3
Fig/fig wasp, yucca/yucca moth, orchid/orchid bee: co-evolutionary relationship, Reciprocal mutualism (Eltz et al. 2011)
4
Indirect facilitative pathway b/w plants and pollinators = neighbour-mediated pollinator facil.(magnet hypothesis) (Laverty 1992). Competition b/w plants for resources incl. pollinators. Diverse floral displays = net increase in pollinator frequency/diversity (Wirth et al. 2011)
5
Data includes observational counts/abundance data, est. of pollen lim., seed/fruit set, visitation rates & diversity measures. Would benefit highly from synthesis to determine most effective methodologies for dealing with this data
6
Way of understanding the level of complexity of stats approaches used in pollinator facil. lit. and allows for ID of stats methods that have potential for increased citations
7
Thomson Reuters web of science used to conduct search using the term “plant AND pollinat* AND facilitate*”. Refined by research areas: environmental sciences, ecology, plant sciences.
8
Studies without primary data (notes, reviews, meta-analyses) and studies on non-pollinating insects were excluded
10
Fig1. Multiple polynomial regressions of the change in the use of tests over time. (n=245, ANOVA: R2=0.46308, mean (SD/SE): R2=0.32680, regression/correlation: R2=0.50553, t-test: R2=0.16791, mixed effects model: R2=0.41467, chi-square R2=0.34096, re-sampling: R2=0.22495, ordination: R2=0.38394, none: R2=0.15419, other: R2=0.16871).
P facilitation primarily used only mean/SE/SD. ANOVAs used most frequently from 2000-14. Regression= steepest increase in use and along w/ means (SD/SE) were second most highly used test after ANOVAs. Mixed effects models, ordinations &chi-square tests also highly cited. Studies that used no/other statistics were low frequency.
11
Fig2. Breakdown of the tests used in sub-disciplines of pollinator facilitation: neighbour-effect, pollination biology, invasion biology, agriculture and other (herbivory, behaviour, trophic cascades and evolution). A) Total frequency of studies that utilized these statistical test groups in each sub-discipline (N=245). B) Relative proportion of studies that utilized these statistical test groups in each sub-discipline. The number of studies in each category (n) is shown (N=245).
12
Fig3. Frequency and average number of total citations for each group of statistical tests used in pollinator facilitation studies (n=95). Tests are grouped according to having A) relatively high citations compared to publication frequency, B) relatively equal citations compared to publication frequency, C) Relatively low citations cf. to publication frequency.
13
Fig4. A linear regression for the relationship between the diversity of statistical tests used (the number of unique statistical tests used per study) and the average citations per year in the discipline of pollinator facilitation (n=95, R= 0.11517)
14
ANOVAs, regressions/correlations and ordinations increasing quickly. St
2. Introduction
Many plants rely on insects for pollination and co-evolution is common
Facilitation between plants and pollinators = shared access to resources
3. Introduction – obligate pollinator
mutualisms
Rely exclusively on each other for
pollination/reproduction
E.g. fig & fig wasp, yucca & yucca moth, orchids &
orchid bees
5. Purpose
Data in this field takes on many forms…
This review examines the statistical scope of the
literature to date on the study of pollinator
facilitation.
6. Objectives
1) Determine the change in use of statistical tests
over time
2) Compare use of statistical tests for various sub-disciplines
of pollinator facilitation
3) Compare publication frequency and average
citations per publication for different test groups
4) Contrast the relationship between diversity of
statistical tests and citation rate
7. Search technique
plant AND pollinat* AND
facilitat*
Refined by: Environmental science, ecology,
plant science, English-only
496
8. Search technique
plant AND pollinat* AND
facilitat*
Refined by: Environmental science, ecology,
plant science, English-only
95
10. Test usage over time…
12
10
8
6
4
2
0
1979 1984 1989 1994 1999 2004 2009 2014
Frequency
Year
ANOVA mean (SD/SE) regression/Correlation t-test
mixed effects model chi-square re-sampling ordination
none other
12. Publications vs. citations by test
60
50
40
30
20
10
0
Publications Average citations
Number of studies
60
50
40
30
20
10
0
Publications Average citations
Number of studies
60
50
40
30
20
10
0
Publications Average citations
Number of Studies
13. Test diversity vs. citation rate
R² = 0.1152
18
16
14
12
10
8
6
4
2
0
0 1 2 3 4 5 6
Average citations/year
Diversity of tests
14. Key Results
ANOVAs, means, and regressions/correlations
used most frequently
Neighbour-effect and invasion biology studies
used more extensive tests compared to pollination
biology
Ordinations, re-sampling techniques and t-tests
received the most citations
15. Novel pathways…
Rarefaction curves/ordinations should be implemented
Effect size estimates rarely used;
useful for teasing apart facilitative vs.
competitive effects
Network analyses to highlight
important pathways and
unanticipated interactions
16. But…
Must still take experimental design into account
Not all tests can be applied in every scenario
Test assumptions must be met
Notes de l'éditeur
Fif/fig wasp and yucca/yucca moth: pollinator collects and distributes pollen between plants, while utilizing the fruit of the plant for oviposition and as a larval food source (Pellmyr et al. 1996, Fleming and Holland 1998). Many orchid species and orchid bees (tribe: Euglossini) have also developed a co-evolutionary relationship in which the plant produces scent compounds that attract male bees to the inflorescences for pollination (Eltz et al. 2011). The males collect these emitted compounds on their hind tibiae for use as mating cues for females, forming a reciprocal mutualism that benefits both parties (Eltz et al. 2011).
Indirect facilitative pathway between plants and pollinators of neighbour-mediated pollinator facilitation, i.e. the magnet hypothesis (Laverty 1992). While there’s competition between plants for resources including pollinators, diverse floral displays can encourage a net increase in pollination frequency and pollinator diversity (Liao et al. 2011, Wirth et al. 2011)
Data collected takes on many forms including observational counts and abundance data, estimations of pollen limitation, seed and fruit set, visitation rates, and diversity measures, among others. Thus, this body of literature would benefit highly from synthesis to determine the most effective methodological approaches when dealing with these types of data
This novel investigation provides a unique way of understanding the level of complexity of statistical approaches used in pollinator facilitation literature, and allows for the identification of statistical methods that have the potential to increase the value of these publications to the scientific community.
Thomson Reuters web of science used to conduct search using the term “plant AND pollinat* AND facilitate*” and resulting articles were refined to the research areas of environmental sciences, ecology or plant sciences.
Studies without primary data (e.g. notes, reviews and meta-analyses) and studies on non-pollinating insects such as those focusing only on seed dispersal, herbivory or plant pathogens were excluded
Figure 1. Multiple polynomial regressions depicting the change in the usage of ten main categories of statistical tests used in pollinator facilitation studies over time (1979-2014). (n=245, ANOVA: R2=0.46308, mean (SD/SE): R2=0.32680, regression/correlation: R2=0.50553, t-test: R2=0.16791, mixed effects model: R2=0.41467, chi-square R2=0.34096, re-sampling: R2=0.22495, ordination: R2=0.38394, none: R2=0.15419, other: R2=0.16871).
Early on, studies on pollinator facilitation primarily utilized measurements of only mean and SE/SD. \ANOVAs used most frequently from 2000-2014. Regressions had steepest increase in use and along with means (SD/SE) represent the second most highly used tests after ANOVAs. Mixed effects models, ordinations and chi-square tests also increased in use over the last 35 years. Studies that used no statistics and ‘other’ statistics were low in frequency.
Figure 2. A breakdown of the statistical tests used in various sub-disciplines of the study of pollinator facilitation: neighbour-effect, pollination biology, invasion biology, agriculture and other (herbivory, behaviour, trophic cascades and evolution). A) The total frequency of studies that utilized these statistical test groups in each sub-discipline (N=245). B) The relative proportion of studies that utilized these statistical test groups in each sub-discipline. The number of studies in each category (n) is shown (N=245).
Figure 3. The frequency and average number of total citations for each group of statistical tests used in pollinator facilitation studies (n=95). Statistical tests are grouped according to having A) relatively high citations compared to publication frequency, B) relatively equal citations compared to publication frequency, C) Relatively low citations compared to publication frequency. See Methods for a full description of the search method and statistical test classification.
Figure 4. A linear regression for the relationship between the diversity of statistical tests used (the number of unique statistical tests used per study) and the average citations per year in the discipline of pollinator facilitation (n=95, R= 0.11517). See methods for a full breakdown of study selection procedures and analyses.
The frequency of statistical analyses in the field has increased over time, with certain tests such as ANOVAs, regressions/correlations and ordinations increasing at a much higher rate. Discipline biases are also clear, with studies on the neighbour-effect and invasion biology forming the greatest proportion of studies that use more extensive statistical methods such as t-tests, ANOVAs, mixed effects models, re-sampling techniques, network analyses and effect size estimates. Other sub-disciplines such as pollination biology tended to do simpler analyses, with the greatest proportion of them using only calculations of means and standard deviation/standard error/no statistics at all.
Rarefaction curves allow for the standardization of different sample sizes, which is often the case with ecological data (Simberloff 1972, Melo et al. 2003). This standardization allows for the production of richness/diversity estimates for large samples that represent their expected values for sample sizes equal to those of the smallest sample-size used (Simberloff 1972, Melo et al. 2003). Ordinations have typically been used in the field to determine the factors that effect pollinator species compositions, floral neighbourhood assemblages, or species trait similarities and/or differences (e.g. Castillo et al 2014). CCAs are also especially useful tools when dealing with species abundance data, which was encountered frequently throughout the processing of these studies (Anderson and Willis 2003). Effect size estimates are useful for answering the question of whether facilitation or competition is occurring, which is a dominant theme throughout the pollinator facilitation literature. This is especially useful for studies on the effects of invasive species on natives and for studies that test the magnet hypothesis for plants with co-flowering neighbours. Network analyses are capable of highlighting important pathways and interactions that otherwise may not have been anticipated