Performance responses from stem-meta-analysis

We searched for experimental and observational studies that investigated effects of insect stem herbivores on plant performance, including studies that compared plant performance at sites where the stem herbivore was present to plant performance at sites where the stem herbivore was absent. Only woody plants were considered. All types of attack to living stems were considered including sap-suckers, bark beetles, gallers, stem borers, bark chewers. We included ovipositional damage and species that attacked other parts of the plant in addition to the stem (often in different life-history phases). We also included studies where herbivory was simulated to mimic attack by a particular species of insect herbivore. Studies that simulated attack without specifying a herbivore model or where the model was a mammal were excluded. The studies that we used compared woody plant survival, growth, reproduction or photosynthesis on plants (or parts of plants) with and without stem herbivore attack. We also recorded changes in allometry, phenology, defence and carbon and nitrogen chemistry. Google Scholar was used to locate studies. Only studies in English were included and our final search for studies was in late April 2014. We searched for the terms "herbivor-" and "damage" with a range of possible terms for different parts of the plant stem. In addition, we searched for each of the major insect families that attack plant stems together with the terms "herbivor-" and "damage". A full list of search terms is provided in the Supplementary Material. Reference sections of papers were also scanned for additional studies. Data were obtained from text or tables in the paper or, where necessary, by extraction from graphs using Plot Digitizer (Huwaldt 2010). We recorded the mean and total sample size for each variable measured. Where multiple levels of herbivory were assessed, we used only the most severe treatment. This might be the treatment that involved the most insects, removed most biomass, or exposed plants to herbivores for the longest period of time. If impacts of additional variables (e.g. drought, nutrients) were assessed in conjunction with herbivory, the control of this treatment was taken. If different sites were presented, we calculated the mean across sites for each response variable and if the same response was measured multiple times, we used either the final measurement (e.g. seed production in the final year) or the peak measurement (e.g. photosynthesis rate at solar noon). We removed all response variables where the number of replicates was less than three for either the treatment or the control. We categorised response variables into performance variables and ‘other’ variables. Performance variables were then further categorised into vegetative performance (e.g. height, number of branches), reproductive performance (e.g. number of seeds), photosynthetic performance and survival (e.g. number of dead branches or plants). Almost all of the analyses were done using the performance variables only. 1. Lateral vs. leader responses Where possible, we attributed responses to either the leader or a lateral stem to determine the importance of compensatory growth. Leader responses included plant height and length of terminal shoot, while lateral responses included number of branches, length of lateral shoots. Most responses could not be categorised as either leader or lateral. 2. Plant attributes Plant attributes or traits may predict their susceptibility to stem attack. We considered whether plants in different classes (Gymnosperm vs. Angiosperm), different life-forms (evergreen vs. deciduous), different functional groups (trees vs. shrubs vs. lianas) and at different ontogenetic stages (adults vs. juveniles) were differently affected by stem attack. We also considered the effect of plant family, following taxonomy from the USDA plant name database (www.plants.usda.gov) and NSW online flora (http://plantnet.rbgsyd.nsw.gov.au). 3. Insect attributes The impact on plant performance may depend on attributes or traits of the insect. We asked if the overall impact on plant performance varied with location of attack (growing tip, bark or wood), insect order, or feeding guild. Insect feeding guilds were categorised as 1. bark beetles; 2. borers or miners; 3. bark removers (through either feeding or oviposition damage); 4. apical tip/bud clipping (typically in simulated herbivory studies); 5. gallers and; 6. sap-suckers. Each species was assigned to only one group. 4. Study procedures We recorded 1. type of study (experimental vs. observational); 2. unit of replication (individual plants or groups of plants vs. comparisons within a plant vs. between sites); 3. herbivore manipulation method (added vs. removed vs. simulated vs. observation); 4. study location (field vs. greenhouse vs. common garden); 5. length of study (snapshot vs. <= 1 growing season vs. > 1 growing season); and 6. latitude of the field site or common garden (greenhouse studies were excluded). We also assessed whether the plant was considered a forestry or crop species with the insect as a pest, a weed with insect as an actual or potential biological control agent, or without economic significance (= ‘native’). Researchers in applied systems may be more likely to focus on species with conspicuous effects, a potential source of research bias (Gurevitch & Hedges 1999). As economic status can vary, we followed the categorisation of the original study.