The effectiveness of classical biological control of invasive plants


Correspondence author. Imperial College London, Silwood Park, Buckhurst Road, Ascot, Berkshire SL5 7PY, UK. E-mail:


  1. Invasive alien plants have serious economic and ecological impacts, for example, by displacing native plants and invertebrates, and their management is often costly and ineffective in the long term. Classical biological control using specialized, coevolved natural enemies from the native region of the invader is often advocated as a preferred alternative to chemical and mechanical control, but there is a lack of quantitative assessment of control of the target species and subsequent establishment of native vegetation and invertebrates.
  2. Meta-analyses were carried out combining the results of 61 published studies (2000–2011) that quantified the impact of classical biocontrol at the level of individual target plants, target populations or non-target vegetation. Factors associated with the control programmes (invasive region, native region, plant growth form, target longevity, control agent guild, taxonomy and study duration) were analysed to identify patterns in control success.
  3. On average, biocontrol agents significantly reduced plant size (28 ± 4%), plant mass (37 ± 4%), flower and seed production (35 ± 13% and 42 ± 9%, respectively) and target plant density (56 ± 7%). Beetles in the Chrysomelidae and Curculionidae families were more effective at reducing plant size than other groups.
  4. Non-target plant diversity significantly increased by 88 ± 31% at sites where biocontrol agents were released, but it was largely unclear whether the replacement plant species were native or invasive.
  5. Synthesis and applications. The number of studies that provide quantitative indications of the effectiveness of biocontrol and the response of non-target taxa has increased over the past decade, but remains small compared to the total number of publications on the classical biocontrol of invasive plants. Nonetheless, this study demonstrates the positive impacts of classical biocontrol and the re-establishment of native plants in a broad range of systems and establishes the value of classical biocontrol for the control of invasive alien plants. The Chrysomelidae and Curculionidae families were the most effective agents and we recommend these be prioritized in cases where potential agents of different taxa have also been identified. In addition, data on the recovery of native plant species and the invertebrate community remain sparse and it is recommended that future studies report the identity of plant species that replace target species as well as invertebrate community responses.


Invasive species are a key threat to biodiversity (Pimentel 2001) and the Convention on Biological Diversity stipulates the prevention, control or eradication of invasive species and the mitigation of their impacts (United Nations 1992). Invasive plants are among the most numerous and prominent species in invasion ecology (Callaway & Maron 2006). Only a small proportion of non-native plants have a detrimental impact in their introduced ranges (Crawley, Harvey & Purvis 1996), often estimated to be around 1% (Williamson 1996). However, those that do have the potential to cause considerable economic damage to agriculture, forestry and infrastructure (Pimentel, Zuniga & Morrison 2005; Williams et al. 2010), threaten native biodiversity and impact ecosystem services (Pejchar & Mooney 2009; Vilà et al. 2011). Control of invasive species using mechanical and chemical methods can often be expensive. For example, in Great Britain, an estimated £90 million is spent annually on herbicides for use against invasive weeds in agriculture alone (Williams et al. 2010). More importantly, chemical and mechanical methods are often not sustainable in the long term as control efforts remain necessary over a prolonged period. Classical biological control (biocontrol) is advocated as an alternative to conventional invasive species management that has the potential for long term, self-perpetuating and effective control, especially in more sensitive environments such as protected areas or riparian habitats (McFadyen 1998).

One criticism of classical biocontrol has been the lack of quantitative assessment of effectiveness, especially post-release (McEvoy & Coombs 1999). Many classical biocontrol programmes have been concerned with the identification, safety-testing and release of control agents, but less with the control success once the agent was released (McEvoy & Coombs 1999) where often little information is available (Julien & Griffiths 1998). However, without thorough assessment of the effectiveness of control efforts, it is difficult to understand why these programmes worked (or not) and how to prioritize potential target species, potential control agents and resources. Detailed monitoring is also required to understand non-target impacts, such as the floral community changes after the creation of free space (Barton et al. 2007). To objectively assess the effectiveness of biocontrol, standardized measures of success are needed. However, most studies have used subjective or imprecise descriptions of success, such as classifying the degree of control as ‘slight’ or ‘marked’ (Crawley 1989), or focused entirely on impact on the target species (Hoffmann 1995; Fowler, Syrett & Hill 2000).

Ultimately, the measure of success should be whether specific aims, defined at the start of a biocontrol programme, are met (van Klinken & Raghu 2006). Aims and objectives will vary depending on context, such as the impact on the target species and costs and benefits of biocontrol relative to conventional control measures. For example, success in an agricultural setting may depend on the reduction in the cost of conventional control methods or increased crop yields as a result of effective biocontrol of the invasive plant. In a natural environment, by contrast, the invasive species may not be managed and its economic cost may be unknown. Here, an increase in the abundance of native species after biocontrol of the invasive plant, or the establishment of a community similar to the original community that was displaced by the invasive species may indicate success. If the aims of a control programme are to mitigate or reverse the detrimental environmental impacts of an invasive plant, only monitoring the reduction in abundance of the target species may not be sufficient when the negative impacts include the loss of native species. In this case, the impact on native species and communities and their functioning should also be recorded. However, few control programmes have been quantitatively assessed regarding the community composition and the results of such assessments have rarely been reported. Recently, the importance of studying post-control changes in the plant and invertebrate community has been discussed and components of effective study designs, such as control sites where no biological control agents are released, have been suggested (Blossey & Skinner 2000; Carson et al. 2008; Morin et al. 2009), but no analysis of the combined results has occurred. Furthermore, setting specific goals based on the impacts of the target species prior to release, for example, setting an acceptable target density, against which the effectiveness of control can be measured, can be beneficial (Paterson et al. 2011). This will help to distinguish between success and impact as a successful biocontrol project need not necessarily demonstrate changes in all possibly measures of impact, for example, if the goal was solely to reduce seed production (Norambuena & Piper 2000).

Meta-analysis provides an excellent tool to combine data from multiple studies to identify broad patterns (Gurevitch & Hedges 1993; Harrison 2011). There have been two previous meta-analyses on the effectiveness of biocontrol agents on invasive plants. Stiling & Cornelissen (2005) considered the impact of weed biocontrol agents as part of a broader analysis of biocontrol. They reported significant declines in plant performance but only impacts on individual plants were considered. However, depending on the target plant biology and ecology, the impact on individual plants may not relate to population scale impacts and therefore the results of such studies may have limited value. Thomas & Reid (2007) addressed this problem in their analysis of biocontrol agents in Australia by including measures of target plant density and indirect non-target plant response as a result of target removal. Whether or not the non-target species were native or not was not clear. They found significant reductions in plant performance and target density but no firm conclusions could be drawn for the impact on abundance of non-target plants due to a sample size of two studies, highlighting the scarcity of available and detailed post-release monitoring data.

The aim of the present study was to quantitatively assess the effectiveness of classical biocontrol of invasive plants, not only at the level of individual plants, but also at the level of the plant population and, ultimately, to assess changes in the associated plant community using post-release data. In the present study, only results published after 2000 were included to provide a reflection of modern biocontrol practise and to assess the effectiveness of classical biocontrol over relatively short time scales.

Materials and methods

Literature search

Data were primarily gathered through a literature search using both ISI Web of Knowledge ( and CAB Direct ( The search was restricted to publications from the years 2000 to 2011 and to the key words ‘weed biological control’ and ‘invasive plant biological control’. The search included sub-searches for ‘community impact’ and ‘population impact’. Also, a systematic search through the journals Biological Control (vol. 17–56) and BioControl (vol. 45–56) and through the Proceedings of the XI and XII International Symposia on the Biological Control of Weeds was carried out. Any appropriate studies found in the references of those papers found as above were also included.

Studies were included if they met the following criteria. The biocontrol agent must be non-native to the region where it has been intentionally introduced, and the study must be carried out after the release of the biocontrol agent. Pre-release and native range studies that only reflect potential impacts were not included. There had to be an appropriate design with control and treatment groups. Studies with biocontrol agent absence/presence comparisons from control and release sites and pre- and post-release comparisons from the same sites were pooled together only if the nature of response variables measured (magnitude and direction) was not significantly different from each other using heterogeneity analysis. If significant differences were found then only control and release site comparisons were included in the analysis.

Data extraction

Of the total 151 studies identified, 61 studies were included in the final analyses that covered the period 1992–2008 in which the studies were carried out. These produced 173 independent comparisons that represent 49 different biocontrol agents used against 28 target invasive plants from across the world (Appendix S1, Supporting Information). Studies were most often rejected due to being carried out in the native range, with unintentionally released biocontrol agents or poor reporting of variance and sample sizes.

If multiple samples were taken over time or various agent densities tested, then only the most recent sample and highest density treatments were used in the analyses. If multiple measurements were taken from the same plants that could be included in the same analysis (e.g. above- and below-ground biomass), then they were combined to avoid pseudoreplication. Additionally, if multiple studies reported data on a target plant species with the same biocontrol agent, response variable and geographical region, they were combined to produce an average effect size for that species. Classical biocontrol programmes from any country were included, and data used were collected predominantly under field conditions (64% of studies: Appendix S1, Supporting Information). Field cage and laboratory studies were only included if they demonstrated the efficacy of biocontrol agents already released into the field. These studies were excluded if the effect sizes differed significantly from those under field conditions determined using heterogeneity analysis. The response variables related to the target plants were plant size (height or diameter), plant mass, flower production, seed production (number produced) and target density. Response variables for the associated plant community were the abundance and diversity of non-target plant species. If the data were available, changes in the abundance or diversity of only native non-target plant species were analysed separately.

The means from control and experimental treatments and their respective standard deviations (calculated from the standard error if necessary) and sample sizes were extracted from Tables and Figures in each study. Data were extracted from graphs using uthscsa Image Tool software (University of Texas, San Antonio, Texas, USA).

The extracted data were converted to a common effect size, Hedges’ d (Gurevitch & Hedges 1993) for each comparison, which is based on a standardized mean difference between treatments (eqn 1). Hedges’ d can be negative or positive indicating the general direction of the effect and when = 0, there was no difference between treatments.

display math(eqn 1)

where inline image is the mean of the experimental treatment (with control agents) and inline image the mean of the control treatment (without control agents). The SDPooled indicates the pooled standard deviation of both treatments, which was calculated as per eqn 2. The term J weights studies based on their sample size (eqn 3).

display math(eqn 2)

where SDE and SDC are the standard deviations for the experimental and control groups, respectively, and nE and nC are the sample sizes of those groups.

display math(eqn 3)

Various characteristics of the studies were recorded and used as additional explanatory factors in the analyses. These were the invasive (study) region, the native region (continental scale), growth form [defined using the United States Department of Agriculture (USDA) growth habit codes: forb, shrub, tree or vine, plus the additional ‘aquatic’] and longevity (annual, biennial or perennial) of the target weed, the feeding guild (miner or borer, sap-feeder, root-feeder, folivore, gall-former, seed-feeder, flower-feeder, pathogen or a mixture of strategies) and taxonomic group (Order and Family) of the biocontrol agent, and the study duration or period in years between the earliest and latest sampling dates (Appendix S1, Supporting Information).

The majority of studies were carried out on plant species invasive in North America (66%), followed by Oceania (18%). Most target plant species were native to Europe (56%) or Central and South America (28%). Arthropods represented the predominant control agents, used in 93% of studies and 10% using fungal pathogens. Three percent of studies used both arthropods and pathogens. Over half of the studies used coleopteran biocontrol agents (57%), with the Curculionidae (38%) and Chrysomelidae (17%) as the most common families. One-third of all studies used folivores (31%), followed by internal above-ground feeders (15%). There has been a strong emphasis to target perennial plants (82%) and most targeted plant species were forbs (54%).

Data analysis

The change in the annual number of publications available for analysis from 2000 to 2011 was analysed using linear regression. The influence of the biocontrol agent on the target plants and on the associated plant communities was analysed in separate meta-analyses.

The uniformity of effect size across the studies in the analysis of each response variable was assessed using the Q test of homogeneity (Gurevitch & Hedges 1993). A significant Qtotal statistic indicates more variation across the effect sizes than expected by sampling error alone, that is, a significant heterogeneity between the biocontrol agent and the control treatment. This variation can be explained by the magnitude of the effect sizes or the direction, identified by the respective number of studies that showed an increase or decrease in the response.

Analyses were carried out using random-effects models, as the results were from studies that represent a larger population of possible effect sizes of interest. Fixed-effects models are only appropriate for a set of studies with no heterogeneity in effect size and when the inference of results is limited to the observed studies only (Hedges & Vevea 1998). To study the variation between studies, the additional explanatory factors were included as fixed factors in a mixed-effects model. For this, the Qbetween statistic was calculated (Gurevitch & Hedges 1993). A sensitivity analysis was carried out to identify any extraneous studies if any significant heterogeneity was found. This approach reiterates the analysis n times, omitting each study in turn. Excluding studies through the sensitivity analysis did not change any of the observed results, which justified the inclusion of all studies.

Many studies are likely not to be published, especially if they demonstrate non-significant effects or represent failed biocontrol attempts, an issue known as the ‘file-drawer problem’ (Rosenthal 1979). To account for this publication bias, fail-safe numbers were calculated for each analysis. The fail-safe number indicates how many additional, non-significant unpublished studies would have to be included in the analysis to change the result from a significant one to a non-significant one. If the fail-safe number is larger than 5n + 10, where n is the number of comparisons, then the analysis is considered robust (Rosenberg 2005).

All analyses were carried out using R version 2.12.1 (R Development Core Team 2010) and the ‘metafor’ package (Viechtbauer 2010).


Over the past decade, there has been a significant, linear increase in the number of publications reporting quantitative assessments of the impact of biocontrol and which met our criteria for inclusion in the meta-analyses, (n = 11, R2 = 0·49, < 0·05; Fig. 1). There was significant heterogeneity in the effect sizes within each analysis (Table 1), indicating variation in the magnitude, and direction for some analyses, of the responses to biocontrol. Overall, the results revealed the impact of biocontrol on the target species: all measures of individual plant performance decreased significantly (Fig. 2). The observed reduction in plant size was 28 ± 4% (mean ± 1 SE), plant mass 37 ± 4%, flower production 35 ± 13% and seed production 42 ± 9%. There were also significant reductions in the target plant density with mean declines of 56 ± 7%, demonstrating impacts at the population level. All target level analyses were robust, as all the fail-safe numbers were high.

Figure 1.

The annual number of publications over the past eleven years that met the criteria for inclusion in the analysis. The line indicates a significant linear relationship (y = 2·82 + 0·53 x,< 0·05).

Figure 2.

Mean effect size (±95% Confidence Interval) for the use of biocontrol agents against invasive plants. The response variables are detailed on the left followed by the number of independent comparisons in each analysis (n) and fail-safe numbers on the right. The effect is considered significant if the 95% Confidence Interval does not overlap with zero. Analyses are considered robust if the fail-safe numbers are larger than 5n +10. Significant and robust results are indicated with an asterisk.

Table 1. Heterogeneity (Qtotal) for each variable of observed effect sizes from studies of the impact of biocontrol on invasive plants. Nincrease and Ndecrease refer to the number of studies with an effect size in a particular direction between the treatment group relative to the control. Native species data were a subset of non-target species
Variable Q total N Increase N decrease P
Plant size93·3131<0·001
Plant mass88·3236<0·001
Seed production84·8323<0·001
Flower production104418<0·001
Target density169·3229<0·001
Non-target abundance66·1123<0·001
Non-target diversity35·2100<0·001
Native species abundance54·942<0·001
Native species diversity5·341ns

There was an increase in non-target plant species diversity (88 ± 31%) as a result of the release of classical biocontrol agents (Fig. 3). Non-target abundance also increased markedly by 227 ± 125%, but the low fail-safe number indicates that this result should be interpreted with care. A subset of the non-target data including only plants native to the study region (using the studies where this information was available) was analysed. There was a significant increase in the diversity of native non-target species (342 ± 268%), but with a high variance and a very low fail-safe number. No significant effect on the abundance of native plant species was found (Fig. 3).

Figure 3.

Mean effect size (±95% Confidence Interval) for the change in non-target plant diversity and abundance following the introduction of biocontrol agents against invasive plants. The number of independent comparisons in each analysis (n) and fail-safe numbers are on the right. Fail-safe numbers are considered robust if they are larger than 5n +10. The effect is considered significant if the 95% Confidence Interval does not overlap with zero. For the analyses of native plant responses, a subset of all the non-target data was used. Details on the native status of the non-target plants were not available for all the studies. Significant and robust results are indicated with an asterisk.

Some of the observed heterogeneity within the analyses of the effect of biocontrol was explained by the additional explanatory factors (Table 2). Classical biocontrol caused significant declines in the size of invasive plants native to Oceania, Europe or South America (−94% n = 1, −34 ± 5% n = 13 and −24 ± 4% n = 15, respectively), but not of those native to Asia or Africa (Fig. 4). There was also a significant difference in plant size responses as a result of agent families (Fig. 5): agents from the Chrysomelidae, Curculionidae, Psyllidae and Tingidae caused significant declines (−37 ± 6%, −24 ± 4%, −94% and −40%, respectively. All < 0·05, except Curculionidae: = 0·054). There was a significant difference between agent families in their impacts on flower production. An agent from the Agromyzidae caused a significant decrease (98%) in flower number [Hedges’ d (95% CI) = −2·96 (−5·24, −0·69) n = 1]. No significant effects of agents belonging to any other family were observed.

Figure 4.

Mean effect size (±95% Confidence Interval) for the change in plant size after the introduction of biocontrol agents against invasive plants. The studies are grouped by the native region of the target plant. The number of independent comparisons in each analysis (n) is displayed on the right. The effect is considered significant if the 95% Confidence Interval does not overlap with zero. Significant results are indicated with an asterisk.

Figure 5.

Mean effect size (±95% Confidence Interval) for the change in plant size following the introduction of biocontrol agents grouped according to taxonomic Family. The number of independent comparisons in each analysis (n) is on the right. The effect is considered significant if the 95% Confidence Interval does not overlap with zero. Significant results are indicated with an asterisk.

Table 2. Heterogeneity (Qbetween) between different study characteristics considered to explain variation in each meta-analysis of the impact of biocontrol of invasive plants. Characteristics are invasive region (continental scale), native region (continental scale), feeding guild or attack mode and taxa of the biocontrol agent and the growth form [defined using the United States Department of Agriculture (USDA) growth habit definitions], longevity (annual, biennial or perennial) of the target plant and the study duration. Significant results are in boldface
VariableStudy characteristic
Invasive regionNative regionAgent guildAgent order
Q d.f. P Q d.f. P Q d.f. P Q d.f. P
Plant size1·633NS 12·79 4 <0·05 8·797NS6·176NS
Plant mass5·423NS4·014NS9·616NS7·589NS
Seed production2·443NS2·834NS8·728NS5·026NS
Flower production7·623NS0·363NS0·885NS1·013NS
Target density2·192NS0·933NS12·338NS7·056NS
Non-target abundance0·501NS1·562NS3·304NS0·832NS
Non-target diversity0·501NS0·501NS4·753NS0·501NS
 Agent familyTarget growth formTarget longevityStudy duration
Q d.f. P Q d.f. P Q d.f. P Q d.f. P
Plant size 27·1 12 <0·05 7·755NS0·031NS9·347NS
Plant mass6·6715NS1·415NS5·662NS12·196NS
Seed production8·989NS3·564NS0·092NS2·814NS
Flower production 24·30 6 <0·05 6·063NS2·172NS3·844NS
Target density8·029NS6·084NS0·801NS5·939NS
Non-target abundance5·763NS0·941NS0·411NS 48·28 7 <0·001
Non-target diversity4·722NS0·501NS0·131NS 11·31 3 <0·05

There were significant effects of study duration on non-target diversity and abundance. The longest studies (7 years) showed increases in non-target diversity, but not the shorter studies (Fig. 6). Only one study of intermediate length (5 years) showed a significant increase in non-target abundance [Hedges’ d (95% CI) = 11·64 (7·15,16·14) n = 1], but no such effects were observed in longer or shorter studies.

Figure 6.

Mean effect size (±95% Confidence Interval) for the change in non-target plant diversity following the introduction of biocontrol agents against invasive plants. The comparisons are grouped by the duration in years that the studies were carried out. The number of independent comparisons in each analysis (n) is on the right. The effect is considered significant if the 95% Confidence Interval does not overlap with zero. Significant results are indicated with an asterisk.


This analysis demonstrates the effectiveness of classical biocontrol in invasive plant management. There were significant reductions in all individual plant measures considered, which confirms the results of a previous study (Stiling & Cornelissen 2005). However, the present analysis had more stringent selection criteria and considered broader measures of impact from a larger number and variety of biocontrol programmes. Unlike the previous study, results collected from the native region were excluded here as they do not represent the effectiveness of biocontrol. This analysis also revealed significant declines in target abundance as a result of classical biocontrol, but using data from a larger range of study systems than those included in a previous analysis of biocontrol programmes in Australia (Thomas & Reid 2007).

In 94% of the comparisons that considered the impact on the target plant, the response was a decline in performance or abundance (Table 1), which indicates the generality of these results. The overall reduction of performance at the level of individual plants may help mitigate the impact of the target plant, either through reduced growth or seed production. Crawley (1989) stated in a descriptive review of biocontrol programmes that the Coleoptera were the most widely used and the most successful agents. In the present quantitative study, the Coleoptera were still the most prevalent agents, but there was no clear evidence that they are more effective than agents of other Orders. However, agents of two of the coleopteran families that were considered had a negative impact on plant size, and we conclude that Coleoptera are the most effective agents for reducing plant size. The evidence for impact of families in other Orders was not as consistent and robust. For example, in the Hemiptera, only significant reductions were found for the Psyllidae and Tingidae due to individual studies. Significant heterogeneity was also found in the effect of agent families on flower production. This was due to the result of a single study, where the leaf miner Ophiomyia camarae Spencer (Diptera: Agromyzidae) was released against Lantana camara L. in South Africa and the flower number of the target was reduced by 97·5% (Simelane & Phenye 2005). For all other response variables, only overall impacts of biocontrol were found (Fig. 1). Hence, the studies considered in this analysis provide broad and fairly consistent evidence for the effectiveness of classical biocontrol of invasive plants, but the most effective agents at reducing plant size were beetles in the Chrysomelidae and Curculionidae families, and we recommend these be prioritized in cases where potential agents of different taxa have been identified.

We calculated the fail-safe N value to assess the robustness of each main response variable (Rosenberg 2005). The analyses of the overall effects of biocontrol on individual plants and non-target plant diversity were robust, but the analyses of characteristics of the studies were less robust. The significant variation between the impacts of different agent families was a result of a small number of studies, studies with large effect sizes, or comparisons of groups with very unbalanced sample sizes and the results of these comparisons should be interpreted with care. For example, there were far more studies to control invasive plants with European or South American than Asian origins, and 85% of studies that reported non-target responses were carried out in North America. Similarly, the significant reductions in target plant size by some of the agent families only represent the results of single studies (Agromyzidae, Psyllidae and Tingidae). All the data in the present analyses were from published sources. It is likely that biocontrol programmes that are successful have a better chance of being published (Rosenthal 1979). However, data from more studies and especially where the target plant was not controlled could provide valuable insights into which factors are associated with effective biocontrol efforts. As such, sources that provide information on both successful and unsuccessful control efforts will continue to be particularly useful (Klein 2011).

A key impact of invasive plants is the loss of local plant diversity (Vilà et al. 2011). The reduction in size of plants of the target species following the release of control agents is likely to reduce the competitive strength of the target species and enable other plants to become established. On a longer time-scale, a reduction in seed production may reduce the spread or regeneration of the target species. Prior to this analysis, there has been no quantitative demonstration of the response of the non-target plant community to a broad range of biocontrol efforts. Our analysis revealed that the number of plant species increases following the release of biocontrol agents, but it is unclear whether this was due to an increase in native species, or secondary invasion by other non-native species. The increase in non-target plant abundance is also encouraging, although there were insufficient data for a robust comparison. Many of the studies included were designed before the publication of current post-release monitoring advice (Blossey & Skinner 2000; Carson et al. 2008) or may not have been running long enough to for such impacts to occur. The opportunity for large-scale ecological study in biocontrol systems has long been recognized (Crawley 1989) and much of the work is focused on post-release research. It is likely that in the coming years, more data, collected over longer periods, will become available regarding the impact of biocontrol which may allow separation of long- and short-term patterns. Especially long-term patterns could not be evaluated with great certainty in this study as a result of the paucity of long-term studies. However, slower ecosystem responses, such as re-establishment of native vegetation and invertebrates, would be the best measure of biocontrol success.

The number of studies that met our selection criteria with respect to reporting of data increased consistently in recent years (Fig. 1). This is probably a result of the increasingly common scientific evaluation of biocontrol programmes, demanded by funding agencies, and of the development of biocontrol as an ecological science through the higher involvement of academic researchers. However, despite recent improvements in the reporting of the effect of biocontrol on the diversity and abundance of non-target species, it was often unclear that species were occupying the free space created as a result of biocontrol. Only around one-third of studies detailed the recovery of native species and in the other studies, the identity of the non-target species was either not clear or a combination of native and invasive. If the information was available and only native species were considered, the positive non-target diversity response was still significant, but no longer robust due to the small sample size. The increase in abundance and diversity of non-native non-target species may indicate secondary invasions (Symstad 2004), i.e. that free space created by the removal of one invasive species made possible the establishment of another. This would severely undermine the overall success of a control programme, even one that was ‘successful’ at removing the target plant, as the original impacts may not be alleviated. Clearly, greater detail regarding the identity of non-target plants is needed in future studies. The control of Ageratina riparia (Regel) R. King and H. Robinson in New Zealand (Barton et al. 2007) is a good example, as this study demonstrated a 30% increase in the diversity of non-target vegetation. More importantly, data from paired control and release sites that were monitored over several years were presented, and the establishing native and non-native plant species were identified, revealing that secondary invasion did not happen. Nonetheless, classical biocontrol resulted in a reduction in the target plant and increases in diversity in invaded areas, which is likely to impact other trophic levels due to the provision of additional niches (Florens et al. 2010). These impacts occurred in the absence of additional control and restoration efforts, such as seeding. It is possible that the impact of biocontrol on non-target plant species can be enhanced by an integrated approach (Paynter 2003) or by active restoration by land managers (Seabloom et al. 2003). Reid et al. (2009) considered the impact of invasive plant management using a combination of control methods on ecosystem restoration in Australia through a literature review and land manager surveys. The authors identified similar problems with information provided in the scientific literature as the present study: only 20% of the studies they included considered non-target impacts and many reported only qualitative data. However, the land manager surveys provided more detail. In 52% of cases, there was an increase in both native and invasive plants and 33% of cases reported increases in native plants only. Although these results are still mostly based on qualitative assessments and confirm the lack of this information in the literature.

In conclusion, the results present here show that classical biocontrol of invasive plants significantly reduces the size and the reproductive output of individual plants and the abundance of target species. Thus, these results illustrate that classical biocontrol of plants is a valuable management tool. They also reveal the enhanced establishment of other, non-target plant species. Many of these impacts were recorded within several years of the initial release of the agents, which is encouraging, but long-term monitoring is essential to improve understanding of the effectiveness of classical biocontrol. The outcome of control efforts is not always the same due to the highly variable nature of plant invasions (Shea et al. 2010), and reliable predictors of successful programmes are still lacking. However, the results presented here provide indications for selection of the most promising control agents, as some coleopteran agents were more effective at reducing plant size than other taxa. We therefore recommend that Coleoptera be prioritized when selecting potential agents for future biocontrol projects. Studies presenting a lack of control or recovery are still relatively few (e.g. Butler & Wacker 2010). Yet, reporting the outcome of unsuccessful projects, which has long been advocated (Crawley 1989), will be invaluable to inform future control efforts, and we recommend that post-release monitoring not be restricted to the biocontrol agent and its target and that they include a range of non-target species (Denslow & D'Antonio 2005; Carson et al. 2008). Especially as some key functional groups, such as predators, pollinators and decomposers, may respond quicker to target control than non-target plant species (Fiedler, Landis & Arduser 2011).


We would like to thank Julia Koricheva for guidance on meta-analyses. Thanks are also due to the comments of two anonymous reviewers on an earlier draft of this manuscript. This work was funded through a postgraduate scholarship of the Biotechnology and Biological Sciences Research Council to G.D.C.