For floral resource specialization, for each pollinator species in our data, we calculated the metric d 0 , which measures the deviation of the observed interaction frequency from a null model in which all partners interact in proportion to their abundances ; thus, it is not confounded with abundance as is linkage . It ranges from 0 for generalist species to 1 for specialist species. Body size metrics and abundance were log-transformed. For syrphid flies, larval diet is entirely distinct from adult floral resource use; thus, larval diet type and d 0 provide non-overlapping information. However, for bees, measurements of d 0 include floral visits both for pollen to provision larvae and for nectar and pollen for adult food, reflecting both larval and adult diet breadth. We therefore used only d 0 and not assessments of lecty classes , since these traits would constitute overlapping measurements. Since d 0 is measured from our network data, it is available for all of our bee species, whereas data on lecty are poor or absent for a number of our species. We were able to measure or obtain all traits for 80 of 97 bee species in our data set and for 26 of 30 syrphid fly species .To evaluate the effect of habitat restoration over time on bee communities and traits, we model species occurrence data as a function of the number of years post-restoration for a particular site in a particular year. ypr values for restoration sites begin at 0 and increase each year following restoration, but remain at 0 for controls in all years. Thus,raspberry plant container sites restored in 2007 have a value of ypr = 0 in 2006 and 2007 and a value of 6 in 2013. Use of the continuous ypr variable permits more flexibility in analyses then a classic before–after coding scheme.

The before–after coding is better suited for analysing a pulse disturbance, whereas we studied a press disturbance . Further, since different sites were restored in different years, the ypr variable permits us to isolate changes associated with restoration from annual fluctuations in insect population dynamics. Bee and syrphid fly data sets were analysed separately. In order to maximize the number of species that could be included in analyses, we first analysed each trait separately and then considered the subset of species with full trait data in a multitrait analysis. All quantitative traits were centred and scaled /2 SD to facilitate comparison of effect sizes . All analyses were conducted in R v. 3.1.1. using ‘LME4’ . For single-trait analyses, we used generalized linear mixed effect models with a binomial error and a logit link function to model species occurrences for each site and date, with ypr , a trait and the interaction between ypr and that trait as fixed effects. We were specifically interested in this interaction because, for a given trait, a significant interaction indicates that restoration differentially affects species differing in that trait. Site, species and year were all included in each analysis as random effects. Using Akaike information criterion values, we compared each single-trait model to a ‘no-trait’ model based on the same species set , constructed as before but with only ypr included as a fixed effect. Comparison of these two AIC values enabled us to assess whether the trait or its interaction with ypr contributed substantially to the model. We considered models with DAIC ≤4 to be equivalent . Using the same basic model structure, we also constructed multitrait models using the subset of species for which we had a complete set of trait values. Here, we included each trait and an interaction between that trait and ypr in a single model, with species, site and year as random effects, as above. The advantage of including all traits within the same model is that one can assess the relative importance of each trait while also accounting for their combined effects.

However, since functional traits are intercorrelated , we used variance inflation factors , calculated using the AED package to remove collinear variables from the model. We successively removed the covariate with the largest VIF exceeding 3 and recalculated VIFs until all VIFs were <3 , following Zuur et al. . This covariate set was then used in the multitrait model. By combining data from all of our species into a single analysis and including species identity as a random effect, we were able to accomplish our goal of making inferences at the community level. While some species occurred infrequently in the data, such species only exert a small influence on the estimation of effect sizes. Analyses with infrequent species removed produced similar results to analyses including all species, except for lack of convergence in one of the 12 analyses; therefore, we present only the analyses with all species included. Since no species-level phylogeny of our specific taxa yet exists, we could not fully account for potential phylogenetic non-independence in our analyses. However, Bartomeus et al. recently showed that, for bees, nesting species within genus and genus within family as random effects produced essentially the same results as a more sophisticated analysis that accounted for phylogenetic non-independence using generic-level phylogenetic trees created from GenBank sequences. Therefore, we also conducted analyses nesting species within higher-order taxonomy . For all single-trait models, these analyses yielded equivalent out-comes. Therefore, for all analyses, we present only the analyses without taxonomy.If habitat restoration chiefly benefits the common generalists that are able to survive in intensive agricultural landscapes, then we would expect to see increased occurrence of species between restored and control sites, but no increases in the occurrence of the species that are more sensitive to disturbance. In contrast, our results show that hedgerows not only significantly enhanced occurrences of native bee and syrphid fly species but differentially promoted occurrence of species with greater floral specialization, more specialized habitat requirements and smaller body sizes . These results suggest that small-scale habitat restoration within intensive agricultural landscapes has the most positive effects on species whose response traits may render them more vulnerable to habitat degradation, namely more specialized and less mobile species. 

Thus, these plantings may be partially reversing the community disassembly that has occurred in response to agricultural intensification in this region . It is important to note, however, that we did not compare communities at hedgerows with a reference natural or semi-natural community and, therefore, we cannot say to what extent hedgerows promote more specialized or less mobile species relative to the full complement of species from the region. A study on bee functional trait composition in the same biogeographic region found that farms impose strong environmental filters limiting species occurrences relative to semi-natural habitats . This finding, coupled with our finding of enhanced success of cavity nesters with restoration, suggests that providing shrubs and trees on farms is the key to re-establishing the cavity-nesting component of native bee communities. We found support not only for our general hypothesis that habitat enhancements differentially promote species that may be more sensitive to disturbance, but also for some of our specific predictions on response traits. For bees, however, several specific predictions were not borne out. We predicted that hedgerows might differentially promote large-bodied species, based on previous work in this region . Instead, we found either no interaction with body size or that smaller bees were promoted . However, both of the other traits that were promoted by hedgerow maturation,container raspberries cavity nesting and floral specialization were strongly associated with larger body size . These results suggest that, for bees, body size alone may not be an ideal indicator of species responses to small-scale habitat restoration, although it may be correlated or interacted with other traits . Also contrary to our prediction, we did not find that hedgerows differentially supported parasitic bees. Parasitic bees tend to be uncommon in our collections , so it is possible that we are simply unable to detect such a trend, if it occurs, or that insufficient time has elapsed post-restoration for a trophic-level trend to emerge. Finally, we did not find that hedgerows differentially supported less common bee species, although cavity nesting bees tended to be less common , and a previous study in the same area did find greater abundances of less common species at mature hedgerows than at controls . For bees, our principle finding – that hedgerows differentially promote more specialized flower visitors with more specialized nesting requirements – was consistent between single- and multi-trait analyses. The importance of both variables in the multi-trait models was evident even though cavity-nesting bees also were more specialized in floral resource use . For flies, hedgerows differentially promoted more specialized flower visitors, but only the body size effect was consistent between single- and multi-trait analyses. In bees, the main effect of hedgerow maturation became non-significant or marginally significant when traits with significant interactions were included in the single- or multi-trait analyses, suggesting that hedgerows do not promote abundances of bees uniformly, but rather, a subset of bees with specific traits. In flies, the main effect of hedgerow maturation remained significant even when significant interactions were included in the models, suggesting either that our analysis failed to include some key response traits of the fly community, or that hedgerows promote the abundances of all fly species, while promoting species with certain response traits more than others. For both bees and flies, significant interactions between hedgerow status and various response traits emerged between 4 and 5 years post-restoration .

Some evidence suggests that the European Union’s ‘agri-environment schemes’, which subsidize growers to implement small-scale habitat enhancements and other presumed wildlife-friendly farm management techniques, increase species richness and abundance on farms primarily by promoting common and/or resilient species rather than uncommon or endangered species and are effective in simple but not in cleared landscapes . In the United States, Farm Bill conservation programmes are the analogue to the EU’s agri-environment schemes. Several of these programmes, such as the Environmental Quality Incentives Program and the Wildlife Habitat Incentives Program, include specific provisions to promote pollinator conservation through habitat enhancements like native plant hedgerows or insectary strips. Our results suggest that such programmes can promote not just common, resilient species, but also some disturbance sensitive species, even in cleared landscapes. It is important to note, however, that the hedgerow plantings we studied here were specifically designed to support flower visitor communities in the region. Plant palettes were selected using bee–flower network data from the same area to obtain bee-attractive plant species that would provide a sequence of floral resources throughout the flight season. Therefore, the conservation benefits that we observed from farm-scale habitat enhancement in our study area might only be realized in other regions if planting palettes are specifically tailored for the flower visitors found there. Similar conclusions about the need for tailoring agri-environment schemes to specific conservation objectives were reached through assessments of EU agri-environment schemes . Flower-rich patches in intensive agricultural landscapes may simply concentrate existing flower visitors from the surrounding landscape, rather than promote their population growth . Studies of species abundances or occurrences cannot distinguish between concentration vs. population effects, and demographic data instead would be needed. However, several lines of evidence suggest that our results are not simply due to concentration effects. First, on other native plant hedgerows in the same landscape, we observed increases, not decreases, in the abundances of flower visitors in fields immediately adjacent to hedgerows, a pattern consistent with exportation, rather than concentration, of flower visitors from hedgerows . Secondly, in multi-season occupancy analyses of this same data set, we found that, relative to controls, hedgerows enhance rates of persistence and colonization, particularly for more specialized species, suggesting that hedgerow resources promote the establishment of populations at these sites . Restoring habitat for flower visitors in agricultural landscapes might also promote important ecosystem functions and services on adjacent farm fields like pollination and pest control . While some direct evidence supports a positive role of native plant restoration in promoting pest control and crop pollination in adjacent fields , it remains to be determined whether this differential effect of restoration on response traits of flower visitor communities would translate into measurable improvements in ecosystem services.