1. Introduction

Animal pollination is often held up as an ideal example of beneficial mutualism for all parties involved, but resource theft from plant partners through floral larceny seemingly threatens the balance of these interactions (Irwin et al., 2001, 2010; Burkle et al., 2007). Nectar robbers, a common type of floral larcenist, access nectar by puncturing flowers and bypassing the usual pathways used by pollinators (Fig. 1a–d; Inouye, 1980; Irwin et al., 2010). Similarly, nectar thieves, who exploit floral structures in ways that mismatch with typical pollinator behavior, represent another form of floral larcenist (Fig. 1e and f; Inouye, 1980). Floral larcenists, who are widespread in plant-pollinator mutualisms, not only complicate plant-pollinator interactions but also entail implications for the evolutionary strategies and ecological adaptations of plants (Irwin et al., 2010).

Although it has long been expected that floral larcenists universally harm plants (Darwin, 1877), numerous studies have shown that their impacts are highly variable, ranging from negative to neutral, and even rarely positive (Irwin et al., 2010). Floral larcenists consume nectar and may destroy reproductive organs, directly causing energy loss and physical harm to the plant (Galen, 1983; Navarro, 1999). This reduction in nectar availability, along with floral damage, can decrease the attractiveness of flowers, reducing subsequent visitation rates by pollinators (Irwin et al., 2015). Pollinators may therefore avoid these unrewarding flowers, leading to reduced pollen donation and stigmatic pollen deposition, which, in turn, diminishes both male and female reproductive fitness (Irwin and Brody, 1999). However, decreased in nectar levels may shorten individual visit durations, reduce the number of flowers visited per plant, and increase pollinator flight distances between plants, potentially enhancing pollen flow and reducing geitonogamy (Irwin, 2003). Therefore, the impact of floral larceny on plants is not straightforward; it may have both harmful and beneficial effects.

The intensity and outcomes of floral larceny are highly dependent on various contextual factors, including environmental conditions, plant characteristics, and the attributes of floral visitors (Irwin et al., 2010). Key determinants, such as climatic region (tropical vs. non-tropical), growth form (perennial vs. annual), mating system (autonomous selfing or self-(in)compatible), and pollen limitation, may potentially influence plant responses to floral larceny (Irwin et al., 2001; Burkle et al., 2007). Additionally, floral visitors, including the identity and behavior of primary robbers, thieves, secondary robbers and pollinators, have been shown to play a critical role in the plant-visitor dynamics during floral larceny events (Irwin et al., 2010). For instance, robbers and thieves, due to their different nectar access behaviors, can impact plant-pollinator interactions, while the presence of secondary robbers may result in chronic nectar removal from flowers (Richman et al., 2018). Furthermore, experimental design, for example, focusing on individual flowers or the entire plant, is also known to affect observed outcomes (Irwin and Brody, 1999; Irwin et al., 2001). Lastly, given that closely related plant species may exhibit similar responses to the stress of floral larceny, the phylogeny of tested species can significantly influence the consequences (Chamberlain et al., 2012).

Here, we performed a phylogenetically controlled hierarchical meta-analysis to assess the effects of floral larceny on plant reproductive success and examine how these effects are influenced by key moderators. Our analysis included 153 publications on floral larceny, covering 120 plant species, representing a 460% increase in the number of papers compared to previous meta-analyses (Fig. S1). In addition to assessing plant female fitness, we have expanded the evaluation to include flower traits (e.g., nectar volume), pollination, and plant male fitness as three new categories of response variables, enabling a more comprehensive assessment. In addition, we have expanded factors that moderate the impact of floral larceny, including abiotic factors such as climatic region and experimental design, as well as biotic variables such as larcenist type and the presence of secondary robbers. Specifically, we hypothesize that (1) Floral larceny negatively impacts flower traits, which, in turn, lowers pollinator visitation rate and pollen donation, ultimately reducing plant female fitness; (2) In addition to decreasing visitation rates, floral larceny alters pollinator foraging behaviors, potentially benefiting both male and female fitness by reducing geitonogamy; (3) The effects of floral larceny are influenced by various moderators. For example, self-incompatible and pollen-limited plants are particularly vulnerable to floral larceny (Burkle et al., 2007). By testing these hypotheses, we aim to enhance the understanding of the ecological dynamics and long-term persistence of nectar larceny.

2. Materials and methods 2.1. Literature search and inclusion criteria

We conducted a comprehensive literature review (15 May, 2024) using multiple databases, including the ISI Web of Science (all database), Scopus, Google Scholar and China National Knowledge Integrated database (CNKI, http://www.cnki.net). For the ISI Web of Science and Scopus, we entered the following keyword combinations into the "Topic" and in "All fields" search areas respectively: nectar robb^* OR nectar thie^* OR nectar larceny. We searched "TS = (nectar robb^* OR nectar thie^* OR nectar larceny)" in Google Scholar, and "nectar robbing" (in Chinese) in the "Topic" field for CNKI, respectively. This search strategy yielded 716, 698, 583 and 62 publications on the Web of Science, Scopus, Google Scholar and CNKI, respectively. In addition, we surveyed two prior meta-analyses (Irwin et al., 2001; Burkle et al., 2007) to identify studies that were not included in the search results.

The article screening procedure followed PRISMA guidelines (Fig. S2). After removing duplicates, we applied the following criteria to further exclude studies or cases that might have any confounding effects: (1) Studies that only reported the occurrence of floral larceny without providing any statistical comparisons of its impact on controls or that lacked information on sample size, mean or variance; (2) Studies that focused on the impact of floral larceny on nectar microbes rather than plant reproductive success. A total of 153 publications met our criteria (Note S1).

2.2. Data extraction and effect size calculation

For each study, when comparing robbed plants with their control counterparts, we examined four categories of response variables: (1) Flower traits, including nectar volume, sugar concentration and flower lifespan; (2) Pollination metrics, including visitation rate, number of flowers per bout (number of flowers visited on each inflorescence in one bout), flower handling time and flight distance; (3) Male fitness indicators, including pollen donation, pollen dispersal and pollen quality. (4) Female fitness indicators, including fruit set, seed set, seed quality and pollen deposition. Detailed definitions and criteria for collecting these response variables are listed in Table S1.

We extracted sample sizes (N), means and standard deviations (SD) from texts or tables. When only standard errors (SE) or 95% confidence intervals (CI) were present, we calculated SD following these formulas:

(1)

(2)

where CI_u and CI_l are the upper and lower limits of the 95% CI, respectively. If these parameter values were reported in graphs, we digitized the data using GetData Graph Digitizer 2.26 (http://getdata-graph-digitizer.com/), and unspecified error bars were considered to be SE. When SD were indicated through box plots with minimum, maximum, first quartile, median and third quartile, we estimated SD using an online calculator based on formulas by Wan et al. (2014) (https://www.math.hkbu.edu.hk/∼tongt/papers/median2mean.html). For cases where both SD and its conversion channels were unavailable, we used the "Bracken1992" approach to impute SD based on all complete cases using the impute_SD function in R package "metagear" (Lajeunesse, 2016). We only extracted the highest nectar robbing level from manipulative experiments testing more than two nectar robbing intensities.

We used the natural log response ratios (ln RR) as measures of effect size (Hedges et al., 1999) to assess the impact of floral larceny on plant performance:

(3)

where X_t and X_c are the numerical mean values of plant performance in the floral larceny treatment and the control, respectively. A negative lnRR value signifies adverse effects of floral larceny on plant performance, whereas a positive value indicates the opposite. Each lnRR was weighted by the inverse of the sampling variance (v), which was calculated as:

(4)

where SD_t and SD_c represent the SD of X_t and X_c, respectively, with N_t and N_c as sample sizes. Individual effect size was calculated using the escalc function in R packages "metafor" (Viechtbauer, 2010) in R (4.2.2) (R Core Team, 2022).

2.3. Moderator variables

To identify potential factors that influence the impact of floral larceny on plant reproductive success, we selected 14 moderators encompassing plant traits, traits of floral visitors and experiment level (Table S2). We also defined a set of criteria for data extraction for these moderators.

2.3.1. Plant traits that moderate the effect of floral larceny

The response of plants to floral larceny is significantly influenced by their climatic region and inherent characteristics. For instance, tropical plants, which thrive in humid and warm conditions, tend to produce a large number of flowers, potentially mitigating the impact of nectar robbing (Deng et al., 2005). Here, we classified each study as tropical or non-tropical according to its latitude (tropical < 23°26′13.4″, non-tropical > 23°26′13.4″). For studies covering multiple populations, we used the centroid of each region to simplify and avoid overplotting. Additionally, perennial plants have a longer lifespan and store more energy compared to annual herbs (Vico et al., 2016), potentially reducing the impact of energy loss caused by floral larceny. Therefore, we categorized growth forms into three groups according to longevity and growth habit: annual herb, perennial herb and wood. Moreover, exotic plants, as previous studies suggest, generally do not benefit from nectar robbing for they are more susceptible to enemies in novel habitats (González-Browne et al., 2016). We, therefore, classified plants as native or non-native. If the origin of a plant was not available in the publication, we supplemented this information by searching its name in Plant of the World Online (https://powo.science.kew.org/).

The mating system of plants is also crucial in determining the impact of robbing on female reproductive success. For instance, autonomous selfing or self-compatible plants may be less affected by changes in pollinator behavior compared to self-incompatible species that rely heavily on pollinators (Burkle et al., 2007; Zhang et al., 2009). To evaluate plant mating systems, we used two independent categories: self-compatible vs. self-incompatible, and autogamous vs. non-autogamous. Data on mating systems for each species were extracted from the original study or from relevant publications investigating their reproductive ecology.

The degree of pollen limitation is another important predictor, with pollen-limited plants being particularly vulnerable to the effects of floral larceny (Burkle et al., 2007). We extracted information on pollen limitation directly from studies. If this information was absent, we examined the cited references of the study or conducted searches on Google Scholar for such details. Importantly, we only considered pollination experiments conducted concurrently and at the same location as the original study, given the high variation in pollen limitation across different years and sites (Knight et al., 2005).

Moreover, we gathered data on the nectar standing crop as a continuous moderator to examine whether higher nectar production enhances plant tolerance to floral larceny (Irwin et al., 2008). For studies reporting multiple measures or ranges of nectar production, we calculated the average values.

2.3.2. Flower visitor characteristics that moderate the effect of floral larceny

To comprehensively characterize floral visitors, we reviewed each study and collected the following factors: (1) larceny type, categorized as either robber or thief; (2) presence of secondary robbers, classified as yes or no; (3) larcenist type and (4) pollinator type, both of which were grouped into bird, insect or mixed (i.e., including multiple taxonomic classes). Notably, studies have indicated that hummingbird pollinators are more likely to avoid robbed flowers compared to bees (Bergamo and Sazima, 2018), suggesting that plants pollinated by birds are more vulnerable to nectar robbing (Irwin et al., 2001).

Although direct effects, such as damage to reproductive organs are significant, a growing body of literature highlights the undeniable importance of indirect effects mediated through changes in pollination (Irwin et al., 2015; Maidana-Tuco et al., 2024). To further understand these dynamics, we reviewed the texts of each study to determine whether primary larcenists caused damage to reproductive organs such as pistils and anthers. Although Burkle et al. (2007) in a previous meta-analysis excluded cases of direct damage to focus on the indirect effects of robbing on female reproduction through changes in pollination, we retained these studies to expand our dataset and more importantly, to include analyses of the direct negative effects of floral larceny.

We also collected data on the floral larceny rate within the population as a continuous moderator to examine whether the prevalence of floral larceny correlates with the plant's susceptibility to larcenists. When studies reported multiple instances or ranges of larceny rates, we calculated the average values.

2.3.3. Experimental designs that influenced reported effects of floral larceny

Floral larceny might not have a significant overall effect on the entire plant if pollinators do not distinguish between robbed and unrobbed flowers, and if unrobbed flowers compensate for the energy loss caused by the robbed ones (Irwin and Brody, 1999). However, individual robbed flowers may still experience reduced success due to direct damage or decreased pollinator visitation. To better understand these dynamics, we assessed how each study examined the effect of floral larceny and classified these experimental designs as either whole-plant (comparisons of plants with high versus low robbing rate) or individual-flower level (comparisons of robbed versus un-robbed flowers). However, we only examined how experimental design affected sub-response variables of female and male fitness indicators, since measuring nectar volume at the whole-plant level or assessing pollinator foraging behavior at the individual-flower level is considered unreliable.

2.4. Data analysis

We conducted our meta-analysis in two phases. First, to evaluate the overall impacts of floral larceny on plant reproductive success, we used a random-effect model with a restricted maximum likelihood (REML) approach to calculate the weighted mean of individual effect sizes via the rma.mv function in R package "metafor" (Viechtbauer, 2010). Cumulative effect size was considered statistically significant if its 95% CI did not overlap with zero. Given that some publications provided more than one effected size, we ran a hierarchical meta-analysis to deal with this non-independency (Tuck et al., 2014). To carry out the process, the id of the effect size nested in the reference were treated as random factors in the mixed-effect model. In addition, we incorporated a variance-covariance matrix based on phylogenetic relationships (Fig. 2c) to control for the potential non-independence among tested plant species. To create this matrix, we constructed a phylogenetic tree for the 120 plant species using the phylo.maker function in R package "U.PhyloMaker" (Jin and Qian, 2023). The phylogenetic tree was then transformed into a variance-covariance matrix using the vcv function in the "ape" package (Paradis et al., 2004) and was included as a random factor in the mixed-effect model. Second, to ascertain which factors robustly modulate the effects of floral larceny on plant reproductive success, we incorporated moderators of plant traits, floral visitor characteristics and experimental design, respectively. To evaluate the level of heterogeneity of effect sizes, we calculated the P-value of the Q_t statistics. When they were statistically significant (p < 0.05), the influence of moderators on the effects of floral larceny was examined using Q_m (Koricheva et al., 2013). If a moderator had more than two levels, we analyzed the differences between levels using the glht function in R package "multcomp" (Hothorn et al., 2008). Note that moderators were not included for flower lifespan, flight distance, or pollen quality due to limited sample sizes. Similarly, moderator levels with a sample size < 2 were excluded from our analysis. For continuous moderators, we conducted linear meta-regression models to estimate whether a change in larceny rate or nectar production could predict a change in response variables of plant fitness.

After completing the meta-analysis pipeline, for easier interpretation, we back-transformed the cumulative effect size () of each response variable into percentage change (%) (Tang et al., 2023):

(5)

The lower and upper limits of 95% CI were also transformed using the formula above.

Additionally, we proceeded with further data analyses to present an overview of our results. We categorized the 153 examined publications by the taxonomy of their primary floral larcenist (e.g., bumblebee, hummingbird, mammal) and calculated the proportion of studies in each taxonomic category. Similarly, we summarized the proportion of each plant taxonomic order (e.g., Lamiales, Ericales) by dividing the number of plant species in each order by the total of 120 plant species.

2.5. Publication bias

To evaluate publication bias against negative results, we used funnel plots and statistically inspected asymmetry using Egger's regression tests (Egger et al., 1997). Significantly asymmetric results were then adjusted using the Trim and Fill method (Duval and Tweedie, 2000). Moreover, we computed Rosenberg's fail-safe number to gauge the potential impact of unpublished studies with non-significant results on our conclusions (Rosenberg, 2005). A fail-safe number exceeding 5n + 10 (where n is the number of cases used in the meta-analysis) indicated robustness (Rosenthal, 1986).

3. Results 3.1. Study characteristics

We identified a total of 153 publications (n_p) and 776 observations (n_o) examining the effects of floral larceny on plant reproductive success. Studies on insect floral larcenists comprised the majority at 79.7% (n_p = 122, n_o = 628), while those on birds accounted for 13.1% (n_p = 20, n_o = 105) and mixed larcenists for 8.5% (n_p = 13, n_o = 43). The distribution of larcenist types was closely tied to their global geographic locations; for example, 72.1% studies on insect larcenists were distributed in non-tropical regions (n_p = 88), while 46.7% studies involving bird larcenists were found in South America (n_p = 14; Fig. 2a). For studies that examined larcenist taxonomic category, bumblebees (37.3%, n_p = 57) and carpenter bees (15.7%, n_p = 24) comprised the two most common insect larcenists, while hummingbirds (8.5%, n_p = 13) were the most investigated among vertebrate larcenists (Fig. 2b). Moreover, we found that the incidence of floral larceny was particularly high in certain plant taxonomical orders, with Lamiales and Ericales being the most frequently targeted among the 120 studied plant species (Fig. 2b and c). Studies recorded that floral larceny occurred predominantly in corollas (69.2%), with 13.3% and 17.5% of plant species suffering spur robbing and nectar thieving, respectively (Fig. 2c).

3.2. Effects of floral larceny on plant reproductive success

Floral larceny generally exerted negative impacts on flower traits (−51.0%, 95% CI: −62.5% to −36.2%, p < 0.0001; Fig. 3e and Table S3). Regarding flower traits, nectar volume and flower lifespan decreased by 57.9% and 27.4%, respectively, while the sugar concentration of nectar remained unaffected (+8.6%, p = 0.39; Fig. 3a and Table S3). Consistent with these negative impacts on flower traits, both pollinator visitation rate (−26.0%, p = 0.002) and flower handling time (−29.1%, p = 0.01) were adversely affected (Fig. 3b and Table S3). Floral larceny led to a decrease in the number of flowers per bout, although the impact on the latter was only marginally significant (−19.5%, p = 0.061; Fig. 3b and Table S3). However, floral larceny significantly increased pollinator flight distance (+100.7%, p = 0.001; Fig. 3b and Table S3). For male fitness, floral larceny had a neutral overall impact (−16.4%, 95% CI: −38.1% to +12.7%, p = 0.23; Fig. 3e and Table S3) and did not significantly influence any of its sub-response variables (Fig. 3c and Table S3). However, female fitness was significantly reduced by floral larceny (−17.9%, 95% CI: −28.1% to −5.8%, p = 0.004; Fig. 3e and Table S3). Fruit set and pollen deposition was decreased by 22.4% (p = 0.0005) and 25.6% (p = 0.02), respectively. Unexpectedly, there were no significant effects observed on seed set (−11.3%, p = 0.18) or seed quality (+3.0%, p = 0.56; Fig. 3d and Table S3).

3.3. Factors regulating the impacts of floral larceny on plant reproductive success

The heterogeneity in effect sizes was statistically significant for both nectar volume (Q_t = 4130.20, df = 166, p < 0.0001) and sugar concentration (Q_t = 2677.77, df = 48, p < 0.0001; Table S4). Overall, floral larceny negatively affected nectar volume, except for plants suffering floral damage, plants with mixed larceny type and non-native plants (Figs. 4a and S3a; Table S6a). Floral larceny did not appear to affect sugar concentration, which only showed a significant increase in the absence of secondary robbers (Fig. 4b and S3b; Table S6a). Conversely, we found that the effect of floral larceny on sugar concentration is moderated by plant growth form, with woody plants responding more positively (Fig. S3b and Table S5a). Meanwhile, sugar concentration decreased as larceny rate increased (Fig. S4a and Table S5a).

No factors were identified that could explain why the impact of floral larceny on visitation rate varied (Fig. 5a and Table S5b), although the heterogeneity in effect sizes observed was statistically significant (Q_t = 750.96, df = 77, p < 0.0001; Table S4). Yet, the effects of floral larceny on visitation rate were found to be specific to different floral visitors, with insect larcenists and the presence of secondary robbers showing significantly negative impacts (Fig. 5a and Table S6b). The impact of floral larceny on the number of flowers per bout was also moderated by several characteristics of floral visitors, including larceny type, larcenist type, and pollinator type (Fig. 5b and Table S5b). The number of flowers visited per bout was reduced significantly more by bird pollinators than by insect pollinators; and this reduction was also observed in cases of floral larceny by nectar thieves or bird larcenists (Fig. 5b and Table S6b). The response of flower handling time also varied with different types of floral visitors (Table S6b). Floral larceny had a significantly negative effect in cases involving insect pollinators, insect larcenists or nectar robbers, or in the absence of secondary robbers (Fig. 5c and Table S6b). Moreover, the impact of floral larceny on flower handling time diminished as the larceny rate increased (Fig. S4b and Table S6b).

The effects of floral larceny on male fitness were strongly influenced by pollinator type (Q_m = 22.55, p < 0.0001) and pollen limitation (Q_m = 3.88, p = 0.0488; Table S5c). The effect of floral larceny on pollen donation was significantly negative in bird-pollinated plants and similarly pronounced in pollen-limited plants (Fig. 6a and Table S6c). Moreover, the adverse impact of floral larceny on pollen donation dropped as larceny rates increased (Fig. S4c). Pollen limitation emerged as the key predictor for the response of pollen dispersal to floral larceny (Fig. 6b and Table S5c). Interestingly, floral larceny extended pollen dispersal distance in plants without pollen limitation, but this effect was insignificant in pollen-limited plants (Fig. 6b and Table S6c).

Among female fitness indicators, climatic region explained variations in effect sizes of pollen deposition, with significantly more detrimental effects in tropical regions (Fig. 7a and Table S6d). Pollen deposition was also significantly reduced in cases involving woody plants, pollen-limited plants or plants suffering from nectar robbing (Figs. 7a and S3g). The impact of floral larceny on fruit set was largely controlled by self-compatibility (Q_m = 10.29, p = 0.0013) and pollen limitation (Q_m = 9.80, p = 0.0017; Table S5d). Self-incompatible and pollen-limited plants were more vulnerable to the adverse effects of floral larceny (Fig. 7b and Table S6d). Similar to self-incompatible plants, floral larceny notably decreased fruit set in non-autogamous plants (Fig. 7b and Table S6d). Plants in tropical regions and those affected by direct damage or mixed larcenists experienced greater reductions in fruit set when subjected to floral larceny (Fig. 7b and Table S6d). Significant negative effects of floral larceny on fruit set were also observed in cases involving plants suffering from nectar robbing, the presence of secondary robbers, plants pollinated by birds, or when the effect sizes were measured at the whole-plant level (Fig. 7b and Table S6d).

Seed set was generally less affected by floral larceny, except in plants that were self-incompatible, pollen-limited or directly damaged by larcenists (Fig. 7c and Table S6d). An additional factor that modified the impact of floral larceny was whether researchers examined whole plants or individual flowers (Q_m = 3.85, p = 0.0496; Table S5d), with effect sizes at the whole-plant level being significantly negative (Fig. 7c and Table S6d). Similarly, seed quality presented overall neutral responses to floral larceny (Figs. 7d and S3j), although the heterogeneity in effect sizes was statistically significant (Q_t = 255.65, df = 42, p < 0.0001; Table S4). Specifically, pollen limitation explained variation in effect sizes, with pollen-limited plants exhibiting marginally higher seed quality (p = 0.09; Fig. 7d and Table S6d).

3.4. Publication bias

There was no evidence of publication bias for most of the 14 response variables (Fig. S5 and Table S7). Although the funnel plots of nectar volume, sugar concentration, pollen donation, fruit set and seed set were asymmetric (Egger's test: p < 0.05; Fig. S5 and Table S7), the Rosenberg fail-safe numbers greatly exceeded 5n + 10 (Table S7). Note that the funnel plot of pollen quality was asymmetric, with a fail-safe number (N = 23) below the required threshold (N = 40). We considered that this asymmetry was likely due to the small sample size of pollen quality (n_p = 3, n_o = 6; Fig. 3c). These results suggest that publication bias can be safely ignored.

4. Discussion

Our results revealed that floral larceny had an overall negative impact on flower traits and pollinator visitation rate, whereas its effect on male fitness was neutral. Notably, although floral larceny negatively affected female fitness, its impact on seed set and seed quality was neutral. Our analyses also showed that pollinators reduce handling time and travel farther after visiting robbed plants, indicating potential benefits in decreasing geitonogamy. Furthermore, we identified several key moderators of the impact of floral larceny, such as pollinator type, pollen limitation and plant mating system, that can explain effect size heterogeneity.

4.1. Effects of floral larceny on flower traits

In line with prevailing views, our analyses confirmed significant negative effects of floral larceny on both nectar volume and flower lifespan (Fig. 3a and Table S3). Numerous studies have demonstrated that floral larceny decreases standing nectar crop (Lara and Ornelas, 2002; Dedej and Delaplane, 2005; Newman and Thomson, 2005). Floral larcenists often cause visible damage to flower structures while accessing concealed nectar, which can sometimes lead to early flower abortion (Carbonari et al., 2009). Consistent with most previous studies (Navarro, 1999; Fumero-Caban and Melendez-Ackerman, 2013), our results also showed that floral larceny did not significantly affect nectar concentration (Fig. 3a and Table S3). However, some earlier studies have reported increased nectar concentration after floral larceny, possibly due to evaporation through the openings created by larcenists (Pleasants, 1983). Additionally, most moderators of floral larceny we examined did not significantly explain variations in nectar volume or sugar concentration (Fig. 4 and Table S5a). This suggests that floral larceny primarily reduces nectar volume but has no effect on sugar concentration, and that these effects are largely independent.

Nectar serves as an investment by plants to attract pollinators, facilitating their reproduction (Navarro, 2001). Consequently, plants must develop effective strategies to cope with floral larceny. Some plants continue to secrete nectar in flowers that have been robbed, which may account for instances where nectar volume remains unchanged (Fumero-Caban and Melendez-Ackerman, 2013; Maidana-Tuco et al., 2024). However, most plants tend to reduce resource allocation or even stop nectar production in robbed flowers (Navarro, 2000), and in some cases, they may abandon these flowers entirely, reducing robbed flower longevity (Carbonari et al., 2009). If plants persist in supporting robbed flowers without lowering resource supplement, the energetic cost can be considerable, potentially leading to reduced seed production due to resource reallocation (Pyke, 1991; Navarro, 2001).

4.2. Effects of floral larceny on pollinator visitation rate and foraging behavior

Pollinators may not meet their daily energetic requirements after visiting poor nectar resources, leading them to avoid robbed flowers with reduced nectar volume (Gass and Montgomerie, 1981). Our analyses confirmed that floral larceny significantly decreased pollinator visitation rate (Fig. 3b and Table S3). Floral larceny also reduced flowers visited per bout and flower handling time, although the effect on number of flowers visited per bout was only marginally significant (Fig. 3b and Table S3). These results support the hypothesis that pollinators spend less time probing robbed flowers and visit fewer flowers on robbed plants (Irwin, 2003; Irwin et al., 2015). Notably, we found that pollinator taxonomic groups differed in their sensitivity to floral larceny, with bird pollinators being more affected than insect pollinators in terms of both visitation rate and number of flowers visited per bout (Fig. 5a and b and Table S6b). This could be because birds might discriminate between robbed and un-robbed flowers better than insects (Bergamo and Sazima, 2018). Additionally, some studies observed that some aggressive nectar robbers, such as stingless bees (Trigona sp.), chase away bird pollinators, which may also contribute (Roubik, 1982).

Notably, our results support the widely accepted hypothesis that floral larceny may reduce geitonogamy by modifying pollinator behaviors. Numerous studies have suggested that floral larceny can induce pollinators to spend less time on individual flowers, visit fewer flowers per plant, and travel longer distances, thereby reducing self-pollen transfer, and promoting outcrossing within populations (Maloof and Inouye, 2000; Irwin, 2003; Irwin et al., 2015). Taken together, our results confirm that floral larceny alters pollinator foraging behavior, which potentially reduces geitonogamy.

4.3. Effects of floral larceny on male fitness

In our meta-analysis, the overall impact of floral larceny on male fitness was neutral, and neither pollen donation, pollen dispersal nor pollen quality were significantly affected (Fig. 3c and Table S3). Pollinator type proved to be an important moderator for pollen donation, with floral larceny having a strong negative effect on bird pollinators (Fig. 6a and Table S6b). This suggests that bird pollinators are more adversely affected by floral larceny (Bergamo and Sazima, 2018), consistent with our observation and analysis on pollinator visitation rate. Similarly, pollen-limited plants experienced a greater reduction in pollen donation (Fig. 6a and Table S6b), likely due to more pronounced scarcity of pollinators following floral larceny (Irwin, 2003). In addition, we did not find significant detrimental effects on pollen quality (Fig. 3c and Table S3), which is in line with observations from a previous study (Irwin et al., 2015).

Theoretically, floral larceny affects two different aspects of male fitness. First, reduced nectar volume can lead to fewer pollinator visits and less pollen donation, thereby negatively impacting male reproduction (Irwin, 2003). Second, decreased nectar availability can lead pollinators to travel greater distances between flowers, improving pollen transfer quality and benefiting male fitness while simultaneously reducing geitonogamy (Irwin, 2003). Our results confirmed that floral larceny reduces nectar volume and pollinator visitation rate, but no significant reduction in pollen donation was observed. Our analysis supports the second theory, as floral larceny appears to reduce geitonogamy by altering pollinator foraging behavior. In addition, the cumulative pollen dispersal estimates in our study were positive, particularly for plants that were not pollen-limited (Fig. 6b and Table S6c). Overall, our results suggest that floral larceny does not negatively affect male fitness and may even reduce geitonogamy, consistent with previous studies (Maloof, 2001).

4.4. Effects of floral larceny on female fitness

There was a significant negative impact of floral larceny on female reproductive success, particularly in terms of pollen deposition and fruit set (Fig. 3d and Table S3). Robbed plants often receive less pollen due to reduced pollinator visits, with a mechanism resembling that of reduced pollen removal (Irwin, 2003; Irwin et al., 2015). This decrease in pollen deposition can further result in reduced fruit set, since plant reproduction is positively correlated with the amount of pollen receipt (Cayenne Engel and Irwin, 2003). Previous meta-analyses have examined factors that may moderate the impact of floral larceny on female fitness, including larcenist type, mating system, and pollen limitation (Irwin et al., 2001; Burkle et al., 2007). In addition to these moderators, our results suggested that the impact of floral larceny on female fitness can also be influenced by floral damage and climatic region (Fig. 7a, b and Table S5d). Specifically, floral larceny-induced damage to reproductive organs, such as pistils and anthers, hampers plant reproduction directly (Galen, 1983). For tropical plants, the negative effect of floral larceny on fruit set can be partially attributed to the highly sensitivity of bird pollinators to floral larceny, which have a wide distribution in tropical regions (Cronk and Ojeda, 2008).

In contrast, floral larceny has a neutral impact on seed set and seed quality (Fig. 3d and Table S3). The presence of floral damage was an important moderator for the effect of floral larceny on seed set, while pollen limitation explained the heterogeneity in effect sizes of seed quality (Fig. 7c and d and Table S5d). Notably, floral larceny had a more severe effect on seed set when studies were conducted at the whole-plant level (Fig. 7c). We deem that if a plant experiences extensive flower robbing (e.g., 80% of flowers robbed; Irwin and Brody, 1999), it may suffer resource limitation caused by energy loss. It may even become significantly less attractive to pollinators, thereby exacerbating the negative impact compared to the effects observed in within-plant comparisons of robbed and unrobbed flowers. Furthermore, floral larceny may influence pollinator behavior in ways that increase outcrossing rates (or reduce geitonogamy), which often enhances seed quality (Darwin, 1877; Maloof and Inouye, 2000). This could help explain why seed quality was not significantly affected by floral larceny.

Our study shows that fruit set is significantly reduced by floral larceny whereas seed set is unaffected, indicating that seed set tends to be less sensitive to the negative effects of floral larceny than fruit set. There are several possible explanations for this discrepancy. First, during the fertilization process and fruit development, flowers typically require adequate pollen to mature into fruits (Burd, 1994). In this context, reduced pollinator visitation caused by floral larceny may decrease the likelihood of flower maturation, as pollinator visitation rate is positively related to pollen receipt and plant reproduction (Cayenne Engel and Irwin, 2003). However, while floral larceny reduces pollen deposition and fruit set, once a flower with intact ovaries has received sufficient pollen for fertilization, further floral larceny does not hinder seed development. This points to a level of resilience in plants' reproductive systems, where floral larceny does not critically affect the ability to produce viable seeds. Second, plants may compensate for the effect of floral larceny by enhancing seed production in pollinated flowers or producing more flowers, thus offsetting losses in damaged or un-pollinated flowers (Irwin et al., 2008; Irwin, 2009). This strategy can be viewed as a tolerance mechanism that allows plants to cope with antagonists, such as floral larcenists, without significant reductions in reproductive fitness (Núñez-Farfán et al., 2007).

In summary, while floral larceny imposes a significant negative effect on pollen deposition and fruit set, its overall impact on female fitness is not purely negative. The neutral effects on seed set and seed quality suggest that the ultimate reproductive success of plants is not substantially compromised. These findings confirm an ultimate neutral impact of floral larceny, and reveal the complex interplay between plants and their floral visitors, which together shape the reproductive outcomes of plants. This balance may contribute to the evolutionary stability of floral larceny, as its negative impacts are sufficiently mitigated by plant tolerance mechanisms and compensatory reproductive strategies. Moreover, from the evolutionary perspective, plants have evolved various morphological and chemical traits (e.g., thickened calyx, secondary metabolites) that may protect them against floral larcenists, indicating that floral larceny has played a role in shaping plant evolution and speciation (Irwin et al., 2004). For floral visitors, the persistence of floral larceny may enhance foraging efficiency and energy savings (Irwin et al., 2010). Driven by these benefits, floral visitors may learn and even evolve new foraging strategies (Richman et al., 2021).

5. Conclusion

This study provides a thorough reevaluation of the impacts of floral larceny on plant reproductive success, confirming an ultimate neutral outcome. While floral larceny reduces certain flower traits and decreases pollinator visitation, it has a neutral effect on key reproductive outcomes such as male fitness, seed set, and seed quality. By promoting geitonogamy through altered pollinator foraging behaviors, floral larceny could even enhance genetic diversity, suggesting a complex and balanced role in plant-pollinator interactions. The evolutionary persistence of floral larceny suggests that plants have developed tolerance mechanisms that enable coexistence with larcenists without significant reproductive harm. This balance between antagonism and mutualism indicates that floral larceny, rather than destabilizing plant-pollinator interactions, may contribute to long-term evolutionary stability by promoting genetic diversity and resilience in plant populations in fascinating and unexpected ways. In fact, floral larceny is quite ubiquitous in nature, observed in 214 plant species in 59 different families, and nearly all plants with tubular flowers or nectar spurs experience floral larceny (Irwin and Maloof, 2002). Thus, our study provides a reliable explanation for the ubiquity of floral larceny. For future research in this field, studies from the animal's perspective—particularly the evolutionary implications of foraging strategies involving floral larceny—could provide deeper insights into this widespread ecological phenomenon.

Acknowledgements

We thank all authors of the studies included in this meta-analysis; this research would not be possible without their previous work. We acknowledge the financial support by the National Natural Science Foundation of China (32170241, 32160054, and 32470241). MCO was supported by the Chinese Academy of Science's PIFI Fellowship Initiative (2024PVC0046).

CRediT authorship contribution statement

Jin-Ru Zhong: Writing – original draft, Visualization, Formal analysis, Data curation. Xiao-Fang Jin: Writing – review & editing, Funding acquisition, Conceptualization. Michael C. Orr: Writing – review & editing. Xiao-Qing Li: Data curation. Yong-Deng He: Validation. Sheng-Wei Wang: Writing – review & editing. Qing-Feng Wang: Writing – review & editing. Chun-Feng Yang: Writing – review & editing, Supervision, Conceptualization. Zhong-Ming Ye: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Data availability statement

All meta data generated in this study and used to create the Figures have been deposited the Figshare database: https://figshare.com/s/99df84fe5c473dc85d85.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2024.12.004.