Introduction

Seafood is a major contributor to global food security and provides critical macronutrients and micronutrients for human nutrition1,2,3,4,5,6,7,8,9. Seafood, which includes aquatic foods from marine and freshwater environments, contributes over 15% of total animal protein consumption globally, and 3.3 billion people worldwide rely on seafood for at least 20% of their animal protein intake10. In developing countries, seafood can be one of the few accessible sources of key micronutrients and animal protein4,5,6. Seafood is also the most internationally traded food group in the world, and this trade results in a modest net flow of seafood from developing countries to developed countries by volume11. Nonetheless, developed countries as a group have a large seafood trade deficit measured in value because developing countries tend to export high-value seafood and import relatively low-value seafood12. This value imbalance raises questions about whether this low-value seafood is less nutritionally dense than the healthy seafood exported by developing countries, such that seafood trade may exacerbate the nutritional deficits of developing countries5,9. Locally sourced fish in developing countries have higher micronutrient content than some major globally traded fish such as tilapia, suggesting that more nutritious fish could be displaced from diets through trade6. Moreover, at the country level, species composition is a more important determinant of nutrient content than total fishery yield5.

Studies find mixed results of the impacts of the seafood trade on development and nutritional security. For example, Béné, Lawton, and Allison13 use aggregate country-level data to show that the fish trade is neither positively nor negatively correlated with development outcomes for African countries. Nash et al.9 examine net nutrient gains and losses from reciprocal trade in wild-caught seafood, using country-level catch data. They use a statistical model that predicts nutrient content at the species level based on other species in the same family5. Nash et al.9 find that 63% of countries experience net gains in nutrients through trade, but of the countries that experience net losses, half of them are small-island developing states and/or African nations. There is limited understanding of whether trade in farmed seafood exhibits similar patterns and the extent to which seafood processing mediates nutrient flows. In the broader food system, international trade increases macronutrient availability and access in low-income countries14. However, there is limited understanding of whether this finding holds within the seafood sector.

The literature on nutritional security also finds that food prices are an essential element of access to nutrition9,15,16, but limited information exists linking seafood prices to nutrient content. Lower seafood prices could not only increase the availability of seafood-based nutrition but also displace less healthy red meat and other animal proteins8. Yet, concerns persist that globalized seafood markets may be driving up prices for low-income consumers, reducing nutrient accessibility15,17,18. Higher prices potentially benefit fishers and fish traders at the expense of fish consumers18,19,20. On the whole, global seafood prices have remained roughly constant over the past several decades21,22, but there is limited understanding of how trade-driven differences in seafood prices between developed and developing countries have affected nutritional affordability.

To assess the impact of the global seafood trade on food and nutritional security, it is necessary to quantify the embodied nutrients in traded seafood products and evaluate the nutritional affordability of these products in developed and developing countries. This assessment presents four research challenges. First, quantifying nutrient density per dollar in traded seafood requires detailed information at the species level, as nutrient content varies widely across species5. Second, embodied nutrients vary by product form and degree of processing, such as a whole fish or a fillet of the same species. We have a limited understanding of the impacts of processing form—e.g., whole, filleted, or meat—on nutritional affordability despite considerable evidence showing that these attributes affect seafood pricing17,23,24,25,26. In general, the nutritional content in seafood is bundled with other seafood product attributes that high-income consumers demand17,23,24,25,26. As a result, the high degree of international trade in seafood raises the theoretical possibility that developing countries are outbid in the market for nutrients17. High-income consumers who are concentrated in developed countries may simply exhibit higher willingness to pay for nutritional attributes than their counterparts in developing countries. Alternatively, high-income consumers may inadvertently draw nutrition away from consumers in developing countries due to their higher valuations of other product attributes that are correlated with nutrition17. Third, assessing nutritional affordability requires an analysis of both wild-caught and farmed seafood. Farmed seafood is roughly half of global seafood supply10 and is an increasingly important component of global food security27,28,29. While a large share of farmed seafood is consumed domestically, there is still substantial international trade29,30. Moreover, farmed and wild-caught seafood markets are often highly integrated and jointly determine consumer prices19,31,32. Fourth, testing hypotheses about nutritional affordability for developing countries requires an operational definition to distinguish between developing and developed countries. Because there are many ways to characterize development status of countries—including categories based on income per capita33, food insecurity34, specific geographies such as small island developing states9, and measures that combine standard of living with health, longevity, and knowledge35—it is important to establish that results about nutritional affordability are robust across reasonable alternative definitions.

Here we analyze prices and nutritional content of internationally traded seafood and assess whether international trade is likely to enhance or erode nutritional affordability for developing countries. Importantly, we find that the global seafood trade may actually increase nutritional affordability for developing countries. To establish this finding, we construct a database that links global trade data containing product information from both wild-caught and farmed seafood with a data set containing detailed nutritional information by fish species and product form. We analyze seafood import prices and nutrients using data from the top 41 exporting countries that together constitute 90% of the global seafood trade. The bilateral trades of these countries contain transactions with 237 unique importing countries and independent territories, 266 unique seafood products, and 267,505 total observations. Recognizing that there are competing definitions and cutoff points for distinguishing between developing and developed countries, we use six different ways of categorizing developed and developing countries (two based strictly on gross domestic product per capita, one using the United Nations Human Development Index, one based on Food and Agriculture Organization’s list of Low-income Food Deficit Countries, and two using World Bank categories), and we compute nutrition per dollar for each of nine different nutrients (three macronutrients and six micronutrients). From here on, developed/developing refers to these alternative definitions, with the particular definition being clear from context. We consistently find that developing countries receive an overall discount on the nutrients analyzed. That is, developing countries receive more nutrition per dollar spent than developed countries. This result holds across all of the micronutrients and macronutrients that we analyze and is robust to the six different ways of dividing developed and developing countries, trimming the sample to exclude potential outliers, excluding observations from major re-exporting countries, and limiting the sample to products with exact one-to-one matches between the trade and nutrition databases. We find that the net exchange of high-value exports from developing countries for low-value imports does not reflect lower quality in the form of lower nutritional content; instead, a substantial portion of the value difference stems from non-nutritional attributes that differ across developed and developing country importers such as type of preservation and product form.

Results

Nutrition per dollar

Our main finding is that developing country importers pay lower prices for nutrition than developed country importers in the global seafood trade. For both macronutrients and micronutrients, and using our first categorization of developed/developing (Dev 1), developing countries receive more nutrition per dollar in seafood imports (Fig. 1, Supplementary Fig. 1). For macronutrients, pooling all years, we find that developing countries receive 55% more calories (Mann–Whitney Z = 66.97, p < 0.0001), 50% more protein (Mann–Whitney Z = 64.96, p < 0.001), and 66% more fatty acid content (Mann–Whitney Z = 51.43, p < 0.001) for the same expenditure compared to developed countries. For micronutrients, we find that developing countries receive 52% more vitamin B-12 (Mann–Whitney Z = 63.59, p < 0.001), 42% more calcium (Mann–Whitney Z = 45.10, p < 0.001), 55% more iron (Mann–Whitney Z = 57.06, p < 0.001), 43% more zinc (Mann–Whitney Z = 56.54, p < 0.001), 49% more potassium (Mann–Whitney Z = 60.62, p < 0.001), and 52% more magnesium (Mann–Whitney Z = 66.10, p < 0.001) for the same expenditure compared to developed countries (Supplementary Data 1). Our main finding also holds when analyzing individual years instead of pooling all years (Fig. 1, Supplementary Fig. 1a, Supplementary Data 1).

Fig. 1: Developing countries pay lower prices for nutrition in the international seafood trade.
Fig. 1: Developing countries pay lower prices for nutrition in the international seafood trade.
Full size image

Percent differences in nutrient concentration per dollar for developed and developing country importers for two macronutrients and two micronutrients. Positive percent differences imply that developing countries receive more nutrition per dollar. Point estimates (filled circles) and 95% confidence intervals (shaded areas) are depicted for each nutrient using the full sample for Dev 1, which divides Gross Domestic Product Per Capita in Purchasing Power Parity (GDPPCPPP) into quintiles, assigns the bottom two quintiles to developing, and assigns the top three quintiles to developed. Blue indicates the full sample. Orange excludes major re-exporters from the sample. Other nutrients and the remaining five development status distinctions show similar patterns, namely that developing country importers receive more nutrients per dollar than developed country importers (Supplementary Fig. 1). All differences are statistically different from zero based on Mann–Whitney tests (Supplementary Data 1).

Qualitative results are the same when we consider alternative definitions for developed/developing countries (Dev 2, Dev 3, Dev 4, Dev 5, and Dev 6) (Supplementary Fig. 1b–f). Using our full sample, Mann–Whitney tests show that for each nutrient developing countries receive more nutrition per dollar for all classifications of development status (Supplementary Data 1). Results are consistent for parametric tests (t tests and volume-weighted t tests) with just a handful of counterexamples for individual nutrients in some years (Supplementary Data 1). Results are also consistent when using trimmed or untrimmed data (See Methods; Supplementary Data 1).

As robustness checks, we assess the potential impact of the quality of match between the trade and nutrition data on our results. We specifically analyze the differences in nutrition per dollar based on two alternative matching criteria: 1) records for which there is an exact one-to-one species match and 2) records that exclude broad composite fish categories such as “Fish; fresh or chilled” and subjective matching to similar species, but include products in the trade data with multiple species listed. The first criterion is most restrictive because it only includes records from the trade data that indicate a single species. The second criterion is less restrictive in that it includes all of the one-to-one matches as well as multiple-to-one matches. Multiple-to-one matches exist because the trade data sometimes bundles multiple species in the same general market category. For instance, “Fish; Atlantic salmon and Danube salmon, frozen, excluding fillets, livers, roes and other fish meat of heading no. 0304” meets our second criterion because two species are listed. In this case, we match to the first species listed, Atlantic salmon. By contrast, “Fish; frozen, yellowfin tunas, excluding fillets, fish meat of 0304, and edible fish offal of subhheadings 0303.91 to 0303.99,” meets our first criterion because there is a one-to-one match to yellowfin tuna in the nutrition data. Our main finding is robust to both of these exclusion criteria (Supplementary Fig. 2, Supplementary Data 2, 3).

As additional robustness checks, we analyzed restricted samples by excluding the importer being a major re-exporter. Re-exports involve the export of foreign goods previously imported. In seafood markets, re-exports typically involve some added processing such as turning a whole fish into a fillet36. By excluding re-exporters, we can rule out the possibility that our main finding is driven by processing margins in large re-exporting countries such as China. Our findings are qualitatively the same excluding re-exporters and quantitatively similar (Fig. 1, Supplementary Figs. 1, 2). The statistical tests on the restricted samples support our main finding and indicate that developing countries receive more nutrients per dollar than developed countries (Supplementary Data 1–3). We also re-run the Mann–Whitney tests on a dataset that pools all years and excludes transactions with China to address China’s particularly important role as an importer of seafood for re-export36,37 and to examine robustness when a large seafood trader in the middle of the income distribution is excluded. Our main findings are robust. Tests indicate that developing countries receive more nutrition per dollar than developed countries (Supplementary Data 1–3).

Nutrient levels per dollar are not strictly independent across different products, e.g., the amount of protein per dollar in a seafood product is positively correlated with the amount of vitamin B-12 per dollar (Supplementary Table 1). All nutrient per dollar measures are positively correlated with each other and statistically significant with some having substantially higher correlations than others (Supplementary Table 1). To examine the potential mediating effect of dependence across nutrients, we test our main result in a stacked model using Seemingly Unrelated Regression for each nutrient and each developed/developing categorization (Supplementary Table 2). Our main finding is unchanged. Across all models and all nutrients, developing countries receive more nutrition per dollar than developed countries, we reject cross-equation restrictions of equal differences across the nutrients, and the percent differences are quantitatively similar to those found assuming independence across nutrients (Fig. 1, Supplementary Data 1, Supplementary Table 2).

Explaining differences in nutrition per dollar

The finding that developing countries systematically pay lower prices for nutrition naturally raises the question of what could explain these differences. A challenge in explaining developed/developing country differences is that consumers do not buy food nutrition directly but instead buy composite goods— e.g., a fresh cod fillet or a frozen whole anchovy. Each composite good has its own nutritional profile as well as a bundle of non-nutritional characteristics. Previous studies show consistent patterns for implicit prices of non-nutritional characteristics such as type of preservation (e.g., fresh or frozen) and product form (e.g., whole or filleted)23,24,25,26. In particular, this literature consistently finds a price premium for fresh and filleted products relative to frozen and whole products, suggesting that these non-nutritional attributes may be a major driver of the robust gap in nutrition per dollar between developed and developing nations.

To assess this possibility, we estimate Oaxaca-Blinder decompositions for each nutrition per dollar variable. Results show that differences in product characteristics and market values of these characteristics explain differences in nutrition per dollar (Fig. 2, Supplementary Fig. 3). For both macronutrients and micronutrients, fresh and filleted consistently account for large shares of the explained variation from the Oaxaca-Blinder decompositions (Fig. 2, Supplementary Fig. 3).

Fig. 2: Contributions of product attributes to differences in nutrition per dollar.
Fig. 2: Contributions of product attributes to differences in nutrition per dollar.
Full size image

Normalized contributions to developed and developing differences in nutrient concentration per dollar for two macronutrients and two micronutrients based on Oaxaca-Blinder Decomposition using Dev 1. Dev 1 divides Gross Domestic Product Per Capital in Purchasing Power Parity (GDPPCPPP) into quintiles, and it assigns the bottom two quintiles to developing and the top three quintiles to developed. Positive (negative) values reflect contributions to more (less) nutrient content per dollar for developing countries. The base category is fish that is frozen and whole. The attribute contributions include crustacean (gray), mollusk (pink), live (dark orange), fresh (dark blue), other preservation (yellow), other portion (green), filleted (light blue), and meat (light orange). The attributes filleted and fresh are consistently associated with positive differences. Supplementary Fig. 3 depicts results for nine nutrients.

The Oaxaca-Blinder decompositions also allow for statistical testing of explained variation (Tables 1 and 2). Each coefficient is the difference in each nutrient content per dollar between developed and developing countries explained by the product characteristic. We find that the attribute fresh contributes a statistically significant positive difference in nutrition per dollar for all development status indicators and for all nutrients tested (Table 1). Similarly, the attribute filleted contributes a statistically significant positive difference in nutrition per dollar for all development status indicators and for all nutrients tested (Table 2).

Table 1 Explaining differences in nutrients per dollar using fresh
Full size table
Table 2 Explaining differences in nutrients per dollar using filleted
Full size table

To understand these results, consider the equivalency in total nutrition for a fresh fish compared to a frozen one. While the nutrition is the same, the fresh fish is more expensive23,24,25,26. That means fresh fish have lower nutrition per dollar (same nutrition but higher price) when holding other characteristics such as species constant. To the extent that developed countries tend to import more fresh fish relative to developing countries (as shown below), they receive less nutrition per dollar (positive difference in the Oaxaca-Blinder explained). Comparing these findings to recommended daily allowances also facilitates interpretation. Using the example of Dev 1 and protein, and pooling all years, the total difference between developing and developed is 15.23 g/$ (Table 1). Using an RDA of 50 g38, that constitutes 30.5% of the RDA. The difference explained by only the attribute fresh is 1.8 g/$ (Table 1), or 3.6% of the RDA. For the micronutrient vitamin B-12, the percentages are even larger. The total difference between developing and developed is 3.83 mcg/$ (Table 1). Using an RDA of 2.4 mcg38, that constitutes 159.8% of the RDA. The difference explained by only the attribute fresh is 0.19 mcg/$, or 7.9% of the RDA.

To evaluate the plausibility of the Oaxaca-Blinder findings, we directly test for differences in non-nutritional attributes of seafood product imported by developed and developing countries. Indeed, non-nutritional seafood attributes that are more highly valued in the market also tend to be more associated with developed country imports (Table 3). The shares of live, fresh, or other preservation are higher for developed country importers compared to developing country importers, whereas the share of frozen is lower (Table 3). Similarly, the share of filleted is higher for developed country importers compared to developing country importers, whereas the share of whole fish is lower (Table 3). These qualitative findings are robust to whether or not the transactions are volume weighted, but quantitatively the differences are larger when applying volume weights (Table 3). The findings are also consistent across all six development status indicators (Table 3, Supplementary Tables 3–7). These product characteristic differences provide supporting evidence that non-nutritional attributes contribute to the differences in nutrients per dollar between developed and developing country importers.

Table 3 Comparing covariates of developed and developing importers
Full size table

As further supporting evidence that non-nutritional characteristics contribute to differences in nutrition per dollar for developed and developing countries, we run hedonic regressions that decompose price per kg into implicit prices for each product attribute. Consistent with previous hedonic studies of seafood markets23,24,25,26, we find that fresh and filleted products are more highly valued than frozen and whole products (Table 4). The results are robust to including interactions with each of our six development status indicators (Table 4). These patterns of implicit prices, combined with covariate differences (Table 3, Supplementary Tables 3–7), suggest that discounts on nutrition to developing nations partially reflect observable differences in non-nutritional attributes of products imported by developed versus developing importers that are nonetheless bundled with nutritional attributes.

Table 4 Multivariate hedonic regressions of natural log of price per kilogram of fish imports
Full size table

Discussion

We find that developing countries pay less for nutrition in seafood than developed countries. There is broad consensus that seafood plays an important nutritional role globally, and that role is even more pronounced in developing countries1,2,3,4,5,6,7,8,9. There is a tendency for developed countries to attach a high value to fresh seafood, high-value product forms such as filleted, and crustaceans23,24,25,26. These features add to an already elevated willingness to pay for seafood based on income. Our results suggest that because of this exchange, developing countries, which tend to be lower income, have access to seafood nutrition in imports at much lower prices. Understanding why some species and product forms are more highly valued than others in the global seafood trade and whether these values are associated with embodied nutrient content is important future research.

Focusing on exports and the potential for developing countries to compromise nutritional security by trading away healthy seafood5,6 overlooks the potential gains on the other side of the market. That is, developing countries as importers gain access to low-price nutrition because many species and product forms are less valued in developed countries. In terms of quantity, some countries experience net increases of wild-caught seafood nutrition through reciprocal trade (63%), while others experience net losses9. The growing importance of aquaculture globally28,30,39, which has previously been omitted from the seafood trade and nutrition literature, provides a potential channel through which reciprocal trade benefits even countries that experience net losses in wild-caught seafood nutrition. Our results, which include both wild-caught and farmed seafood, suggest another important channel through which trade can enhance nutritional security, by lowering prices of macro and micronutrients for developing countries. Indeed, a developing country with a seafood nutrition trade surplus (i.e. with greater exports of seafood nutrition than imports) may nevertheless be able to satisfy its seafood nutrition needs at lower cost through trade than through domestic supply—utilizing the monetary surplus to satisfy other nutritional or non-nutritional development needs12. Testing whether increased (decreased) trade increases (decreases) the net availability of seafood nutrients at the country level is important future work.

Although our main result is highly robust across different categorizations of developed/developing, matching criteria, subsamples that exclude major re-exporters, individual years, and methods of trimming the data, we do not establish a causal link between the globalized seafood market and lower nutrient prices for developing countries. One possible mechanism is that increased globalization and seafood trade volume allow both low-income and high-income consumers to express demand for different product attributes such that the exchange benefits both groups. However, we do not specify a counterfactual that indicates what would have happened in a less globalized world or with increased or reduced trade liberalization40,41. This suggests that future research could investigate how variation in trade policy or shocks to the trading system such as COVID-19 affect the resulting prices of nutrients as well as the resulting flows of nutrients across borders.

Our results suggest that restrictions on trade, such as tariffs, could have negative consequences for nutritional security. Because of price and product form differences, developed and developing countries may benefit from reciprocal trade even within the seafood sector. Moreover, it may be difficult to develop fishery-specific trade policy such that a country internally consumes its healthy exports while maintaining access to low-price nutrition from seafood imports. One developing country’s attempts to restrict exports could undermine another developing country’s access to nutrition in imports, and attempts to discriminate between developing and developed countries could run afoul of WTO rules. Similar attempts to fine-tune trade policy to granular sustainability objectives in aquaculture are also not likely to be feasible42.

By focusing on the country level, our analysis is unable to examine within-country heterogeneity. Demand for seafood differs across income strata within counties, across regions, and across rural and urban settings17,43,44. As supermarkets have become common in many developing countries45, there are questions about how consumers are accessing seafood and who is purchasing it. It may be that benefits of greater access to nutrition through imports flow to urban consumers, while there are costs to rural consumers of exporting high-value nutritional seafood. This is an important area for future study. Further research is also needed to understand how and why the gap between developed and developing country nutrition per dollar changes over time (Fig. 1, Supplementary Fig. 1). Over a longer time horizon, imports as a share of total seafood food consumption globally rose from 16% in 1961 to 32% in 201910. In general, we expect globalization to expose more seafood to trade competition, integrate seafood markets, and harmonize prices17,19,21,31, but even with increased globalization and market integration, market segmentation based on product characteristics such as fresh or filleted persists19,23,24,25,26.

Ultimately, lower prices and increased access to seafood are important for food and nutritional security, but sustainability of the fisheries and aquaculture operations that supply that nutrition are also critical11,17,46,47. Some researchers are even beginning to re-examine sustainable fisheries management through the lens of nutrition48. Sustaining and growing the contributions of seafood to human nutrition and food security require tackling a diverse set of governance challenges across commercial capture fisheries, subsistence fisheries, recreational fisheries, and aquaculture17,22,46,49,50.

Methods

Data

We construct a database by linking UN Comtrade51 with FAO’s seafood nutrition database52. Comtrade contains bilateral international trade data that are finely resolved at the product level, while the FAO nutrition database contains species-level nutritional content for a wide range of macro- and micronutrients. The Comtrade data specifically includes: trade flow type (import vs. export), nation of import, nation of export, quantity, dollar value, and species or group of species traded. Additionally, Comtrade reports unique characteristics of each product, such as fillet vs. whole, fresh vs. frozen, and other preparation types such as salted, dried, or smoked.

We begin to build the dataset by rank-ordering all countries based on 2019 seafood exports and selecting all countries that collectively constitute 90% of global seafood exports53. This yields 41 countries: China, Norway, Russian Federation, Viet Nam, Peru, United States of America, Netherlands, Thailand, India, Chile, Indonesia, Spain, Denmark, Ecuador, Mauritania, Germany, Sweden, United Kingdom, Morocco, Taiwan Province of China, Iceland, Japan, Republic of Korea, Canada, Myanmar, Faroe Islands, Poland, Argentina, Namibia, France, Mexico, Senegal, Malaysia, Ireland, New Zealand, Portugal, Oman, Pakistan, Türkiye, Philippines, and Greenland. We choose 2019 for the rank ordering as the last year prior to COVID-19.

We download all Comtrade data for these countries in the six-digit commodity codes HS030000 through HS030999 for the years 2015 through 2021, which includes all products listed as seafood in Comtrade. As Comtrade excludes all varieties of seaweed from this category, the analysis omits seaweed. Additionally, all products not directly consumed by humans are removed. This includes items directly labeled as “Not Fit for Human Consumption” or “Ornamental” as well as menhaden, krill, corals, and sponges. We also exclude fish-based flours. Other countries beyond our 41 exporting countries are well represented in the data through their relationships as import partners. The resulting dataset contains 267,505 total observations, 266 unique products, 237 unique importing countries and independent territories, and 1267 unique trading partner combinations.

The nutritional database is the FAO/INFOODS global food composition database for fish and shellfish, version 1.0 – uFiSh1.052. The dataset includes “Food Name in English,” which we use to match to Comtrade; the percentage edible coefficient; and a wide range of nutrients summarized in amount per 100 g of edible seafood. For each of these individual product categories, the dataset affixes 181 unique nutritional components. This breakdown includes macronutrients such as calories, protein, and fatty acids, as well as micronutrients such as iron and calcium.

We link the Comtrade data to the nutritional data by using the full text description of each commodity code in Comtrade. Our matching protocol includes four match types (ordered from most to least precise): one-to-one exact match, multiple-to-one exact match, one-to-similar match, and composite goods.

One-to-one exact match (n = 91,635). When possible, we directly match to the nutritional dataset with the same species (e.g. Atlantic cod) and same product form (e.g. fillet). If the species is available with complete nutritional information but not the same product form, we match the nutritional information with the species in the nutritional data having complete nutritional information. Then we choose another record in the nutritional dataset for the same species and product form to identify the edible coefficient. This allows us to use the edible coefficient for species and product form combinations when some of the nutritional information is missing from the FAO record. For cases in which edible coefficients are unavailable in the nutritional data, we assign the following values: 1.0 for fish meat, livers, roes, and smoked; 0.75 for fish fillets and salted or brined products; and 0.5 for whole fish. One-to-one exact matches are the highest quality matches, and we test our main results using only these records as the most restrictive dataset.

Multiple-to-one exact match (n = 96,796). Often records in the Comtrade database report multiple species for the same commodity code. For a given trade, it is not specified what share of the trade consists of the individual species in the list. For these cases, we use the first species listed in Comtrade to match to FAO nutrition data. If that species is missing in FAO nutrition data, we go to the second in the list, and so forth. We follow the same rule as the one-to-one matching for the edible coefficient.

One-to-similar match (n = 20,634). In some cases, the species in the Comtrade data is unavailable in the FAO nutrition data, or the Comtrade commodity code does not provide sufficient detail on the species. Here we choose a similar or representative species based on expert judgement of a species in a similar product category. The full list of these matches is as follows: bigeye tuna uses southern bluefin tuna; toothfish uses sea bass, mixed species; fish livers uses cod liver; frozen crabs uses blue crab; Nile perch uses snake head; fish fins (other than shark fins) uses shark fins; and we exclude fish, edible offal from the analysis. We follow the same rule as the one-to-one matching for the edible coefficient.

Composite goods (n = 58,440). If the species is unspecified (denoted NSPF) in the Comtrade data, we create a composite species and weight the nutritional variables by the shares of the composite. We constructed weighted averages of the most frequently caught species within each category based on 2016 landings data and using species ranked highest in each category53. Because landings for these broad categories are widely available for the United States, namely landings of different species of groundfish, flatfish, mollusk, and crustaceans, we use U.S. landings to create nutritional weights for composite goods. Within each grouping, the total quantity of landings was calculated, and each species contribution to that value was generated as a decimal proportion. These proportions acted as modifiers to a species’ nutritional value, an all values were summed to create a weighted-average nutritional profile. As fish and groundfish constitute much of the top landings by species, each of these used ten individual species to generate a nutritional profile. For flatfish, crustacean, and mollusks, species were only included that would have a noticeable effect on a nutritional profile, ranging from three to six species per category54 (Supplementary Table 8). For several composites, we supplemented the FAO nutritional data with nutritional data from Japan’s Ministry of Education, Culture, Sports, Science, and Technology55.

For our main set of tests – whether developing country seafood purchases are less nutritionally dense—we compute nutrition per dollar in each transaction by each nutrient. We define a transaction at the importer level so that trades between the trading partners of our 41 exporting countries are not double counted. To this end, we first compute total edible seafood in each transaction by multiplying the total quantity by the edible coefficient. We then create total nutrients for each macronutrient and micronutrient that we analyze. These include calories, protein, fatty acids, calcium, iron, magnesium, potassium, vitamin B-12, and zinc. We multiply each nutrient per 100 g by 10 to so that we have total nutrient content per kg of edible seafood and then multiply by total edible seafood to obtain total nutrients. We arrive at nutrition per dollar by dividing each total nutrient by the total dollar value of the transaction. Note that nutrition per dollar is the inverse of price. It is convenient to use the inverse of price because different nutrients are measured in different units such as grams of protein, kilocalories, milligrams of calcium, or micrograms of vitamin B-12. By using the inverse of price, we preserve the conventional scaling of different units for different nutrients.

For each transaction, we also link data for importing country gross domestic product per capita in purchasing power parity terms (GDPPCPPP)56.

We categorize developing and developed six different ways: Dev 1 divides Gross Domestic Product Per Capita in Purchasing Power Parity (GDPPCPPP) into quintiles, and it assigns the bottom two quintiles to developing and the top three quintiles to developed; Dev 2 uses country-level GDPPCPPP and assigns the bottom two quintiles to developing, the top two quintiles to developed, and drops the middle quintile; Dev 3 divides the United Nations Human Development Index (HDI) into quartiles, assigns the first and second quartiles to developing and the third and fourth quartiles to developed35; and Dev 4 uses the UN FAO’s Low Income Food Deficit Country (LIFDC) list to assign developing, and countries not listed are considered developed34; Dev 5 assigns a country to developing if the World Bank GNI indicator is in the bottom two groups (low or lower-middle income) and assigns all others to developed; Dev 6 assigns a country to developing if the World Bank GNI indicator is in the bottom group (low income), assigns the country to developed if it is in the top group (high income), and drops the other countries from the sample33.

We create categorical variables based on the detailed product descriptions to describe preservation form, including fresh, frozen, live, and “other,” which includes dried, salted, packed in oil, or smoked. We also create categorical variables to describe portion, which includes whole, fillet, meat, and “other portion.” The latter includes livers, roe, and fins. We also create categorical variables to describe fish, crustaceans, or mollusks. For the regression analysis, we compute price per kg at the transaction level (year, commodity code, and exporter/importer combination) by dividing total value by total quantity.

Analysis

To compare nutrient per dollar for developing and developed countries, we test for differences between developed and developing nutritional content using all six ways of categorizing countries as developed or developing. Specifically, we compute one-sided Mann–Whitney non-parametric tests for percent difference in nutrients per dollar. We summarize mean percentage differences in nutrients per dollar in figures and compute analytical confidence intervals57. We also conduct parametric t tests for differences in two ways: 1) treating each transaction equally and 2) weighting nutrients per dollar by the volume of the transaction.

All statistical tests (Mann–Whitney, t test, and weighted t test) are done on each nutrient using the full sample and a trimmed sample that drops the top 1% to eliminate potential outliers. To trim the sample, we iteratively sorted each nutrient per dollar in descending order and flagged observations that fall in the top 1%. We then trimmed the observations that are flagged at least once in the above iterative process, which leads to trimming about 3% of the total observations. The purpose of trimming the sample is to evaluate robustness of the main findings to dropping extreme observations. We also report all tests for the pooled sample (all years) and for each nutrient separated out by year (2015 through 2021). We furthermore conduct the tests while dropping China from the analysis, since China is the largest producer, consumer, exporter, and importer of seafood and is heavily engaged in re-exporting36,37. We also create an indicator for re-exporter that includes a broader set of important seafood re-exporting countries—China, Netherlands, Poland, Thailand, and Vietnam—and report the tests when these countries are dropped from the sample. Moreover, all tests are conducted for each of our six development status classifications. This leaves us with 2592 sets of statistical tests: 2 (trimmed and untrimmed) × 9 (nutrients) × 8 (years, including 7 individual years and all years combined) × 3 re-export samples (drop none, drop China, drop broader set of re-exporters) × 6 (Dev 1 through Dev 6 development indicator).

To evaluate match quality (matching Comtrade to the FAO nutrition data), we conduct the 2592 different tests on three different subsamples: the full sample, exact one-to-one matches only, and a combination of one-to-one matches and multiple-to-one matches.

We run Seemingly Unrelated Regression (SUR) models to allow for correlation across nutrients in the nutrient per dollar outcome variables58. We use natural log of nutrition per dollar and test for cross-equation restrictions that force all nutrient differences to be the same. To compare to other results, we convert the regression coefficients to percent differences: % change = (\({e}^{\beta }\) − 1) × 100.

To explain our main findings, we compute Oaxaca-Blinder decompositions. Specifically, we compute two-part Oaxaca-Blinder decompositions for each of nine nutrient per dollar measures using the omega option in the oaxaca command in Stata59,60,61. Oaxaca-Blinder decompositions, which were developed originally to analyze wage differences in labor markets and diagnose wage discrimination, decompose differences in a dependent variable between two groups into differences that can be explained by differences in the covariates and a residual amount that cannot be explained by these differences. In our application, for example, the higher protein per dollar for developing country importers compared to developed country importers could be explained in part by developing countries having a larger share of frozen seafood in imports, and frozen product is less valued in the market than fresh, live, or other preservation.

We analyze covariate differences to assess supporting evidence for the Oaxaca-Blinder decompositions. Specifically, we compare differences between mean non-nutritional covariate levels for developed and developing countries. By comparing developed and developing country importer values for these characteristics, we can begin to understand differences in prices paid for seafood and associated nutritional content. We compare both unweighted and volume-weighted means of these characteristics.

To assess findings from the Oaxaca-Blinder decompositions further, we analyze the determinants of seafood prices using hedonic regressions. Specifically, we estimate multivariate hedonic regressions to decompose price per kg into the effects of individual attributes, including type of seafood (e.g., fish, crustacean, or mollusk), preservation form (e.g., frozen), portion (e.g., fillet), and year (to allow for inflation and changes in demand and supply conditions). The base category is frozen whole fish. We run a baseline model with no interactions and six other models that interact each of our six development indicators with the covariates in order to examine whether developed countries pay a premium or discount for a given product characteristic. Standard errors are clustered at the year level.