Although scientists strive to accurately communicate their research, disconnects can arise between results and rhetoric. Some have regarded invasion scientists as particularly prone to using value-laden language incommensurate with the scientific facts or results. We addressed how authors used 10 near synonyms (words for which usage is similar but not completely overlapping) of the negative-value word invasive. We asked whether study findings (effect sizes) or other factors predicted language use. The use of negative-value words such as invasive was not associated with study findings but, instead, with contextual factors. For example, plant and invertebrate biologists used more negative language to describe nonnatives than did those studying vertebrates. The authors also tended to use more negative language in recently published papers than in older studies. Although many have called for impartial language when communicating research, some scientists use language imbued with value that may be inappropriate. Such use may affect how the public perceives scientific findings.

Our manuscript tested our prediction that value-laden words are used more often in studies reporting stronger empirically determined impacts, we collected data from primary studies quantifying the effects of biological invasions on natural systems. We began with the papers included in a meta-analysis of impacts by aquatic non-native species (Gallardo et al. 2016) and followed the original authors’ criteria to add 52 more recent papers for a total of 202. From each paper, we considered communicated impact as the frequency of 10 words extracted from the full text of each paper, focusing on commonly used near-synonyms of “invasive:” alien, coloni(-ser/-zer/-sing/-zing, hereafter colonisz), exotic, introduced, inva(-der/-ding/-sive, hereafter invasd), naturalized, nonindigenous, nonnative, pest, and weed (an additional near-synonym, neobiota, was absent from all papers in our dataset). Among invasion scientists and others, determining the value implied by these near-synonyms has been fraught, thus we ranked the 10 words according to an independent metric, linguistic sentiment (Rinker 2017). This metric attributes a connotation, from negative (−1) to positive (+1), to common English-language words. Sentiment therefore captures value as perceived by a wide audience and is objective (outside the field of invasion science) and quantitative (numeric). In interpreting our results, we consider how sentiment scores compare to the values implied by researchers’ usage of these words.

We also collected from each paper ecological impact data (effect sizes) and other potential predictors we hypothesized may influence communicated impact. These other predictors, described in greater detail in Supplemental Table 1 (Supplementary Materials), were: habitat type (river, lake, or estuary), study type (observational, manipulative, or mesocosm), non-native species identity, trophic level of the invader (primary-producer, filter-collector, omnivore, herbivore, or predator), citation rate, journal impact factor, publication year, journal category (general, ecosystem-specific, conservation-focused, invasions-focused), indicators for types of funding (general, conservation, natural resources), development status for the country of the last author’s home institution, and mean linguistic sentiment averaged across all words in an entire paper (Rinker 2017).

We first tested whether authors used some near-synonyms more frequently based on the measured ecological impacts reported in their studies (Question 1) using linear mixed models or generalized linear mixed models (Bates et al. 2015) and AIC-based model selection. We then described patterns in word use (Question 2) across all papers that used at least one of the 10 near-synonyms of “invasive” (194/202) by summarizing frequencies of each word in each paper using non-metric multidimensional scaling (NMDS) ordination (Oksanen et al. 2015). To determine which predictor variables, including effect size and other features of studies, best explained word use (Question 3), we used two approaches. First, we used permutational multivariate analysis of variance (PERMANOVA, Oksanen et al. 2015), which evaluates the significance of potential predictors of the NMDS ordination solution. Second, we used random forest analysis (Liaw and Wiener 2002), to analyze NMDS axis 1 and axis 2 scores as indicators of word use. We describe additional methodological details in the Supplementary Materials.

See README file