Why flower visitation is a poor proxy for pollination: measuring single-visit pollen deposition, with implications for pollination networks and conservation


Correspondence author: E-mail: pgw@st-andrews.ac.uk


  1. The relative importance of specialized and generalized plant-pollinator relationships is contentious, yet analyses usually avoid direct measures of pollinator quality (effectiveness), citing difficulties in collecting such data in the field and so relying on visitation data alone.
  2. We demonstrate that single-visit deposition (SVD) of pollen on virgin stigmas is a practical measure of pollinator effectiveness, using 13 temperate and tropical plant species. For each flower the most effective pollinator measured from SVD was as predicted from its pollination syndrome based on traditional advertisement and reward traits. Overall, c. 40% of visitors were not effective pollinators (range 0–78% for different flowers); thus, flower–pollinator relationships are substantially more specialized than visitation alone can reveal.
  3. Analyses at species level are crucial, as significant variation in SVD occurred within both higher-level taxonomic groups (genus, family) and within functional groups.
  4. Other measures sometimes used to distinguish visitors from pollinators (visit duration, frequency, or feeding behaviour in flowers) did not prove to be suitable proxies.
  5. Distinguishing between ‘pollinators’ and ‘visitors’ is therefore crucial, and true ‘pollination networks’ should include SVD to reveal pollinator effectiveness (PE). Generating such networks, now underway, could avoid potential misinterpretations of the conservation values of flower visitors, and of possible extinction threats as modelled in existing networks.


Pollination ecology has recently been invigorated by a strong community-level approach, often linked with concern over pollinator declines and conservation, and hence a need to understand how particular pollinator deficits may affect plant and animal populations and interactions (Waser et al. 1996). This has led to many analyses of ‘pollination webs’ or ‘pollination networks’, aiming to understand network structure and resilience to change. As networks have become embedded in ecological and evolutionary thinking (Proulx, Promislow & Phillips 2005), ‘plant–pollinator networks’ proliferate and associated methodologies and terminologies become more sophisticated. Core problems of inadequacy of the underlying data sets (incomplete sampling or varied relative sampling intensity, Blüthgen et al. 2008; Gibson et al. 2011), and of inadequate temporal/spatial spread of sampling (Olesen et al. 2008; Dupont et al. 2009) have been addressed. The resultant more complex models are often in turn used in meta-analyses: comparisons with other mutualistic communities (Olesen et al. 2007; Aizen, Morales & Morales 2008; Pocock, Evans & Memmott 2012), or assessing effects of invasive species (Memmott & Waser 2002; Bartomeus, Vila & Santamaria 2008; Valdovinos et al. 2009), of potential extinction rates and patterns (Memmott, Waser & Price 2004; Kaiser-Bunbury et al. 2010), or of resilience to anthropogenic factors such as climate change (Memmott et al. 2007; Willmer 2012).

This modelling activity has become linked with issues of specialization and generalization in plants, pollinating animals and their interactions (Waser et al. 1996; Johnson & Steiner 2000; Gibson et al. 2011). ‘Plant–pollinator networks’ appear to have flower visitors that are mostly generalized in their flower choices (Vazquez & Aizen 2004; Petanidou & Potts 2006), in turn suggesting that the concept of specific ‘pollination syndromes’ is less useful than the earlier literature had indicated (Waser 2006; Ollerton et al. 2009).

These issues have been highlighted in several key papers (Waser et al. 1996; Fenster et al. 2004) and a recent book (Willmer 2011), though the network and syndrome approaches can potentially be synergistic. Many ‘pollinator networks’ suggest preponderant generalization with high connectance, but merely eliminating cheats can make a network register as more specialized (Alarcón 2010), and levels of apparent generalization can vary across populations or even individuals of a given plant species (Herrera 2005). Meanwhile many pollination case studies report rather high levels of specialization, and/or a good match of selective pressures on flowers to particular functional groups of visitors acting as pollinators (Johnson & Steiner 2000; Fenster et al. 2004).

But a key issue still goes largely untested: the ability to distinguish between mere flower visitors and effective pollinators. This problem is well documented (Fishbein & Venable 1996; Ne'eman et al. 2010; Popic, Wardle & Davila 2013), and many ‘pollination networks’ explicitly or implicitly recognize the potentially misleading title used, in relying on simple visitation records. But variations on the claim that ‘pollination can be inferred if quantitative data is available on visitation’ (Hegland et al. 2010) remain prevalent.

Testing this requires incorporation of measures of effective pollination into community studies and thence into networks. Some studies add a more realistic ‘pollination’ slant to visitation data using various added measures (reviewed in Ne'eman et al. 2010; most recently using visitor pollen loads, Popic, Wardle & Davila 2013), but as yet sidestep measuring effectiveness of visitors as true pollinators. Here, we quantify the ‘pollinator vs. visitor’ problem to show that the distinction matters greatly and may undermine some existing literature. We use the term pollinator effectiveness (PE) throughout, rather than other variant terms (Inouye et al. 1994; Ne'eman et al. 2010), agreeing that it best describes the character of the measure needed. Ne'eman and his co-authors supported (from first principles) measuring numbers of conspecific pollen grains deposited on a virgin stigma in one visit – single-visit deposition, hereafter SVD. This measures both an animal's ability to acquire pollen in earlier visits to the plant species (thus incorporating visit constancy), and to accurately deposit it where it can potentially lead to fertilization. It avoids hazards of measurements of seed- or fruit-set that bring postpollination factors into play, and it gives species-specific values for PE. It can be expanded to give SVD per unit time (hour, or day), or through the life of the flower, or plant, or population. Some early papers had shown that this field measurement could indeed clarify the visitor–pollinator distinction. Good models of best practice exist (Primack & Silander 1975; Motten et al. 1981; Wilson & Thomson 1991), and examples occur for bees, flies, lepidopterans and vertebrates (Willmer 2011).

Our field measurements demonstrate that pollinator effectiveness (PE) is reliably and relatively easily determined using SVD, for 13 plant species from various traditional ‘syndromes’. True pollination networks are therefore feasible and much needed, and this ongoing work will improve understanding of the pressing issues of pollination ecosystem services and pollinator conservation.

Materials and methods

Plants and study sites

We used 13 plant species (Table S1), from two temperate Scottish sites (scrubby woodland, West Quarry Braes, Fife (NO 597 088) and meadowland near Loch Tay, [NN 669 358)], and from deciduous forest in Costa Rica (Santa Rosa, 10°50′N, 85°40′W). Plants were selected for their flowers’ apparent conformity to particular pollination syndromes with a broad range of morphological and reward traits.

Measuring pollinator effectiveness

Flowers were selected as buds, usually in the evening, and covered (individually, in small groups, or as inflorescences) in 2 mm netting to exclude flower visitors but avoid excessive environmental modification. Once flowers had fully opened the next day they were uncovered and observed until a single visitor landed and foraged. Visitors were identified immediately, or photographed, or captured for later identification. Each visit's duration was timed using a stopwatch, or by estimation (to nearest 10 or 30 s) where a visitor fed successively at several flowers on an inflorescence (mean duration shown without SE in Table 1), or where several visitors were active concurrently. For hummingbirds, hovering between flower visits, durations were corrected to give mean time spent feeding using video recordings. Visitor feeding (nectar, pollen or both) was also recorded.

Table 1. Mean single-visit deposition (SVD) values (±SE) for each visitor group, and all visitor species where n > 5 or P-value significant, for the 13 plant species, with significance indicated as the difference in SVD (corrected for mean pollen on unvisited control flowers, value in parentheses alongside plant name) from zero. P-value bold where P < 0·05 (*where significance also meets the criteria of Bonferroni's correction). Final column shows mean visit duration (±SE). Spearman's rank correlations for SVD/duration comparisons (overall, and split by visitor species) are also shown
 Mean SVD n P-valueMean visit duration (s)
Malvaviscus (10·6)
 Hummingbirds (Amazilia rutila)104·4 ± 9·821 <0·0005* 6·1 ± 1·2
 Bees29·035 <0·0005* 92·1 ± 9·2
Agapostemon sp.53·1 ± 15·38 0·008 91·9 ± 23·8
Trigona fulviventris21·9 ± 5·513 0·018 110·8 ± 17·8
Tetragonisca angustula21·9 ± 4·314 0·008 75·0 ± 8·2
 Butterflies5·8 ± 1·7120·180122·5 ± 21·0
 Ants (Camponotus novograndensis)11·1 ± 1·580·066180·0 ± 29·9
SVD vs. duration: r = −0·64, n = 76, P < 0·001. Split by visitor species: NS
Helicteres (89·0)
 Hummingbirds (Phaethornis guy)1517·1 ± 97·521 <0·0005* 1·73 ± 0·4
 Bees441·8105 <0·0005* 202·0 ± 10·0
Trigona fulviventris443·4 ± 29·992 <0·0005* 232·5 ± 10·8
Agapostemon sp.400·0 ± 101·46 0·028 80·0 ± 24·1
Tetragonisca angustula162·9 ± 26·07 0·028 68·6 ± 14·2
SVD vs. duration: r = −0·41, n = 127, P < 0·001. Split by visitor species: NS
Geranium (16·7)
 Bees33·956 <0·0005* 23·8 ± 2·5
Bombus pratorum31·2 ± 6·752 <0·0005* 25·2 ± 2·6
 Flies19·825 0·027 48·0 ± 7·9
  Rhingia campestris 19·0 ± 5·819 0·012* 42·6 ± 5·8
SVD vs. duration: r = 0·19, n = 75, P = 0·103. Split by visitor species: B. pratorum (r = +0·32; P = 0·019)
Digitalis (19·4)
 Bees58·238 <0·0005* 16·1 ± 1·6
Bombus hortorum73·2 ± 16·725 <0·0005* 11·4 ± 1·3
Bombus muscorum31·0 ± 4·412 0·005* 26·3 ± 2·6
SVD vs. duration: r = −0·15, n = 37, P = 0·362. Split by visitor species: NS
Byrsonima (48·5)
 Bees313·982 <0·0005* 65·9 ± 6·0
Exomalopsis sp.1686·7 ± 121·730·10920·0 ± 5·8
Centris nitida381·7 ± 96·86 0·043 45·0 ± 5·5
Trigona fulviventris254·5 ± 29·961 <0·0005* 64·9 ± 5·3
Tetragonisca angustula238·8 ± 41·312 <0·003* 92·5 ± 29·7
SVD vs. duration: r = −0·14, n = 82, P = 0·202. Split by visitor species: NS
Agrimonia (8·5)
 Hoverflies36·2139 <0·0005* 24·1 ± 1·4
Rhingia campestris55·2 ± 21·915 0·005* 20·0
Platycheirus scutatus52·8 ± 8·119 <0·0005* 30·0
Platycheirus albimanus47·6 ± 19·210 0·008 63·5 ± 7·9
Leucozona laternaria43·5 ± 10·512 0·008 20·0
Episyrphus balteatus27·6 ± 2·963 <0·0005* 19·9 ± 1·6
Meliscaeva auricollis23·2 ± 6·713 0·012 16·5 ± 1·3
SVD vs. duration: r = 0·11, n = 141, P = 0·177. Split by visitor species: NS
Cirsium (0)
 Bees (Bombus terrestris)1·8 ± 0·222 0·038 19·1 ± 2·4
 Hoverflies2·953 <0·0005* 8·8 ± 1·7
Episyrphus balteatus3·8 ± 0·826 <0·0005* 8·7 ± 3·5
Platycheirus manicatus2·1 ± 0·316 0·002* 7·5 ± 0·5
Melanostoma mellinum2·1 ± 0·811 0·001* 10·9 ± 0·3
 Other Flies1·231 <0·0005* 20·6 ± 1·6
Empis sp.1·8 ± 0·55 <0·0005* 36
Calliphora vomitoria1·2 ± 0·115 <0·0005* 22·7 ± 0·8
SVD vs. duration: r = −0·22, n = 106, P = 0·021. Split by visitor species: B. terrestris (r = +0·63, P = 0·001), M. dubium = (r = +0·77, P < 0·001): C. vomitoria (r = +0·53, P = 0·040)
Centaurea (14·0)
 Hoverflies217·9240 <0·0005* 11·4 ± 0·7
Episyrphus balteatus273·7 ± 41·7158 <0·0005* 8·2 ± 0·2
Eupeodes corollae115·0 ± 23·612 0·002* 15
Rhingia campestris114·1 ± 13·965 <0·0005* 18·6 ± 2·4
Platycheirus manicatus50·4 ± 25·850·1096
SVD vs. duration: r = −0·25, n = 240, P < 0·001. Split by visitor species: R. campestris (r = −0·60; P < 0·001)
Knautia (0)
 Bees4·966 <0·0005* 6·7 ± 0·7
Bombus pratorum6·0 ± 0·921 <0·0005* 4·3 ± 0·8
Bombus (Psithyrus) bohemicus5·9 ± 1·319 0·001* 1·6
Bombus lucorum4·8 ± 0·712 0·002* 10·0
Bombus terrestris2·1 ± 1·014 0·018 14·3 ± 0·5
 Hoverflies5·8303 <0·0005* 3·2 ± 0·2
Rhingia campestris7·4 ± 1·454 <0·0005* 2·2 ± 0·1
Episyrphus balteatus6·4 ± 0·6203 <0·0005* 3·6 ± 0·3
Syrphus ribesii1·0 ± 0·242 0·018 1·8 ± 0·1
 Other dipterans (Empis sp.)6·1 ± 0·6147 <0·0005* 7·9 ± 0·5
SVD vs. duration: r = −0·11, n = 516, P = 0·016. Split by visitor species: R. campestris (r = +0·64; P < 0·001), E. balteatus (r = −0·41; P < 0·001)
Trifolium (0·6)
 Bees12·2371 <0·0005* 3·2 ± 0·1
  Bombus lucorum 25·1 ± 2·231 <0·0005* 1·3 ± 0·1
  Bombus terrestris 13·3 ± 1·534 <0·0005* 1·5 ± 0·1
  Bombus hortorum 10·8 ± 0·6275 <0·0005* 3·7 ± 0·1
  Bombus muscorum 10·0 ± 1·831 <0·0005* 2·3 ± 0·1
 Hoverflies (Criorhina sp.)28·8 ± 2·418 <0·0005* 5·0
SVD vs. duration: r = −0·04, n = 389, P = 0·47. Split by visitor species: B. terrestris (r = +0·75; P < 0·001)
Ipomoea (52·8)
 Bees108·7119 <0·0005* 76·0 ± 6·0
Andrena sp.155·7 ± 15·919 <0·0005* 44·3 ± 9·8
Agapostemon sp.118·5 ± 10·155 <0·0005* 103·6 ± 9·8
Partamona musarum113·5 ± 9·411 0·003* 50·9 ± 7·6
Tetragonisca angustula70·4 ± 12·616 0·008 32·5 ± 3·4
Trigona fulviventris35·8 ± 9·2120·10978·7 ± 18·2
Ants65·037 <0·0005* 142·7 ± 13·7
Pseudomyrmex gracilis69·0 ± 10·728 0·001* 148·9 ± 15·5
Camponotus novograndensis52·6 ± 12·190·068123·3 ± 29·8
 Beetles93·540 <0·0005* 578·3 ± 86·9
Notoxus sp.87·8 ± 9·836 <0·0005* 556·7 ± 92·1
SVD vs. duration: r = −0·14, n = 194, P = 0·047. Split by visitor species: P. gracillis (r = −0·48; P = 0·010)
Heracleum (16·8)
 Hoverflies43·7239 <0·0005* 6·8 ± 0·3
Epistrophe grossulariae61·8 ± 12·722 <0·0005* 7·1 ± 0·3
Episyrphus balteatus55·8 ± 5·5100 0·005 7·8 ± 0·4
Syrphus ribesii32·1 ± 3·052 <0·0005* 2·7 ± 0·1
Eupeodes corollae22·5 ± 4·012 0·007 10·0 ± 1·5
Platycheirus albimanus25·8 ± 12·960·10920
Other Syrphini sp.28·0 ± 1·542 <0·0005* 10·2 ± 0·8
 Other dipterans80·5152 <0·0005* 7·2 ± 0·5
Lucilia sericata116·1 ± 12·833 <0·0005* 4·7 ± 0·1
Platypezidae sp.79·9 ± 7·837 <0·0005* 3·5 ± 0·2
Anthomyiidae sp.62·8 ± 19·860·0681·7
Phaonia subventa67·4 ± 8·176 <0·0005* 9·1 ± 0·7
SVD vs. duration: r = −0·04, n = 390, P = 0·449. Split by visitor species: E. balteatus (r = +0·23; P = 0·032), L. sericata (r = −0·40; P = 0·020), platypezid sp. (r = +0·54, P = 0·001)
Rubus (52·7)
 Bees256·242 44·3 ± 8·2
Bombus lucorum343·3 ± 40·26 0·026 30·0 ± 3·4
Bombus terrestris295·5 ± 53·216 <0·0005* 55·3 ± 10·3
Bombus pratorum223·0 ± 82·570·06877·1 ± 39·8
Bombus pascuorum142·0 ± 21·25 0·043 12·2 ± 5·4
Apis mellifera270·0 ± 49·840·06812·5 ± 3·2
 Hoverflies136·635 0·001* 99·9 ± 13·6
Rhingia campestris172·6 ± 46·019 0·005 111·3 ± 14·3
Eristalis horticola87·0 ± 40·650·31714·8 ± 4·8
Episyrphus balteatus80·0 ± 11·470·068112·9 ± 26·0
 Muscoid dipterans54·6130·18067·8 ± 35·6
 Wasps (Vespula vulgaris)80·9 ± 8·260·06621·5 ± 5·4
SVD vs. duration: r = −0·08, n = 92, P = 0·428. Split by visitor species: B. terrestris (r = +0·57; P = 0·021)

Stigmas from each visited flower (or each floret visited in a composite) were then removed with forceps and stored in separate cells of plastic cell-culture arrays, kept covered and cool. Numbers of adherent pollen grains per stigma were counted immediately using a dissecting microscope; or the array was stored frozen for later counting. Pollen grains were only counted if morphologically conspecific.

For each plant species, unvisited flowers were also netted as controls, and pollen grains on their stigmas recorded to account for self-pollen transfer by wind or by flower handling. A value of mean SVD was determined for each visitor species for which sufficient data were available, and compared to the control SVD. A pollinator was defined as any species with an SVD significantly greater than controls. All other visitor species were deemed ineffective pollinators (including, but not synonymous with, floral thieves) and excluded from further analysis.

Sampling periods

Sampling occurred throughout a day where possible, to detect temporal variations in visitor assemblage and performance. Observations were restricted to dry calm weather conditions, when previously protected flowers were unaffected by rain. Sampling sessions were 1–3 h, depending on visit frequency and thus how long it took all previously-protected newly-opened flowers to be visited.

Visitation surveys: Scaling up SVDs and pollinator effectiveness

Observations of flower visits necessarily only applied to the first visitor to previously-netted flowers, so cannot accurately represent overall visit numbers or frequencies. To record both visitation patterns and SVD separately, we chose Scottish populations of Agrimonia eupatoria, with large well-spaced flowers on adjacent stems. Flowers were observed for twelve 45-min intervals daily (06:45–18:30, with all flowers by then pollen-depleted) in July 2009. Visit frequencies, durations and behaviours of each visitor were recorded. Since visitors were undisturbed they visited a sequence of flowers freely, and their chosen flowers were noted. Visitors were mainly hoverflies, taking only pollen; most were identified to species (but to tribe for Bacchini and Syrphini) and a mean SVD was calculated. Combined with visitor frequencies this generated a per-hour and per-day pollinator performance value from existing formulae (Ne'eman et al. 2010).

Statistical analyses

Control pollen values for each plant species were subtracted from SVD values, with any resulting negative values set to zero for the purposes of statistical analyses. Since data for some plants were normally distributed but other data sets were not, nonparametric Wilcoxon Signed Ranks testing was used for consistency to compare SVD values with zero for each of the 13 plants. We show P levels as significant where they are below 0·05; Bonferroni corrections were routinely used, but since application of these is often regarded as too conservative, we merely indicate with an asterisk where they remain significant after Bonferroni corrections. SPSS Statistics for Windows, Version 17.0 (SPSS Inc., Chicago, USA) was used for all statistical analyses.


Measuring SVD and pollinator effectiveness

For every plant species studied, SVD values were calculated for ‘visitor groups’ defined according to traditional pollination syndromes (Willmer 2011), and for each visitor species separately where numbers of recordings allowed (Table 1; expanded details in Table S2). Those animal groups that a syndrome approach (Table S1) would predict as major pollinators generally had the highest SVDs, while for the more generalist plants several groups had high SVDs. For each one of the 13 species, the predicted syndrome was well matched with SVD findings, making SVD demonstrably a good measure of ‘expected’ pollinator effectiveness (PE). Of 105 plant–visitor combinations across the 13 plants, only 63 produced effective pollination.

Testing proxies for pollinator performance

Visit duration

Mean visit durations are included in Table 1, with Spearman's Rank Correlations (visit duration vs. SVD) for all visitors combined. Seven plant species showed no correlation, while the remaining 6 (Malvaviscus, Helicteres, Cirsium, Centaurea, Knautia and Ipomoea) showed a significant negative correlation. However, when visitor species were considered separately (Table 1) an overall relationship between SVD and visit duration was rarely preserved; duration could vary substantially within ‘visitor groups’, and across plant species for a given visitor, so was on its own an unreliable measure of PE.

Visit number or frequency

For Agrimonia eupatorium, visit numbers and rates, and hence pollinator performance for each major visitor, were calculated per hour and per day (Fig. 1). Episyrphus balteatus had the lowest SVD at the single-visit scale, but its high visitation rate gave it the highest SVD at per-hour and per-day scales; it would often be the ‘best’ pollinator. Conversely, Rhingia campestris had the highest SVD but the lowest per-hour and per-day SVD. But neither measure on its own gives a clear picture, whereas using visit frequency with SVD data can substantially affect the perception of ‘most important pollinator’ (cf. Olsen 1997; Ne'eman et al. 2010).

Figure 1.

Single-visit deposition (SVD) values for visitors to Agrimonia eupatoria scaled up to the ‘per hour’ and ‘per day’ level using visitation frequency data.

Combining visit duration, feeding type and visitor species with SVD measures

A Generalized Linear Model was constructed (Table S3) to test the combined utility of typical measures of a good pollinator (visit duration, and type of feeding: nectar/pollen/both, or for Byrsonima oil/pollen/both) as proxies for pollination effectiveness; ‘visitor species’ was also included since variation in SVD between species but within functional groups is evidently important. In 8 of the 13 plants, the only factor significantly related to pollen deposition was visitor species, through its direct association with SVD; for the remaining species, other factors were inconsistently and rarely significant.


Not all visitors are pollinators of a given plant species; a pollinator must deposit sufficient pollen on the correct and receptive stigma, and that pollen must be conspecific and viable. Our SVD protocols address the first two requirements, and any visibly heterospecific pollen grains were discounted. We show that SVD measures are relatively simple to incorporate into pollination studies, giving an accurate value for pollinator performance, and highlighting the effective visitors which in all 13 species largely correspond to expectations from a syndrome approach. Combined with visitation records, SVD can assess ‘pollinator effectiveness’ per hour, per day or per season, and can indicate ‘pollinator importance’, as with Agrimonia.

Only 63 of 105 plant–visitor interactions produced effective pollination (Table 2); and ineffective visits were not just the traditional ‘illegitimate’ visits, as many involved a normal route into the corolla by visitor species of similar size to the effective pollinators.

Table 2. Summary of visitor–pollinator analyses in relation to floral syndromes (ST, LT = short- or long-tongued). Further details on syndrome-related traits are in Supporting Material, Table 1
PlantSyndrome based on traitsFunctional groups of all visitorsSpecies of all visitorsFunctional groups of pollinatorsSpecies of pollinatorsSpecies of ineffective visitorsSyndrome based on SVD analysis
Malvaviscus Hummingbird47243Hummingbird (bee back-up)
Helicteres Hummingbird24240Hummingbird (bee back-up)
Geranium Bee38226Bee
Digitialis Bee13121Bee
Byrsonima Oil-bee14131Oil-bee (pollen-bee back-up)
Agrimonia Hoverfly29163Hoverfly
Cirsium LT bee/hoverfly37361LT bee/hoverfly (ST insect back-up)
Centaurea MT bee/hoverfly14131LT bee/hoverfly (ST insect back-up)
Knautia MT bee/hoverfly39381LT bee/hoverfly (ST insect back-up)
Trifolium LT bee/hoverfly25250LT bee/hoverfly (ST insect back-up)
Ipomoea Generalist/bee615369Generalist/ST insect
Heracleum Generalist312284Generalist, smaller ST insect
Rubus Generalist4183612Generalist, larger insects
All plant–visitor combinations35105266342 

Are proxies for SVD useful or appropriate?

Single-visit deposition is a good direct measure of PE; however, in most existing studies PE is not assessed, being substituted with other parameters such as visitor abundance, pollen load, number of stigma touches, feeding type or visit duration. Visitor abundance alone, though often used (e.g. Olsen 1997), is widely recognized as a poor measure of pollination outcomes (Johnson & Steiner 2000). A positive link may be recorded between abundance or visitation rate and pollen deposition, but can be weak [e.g. only 36% of variation in pollen deposition was explained thus for Ipomopsis aggregata (Engel & Irwin 2003)].

Abundance values for each animal and plant, and their interaction frequencies, can generate quantitative visitation networks, adding qualitative estimates of pollination using visitors’ pollen loads (Popic, Wardle & Davila 2013); and assessing pollen fidelity (% conspecific pollen carried) can refine visitor importance further (Forup et al. 2008) and may encourage using visitor abundance and pollen load fidelity as proxies (Bosch et al. 2009; Kaiser-Bunbury et al. 2010). But pollen on visitors' bodies may poorly represent pollination potential; it can be deposited on incompatible or unreceptive stigmas, or lost before reaching another flower (Inouye et al. 1994; Harder & Routley 2006), so giving no correlation with pollen deposited on conspecific stigmas (Adler & Irwin 2006). Other possible proxies such as ‘contact with reproductive structures’ (Petanidou & Potts 2006; Gibson et al. 2011), number of stigma touches (Olsen 1997), measurements of visit duration (Fishbein & Venable 1996; Kaiser-Bunbury et al. 2010) and of pollen removal (Ivey, Martinez & Wyatt 2003) are similarly subject to problems of pollen loss. We therefore sought explicit relationships between these proposed measures and our direct SVD assessment.

Correlation of visit duration and pollen deposition

There were no significant correlations between visit duration and SVD for all visitors combined for 7 of our 13 species, but 6 showed a significant negative correlation (Table 1). In theory, longer visits could increase visitor contact with, and/or transfer of pollen to, a stigma; but they could also indicate ‘ineffective’ feeding (excessive grooming, eating pollen or floral tissues, avoiding anther or stigma contacts). SVD and PE will be higher for visitors which ‘fit’ the flower, feed rapidly on nectar and/or pollen, and quickly acquire body pollen. Short efficient visits will often predominate early on, when pollen is more abundant, and visitor groups show very different diurnal activity patterns (Willmer & Stone 2005). Thus, when visitor species are treated separately the correlations can change markedly, and only 3 of 13 species did not show such changes (Table 1). For the two ornithophilous plants (Malvaviscus, Helicteres), negative correlations disappeared, largely because visit duration and variance were low, and birds received the most pollen grains of any group. Trifolium and Geranium had significant overall negative correlations, but bumblebees showed significantly greater SVD in longer visits. In Knautia, with no overall relationship, Rhingia campestris showed a significant positive correlation and Episyrphus balteatus the opposite; these differing interactions are masked when visitor species are pooled.

Within all these comparisons, the common visitor species E. balteatus is instructive, showing positive or negative correlations between visit duration and SVD in different plants, though its mean visit duration did not vary greatly (Table 1). Evidently, the varying behaviour and PE of this species on each flower matters, rather than visit duration alone. This reinforces the problems with using visit duration as a proxy in its own right; no particular ‘kind’ of relation between visit duration and SVD can be assumed, for a visitor group or for a single visitor species.

Combined measures as proxies for pollination effectiveness

Our GLM showed that in 7 of 13 plant species the only factor significantly contributing to SVD was visitor species; feeding behaviour and visit duration were unimportant even where duration did affect pollen deposition (Table 1: Malvaviscus, Helicteres, Ipomoea). Duration and feeding behaviour never accounted for more than a small percentage of SVD variation, and in Centaurea, Digitalis, and Geranium no factor significantly explained SVD variation. Overall, in 11 of our 13 plants by far the largest predictor of variation in pollen deposition was visitor species.

Possible criticisms and drawbacks of SVD and of this study

Firstly, measures of SVD are undoubtedly context-dependent, potentially affected temporally and spatially by environmental variation and relative species abundances. Hence, extrapolation between studies is dangerous, and SVD should be measured for a given interaction at a given site (as with many measures in pollination ecology, since phenology and rewards vary between sites).

Furthermore, SVD does not relate to the final female reproductive success of a flower, manifested in seed-set. But postpollination events have little to do with assessing pollinators, and reliance on seed-set may show the same effects described here (Spears 1983) or give contradictory results (Olsen 1997). Equally, SVD does not include estimates of pollen viability or germination, and some deposited pollen grains even though conspecific may not germinate, especially if large numbers clog up a small stigma.

Single-visit deposition measures may also be time-biased, tending to accentuate early visitors. Delayed removal of bags may help, so that ‘first visits’ occur later; but then an uncovered flower may have unusual rewards for that time of day, giving abnormal visit durations or frequencies. Elsewhere, we analyse time dependence of SVD more closely (King & Willmer, in prep.). We also note that all Scottish sites experienced very poor summer weather in 2008–2010 (high rainfall, poor sunshine), so visitor profiles were unusual: very low bee numbers (Apis and Bombus) occurred in eastern Scotland, and bees are under-represented in our data, with perhaps a concomitant increase in hoverfly numbers.

Finally, we considered just 13 plant species, and each in isolation, so proving that SVD methodology is feasible and timely for fieldwork, that it works with varying flower morphologies, and that measuring PE in this way is important because it shows up ineffective visitors. But the required and ongoing step is to use SVD to directly compare ‘visitation’ networks and true ‘pollination’ networks.

Why distinguishing pollinators and visitors matters

Flower visitors are not necessarily pollinators. Some are simple cheats, and their effects have been acknowledged (see Alarcón 2010; Genini et al. 2010). But eliminating obvious cheats is not enough: which apparently legitimate visitors correctly deposit significant pollen on stigmas? Some earlier studies (e.g. Wilson & Thomson 1991) made exactly this point but have been insufficiently built upon. More recent studies have paralleled our own in comparing visitor PE for just one plant genus (Kandori 2002; Stoepler et al. 2012), reaching similar conclusions regarding problems with proxies, and reinforcing the value of SVD (or a near equivalent) as a measure of effectiveness.

Without distinguishing visitors from pollinators, various negative consequences could ensue: conservation efforts could be misled by suggestions that networks are robust and extinctions can be tolerated (e.g. Memmott, Waser & Price 2004; Dupont et al. 2009; Hegland et al. 2010; Kaiser-Bunbury et al. 2010; Burkle & Alarcón 2011), or that visitors acting as ‘hubs’ or ‘connectors’ require most support (Olesen et al. 2007) whereas relationships between connectance and conservation value may be poor (Ruben, Devoto & Pocock 2012). Interpretations of specialization and generalization can also be seriously problematic when only visitation is recorded (see Alarcón 2010; Popic, Wardle & Davila 2013).

Single-visit deposition is a valuable simple and direct means of measuring pollinator effectiveness, for which indirect proxies are unreliable. Here, variation in SVD was poorly related to visit duration or feeding behaviour, but strongly explained by visitor species, the most effective visitors being those predicted as the most important pollinators from syndrome-related floral traits. We are now incorporating SVD into networks to extend this argument; we urge care over extrapolations from existing ‘pollinator’ networks, particularly where these are used to infer consequences for ecosystem management and for modelled extinction threats.


We thank field-site staff in Costa Rica for logistic support, and Will Cresswell, Steve Hubbard and Jane Wishart for statistical advice. CK was supported by an NERC studentship, and GB by a University of St Andrews scholarship. Two anonymous referees helped to improve the analysis and discussion.