Searches

These databases were selected because Scopus is the largest abstract and citation database for peer-reviewed literature. CAB Abstract indexes publications from 150 countries and is among the more comprehensive title and abstract databases for applied life sciences and the environment. AGRIS [32] and AGRICOLA [33] are smaller bibliographic databases that often have unique holdings relevant to the environment. AGRIS and AGRICOLA use basic search strings and hence were not included in the search string development. Several social science databases were considered but ultimately excluded including Education Resources Information Center (ERIC), the social science research network (SSRN), and EBSCO’s social science abstracts. Social science focused databases are excellent for targeted research on a specific topic, but Scopus’ 22,000 titles and CAB abstracts 8.1 million records cover much of the same literature as the social science databases and facilitate rapid searches.

Within each database, the results were modified by Boolean operators, wildcards, and limited to relevant subject areas. Searches were undertaken in English and were not restricted by publication date except for CAB abstracts, which begins with 1973 data. Descriptions of the final searches conducted are included in ‘Additional file 1: evolution of search strings’.

Google Scholar [34] was searched using simple search terms such as ‘gender AND resource management’. Where the search produced more than 200 hits, the first 75 were screened. Otherwise all hits were screened. Searches were also conducted in Google [35] using the same search terms and approach.

We invited the gender experts at the International Union for the Conservation of Nature (IUCN), Conservation International (CI), and World Wildlife Fund (WWF-US), the 1000+ members of the Poverty and Conservation Learning Group (an international network convened by the International Institute for Environment and Development) [36], and the 40+ members of the Gender and Environment Working Group [37] to submit relevant documents and articles, and added those that met the inclusion criteria but were not already in the search results. We also ‘snowballed’ the references of included articles to identify additional relevant articles. Finally, we screened the bibliographies of relevant systematic reviews published by Environmental Evidence [38], 3ie [39], and the Campbell Collaboration [40]. There were no restrictions based on publication status.

Study inclusion criteria

A Kappa analysis was performed on a random sample of 100 titles and abstracts to measure inter-rater agreement in applying the inclusion criteria. Because there were three reviewers, Fleiss’ test [41] was used. After three rounds of Kappa analysis, the inter-rater agreement was above 0.6 at both the title and abstract levels.

The reviewers used a three-stage sequential screening process. We used the screening tool ‘Abstrackr’ to screen the title and then the abstract [42]. We did not use Abstrackr’s semi-automated screening function, though it shows promise for substantially reducing the time needed to screen titles and abstracts [43]. At the title and abstract levels, documents were assessed independently, with each reviewer assigned a portion of the literature. Where there was doubt about whether or not an article met the inclusion criteria, it was retained for assessment at the next stage. After the abstract screening, the full texts of included articles were located, added to an Endnote library, [44] and screened using the inclusion criteria. Full-text electronic or paper copies of the articles and books were obtained.

Potential effect modifiers

In the full-text articles and book chapters, we noted if there was mention of potential ‘effect modifiers’ beyond gender, such as landlessness in the community or within the management group, age of participants, age of resource management group, original state of the resource before community protection began, cultural factors, ethnicity/race, caste, wealth, class, occupation, education level of household head, high levels of political and economic inequality in a community, and ‘other’. These were included in the study coding, as described below.

Critical appraisal

We used the Maryland scientific methods scale to categorize quantitative social science studies likely to have high or low internal validity based on the study design [45]. For each quantitative study and the quantitative portion of mixed-method studies, the design was rated from one to five as per the Maryland scientific methods scale, and if the rating was four or above, it was given a ‘high’ rating for study design. Qualitative studies were rated using the critical appraisal skills programme (CASP) qualitative research checklist [46] which uses ten screening questions to assess quality. Studies that used mixed methods were classified as either predominantly quantitative or primarily qualitative for rating purposes.

Study coding strategy

All data were double extracted by two researchers working independently and then reconciled [47] using Microsoft Excel [48].

Descriptive results

After duplicates were removed, we screened 11,069 records by title and abstract. Common terms in the search string such as ‘gender’ and ‘resource management’ led to a high percentage of irrelevant results such as animal and plant studies. Hence, 10,936 records were excluded at the title or abstract level. Across all sources, we identified 133 studies for full-text review (Fig. 1). We were unable to locate three of the included studies. None of the three is likely to be relevant given that one is an introduction to a book, one is a conference report from a fisheries society meeting, and the other is a newsletter. Ultimately, 17 studies met the inclusion criteria. Of these, nine are peer-reviewed journal articles, four are book chapters, three are conference papers, and one is a thesis. All articles were less than 10 years old, and ten were from the last 10 years. ‘Additional file 2: full-text review results’ provides the references for the 17 included studies and the 116 excluded studies along with the reasons for exclusion.

Of the included studies, four were primarily qualitative [49–52] and 13 were primarily quantitative [8, 17, 18, 53–62]. Of the quantitative studies, the unit of analysis for all but three was the group, and the average sample size was 132 groups. Two studies used households as the unit of analysis [57, 58], and one used fish ponds (‘beels’) [59].

Geographically, the included studies are concentrated in South Asia. Twelve of the 17 studies are from India and/or Nepal. Latin America, East Africa, and southeast Asia had five studies in total.

We identified 14 forest articles and three fisheries articles (Fig. 2). Within fisheries, two studies addressed freshwater fisheries and one dealt with marine fisheries. ‘Additional file 3: study coding results’ gives the geographic and categorical data on each included study.

All 17 included studies identified improvements in local natural resource governance when women participated in the management of the resources, and three studies identified resource conservation benefits. Stricter rules (while still allowing for sustainable extraction), greater compliance with rules, greater transparency and accountability, and better conflict resolution were among the reasons for improvements. Additional file 4 summarizes the key findings from each included study.

Differences in the included studies

The most obvious differences among the studies lie in their design. Among the quantitative studies, seven use regression models with governance indicators as dependent variables and gender composition of groups as the explanatory variable [17, 18, 53–56, 59]. Among the remaining quantitative studies, the designs comprise before-after, control-impact quasi-experimental designs from a single site [57, 58], a before-after, control-impact quasi-experimental design across multiple sites [8], a before-after comparison [61], a control-impact comparison [60], and a frequency distributions of response for women’s, men’s, and mixed-gender groups with a least significant difference test for group averages [62]. For the qualitative studies, two use, inter alia, key informant interviews, focus-group discussions, and participant observations [51, 52], and one uses key informant interviews, focus group discussions, and secondary information [49]. The fourth uses ‘in-depth interviews and discussions with men and women separately’ [50].

Using the Maryland scientific methods scale to assess internal validity based on study design, 11 of the included quantitative studies were rated as having low internal validity given their study designs. One was rated as high because it measures before and after variables in multiple impact and control sites [8], and two studies had designs not covered by the Maryland scientific methods scale: one compared women-headed versus men-headed joint forest management committees [60] and a second compared, inter alia, frequency distributions of response for women’s, men’s, and mixed-gender groups [62]. For the four qualitative studies assessed with the CASP qualitative research checklist, one was high quality (8 out of 10 possible points) [52], one was medium (5 out of 10 possible points) [51], and two were low quality (<5 possible points) [49, 50].

Factors other than the gender compositions of resource management groups can also influence observed outcomes, and another difference among the included studies is which additional negative or positive influences on the outcomes are highlighted. The most frequently mentioned influencer is landlessness. Eight studies note that landlessness may make people more dependent on resources and less compliant with resource regulations [8, 17, 18, 50, 53, 54, 57, 58], and one study notes that including landless women in the governance structure increased regulatory compliance suggesting that governance inclusivity of the landless has benefits [17]. Given that the majority of the included studies were in India and Nepal, not surprisingly caste is a commonly mentioned influencer. For example, having Brahmins, the highest caste, present in a management group may benefit forest protection because they tend to wield traditional authority [8]. Yet a management group in a multi-caste community that is dominated by a high caste can also result in elite capture of the benefits [50]. Four studies highlight wealth among group members as an influencer that often improves resource governance and conservation outcomes [49, 50, 56, 57]. Two studies find differing influences from wealth [17, 18]. Education level of the household head is mentioned by two studies as both a negative and positive outcome influencer [56, 60], and two studies mentions high levels of political and/or economic inequality in a community as a negative influence [8, 56]. To reduce the influence of these factors, several studies use regressions models that control for the impact of factors other than gender and address the issue of omitted-variable bias [8, 17, 18, 53, 54].

Gaps in the available evidence

Forest management is comparatively well studied with 12 of the 17 included articles covering the topic, but most other common forest and fishery interventions are missing from the assembled literature. There are, for example, no included studies that cover reducing emissions from deforestation and forest degradation (REDD), reforestation, or fisheries co-management, and there is only one looking at fisheries protection [61], and two looking at the regulation of fishing [51, 59].

Limitations

The English-language focus may have caused us to miss relevant studies and thus underestimate the available evidence. Two non-English-language studies with English abstracts suggesting relevance were identified during the search (‘Additional file 2: full-text review results’). On bias risk, we found no studies that mentioned negative resource governance or conservation effects from including women in a resources management group. Publication bias in which positive or statistically significant results are more likely to be published could be an issue. Our search strategy aimed to minimise publication bias by including a comprehensive search of both the non-academic and academic literature, and 35 % (n = 6) of the included studies were identified from non-academic sources. This, however, was unlikely to overcome the publication bias towards negative results.

Policy and programme implications

The policy and programme implications beyond South Asia are hindered because the evidence from other regions is limited. At the same time, the conceptual framework provided by existing studies provides a clear case for gathering robust evidence from a diversity of regions on the impact from varying gender compositions of resource management groups.

Research implications