habitats and underlying mechanisms

Meta-analysis of published experimental studies in which the structural complexity of habitats or plants was manipulated or controlled for found a strong overall effect on the abundance of invertebrate natural enemies (Fig. 1A). Notably, effects were detected at two spatial scales (habitat and within-plant), in above-ground vegetation and detritus-based systems (Fig. 1B), and across a diversity of predator and parasitoid guilds (Fig. 2). Although increases in vegetation complexity at the habitat spatial scale (interplanting, no-till practices, or polyculture) promoted higher densities of natural enemies, enhancing the structural complexity of detritus (thatch, leaf litter, or mulch) had an enormous positive impact on natural enemy abundance (Fig. 1B). Similarly, decreasing detrital structure had a large negative effect on natural enemy abundance. Decreasing the structural complexity of vegetation, however, did not have a significant effect on enemy abundance. At the within-plant spatial scale, populations of natural enemies were significantly reduced when the architectural complexity of the plant was reduced by reducing leaf density or excising leaf domatia (Fig. 1B). In addition, increasing plant architectural complexity promoted higher densities of natural enemies, although the magnitude of the effect was relatively small.

Several hypotheses emerge to explain why natural enemies accumulate in structurally complex habitats, although few have been tested rigorously (see Rypstra et al. 1999; Landis et al. 2000; Finke and Denno 2002; Langellotto 2002). Natural enemies may aggregate in complex-structured habitats because they: (1) encounter more abundant prey, (2) gain refuge from predation (cannibalism and intraguild predation), (3) are able to locate and capture prey more effectively, (4) encounter a more favorable microclimate, or (5) gain access to alternative resources (e.g. pollen or nectar). Although the pattern of enhanced natural enemy abundance in complex habitats is robust (Figs. 1 and 2), the underlying ecological mechanisms promoting the accumulation of natural enemies in complex habitats remain largely unexplored (Sunderland and Samu 2000). Of the studies included in this meta-analysis, the ecological mechanism underlying the aggregation of natural enemies in complex-structured habitats was rigorously assessed in only 5 of the 62 total experimental studies we surveyed (see Electronic Supplementary Material). With this small number of assessment, a quantitative analysis of the mechanisms promoting natural enemy accumulation within complex habitats was not possible. Nonetheless, a few insights follow.

For those studies we examined that measured both natural enemy and herbivore responses to altered habitat structure, we found no overall effect of the structural manipulation on herbivore density. However, for generalist predators such as carabid beetles and spiders, when prey (not necessarily herbivores) are abundant, predators are known to achieve high densities (Wise and Chen 1999; Halaj and Wise 2002). Thus, our analysis of herbivore abundance alone likely did not characterize the entire prey assemblage that was available to the predator. In addition, it is difficult to disentangle the cause–effect nature of the relationship between natural enemy abundance and prey abundance from the studies included in our assessment. Herbivore populations may not increase in complex structured habitats because the initial aggregation of natural enemies ultimately inhibits herbivore increase. For example, the wolf spider Pardosa littoralis aggregates in litter-rich habitats independent of prey presence, exhibits a stronger numerical response to prey there, and captures prey more effectively (Döbel and Denno 1994; Denno et al. 2002). Notably, the combination of these factors ultimately results in the long-term suppression of planthopper prey (Döbel and Denno 1994).

There is some evidence that prey capture is facilitated in complex-structured habitats. For example, orb-weaving spiders select complex-structured habitats that provide increased sites for web attachment (McNett and Rypstra 2000). In addition, at a small spatial scale, coccinellid beetles are able to forage better and capture prey more effectively on more architecturally complex plants with pubescence (Kareiva and Sahakian 1990). However, at a larger spatial scale, complex-structured vegetation is known to disrupt the searching behavior and local dispersal of coccinellids (Andow and Risch 1985; Kareiva 1987). Thus, the interaction between structural complexity and the efficiency of prey capture may be scale and predator species dependent.

Evidence is building that refuge from cannibalism or intraguild predation may underlie the accumulation of natural enemies in complex-structured habitats (Gunnarsson 1990; Roda et al. 2000; Norton et al. 2001; Langellotto 2002). For example, experimental studies show that the intraguild predation of hemipteran predators by wolf spiders is vastly reduced in litter-rich habitats compared to litter-free ones (Finke and Denno 2002). Likewise, hunting spiders gain refuge from conspecific cannibals in structurally complex vegetation (Langellotto 2002; but see Wagner and Wise 1996).

Alternatively, refuge from harsh abiotic conditions has been offered as an explanation for enhanced densities of predators in complex-structured habitats and microhabitats (Riechert and Bishop 1990; Grostal and O’Dowd 1994). In particular, predaceous mites achieve higher densities on architecturally complex plants with leaf domatia (Grostal and O’Dowd 1994; Agrawal 1997), perhaps because domatia provide refuge from unfavorable abiotic conditions. In contrast, experimental studies with wolf spiders show that they do not select complex-structured habitats on the basis of microclimate (Langellotto 2002).

Access to alternative food resources may also explain why some predators and parasitoids accumulate in complex-structured habitats, particularly those that supplement their diets with nectar and pollen. For example, enhancing vegetation diversity by encouraging the growth of pollen and nectar sources (e.g. flowering weeds) promotes higher densities of carabid beetles (Lys et al. 1994), syrphid flies (White et al. 1995; Haussmann 1996), and parasitoids (Patt et al. 1997). Obligate predators such as spiders that do not normally feed on nectar or pollen are not consistently associated with diverse vegetation that provides such resources (Riechert and Bishop 1990; Jmhasly and Nentwig 1995). Thus, the elevated density of hunting and web-building spiders in complex-structured habitats (Fig. 2) likely results from other causes.

It is likely that multiple explanations underlie the pattern of accumulation in complex habitats, even for a single natural enemy species. For example, by aggregating in litter-rich habitats, the wolf spider P. littoralis gains refuge from cannibalism (Langellotto 2002), and also gains access to alternative prey (Gratton and Denno 2003). Similarly, carabid beetles appear to increase in intercropped fields because of greater access to prey and alternative food resources (Zangger et al. 1994). Overall, complex habitats are more likely to provide a range of favorable conditions (e.g. microhabitat, alternative prey or food resources, refuge from predation) that not only attract natural enemies, but also decrease the need to move in search of more suitable conditions (Sunderland and Samu 2000). This dynamic interaction between increased immigration (Döbel 1996) and reduced emigration (Langellotto 2002) ultimately promotes the accumulation of an abundant natural enemy community within complex-structured habitats.

Responses of natural enemy guilds to altered habitat complexity

In general, most guilds of natural enemies were significantly affected when the structural complexity of the habitat was altered (Fig. 2). Seven of nine natural enemy guilds were more abundant under conditions of increased habitat complexity. Similarly four of six guilds exhibited a decrease in density when habitat structure was simplified. Although significant, many effect sizes were based on few observations (2–8 studies per guild). Thus, conclusions regarding the response of some enemy guilds should be drawn with caution.

Hunting spiders, web-building spiders, and the pooled spider assemblage all showed striking increases in abundance when habitat structure was increased and large decreases in density when habitat structure was simplified (Fig. 2). Our results agree with those of Sunderland and Samu (2000), who reviewed 24 studies specific to spiders in agroecosystems and found that mean abundance increased 33% and 80% when vegetation diversity and detrital structure were respectively enhanced. The strong response of spiders to changes in habitat structure is not surprising, given the behavioral and physiological sensitivity of both hunting and web-building spiders to structural elements in their habitats (Uetz 1991). Clearly, web-building spiders must anchor their webs to the substrate, and structured habitats provide a greater range of suitable attachment sites (McNett and Rypstra 2000). In addition, because both hunting and web-building spiders utilize substrate-borne vibrations to locate prey and mates (Barth 1985), they may be better able to localize and capture prey and avoid intraguild predators in complex-structured habitats where vibrations are more easily detected (see Uetz and Stratton 1982). Predaceous mites also showed a moderate density response on architecturally complex plants with leaf domatia, a response that is probably attributable to refuge from desiccation (Grostal and O’Dowd 1994) or intraguild predators (Roda et al. 2000; Norton et al. 2001).

Hemipteran predators exhibited a moderate increase in abundance when habitat complexity was enhanced (Fig. 2). Because some hemipteran predators, especially smaller ones such as mirids and mesoveliids, are very susceptible to intraguild predation (Fagan et al. 1998, Finke and Denno 2002), and because complex structured habitats provide refuge for them from intraguild predation (Finke and Denno 2002), it is not surprising that this predator guild was positively affected by increases in habitat structure.

Parasitoids and ants were also sensitive to alterations in the structural complexity of their habitats (Fig. 2). Increased habitat complexity (e.g. floral resources present) promoted higher densities of parasitoids, and ants were less abundant when structural complexity was reduced (e.g. understory cleared). Because parasitoids and ants are so reliant on alternative food such as pollen, nectar, or seeds (Perfecto and Sediles 1992; Patt et al. 1997; Watt et al. 1997), changes in the availability of these resources that are associated with altered habitat complexity can have important effects on their abundance (Gurr et al. 1998).

The only natural enemy guilds that did not show a significant, positive response to increased habitat structure were coccinellid and carabid beetles, for which there were few studies that met our criteria for inclusion in this meta-analysis. Nonetheless, it appears as if the type of manipulation and the spatial arrangement of resources within a habitat appears to influence the aggregative response of both taxa. For example, complex-structured polycultures appear to disrupt local dispersal in coccinellids (Andow and Risch 1985; Kareiva 1987); a problem that can be negated by the spatial arrangement of nectar and pollen-producing vegetation alongside the cropping system (Wetzler and Risch 1984). For carabid beetles, ‘beetle banks’ (islands of vegetation diversity and complexity within a cropping system) have been used successfully to attract and retain carabid beetles in crop systems (Thomas et al. 1991). However, the studies included in our meta-analysis assessed the impact of different tillage regimes on carabid beetles (House and All 1981; Carcamo 1995; Clark et al. 1997). This suggests that no-till management regimes may not be an efficient way to increase carabid density.

Responses of natural enemies to habitat complexity: relevance to ecology and agriculture

For many guilds and at several spatial scales, our analysis found that the abundance of invertebrate predators and parasitoids was positively affected by increases in the structural complexity of their habitats. Moreover, the mechanisms underlying the pattern were diverse. For generalist predators, however, there is growing evidence that refuge from intraguild predation and cannibalism, a common mortality source among invertebrate predators (Polis et al. 1989; Rosenheim et al. 1993; Wagner and Wise 1996; Rosenheim 2001), may be an important explanation for their accumulation in complex-structured habitats (Roda et al. 2000; Norton et al. 2001; Finke and Denno 2002). The consequences of dampened intraguild predation in complex-structured habitats may have profound consequences for trophic interactions, food-web stability, and trophic cascades. Generalist predators are ubiquitous in terrestrial ecosystems where they commonly feed on herbivores, conspecifics and other predators (Rosenheim et al. 1993; Wagner and Wise 1996), and thus form complex and reticulate feeding links throughout the food web (Polis and Strong 1996). In such reticulate food webs, top-down effects are thought to attenuate and trophic cascades are less likely (Polis and Strong 1996; Polis 1999; Schmitz et al. 2000).

A review of manipulative field studies where predator density was increased and the effects on lower trophic levels were measured found that predator assemblages significantly reduced herbivore abundance in 79% of the cases and reduced plant damage or increased yield in 65% of the cases (Symondson et al. 2002). Thus, predators have the potential to enact top-down control in terrestrial ecosystems. However, the studies reviewed by Symondson et al. (2002) included cases where predator abundance was increased to artificially high levels. We suggest that top-down control of herbivores is most likely to occur when complexity is incorporated into a habitat on a local, regional or landscape scale and predators naturally aggregate. If habitat complexity confers favorable microhabitats and refuge from predation for natural enemies, top-down effects on herbivores should increase, not only because natural enemies are more abundant but also because antagonistic interactions among them are reduced. Indeed, there is evidence that the selective accumulation of natural enemies in complex habitats (Andow 1991; Döbel and Denno 1994; Denno et al. 2002) coupled with relaxed intraguild predation intensifies top-down effects and promotes herbivore suppression (Finke and Denno 2002; Langellotto 2002). Thus, this meta-analysis adds to a growing number of assessments demonstrating that basal resources often mediate top-down impacts (Forkner and Hunter 2000; Denno et al. 2002) and suggests that modifications that enhance habitat complexity have the potential to increase the overall effectiveness of natural enemy complexes in agroecosystems (Gurr et al. 1998; Landis et al. 2000; Symondson et al. 2002).

Despite widespread evidence for the accumulation of invertebrate predators and parasitoids in complex-structured habitats, we still know very little concerning the mechanisms that underlie this response (Sunderland and Samu 2000, this analysis), details of dispersal dynamics (Bommarco and Fagan 2002), and the spatial extent of specific predator subsidies as they impact local herbivore and food-web dynamics (Polis et al. 1997). It is encouraging, however, that the four guilds of natural enemies (hemipterans, mites, parasitoids, and spiders) that have historically played important roles in biological control initiatives (Kenmore et al. 1984; Altieri et al. 1993; Nyrop et al. 1998; Sunderland and Samu 2000), respond so positively to structural changes in their habitats (Fig. 2). With a better knowledge of the mechanisms promoting natural enemy aggregation in complex habitats, including a thorough understanding of how specific structural elements moderate antagonistic interactions among members of the predator complex, we may be better able to conserve natural enemies and at the same time increase their collective effectiveness in suppressing pests.