International trade and human movements have accidentally transported thousands of species worldwide at an unprecedented scale. The resulting biological invasions are among the greatest drivers of species extinctions and can cause enormous economic losses. Understanding how globalization affects the accidental transport of species is urgent to prevent new invasions. However, global trade networks have had mixed success so far in explaining intercontinental species movements. Here, we show that commonly used proxies of global trade flows such as general imports and agricultural imports differed greatly from flows of alien ants from their donor regions to the United States. The analysis of 97 individual commodity flows revealed instead that plants and fruit imports, which are a small subset of all agricultural commodities, were associated with invasion flows. All 95 other commodities differed from flows of alien ants, including most “agricultural” commodities which had extremely heterogenous geographic origins. This highlights the need to know precisely which commodities serve as introduction pathways for a particular taxonomic group in order to explain invasion flows and identify likely source regions of future invasions in a world of changing trade relationships.

Data

Species distributions

To determine the number of alien ant species that have established in the continental US (that is, the 48 contiguous states plus Alaska), we used the georeferenced database Antmaps (antmaps.org; an authoritative database maintained and updated regularly by experts based on new records in the peer-reviewed scientific literature; Janicki et al. 2016). Currently, 98 alien ant species have spread across the continental US and have established at least at one outdoor location worldwide (in order to qualify as “alien species”). One of these species, Pheidole guineensis, was excluded from the analyses because its native range is unknown, which prevented the mapping of invasion flows for this species. To calculate the geographic profile representing the “flow” of established ant species from their donor regions to the US, we defined the species’ native range as all countries containing native but no introduced populations. For species whose native range encompassed more than one world region, we weighted the flow from each of the world regions by the number of political regions where the species was recorded as native (that is, non-overlapping country or sub-country polygons representing states, counties, or islands, all of which are more homogenous in size than entire countries; Janicki et al. 2016). The most widespread species are occasionally introduced via bridgehead locations where they are already invasive and do not arrive directly from the native range (Bertelsmeier et al. 2018). Therefore, estimated “flows” between the native range of such species and the US may not reflect their actual introduction pathways. To avoid biases due to the bridgehead effect, we excluded the 46 most widespread species, with invaded ranges covering several world regions, from the analyses (WebTable 1).

Trade data

To calculate import flows to the US for different categories of commodities, we used the UN Harmonized Commodity Description and Coding System (“Harmonized System” or HS), an international nomenclature for product classification. The HS comprises approximately 5300 article or product descriptions grouped into 97 broad categories (WebTable 2). Import data (reported in US dollars) for all 97 broad categories of commodities were sourced from the US Census Bureau. We used import data for the 49 continental US states and 229 trade partner countries across the world, summed over the years 1991–2018. Data for this period comprised most of the exchanged commodity volumes since the beginning of the 20th century. We used records of inflation from the World Bank to convert trade records, which were expressed in US dollars for the year 2017. To standardize import profiles of geographic origins for each commodity, we divided imports arriving from each trading partner (in US dollars) by the total value of this commodity import to the US. This allowed representation of the relative contribution of different parts of the world to each commodity flow and avoided assigning greater weight to commodities with high monetary values. General import flows were calculated by summing all 97 individual commodity import flows, and agricultural trade flows were calculated by summing import flows of commodity categories 01–20. We also calculated a “plant and fruit” trade flow – often considered an important pathway for the transport of ants – by combining commodity categories 06, 07, and 08 (WebTable 3).

 Statistical analyses

To test whether the flows of general trade, agricultural trade, and plant and fruit trade differed from the flow of alien species to the US, we used chi-square tests with 2000 Monte Carlo replicates to determine P values.

Correspondence analysis

To analyze the similarities of import flows of the 97 commodity categories, we performed a correspondence analysis (CA) on the import profiles of geographic origins for all commodity categories, after row standardization (Greenacre and Primicerio 2013). Each row represented the compositional data of import flows for a commodity category, with columns consisting of world regions. Differences among commodity profiles of geographic origins were represented in the two-dimensional space of the factorial map. The first two axes represented 59% of the total inertia. We added profiles of general trade, agricultural trade, and the plant and fruit trade flows, as well as the profile of alien species flows, to the CA space by projecting them as supplementary individuals (ie additional points that do not determine the axes, but can be plotted on the same factorial map using the same transformation).

Classification

To identify commodity flows that share the same geographic origins, we performed a classification of import profiles using a hierarchical cluster analysis. We first calculated Ward distances between the coordinates of the commodity categories in the CA space, keeping all five axes. Next, to test how many clusters constituted statistically significant groups among the 97 commodity profiles and four supplementary individuals (general trade, agricultural trade, plant and fruit trade, and alien species flows), we used a simple permutation test (Greenacre and Primicerio 2013) for determining homogeneous clusters of import profiles. This method allowed determination of the presence of nodes in the hierarchical clustering tree with levels lower than would be expected from dendrograms constructed on random permutations of the data. To generate distribution import flows from each world region under the null hypothesis that no differences exist among flows (ie geographic profiles for different commodities), we randomly permuted the data for each world region across commodity profiles (ie we performed column-wise permutations of the rows, where rows corresponded to commodities’ profiles and columns corresponded to world regions). The test was performed with 999 permutations. The primary objective of the permutation testing was to define the level at which the tree should be cut in order to obtain non-random levels of homogeneous clustering in the set of profiles.

See readme file.