Referendums are crucial for democratic participation but present challenges in clearly and concisely presenting complex issues. The simplification of referendum questions through yes/no options is common, but the 2016 UK EU referendum diverged by using the more complex 'remain' vs. 'leave' options. This shift necessitates a linguistic examination of the potential for bias introduced by different verb choices. Prior research demonstrates the impact of micro-deviations in question phrasing on response variation. This study addresses the research gap regarding the effect of response option framing, specifically examining whether the verbs 'remain,' 'stay,' and 'leave' activate different framing effects due to their etymological roots and inherent semantic structures. While the Brexit referendum serves as a case study, the focus is broader, exploring the general suitability of verbs as alternatives to yes/no options in decision-making processes. The study hypothesizes that these verbs, due to their different etymological origins, might unconsciously activate different framing effects, leading to divergent evaluative structures. The Implicit Association Test (IAT) is employed to assess these potential implicit framing effects.

The paper reviews literature on collective decision-making processes, highlighting the significance of linguistic factors in shaping responses. It references the Electoral Commission's 2015 report acknowledging the impact of linguistic factors in referendum question design. The research notes the previous studies' focus on the requirements for clear and neutral question phrasing and the potential influence of framing effects on rational decisions. The literature shows how even slight changes in question wording can substantially affect responses. The paper distinguishes its approach by examining the etymological and framing effects of verbs as response options, rather than just analyzing their synchronic usage. It considers the diachronic perspective, exploring the historical evolution of word meanings, particularly for near-synonyms, showing how subtle etymological differences can lead to divergent framing effects.

The study utilized a modified version of the Implicit Association Test (IAT) to measure the implicit evaluative associations of the verbs 'stay,' 'remain,' and 'leave.' Two IAT studies were conducted. Study 1 (n=185) employed a convenience sample, while Study 2 (n=355) focused on native English speakers from the UK and USA. Participants were randomly assigned to one of two conditions: (1) remain vs. leave; (2) stay vs. leave. In each condition, participants classified stimuli (words representing the verbs and positive/negative attributes) into dichotomous categories. The D-measure, a standardized measure derived from reaction times, was used to quantify the strength of associations between concepts and attributes. The seven-block IAT procedure involved training blocks for target concepts and attributes, followed by combined blocks where these were linked, with later blocks reversing the assigned key pairings. The algorithm for the D-measure addressed extreme reaction times and incorrect classifications, thereby allowing for a relative measure of evaluative structure. Robustness checks were performed on Study 1 using only native English speakers and utilizing OLS regression analysis to control for confounding factors such as age, gender and native English-speaking status. Ethical considerations regarding data privacy and informed consent were observed and adhered to.

Study 1 revealed that 'remain' was perceived as slightly more positive than 'leave' (μ = 0.189, σ = 0.486, t(91) = 3.739, p < 0.001, r = 0.365, medium effect size), while 'stay' was perceived as significantly more positive than 'leave' (μ = 0.515, σ = 0.331, t(92) = 12.979, p < 0.001, r = 0.804, large effect size). A direct comparison showed that 'stay' was associated with more positive connotations compared to 'remain' relative to 'leave' (t(183) = 5.329, p < 0.001, r = 0.367, medium effect size). These findings held true even when analyzing only native English speakers. Study 2 replicated the pattern, showing 'remain' as more positive than 'leave' (μ = 0.526, σ = 0.410, t(181) = 17.309, p < 0.001, r = 0.790, strong effect size) and 'stay' as even more positive than 'leave' (μ = 0.592, σ = 0.351, t(172) = 22.213, p < 0.001, r = 0.861, strong effect size). However, the difference between 'stay' and 'remain' relative to 'leave' was smaller in Study 2 (t(353) = 1.619, p < 0.53, r = 0.086), although this could potentially be statistically significant if a one-tailed test were justified. No significant differences were observed between native and non-native English speakers, age groups, or genders in the effects across both studies.

The results demonstrate that near-synonyms ('stay' and 'remain') exhibit significant differences in their evaluative structures. This suggests that implicit framing effects rooted in etymology influence perceptions. The more positive evaluation of 'stay' is attributed to its sound-symbolic associations ('strength,' 'stability') and its consistent semantic meaning across time. 'Remain,' on the other hand, carries a negative etymological association ('to be left behind'). The initial expectation that 'leave' would be negatively connoted was mitigated by the context of the Brexit debate and the sample in Study 1, suggesting that synchronic connotations may suppress diachronic associations. The lack of differences in IAT results between L1 and L2 speakers, age groups, and genders suggests that the effect is based on sensorimotor grounding of the concepts. The consistent, sensorimotor-based experience may explain the universal influence. The study acknowledges limitations of the IAT as a relative measure and the potential weak link between implicit measures and overt behavior, but emphasizes that even small effects at the individual level can have a significant impact on overall voting behavior in large-scale elections.

This study demonstrates the potential for bias in political referendums stemming from the choice of verbs as response options. The findings highlight the importance of considering the subtle semantic and etymological properties of words in designing referendum questions to minimize framing effects. While verbs effectively convey information and condense complex issues, the data suggest that, due to their inherent framing effects, they may be less suitable than the standard yes/no option for the formulation of response options. Further research could explore different types of predicates, the role of negation, and scalar orientation of verbs to further delineate the nuanced effects of verbal framing in decision-making.

The study acknowledges the methodological limitations of the IAT, particularly its relative measure of evaluative structure and the potential weak connection between implicit attitudes and actual behavior. The convenience sampling in Study 1 might also limit the generalizability of the findings, although the replication with a more targeted sample in Study 2 provides some reassurance. The potential influence of pre-existing attitudes towards Brexit on Study 1 is discussed, although this is partially mitigated by the findings in Study 2.