Commercially traded fish portfolios mask household utilization of biodiversity in wild food systems

Significance

Wild fisheries have astounding levels of biodiversity, and around the world, people consume and sell very diverse aquatic species. Yet, because accounting for how people use biodiversity is difficult, we often track only commercial species and have only a partial picture of how biodiversity benefits people. In Cambodia’s rice field fisheries, we find ecosystem biodiversity is a key driver of the biodiversity in people’s fish catch, diets, and sold fish—but average species caught and consumed far exceed species sold. Our findings suggest that tracking only commercial species can greatly underestimate the true biodiversity present in wild food systems and, ultimately, the consequences of biodiversity loss for people who rely most strongly on fisheries as a source of nourishment.

Abstract

The global biodiversity that underpins wild food systems—including fisheries—is rapidly declining. Yet, we often have only a limited understanding of how households use and benefit from biodiversity in the ecosystems surrounding them. Explicating these relationships is critical to forestall and mitigate the effects of biodiversity declines on food and nutrition security. Here, we quantify how biodiversity filters from ecosystems to household harvest, consumption, and sale, and how ecological traits and household characteristics shape these relationships. We used a unique, integrated ecological (40 sites, quarterly data collection) and household survey (n = 414, every 2 mo data collection) dataset collected over 3 y in rice field fisheries surrounding Cambodia’s Tonlé Sap, one of Earth’s most productive and diverse freshwater systems. While ecosystem biodiversity was positively associated with household catch, consumption, and sold biodiversity, households consumed an average of 43% of the species present in the ecosystem and sold only 9%. Larger, less nutritious, and more common species were disproportionally represented in portfolios of commercially traded species, while consumed species mirrored catches. The relationship between ecosystem and consumed biodiversity was remarkably consistent across variation in household fishing effort, demographics, and distance to nearest markets. Poorer households also consumed more species, underscoring how wild food systems may most benefit the vulnerable. Our findings amplify concerns about the impacts of biodiversity loss on our global food systems and highlight that utilization of biodiversity for consumption may far exceed what is commercially traded.

More than 25% of globally assessed plant and animal species are threatened with extinction (1). The rapid degradation of Earth’s forests, plains, reefs, and lakes is primarily driven by food production, including overexploitation of fisheries and habitat loss due to agricultural activities (1, 2). The deterioration of biodiversity and the ecosystem services it provides make understanding the role biodiversity plays in global food systems increasingly urgent.

Biodiversity loss is of particular concern for poor households who depend on wild food systems for their food and income. For millions of households, wild foods constitute an important but often poorly quantified segment of the food system (3). The diversity of wild foods accessed is critical because the diversity of diets both across and within food groups is linked to nutrition. Dietary diversity, a measure of diversity across food groups (e.g., eggs, grains), is an established measure that is predictive of nutritional status (4, 5). Biodiversity within food groups has further been linked to improved nutritional adequacy of diets (6–8). Yet, how access to biodiverse natural resources translates to increased household use of that biodiversity to meet food and income needs is unclear. Evidence from agricultural settings suggests that production diversity may have only small effects on consumption diversity, though these relationships vary across settings (9). Further, access to markets plays a key role in mediating the relationship between agricultural production and consumption diversity (10). However, these dynamics are largely opaque within the wild systems that provide a critical food source, support much higher numbers of species compared to agricultural settings, and are experiencing rapid declines in biodiversity.

Fish harvested from Earth’s rivers, lakes, and oceans are among the most important sources of wild food, feeding billions of people worldwide (11). Yet, aquatic biodiversity—and particularly biodiversity within freshwater, inland systems—is rapidly changing (1, 12), which is expected to affect the nutritional status of people that rely heavily on wild fisheries (13, 14). Indeed, fish species vary substantially in their nutritional quality (15, 16), and decreases in dietary species richness can affect micronutrient and fatty acid availability (17, 18). Studies examining the relationship between biodiversity change and nutrient availability are often forced by data limitations to assume that changes in ecosystem biodiversty also reflect changes in access and consumption by people. However, appreciating the role of food system biodiversity requires accounting for how biodiversity available in ecosystems is filtered by households’ choices about what species are caught, consumed, and sold. Such filtering may be particularly complex within food systems that are shifting, for example, through expanding aquaculture production (18, 19) or in response to rising temperatures (20).

Limited data availability similarly positions landing data, meaning tabulations of fish caught at commercial sites, as a simplification for both available fish stocks and fish consumed. Fish landings have been well documented to underestimate stocks (21, 22). This is a particular challenge in data-poor small-scale fisheries (23), in which an estimated 100 million people catch two-thirds of the global fish supply (24). Detailed analyses of household consumption patterns across African fisheries revealed that official catch statistics underreported the quantity of fish consumed by 65.8% (25). While there is a wide diversity of fish species—estimated at more than 31,000 species (26)—a relatively limited number enter markets. Thus, using fish harvest or landing data to represent either the ecosystems where aquatic species originate or the breadth of species consumed on local dinner plates introduces substantial biases.

We examine how biodiversity flows from ecosystems to what people use for consumption and sale (Fig. 1 and SI Appendix). We first examine how the species richness (i.e., number of fish species) available within the ecosystem (approximated by standardized ecological biomonitoring within a regional reserve; see Methods) reflects what households 1a) catch and select for 1b) consumption and 1c) sale (hereafter levels of analysis). Then, we analyze 2) the role of fish ecological characteristics (i.e., body size, nutrient content, commonness) and 3) household characteristics in shaping what households catch, consume, and sell. Together, our analyses illuminate the extent to which natural resource-dependent households use and benefit from biodiversity, and how fish ecological and household characteristics shape these biodiversity relationships.

Fig. 1.

(A) Conceptual diagram of our research approach and questions: How does ecosystem biodiversity shape 1a) catch, 1b) consumption, and 1c) sale biodiversity? How do 2) ecological traits and 3) household characteristics shape biodiversity utilization? (B) Summary results showing the proportion of biodiversity filtering to catch, consumption, and sale.

Our approach integrates ecological and household information within Cambodian rice field fisheries. Fish diversity within this ecosystem is very high, with 135 documented finfish species (27). While engagement in fishing is widespread, livelihoods are relatively diversified and community members around rice field fisheries engage in a suite of activities and often define rice farming as their primary activity (28, 29).

We quantified the overlap between species richness in the ecosystem, household catch, and household sale and consumption using a unique dataset that directly links ecosystem and household data (Fig. 1, Q1a-c). Ecosystem data were collected quarterly from 2012 to 2015 from 40 community fish refuges (CFR) surrounding the Tonlé Sap lake in Cambodia by WorldFish (Methods and ref. 30). CFR are managed, protected water bodies that are designed to increase fishery productivity in rice fields. By design, the CFR are highly connected through inlets/outlets, streams, and channels to the surrounding rice fields and associated ecosystems; many of the species within the system are also migratory (30–32). We elected to sample these refuges, which provide an unfished area as an ecosystem reference point, to understand the potential species available within the ecosystem, but note that reserves may be reasonably expected to support a higher number of species than areas modified by fishing pressure (33). Household data on fishing effort, catch, consumption, sale, and other uses of fish come from a panel of 410 households collected bimonthly from 2012 to 2015 (Methods and ref. 20). We aggregate the household panel across time for our analysis to visualize the distribution of species richness (focal variable) at ecosystem, catch, consumption, and sale levels. Ecosystem biodiversity is defined for each household based on biomonitoring of the nearest CFR.

To examine the association between species richness for the ecosystem, catch, consumption, and sale (Fig. 1, Q1a-c), we used multiple approaches. Using pairwise t tests with Bonferroni corrections, we compared the mean number of species across the ecosystem, catch, consumption, and sale levels. Using generalized linear models with a Poisson distribution and log link function, we examined the relationship between the number of species in the ecosystem and catch and the relationship between catch and consumption and sale.

We analyzed how fish characteristics shape household uses of biodiversity to explain patterns of biodiversity filtering (Fig. 1, Q2). To examine the role of fish characteristics, for every species we obtained total length from FishBase (26); used nutrient information from Heilpern et al. (34) to calculate nutrient density based on supply of protein, iron, zinc, calcium, vitamin A, omega-3 fatty acids docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA); and calculated commonness to provide a metric of abundance within the ecosystem. We used generalized linear models with a Poisson distribution and log link function to compare ecological characteristics (size, nutrient density, and commonness) across the ecosystem, catch, consumption, and sale levels, and used pairwise t tests with Bonferroni corrections to compare mean ecological characteristics across levels.

To examine the role of variation across households in shaping biodiversity use, we analyzed the relationship between ecosystem species richness and caught, consumed, and sold species richness (Fig. 1, Q3). We selected variables a priori, including sociodemographics, asset index, fishing effort, and market access (Methods). Using generalized linear models with a Poisson distribution and log link function, we used stepwise addition of covariates to examine the extent to which the addition of control groups altered the core biodiversity relationships and evaluated these using Wald tests. To compare coefficients, we transformed the model outputs to represent the percent change in the number of species such that they represent, for example, the change in the number of species caught for a 1 unit increase in the number of species in the ecosystem (Methods). Together, these analyses resolved not only the ways biodiversity use varies but identified key patterns within ecological and social spaces of this use.

Results & Discussion

Q1: Biodiversity Filters from the Ecosystem to Human Uses.

We find that the biodiversity of households’ catch and consumption increases alongside the biodiversity in the ecosystem (Fig. 2). On average, 43% of biodiversity present in the ecosystem is consumed by households, though only 9% is sold (Fig. 2, Q1a). While biodiversity is sharply filtered from what is present in the ecosystem to the species caught, of those species that are caught, on average a large majority (94%) appear on people’s plates (Q1b) while only a fraction of species (18%) are sold on average by households (Q1c). While our findings about average harvests point to important patterns in resource use, we also find wide use of available biodiversity. Nearly all species in the ecosystem are harvested by at least one household during at least one time point: 93% (114 of 123 species) of species are consumed at least once and 83% (110 species) are sold at least once.

Fig. 2.

(A) Boxplots depict the number of species in ecosystems and household use portfolios, including catch, consumed, and sold. Diamonds represent the mean, and means are significantly different between all levels of analysis (paired t test with Bonferroni correction). The color gradient represents high (yellow) to low (blue) species richness within the ecosystem (represented by community fish refuges), with colors tracking through to the households surrounding each ecosystem within catch, consumed, and sold portfolios. Scatterplots (B–D) show the bidirectional relationships between each level of analysis and are assessed with generalized linear models with a Poisson distribution and log link function.

Critically, these findings underscore that relying on representations of fisheries’ value and use through commercial harvests or market data could hugely misestimate the extent of households’ use of biodiversity. For example, the United Nations Food and Agriculture Organization highlights 20 commercial freshwater species in Cambodia (35), 16 of which appear in our data. These species represent common species caught and consumed by households and comprise a substantial 51% of catch by households in our data by weight. Within this highly biodiverse system, however, these top commercial species represent only a small subset of ecosystem-level biodiversity: nearly seven times more species are caught and consumed by households. Further, 35% of households consume but never sell fish (27), underlining the potential to undercount biodiversity used by households that is not commercially traded. Considering that geographies where people rely heavily on aquatic foods are highly diverse (36), underestimations of biodiversity used are likely widespread.

Q2: Filtering of Biodiversity for Human Uses Integrates Ecological Traits.

Fish catch is fundamentally a function of a combination of species availability and the choices of households. Households elect to fish, choose their fishing grounds, and select fishing gears, many of which are highly specialized. Similarly, the choice of which species to consume and sell is dependent on a range of factors, including not only ecological traits but also availability, preferences, prices, and market access.

We found that ecological traits—including body size, nutrient profiles, and size—are associated with how biodiversity flows from the ecosystem level to what households choose to consume and sell. On average, households sell species that are larger and more common and consume more nutrient-dense species (Fig. 3 and SI Appendix, Table S2, Q2). The portfolios of species that households catch and those they consume tend to be larger, less nutrient-dense, and less common than the portfolio of species found in the ecosystem (Fig. 3). In comparison to the caught and consumed species, species that are sold are more common, larger, and less nutrient dense than the portfolio of species found in the ecosystem (Fig. 3). Variation in sold species’ ecological traits was higher, potentially due to smaller samples of sold species and selling fish being opportunistic and focused on surplus and especially concentrated in the most productive fishing seasons. While our findings are aggregated to show overall patterns, within this highly seasonal flood-pulse system, the sale of fish is most common in seasons when fishing is most productive (27), and availability of common species may thus drive both higher catches and choices of which species to sell. Further, even when households do not directly use the entire constellation of species available in the ecosystem, the full suite of ecosystem biodiversity underpins aquatic food web functioning and contributes indirectly to the availability of consumed species.

Fig. 3.

Boxplots depict the (A) mean body size, (B) nutrient density score, and (C) commonness index of portfolios of fish species in the ecosystem, catch, consumed, and sold species. The means (represented by white diamonds) are significantly different between all groups except catch and consumed for biomass and nutrient density, and for ecosystem and sold for biomass and commonness (paired t tests with Bonferroni correction).

These findings carry two implications for fisheries sustainability and diets. First, larger species, which are disproportionally represented in sold species portfolios and tend to fetch higher prices, are more vulnerable to overexploitation. Indeed large species are declining in Tonle Sap Lake, even as small species catches have remained stable or increased (37). Targeting of large species and their continued decline could lead to changes in ecosystems and the sustainability of the same fisheries that most support livelihoods and food security. Second, a potentially incidental outcome of selecting large species for sale is that portfolios of consumed species tend to be more nutrient-dense than those that are sold. Pairwise correlations of species traits underline this finding (SI Appendix, Table S3), and similar patterns in the multifaceted relationship between smaller fish, higher nutrient density, and lower prices are observed in global settings (38). The implication of this finding, however, is that relying on commercial information might not only underestimate the biodiversity within food systems but also the nutritional contribution fisheries make to people’s diets. However, the nutrient density of consumed portfolios is lower than that available within ecosystems, underscoring that higher quality diets could be achieved if available species were used in a way that better reflects ecosystem-level patterns.

The nutritional value and potential of drawing from a diverse portfolio of aquatic species is increasingly recognized as a means to advance food security (13, 16, 39). While we observe current patterns of ecological traits associated sold and consumed fish, these patterns are likely driven not only by available biodiversity but also fishers’ values and preferences, and the impact of declining biodiversity on the ecological traits of fish that are consumed and sold remains unclear. Within settings like inland Cambodia where people consume many species, it is critical that fisheries, public health, and conservation policies broaden focus beyond fish quantity to the biodiversity within fisheries to monitor these patterns, especially as aquatic ecosystems are transformed by global change.

Q3: Relationship between Ecosystem Biodiversity and Use of Biodiversity Is Consistent across Household Characteristics.

Integrating household information into our analysis allows us to examine three key dynamics: whether the addition of household characteristics to our models affects the biodiversity relationships we observe, the relative roles of household characteristics and environmental context in driving biodiversity use by households, and whether specific household characteristics affect biodiversity use by households.

The positive relationship between ecosystem biodiversity and household use of biodiversity remained remarkably consistent, even when accounting for variation across households (Fig. 4). After controlling for fishing effort, household characteristics, and market access, the core relationships between biodiversity across levels of analysis remain significant and relatively stable (Fig. 4, Q3), highlighting the importance of ecosystem-level biodiversity in influencing use of biodiversity across households. Ecosystem species richness is associated with caught, consumed, and sold species richness, and catch biodiversity has even larger relationships with consumed and sold species richness. Effort also played an expected role in explaining catch biodiversity and consumption biodiversity (SI Appendix, Tables S4–S8). Effort drives additional sampling of ecosystems, which is likely to increase biodiversity harvest, particularly within fisheries with low selectivity or that are “indiscriminate” as is the case in the Tonlé Sap and other freshwater ecosystems (37, 40).

Fig. 4.

Regression model coefficients for the relationship between (A) ecosystem species richness and B) catch nutrient density and each of the levels of analysis (catch, consumed, sold). Additive models depict relationships with no controls, harvest effort, household characteristics, and market access.

These findings underscore the role of the ecological food environment in driving household use of biodiversity. The concordance between the biodiversity in catch and diets, and the overwhelming role of catch biodiversity in predicting consumption patterns is a striking recognition of the importance of access to fishery biodiversity to provide a diverse portfolio of fish and nutrients to regional diets. We find strong relationships between ecosystem biodiversity and consumption biodiversity compared to what has been observed for biodiversity produced and consumed in agricultural settings (9). The ecological food environment may have important differences that underlie these patterns, such as offering shorter time horizons of food availability relative to agricultural growing seasons or availing culturally appropriate, preferred aquatic foods. That our models do not explain the diversity of species sold nearly as well suggests the importance of other factors (e.g., price, preferences) in driving selection of households’ choices about species to sell. This finding underlines the role of ecological traits discussed above. Further, the global focus on the quantity of fish harvested and those aquatic species involved in commercial trade means that such statistics may not comprehensively reflect the diversity of species caught, consumed, or present in the ecosystem.

Although household amenities and livelihood indices explained a limited amount of the aggregate biodiversity at each level, we also consistently find that poorer households make more use of regional biodiversity. Holding constant the biodiversity in ecosystems and fishing effort, poorer households both catch and consume more species. Natural resources may serve as a “safety net” for vulnerable households (41–43), and this finding suggests that biodiversity could be of particular value for poor households, even if the relationship is driven not by a preference for a biodiverse harvest but that poorer households use more indiscriminate fishing methods. The biodiversity of aquatic foods may thus underpin the supply of a nutrient-dense portfolio of diverse aquatic species for the poorest households around Cambodia’s rice field fisheries. This finding is particularly noteworthy given global patterns of increasing homogeneity in the food system (44) and further underlines the utility of situating aquatic ecosystems as food environments (45).

Our analyses harness a unique and integrated social–ecological dataset to examine how biodiversity filters from ecosystems to household uses, but these analyses have some limitations. First, we use CFR to represent ecological biodiversity, but these are managed protected areas and may not be fully representative of the diverse array of species that households access from a wider range of regional ecosystems, or, conversely, may house higher levels of biodiversity than is available in the surrounding ecosystems that are modified by fishing pressure. Second, the harvest of aquatic foods could affect biodiversity within regional ecosystems, and we are not able to disentangle these complex feedbacks which likely play a role in this system. Third, our analyses are aggregated across seasons, but the Tonlé Sap exhibits strong seasonality, creating a compelling research need to understand whether patterns of biodiversity use will change with ongoing shifts in flood-pulse and climate (46, 47). Fourth, fish processing is common in Cambodia (48) but our data do not capture the ultimate fate of these products as consumed or sold, precluding their inclusion in our analysis. Fifth, though we observe relatively stable coefficients, the ordering of additions in the stepwise regression can impact the outcome. Finally, the households sampled within this study owned rice fields; this inclusion criteria limit generalizability about our findings to poorer households within these communities, notably landless households, and more broadly to those reliant directly on the Tonlé Sap or Mekong Rivers.

Conclusion

Natural resource-dependent households rely on surrounding biodiversity for their food and income. Explicating the ways households use biodiversity is critical to appreciating the true value of diverse ecosystems, the myriad roles biodiversity plays for people, and how food systems will shift in the face of the global biodiversity crises. The rich aquatic biodiversity of the Mekong basin faces formidable threats and has a highly dependent population of resource users (49). While our analyses focus on a single setting, the natural resource dependence is mirrored in small-scale fishing communities around the world and paralleled in a range of wild food environments (forest use, wild meat harvest).

Within data-limited systems like Cambodia’s rice field fisheries, market, or landing, data are often used to stand in for the biodiversity of fish in the system and their use by households. Yet, our findings suggest that many consumed species are rarely sold by households. Consumed species also tend to be smaller and more nutritious than those that are sold, rendering invisible the extent of biodiversity used and thereby the true contribution fisheries make to households’ food security and nutrition.

As global environmental change is progressing amid broad demands on freshwater for energy production, irrigation, and industrial uses, the values assigned to different resource functions are critically important in balancing these competing demands. Misestimating the value of aquatic biodiversity to people or how they use it could have grave consequences for the households that depend directly on biodiversity.

Methods

Data.

Four sources of data contributed to this analysis (for full details, see refs. 20 and 30):

• a)Natural system data collected at 40 CFR every 3 mo over 2 y (13 time points). These data were collected by WorldFish and partner non-governmental organizations between November 2012 and November 2015.
• b)Household fish catch and consumption data collected from 414 households every 2 mo over 3 y (19 time points). These data were collected by WorldFish and partner non-governmental organizations between November 2012 and November 2015.
• c)Household characteristic data collected from 640 households in 2012 and 2015 (two time points). These data were collected by WorldFish and partner non-governmental organizations. We use only the data from 2012 that overlaps with household fish catch and consumption data (n = 410).
• d)Additional species trait information was drawn from FishBase and Heilpern et al. (34). We focused on six nutrients that are central to children’s health and development and are often derived from fish: protein, iron, zinc, calcium, vitamin A, and omega 3-fatty acids.

Analysis.

Methods are summarized below; please see SI Appendix, Supplemental Methods for further details.

Q1 Biodiversity Filters through the System: To examine how biodiversity filters, we aggregated data across time and used pairwise t tests to determine statistical differences in means between the species richness of ecosystem, catch, consumed, and sold portfolios, with Bonferroni corrections for multiple hypotheses. We repeated this process separately with Shannon and Simpson indices (SI Appendix, Supplemental Results). We used generalized linear models with a Poisson distribution and log link function to examine the relationship between the number of species across ecosystem, catch, consumption, and sale levels.

Q2 Ecological Traits: For each level (i.e., ecosystem, catch, consumption, sold), we computed the following ecological characteristics: mean body size, nutrient density score, and mean commonness index. We provide pairwise correlations of these variables (SI Appendix, Supplemental Results).

Portfolio mean body size (B^l,j) was estimated as for each scale l (i.e., ecosystem, caught, consumed, sold), associated with each CFR, j, using each species total length, L_s, as

$$B^{l,j}={\displaystyle \sum }_{s=1}^n(p_s^{l,j}*L_s).$$

Abundance-weighted mean commonness index was estimated as

$$C^{l,j}={\displaystyle \sum _{s=1}^n}(p_s^{CFR,j}*p_s^{l,j}),$$

where $C^{l,j}$ is the mean commonness index at scale l (i.e., ecosystem, caught, consumed, sold), associated with CFR j, and $p_s^{l,j}$ is the relative abundance of species s in scalelassociated with CFR j. This mean commonness index sets a baseline at the ecosystem scale (i.e., for the CFR). When values are lower than the baseline, species portfolios are represented by less common species, whereas when higher, portfolios are composed of more common species.

Portfolio nutrient density, N, was estimated using species-specific nutrient content information ($N_{v,s}$) from Heilpern et al. (34), which indicates the amount of each of nutrient (protein, iron, zinc, calcium, vitamin A, omega-3 fatty acids), $v$, in 100 g of a given fish species s. Using portfolio-specific relative abundance (e.g., $p_s^{l,j}$), we calculate the nutrient content of 100 g of each portfolio as

$$N{C}_{\nu }^{l,j}=\left\{\begin{array}{cc}1,\hfill & if\left(N_{v,s}*p_s^{l,j}\right)>RD{A}_v\\ \frac{N_{v,s}*p_s^{l,j}}{RD{A}_v},& otherwise\end{array}.\right.$$

We used the United States Department of Agriculture Recommended Daily Allowance (RDA) for a child under five for each nutrient (${\mathit{RDA}}_v$) as the threshold for adequacy for a given nutrient. We calculated the Nutrient Density Score, ND, which is the sum of portfolio nutrient content across all nutrients.

Statistical Models: We used pairwise t tests to determine statistical differences in means between the ecological traits (body size, commonness, nutrient density) of ecosystem, catch, consumed and sold portfolios, with Bonferroni corrections for multiple hypotheses. To examine the relationship of ecological traits across ecosystem, catch, consumption, and sale biodiversity levels, we used generalized linear models with a Poisson distribution and log link function for species richness and a Gaussian distribution and log function for other ecological traits.

Q3 Household Characteristics: Data included fishing effort (defined as mean number of person-days spent fishing in the prior 7 d), household size, household dependency ratio (defined as the share of household members <16 or >65), maximum educational attainment by any household member, household amenities (e.g., building materials, water access) index, household livelihood asset index, and market access (defined as the distance to the nearest provincial capital: Battambang, Pursat, Siem Reap, or Kampong Thom).

Statistical Models: To understand the role that household characteristics play in the ways biodiversity filters from ecosystems to household uses (Q3), we examined the role of variation across households using generalized linear models with a Poisson distribution and log link function.

In our models, we added groups of controls in a stepwise fashion and assessed the extent to which the core diversity relationships change with the addition of control groups using Wald tests. generalized linear models cluster standard errors at the CFR level, thus adjusting standard errors to account for correlation between households associated with the same CFR (50). Guided by a priori covariate selection, we included the following covariates in our models: effort, effort squared, household size, dependency share, maximum educational attainment, market access, principle component analysis of a household amenities index (51), and principle component analysis of a livelihood asset index (51). The full model is

$$\mathit{ln}\left(E\left[R_i^{\mathit{catch}}\right]\right)=\alpha +{\rho }_1{R}_j^{\mathit{CFR}}+{{\rho }_{2}E}_i+{\rho }_3{D}_i+{\rho }_4{M}_j R_i^{\mathit{catch}}\sim \mathrm{P}\mathrm{o}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{o}\mathrm{n},$$

where ${\rho }_1$ represents the relationship between CFR-level species richness and catch-level species richness, controlling for effort, $E_i$, household characteristics, $D_i$, and market access, $M_j$. We use Wald tests to statistically test the differences between key coefficients. We interpret the coefficient as predicting percent change in outcome variable given a one unit change in predictor variable as $\left(\%\mathrm{change}\right)=100*(e^{{\beta }_1}-1)$.  In other words, the typical change in outcome for a one unit change in predictor is $(\mathrm{typical}\mathrm{change}\mathrm{in}\mathrm{outcome})=(e^{{\beta }_1}-1)*{\overline{R}}_j^{\mathit{catch}}.$

Models and Wald tests are repeated with $R_i^{\mathit{cons}}$ and $R_i^{\mathit{sold}}$ as outcomes, and $R_i^{\mathit{catch}}$ as the key predictor. We also repeated these analyses with OLS models and used Shapley-Owen decompositions to assess the variance explained by each component of the models. See SI Appendix, Supplemental Methods and Results for further details.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Household and ecological data are available via the CGIAR open data access policy (52).