10.25338/B8KP8P
Rinehart, Shelby
0000-0001-9820-1350
University of California, Davis
Long, Jeremy
San Diego State University
Numerical responses of omnivorous terrestrial arthropods to plant
alternative resources suppress prey populations: a meta-analysis
Dryad
dataset
2021
FOS: Biological sciences
National Science Foundation
https://ror.org/021nxhr62
DGE-1321850
Zuckerman STEM Leadership Program*
Lady Davis Trust*
Minerva Center for Movement Ecology*
University of Alabama
https://ror.org/03xrrjk67
2021-09-19T00:00:00Z
2021-09-19T00:00:00Z
en
155521 bytes
7
CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
Omnivory is ubiquitous in ecological communities. Yet, we lack a consensus
of how plant alternative resources impact the ability of omnivores to
suppress prey populations. Previous work suggests that plant alternative
resources can increase, decrease, or have no effect on the magnitude of
omnivore-prey interactions. This discrepancy may arise from 1) the ability
of omnivores to numerically respond to plant alternative resources and 2)
identity-specific effects of plant alternative resources. We used a
meta-analysis to examine how omnivore numerical responses and the identity
of plant alternative resources affect 1) predation rate by omnivores and
2) omnivore impacts on prey density. Plant alternative resources reduced
omnivore predation rate regardless of identity. The suppression of
predation rate by flowers and flowering plants was magnified when pollen
alone was tested as the alternative resource. Surprisingly, plant
alternative resource availability reduced prey density, suggesting that
omnivore predation increased with plant alternative resources. This
discrepancy (plant alternative resources decreased omnivore predation
rates but also decreased prey density) resulted from experimental
differences in the ability of omnivores to respond numerically to plant
alternative resources. In the presence of plant alternative resources,
allowing omnivore numerical responses decreased prey density, while not
allowing numerical responses increased prey density. Because omnivores
commonly suppress prey density in the presence of plant alternative
resources when numerical responses of omnivores are allowed, the
effectiveness of biological control may depend upon the availability of
such resources and the facilitation of numerical responses.
Literature survey We surveyed the literature using Google Scholar and the
following search terms, (“omnivor*") AND ("consumptive
effects" OR "herbivore interactions" OR “alternative
resources” OR “plant provided foods” OR “non-prey foods” OR “pollen” OR
“flowers”). The search was conducted on 24 October 2019. We used the
preferred reporting practices outlined by PRISMA to structure our overall
literature search (Moher et al. 2009). Our search identified 2,375
potential manuscripts (after duplicate removal). For each potential
manuscript, we read the abstract and determined if the study tested the
interactions between plant alternative resources and omnivorous
terrestrial arthropods. Our initial goal was to include omnivores of all
taxa; however, we obtained few (<5) manuscripts using
non-terrestrial arthropod omnivores (e.g., gastropods) and thus chose to
focus only on terrestrial arthropods for this review. This screening
yielded 426 papers that we read in full to determine if they were eligible
for inclusion in our meta-analysis (see Fig. 1). Studies were deemed
eligible if they included measurements of omnivore top-down effects in the
presence and absence of plant alternative resources. We targeted
manuscripts that measured the effects of plant alternative resources on:
(i) omnivore prey consumption and (ii) prey density (in the presence of
omnivorous predators). Hereafter, these datasets will be referred to as
the “animal prey consumption” and “prey density” dataset, respectively.
Using these criteria, we identified 267 individual studies from 37 papers
to include in our analysis (Fig. 1). Most of the excluded studies tested
the effects of plant alternative resources on omnivore performance (e.g.,
survival, fecundity, and body condition; see for example Eubanks and Denno
1999, Rinehart and Long 2018). Data collection From each paper, we
collected data on omnivore prey consumption and prey density in the
presence and absence of plant alternative resources (Appendix S1: Table
S1). We extracted data from tables, text, and figures (using Web Plot
Digitizer to extract data from figures; Rohatgi 2015). For each relevant
study, we extracted the sample size, mean, and variance (standard error or
standard deviation). Because the mean was not reported for one manuscript
(Robinson et al. 2008), we extracted the sample size, minimum, first
quartile, median, third quartile, and maximum values of prey consumption
for this study. We used this information to estimate the means and
standard deviations for this manuscript’s studies (n = 3, sensu Wan et al.
2014). If manuscripts contained multiple relevant independent studies, we
extracted each individual study. Several of the manuscripts that measured
omnivore impacts on prey density recorded it across multiple,
non-independent timepoints (e.g., repeated measures or timeseries data).
For these studies, we extracted the final timepoint of the dataset for
each relevant study. We chose to use the final timepoint, rather than
using the average across timepoints for three reasons. First, the final
timepoint was the most comparable timepoint across all manuscripts because
it was the only timepoint provided in 65% of manuscripts and 85% of the
individual studies included in our dataset. Second, almost every study in
the animal prey consumption dataset provided only the final timepoint
(accept Choate and Lundgren 2013). Third, we found no effect of timepoint
(final versus time-averaged) on our interpretation of plant alternative
resource effects on prey density (see Appendix S2). This suggests that
despite temporal variation in these data, the final timepoint is
representative of the overall effect of plant alternative resources on
prey density. For each extracted study, we also recorded the 1) plant
alternative resource identity (pollen, flowers, flowering plants, or seeds
and pods), 2) ability of omnivores to display numerical responses, 3)
temporal scale (i.e., days run), 4) experimental spatial scale [i.e.,
replicate size (m2 or m3)], and 5) omnivore taxon. Omnivores were able to
display numerical responses if the experiment 1) allowed omnivorous
predators born outside of the experimental area to freely immigrate into
the experimental area (i.e., no barriers to omnivore dispersal, such as
cages) and 2) contained >1 individual of the omnivore species of
mixed/ undetermined sex or introduced gravid females and allowed offspring
to develop to predatory stages— where they can actively consume animal
prey (i.e., the study did not remove eggs or larvae and ran long enough
for development to occur, see Appendix S1: Table S1) Metanalyses for
effects of plant alternative resources on prey consumption and density. We
conducted our meta-analysis using OPEN MEE software (Build date: 26 July
2016; Wallace et al. 2017). We used both the Hedges’ d and the log
response ratio (hereafter, d and LRR; respectively) to compare the effects
of plant alternative resources (present/absent) on omnivore prey
consumption and prey density (Hedges 1981). We used these two measures of
effect size to increase the robustness of our analysis because d is
sensitive to differences in sample standard deviation and LRR can be
biased by studies with small samples sizes (Osenberg et al. 1997,
Lajeunesse 2003). For both effect sizes, a positive effect size indicates
that plant alternative resources increased the response variable; while a
negative effect size indicates that plant alternative resources decreased
the response variable. In the absence of numerical responses, the effect
sizes of animal prey consumption and prey density should be inversely
correlated— with negative effects on animal prey consumption manifesting
as positive effects on prey density. We used separate meta-analyses
(random-effect models with a Der Simonian-Laird approach) to determine the
overall effect of plant alternative resources on omnivore prey consumption
and prey density. To minimize the effects of small sample sizes, we
excluded covariates (e.g., plant alternative resource identities)
supported by fewer than three separate papers (sensu Rinehart and Hawlena
2020). A synthesis of ecological meta-analyses suggested that three papers
is the minimum number of separate papers that should be included
(Koricheva and Gurevitch 2014). Meta-regressions for the consequences of
experimental methodology on the effect of plant alternative resources on
prey consumption and density We used meta-regressions (random-effect
models with a restricted maximum likelihood approach) to understand the
influence of our extracted covariates (e.g., plant alternative resource
identity, ability of omnivores to display numerical responses,
experimental length, experimental spatial scale [i.e., experimental area
(m2) and volume (m3)], and omnivore taxonomy) on the effect of plant
alternative resources on animal prey consumption and prey density. We
considered extracted covariates eligible for meta-regressions if each
subgroup in the analysis (e.g., pollen vs. flowering plants for plant
alternative resource identity) was supported by at least three separate
papers and five studies (sensu Rinehart and Hawlena 2020). Dataset
variability and publication bias. For all meta-analyses and
meta-regressions, we tested the heterogeneity of our dataset by
calculating both Q (total heterogeneity) and I2 (heterogeneity due to
between-study variance). We tested for potential publication bias by
calculating Kendall’s Rank Correlations (Tb,) between effect size and
pooled variance within each dataset (Begg and Mazumdar 1994). If potential
bias was detected (Tb with p < 0.05), we used funnel plots to
visually identify potential outliers (Begg and Mazumdar 1994, Palmer
1999). Additionally, we calculated the Rosenthal’s fail-safe number, Nfs,
for all significant tests (Rosenthal 1979, Rosenberg 2005). Rosenthal’s
fail-safe number predicts the number of additional studies with neutral
effect sizes (effect size = 0) that would need to be added to the dataset
to lose significance. We classified fail-safe analyses as robust if they
were greater than 5n+10, where n is the number of studies for a given
response variable (Rosenberg 2005). Data extraction method To understand
if the timepoint used in our analysis affected our interpretation of plant
alternative resources effects on omnivore prey consumption and prey
density, we extracted all timepoints from studies using a repeated
measures or timeseries design (i.e., multiple, non-independent
measurements) and time-averaged the extracted data for each study. This
was necessary because several studies in our dataset collected
non-independent data by tracking the same prey populations for several
days to months. Specifically, one study in the prey consumption dataset
used multiple timepoints (Choate and Lundgren 2013), while 40 studies
(56%) in the animal prey density dataset measured prey population density
over multiple time points. Since only a single study in the prey
consumption dataset used this study design, we excluded it from further
time-averaging analyses and focused only on the prey density dataset. To
generate our time-averaged prey density dataset, we extracted the mean
prey density at each timepoint presented in the study. We then calculated
an overall mean and standard deviation for the study using all extracted
timepoint means for a given study. We compared the outcomes of our
time-averaged (described here) and final timepoint (described in main
text) using a meta-regression (random-effect models with a restricted
maximum likelihood approach) which found that data extraction method had
no effect on our interpretation of plant alternative resource effects on
prey density— suggesting that our use of final timepoint does not bias the
findings of our meta-analysis. Additionally, we ran meta-analyses
(random-effect models with a Der Simonian-Laird approach) to compare the
effect of omnivores on prey density. Here, we found further evidence that
extraction timepoint had no effect on our overall conclusions.
For all datasets, Study_ID is the individual study code, while MS_ID is
the code linked to each manuscript (containing multiple Studies).
N_Present, Mean_Present, and SD_Present represent data extracted when
plant alternative resources were availible in the system; while N_Absent,
Mean_Absent, and SD_Absent represent data extracted when plant alternative
resources were not availible to omnivores in the system. Num_Resp denotes
if the study allowed for omnivores to numerically respond (i.e., aggregate
or reproduce). Study_Area and Study_Volume denote the size of experimental
replicates used in each study-- blanks in these columns represent missing
data (data missing due to lack of reporting in individual manuscripts).
Study_Length is the length (in days) that each study was run. d is the
Hedges d effect size calculated for each study, Var (d) is the variation
of the Hedges d calculation. In.Resp.R is the log response ratio
calculated for each study, Var (In Resp.R) is the variation of the log
response ratio calculation. All effect size calculations were preformed
using OpenMEE software (see methods). Any cells filled with na represent
data that was missing from the respective manuscript/study. In the
ExtractionTimpointComparison file, we compared the impact of the timepoint
in the study (final vs. a time-averaged apporach) on our interpretation of
plant alternative resource effects on prey supression. Thus, the
column entitled "Dataset" denotes which approach was taken--
either final timepoint of the study or a calculated time-average across
multiple timepoints in the study. The Data_Summary file outlines the
number of studies extracted from each manuscript included in the
meta-analysis, as well as the Omnivore species, Prey species, Alternative
resource type, and presence/absence of Numerical responses.