Voicing Politics: How Language Shapes Public Opinion 9780691243412

Table of contents :
CONTENTS
ACKNOWLEDGMENTS
Introduction LOST IN TRANSLATION ?
1 Explaining the Language-Opinion Connection
2 Battle of the Sexes
3 Ghosts in the Language Machine
4 Today Is Tomorrow
5 Sensing Ethnic Divisions
6 The Language Premium
Conclusion THE VOICE OF THE PEOPLE
APPENDIXES
NOTES
REFERENCES
INDEX
A NOTE ON THE TYPE

Citation preview

VO I CI N G P O LITI C S

Tali Mendelberg, Series Editor Voicing Politics: How Language Shapes Public Opinion, Efrén Pérez & Margit Tavits Migrants and Machine Politics: How India’s Urban Poor Seek Representation and Responsiveness, Adam Michael Auerbach & Tariq Thachil Native Bias: Overcoming Discrimination against Immigrants, Donghyun Danny Choi, Mathias Poertner, & Nicholas Sambanis Nationalisms in International Politics, Kathleen Powers Winners and Losers: The Psy­chol­ogy of Foreign Trade, Diana C. Mutz The Autocratic M ­ iddle Class: How State De­pen­dency Reduces the Demand for Democracy, Bryn Rosenfeld The Loud Minority: Why Protests ­Matter in American Democracy, Daniel Q. Gillion Steadfast Demo­crats: How Social Forces Shape Black Po­liti­cal Be­hav­ior, Ismail K. White & Chryl N. Laird The Cash Ceiling: Why Only the Rich Run for Office—­And What We Can Do about It, Nicholas Carnes Deep Roots: How Slavery Still Shapes Southern Politics, Avidit Acharya, Matthew Blackwell, & Ma­ya Sen Envy in Politics, Gwyneth H. McClendon Communism’s Shadow: Historical Legacies and Con­temporary Po­liti­cal Attitudes, Grigore Pop-­Eleches & Joshua A. Tucker Democracy for Realists: Why Elections Do Not Produce Responsive Government, Christopher H. Achen & Larry M. Bartels Resolve in International Politics, Joshua D. Kertzer

Voicing Politics How Language Shapes Public Opinion Efrén Pérez Margit Tavits

P R I N C E ­T O N U N I V E R S I T Y P R E S S P R I N C E ­T O N A N D OX F O R D

Copyright © 2022 by Prince­ton University Press Prince­ton University Press is committed to the protection of copyright and the intellectual property our authors entrust to us. Copyright promotes the pro­gress and integrity of knowledge. Thank you for supporting ­free speech and the global exchange of ideas by purchasing an authorized edition of this book. If you wish to reproduce or distribute any part of it in any form, please obtain permission. Requests for permission to reproduce material from this work should be sent to permissions@press​.­princeton​.­edu Published by Prince­ton University Press 41 William Street, Prince­ton, New Jersey 08540 99 Banbury Road, Oxford OX2 6JX press​.­princeton​.­edu All Rights Reserved ISBN 9780691215143 ISBN (pbk.) 9780691215136 ISBN (e-­book) 9780691243412 British Library Cataloging-­in-­Publication Data is available Editorial: Bridget Flannery-­McCoy and Alena Chekanov Production Editorial: Nathan Carr Jacket/Cover Design: Chris Ferrante Production: Lauren ­Reese Publicity: Kate Hensley and Charlotte Coyne Copyeditor: Karen Verde This book has been composed in Adobe Text and Gotham Printed on acid-­free paper. ∞ Printed in the United States of Amer­i­ca 10 ​9 ​8 ​7 ​6 ​5 ​4 ​3 ​2 ​1

CONTENTS

Acknowl­edgments  vii

Introduction: Lost in Translation?  1

1

Explaining the Language-­Opinion Connection  19

2 ­Battle of 3

the Sexes  34

Ghosts in the Language Machine  58

4 ­Today Is Tomorrow 

79

5

Sensing Ethnic Divisions  95

6

The Language Premium  111



Conclusion: The Voice of the ­People 

130

Appendixes 143 Notes 195 References 199 Index 213

v

ACKNOWL­E DGMENTS

How do a comparative politics scholar (Tavits) and a US po­liti­cal psychologist (Pérez) produce a book on the effects of language on po­liti­cal cognition? Our answer highlights the virtues of collaborative science and serendipitous research meetings. It was spring 2013, and Pérez was invited by Professor Jacob Montgomery (Washington University in Saint Louis) to pre­sent a paper on xenophobic rhe­toric and its po­liti­cal influences on immigrants and their co-­ethnics. As part of the visit, Professor Montgomery asked Pérez if he would also lead a small workshop on methodological issues related to interview language and public opinion data. The latter agreed without knowing what he was getting into. As part of that methods workshop, Tavits sat in the audience, patiently asked some questions, and a­ fter the talk shared with Pérez some new research interests she was batting around. She was particularly curious about ­whether the language that ­people speak could impact in mea­sur­able ways what they thought about politics. Tavits’s pitch to Pérez was s­ imple: “­you’re a po­liti­cal psychologist who works with experiments—­would you be interested in collaborating?” Pérez jumped at the opportunity (almost quite literally). But he had one condition: ­there was no guarantee that their collective efforts would yield a paper, much less a book. That summer of 2013, Tavits and Pérez embarked on a long research enterprise that has involved numerous experiments and survey analyses in countries like Estonia, Sweden, the United States, and beyond. But the feature of this proj­ect that both are most proud of is its reconceptualization of language effects. Much of the linguistics lit­er­a­ture outside of po­liti­cal science is mired in smoldering debates about ­whether language determines what we perceive and feel and how we behave. Yet that lit­er­a­ture has made few splashes in po­liti­cal science. This book’s contribution is a new framework to reason about language effects without the more ham-­fisted assertion that language determines ­things, po­liti­cal or not. The key h ­ ere is to view language vii

viii Ac k n ow le d g m e nt s

in the context of how ­human memory and individual minds operate, especially in the realm of public opinion. De­cades of research on the psy­chol­ogy of survey response teaches social scientists that p ­ eople have relatively few po­liti­cal attitudes that are ready-­made and immediately available for reporting. What they instead carry in their heads are considerations—­the values, beliefs, ste­reo­types, knowledge, and other raw materials for individual opinions. Against this backdrop, this book’s claim, from the start, has been that language provides another nudge that shapes the sample of considerations from which ­people draw to assem­ble their po­liti­cal opinions. —­—­— A book that takes this long to produce depends on the generosity and patience of many individuals. With that in mind, we first thank each other, especially for boosting each other’s morale when t­ hings did not look so rosy for this proj­ect. We also thank Jacob Montgomery, who quite unknowingly set in motion a series of events that led to this book, and who has remained a big supporter of this research. We are equally grateful to audiences at several universities inside and outside the United States for providing valuable critiques, as well as affirmation, as we plowed ahead with this proj­ect. ­These institutions include Arizona State University, Columbia University, Cornell University, Georgetown University, Pennsylvania State University, Prince­ton University, UC–­Berkeley, UCLA, UCSD, UIUC, UNC Chapel Hill, University of Georgia, University of Iowa, University of Mary­land, University of Michigan, University of Texas–­Austin, Utah State University, and Yale University. Of all the work e­ ither of us has produced, this has been the most fun to pre­sent b ­ ecause audiences across fields have been so engaged and generous with their feedback. We also owe many thanks to ­those who have read and commented on all or parts of the book manuscript: Michael Bechtel, Mark Beissinger, Amaney Jamal, Cindy Kam, Chris Karpowitz, Jennifer Merolla, Kristin Michelitch, Cecilia Mo, Rory Truex, Nicholas Valentino, Ismail White, and Liz Zechmeister. In addition, t­ here ­were other helpful readers, who, due to the nature of the review pro­cess, remained anonymous to us, but whose comments ­were equally valuable for helping us improve the book. Both authors also thank their colleagues at Washington University and Vanderbilt University (Pérez’s previous institution), who commented on vari­ous iterations of this proj­ect. In par­tic­u­lar, we both publicly acknowledge Cindy Kam’s always incisive advice on the vari­ous experiments conducted on behalf of this proj­ect—up to and including the

Ac k n ow le d g m e nt s ix

androgynous figures in chapter 3 for the Swedish experiments that Pérez drew. The team at Prince­ton University Press has also been incredibly helpful and a model for what a publishing experience should be like. Tali Mendelberg, our academic editor, has been very generous with her advice and support at ­every stage, including organ­izing an amazing book conference for us. Bridget Flannery-­McCoy, Alena Chekanov, and the team they lead have expertly shepherded our manuscript through the submission, review, revision, and production pro­cesses. We are also both grateful to our funders, who took the leap of faith and invested in our aty­pi­cal but incredibly productive collaboration: Vanderbilt University, the Center for New Institutional Social Science, and the Weidenbaum Center on the Economy, Government, and Public Policy (the latter two ­housed at Washington University). We express special thanks to our survey partners, Kantar Emor in Estonia and Enkätfabriken in Sweden, who did not shy away from our unconventional proj­ects and always found creative and clever solutions for our, at times challenging, requests. Several research assistants have helped us along the way and we owe many thanks to them as well: Jonathan Homola, Elena Labzina, Jeremy Siow, and Sydney Weiss. Parts of chapter 2 have been published in “Language Influences Public Attitudes t­ oward Gender Equality,” Journal of Politics 81(1): 81–93, 2019, DOI: https://­doi​.o ­ rg​/­10​.­1086​/­700004 (Copyright © 2018 Southern Po­liti­cal Science Association). Parts of chapter 3 have been published in “Language Influences Mass Opinion ­toward Gender and LGBT Equality,” Proceedings of the National Acad­emy of Sciences 116(34): 16781–16786, 2019, DOI: https://­ doi​.­org​/­10​.­1073​/­pnas​.­1908156116 (Copyright © 2019 National Acad­emy of Sciences). Parts of chapter 4 have appeared in “Language Shapes ­People’s Time Perspective and Affects Support for Future-­Oriented Policies,” American Journal of Po­liti­cal Science 61(3): 715–27, 2017, DOI: https://­doi​.o ­ rg​/­10​ .­1111​/­ajps​.­12290 (Copyright © 2017 Midwest Po­liti­cal Science Association). And parts of chapter 5 have been published in “Language Heightens the Po­liti­cal Salience of Ethnic Divisions,” Journal of Experimental Po­liti­cal Science 6(2): 131–40, 2019, DOI: https://­doi​.­org​/­10​.­1017​/­XPS​.­2018​.­27 (Copyright © 2018 The Experimental Research Section of the American Po­liti­cal Science Association). As any author knows, t­ here are no books without some inspiration. For this necessary motivation, Pérez thanks his parents, Efrén and Maricela Pérez, for insisting that he learn, preserve, and celebrate his native Spanish tongue—­and without any tension with En­glish, his second language.

x Ac k n ow le d g m e nt s

Although only an N = 1, Pérez is still convinced that what he thinks and feels in Spanish is meaningfully dif­f er­ent from what he thinks and feels in En­glish. For Tavits, this book is a culmination of her lifelong fascination with languages. For native speakers of Estonian, like Tavits, whose m ­ other tongue is shared by only about one million p ­ eople, obtaining working knowledge in a handful of other languages is as natu­ral a part of growing up as learning to read or ­ride a bike. Having that knowledge, Tavits is as convinced as Pérez that the world around her appears subtly but noticeably dif­f er­ent depending on what language she speaks. So, let’s make that an N = 2! We conclude by dedicating our work to ­those who have walked this long path with each of us the ­whole way. Pérez dedicates this book to his wife, Tammy Rose Pérez del Cano, who has done as much—if not more—to keep this book (and Efrén) full of life and energy. Part of this has involved raising three young boys to excel in En­glish, even as they pay homage and re­spect to their f­ amily’s tongue, Spanish. When t­ hings go wrong at work (and they eventually do), Pérez is fortunate to have Tammy and the boys to fall back on for solace. For Tavits, t­ here are no words in any language to express the deep sense of gratitude she feels t­ oward her (effortlessly multilingual) husband, Taavi, and their (aspiring polyglot) ­children, Aiden and Linda, for their endless and unconditional love.

VO I CI N G P O LITI C S

Introduction LOST IN TR ANSL ATION?

Does the way we speak affect the way we think? It’s a question that many ­people find intriguing, perhaps b ­ ecause it is so easy to find examples from everyday life where language seems to do just that—­affect what we pay attention to, what we consider impor­tant, how we perceive events, and even ­whether we find jokes funny. For example, imagine trying the following riddle on a Spanish-­speaking friend: Why did the boy throw the butter out the win­dow? Answer: ­Because he wanted to see butter fly! This clever play on words is easy to convey in En­glish. But your Spanish-­speaking friend might find it frustratingly difficult to appreciate the humor in it. That is ­because mariposa, the Spanish word for butterfly, fails to compactly deliver the punchline of butter flying. Alas, it is only in En­glish that the riddle makes sense, since the ­mental associations that are needed for it to resonate are seamlessly contained in that language, but not to the same degree in Spanish. This ­isn’t just a ­matter of vocabulary or lexicon. If you ­were to ask a Russian-­speaking friend, for example, to translate the child ate the ice cream into Rus­sian for you, then you should prepare yourself to answer questions like is the child a boy or a girl? And did the child eat all or part of the ice cream? You might feel flustered by your friend’s impertinence about ­these details. But, in order to express that the child ate the ice cream in Rus­sian, one needs

1

2 I ntro d u c ti o n

to know the gender of the child and ­whether they consumed all or only part of the ice cream (see also Boroditsky, Schmidt, and Phillips 2003). Th ­ ese grammatical quirks mean that your Russian-­speaking friend’s sense of this ice cream–­eating child is more nuanced than yours, simply b ­ ecause of what is demanded by the strictures of their tongue. Other instances carry more social weight. What follows is a fictionalized version of a real-­life experience. One day, the child of one of this book’s authors (the one without the accent marks) came home from school and said, “Mom, my friend, Jordan, d ­ oesn’t want to be he or she. I­ sn’t that how we talk at home about every­one else—­nobody is he or she?” The child was right. At home, they speak Estonian: a language that does not grammatically oblige speakers to denote the gender of objects or ­people. Indeed, he and she are signified by the same pronoun. In the home of Estonian speakers—­ and speakers of other genderless tongues throughout the globe (Pérez and Tavits 2019; Prewitt-­Freilino, Caswell, and Laakso 2012; Santacreu-­Vasut, Shoham, and Gay 2013; Tavits and Pérez 2019)—­people are not expected to grammatically signify the gender of objects or individuals, like they are when they speak Spanish or En­glish. Their language does not direct them to distinguish ­children as he or she. Now consider a similar conversation at an English-­speaking home about a pair of students named Jordan. One Jordan originally self-­identified as a girl, while the other Jordan still self-­identifies as a boy: child: “Mom, we had a meeting t­ oday about Jordan . . .” mom: “Oh, yes? Which one, the boy or the girl?” child: “Eer . . .” mom: “Which one, he or she?” child: “Eer . . . ​, I ­don’t know . . . ​, eer . . .” increasingly frustrated mom: “It’s a ­simple question! HE or SHE?” increasingly frustrated child: “It’s not a ­simple question! Jordan is neither he nor she! That’s the ­whole point, mom! Jordan ­doesn’t want to be ­either a boy or a girl!” The parent in this exchange was harshly struck by the realization that ­simple grammatical distinctions, encouraged by the use of gendered pronouns, can have an effect on one’s expectations of the world. In this case, the m ­ other’s language conditioned her to expect that a child is always e­ ither “he” or “she,” with very l­ittle room for other possibilities, such as being “neither a boy nor a girl.”

Los t i n Tra n s l ati o n ? 3

­These everyday examples underscore that grammatical nuances between tongues can draw our attention to dif­fer­ent features of our environment, perhaps affecting how we construct and interpret the world around us. This is a fascinating prospect that linguists and cognitive scientists have not ignored and, as we claim in this book, one that po­liti­cal scientists should not overlook ­either. ­After all, language is fundamental to the ­human experience. Language production and comprehension is a large part of what we do on a daily basis as h ­ uman beings. For example, Mehl et al. (2017) report that college students produce approximately 16,000 words per day. They listen to and comprehend at least as many words produced by their peers and ­others. Add to this all of the reading, writing, and social media interactions, and it becomes clear that h ­ umans use language incessantly, for large parts of each day. Given this scale, even small language effects could be potentially far-­ reaching and consequential. And ­because of this, a discipline that is studying ­humans, such as ours, cannot afford to ignore them. Furthermore, po­liti­cal scientists readily admit and study the diversity in po­liti­cal structures, ethnic makeup, culture, economic systems, ­etc., yet tend to glance over the vast and systematic differences across languages. According to the most comprehensive cata­logue of the world’s languages, t­ here are about 7,000 distinct tongues in con­temporary use (Eberhard, Simons, and Fennig 2020), with t­ hese languages varying significantly in terms of grammar, meta­phors, lexicons, and other dimensions. This represents a remarkable diversity in the linguistic practices of ­humans, a diversity that our discipline largely ignores at pre­sent. A cursory look at nuances between languages in the following three domains helps illustrate the enormous linguistic variation and the explanatory potential of it, in that even small linguistic nudges seem to have far-­reaching consequences: Space: When setting a ­table, an En­glish speaker might say the fork goes to the left of the plate. However, a speaker of Kuuk Thaayorre, an aboriginal Australian language, would say that the fork goes east of the plate if they happen to be facing south, west of the plate if they are facing north, southwest of the plate if they are facing northwest, and so on, based on cardinal directions (Boroditsky and Gaby 2010). In turn, speakers of Telzatal, a Mayan language, might say that the fork needs to be uphill of the plate (Brown and Levinson 1993). Thus, while En­glish speakers use egocentric frames of spatial reference, the other two languages use absolute frames. ­These nuances ­matter: speakers of languages that use absolute spatial references are more aware of their orientation and display better

4 I ntro d u c ti o n

navigation skills—­for example, they are less likely to get lost even in unfamiliar surroundings (Levinson 2003). Tenses and numbers: En­glish speakers use past, pre­sent, and f­ uture tenses when they speak. Not so in other tongues. In Finnish, for example, ­people regularly merge the f­ uture and pre­sent tense (Casasanto et al. 2004). In turn, speakers of Yagua, a Peruvian indigenous tongue, have five past tenses available, each denoting something that happened within a few hours; one day ago; within a few weeks; within a few months; or in a distant or legendary past (Payne 1997). Number distinctions (words for one, two, three, ­etc., as well as singular and plural) also seem so basic to En­glish speakers. Yet in global comparison they are not. Pirahã, an indigenous language in Brazil’s Amazonas, does not grammatically distinguish between numbers, including through pronouns (Everett 2012). Pirahã’s words for quantities are ambiguous from the ­angle of En­glish speakers. For example, the word hói can mean one or a few. ­These nuances are more than curiosities, for they guide ­people’s temporal outlooks and numerical sense (Gordon 2004). Nouns: Many languages construct nouns on the basis of biological gender. In Spanish, for example, the moon is feminine (la luna), while in German it is masculine (der Mond). Moreover, in French, all weekdays are masculine, while in Rus­sian, Monday, Tuesday, and Thursday are masculine and Wednesday, Friday, and Saturday are feminine (Sunday is neutral, in case you are curious). Yet for speakers of Dyirbal, another Australian aboriginal tongue, nouns are only partially based on biological sex. While one set of nouns is used when denoting men and most other animate entities, another class denotes ­women, some animals, fire, ­water, and violence-­related entities, thus inspiring the linguist George Lakoff ’s (1987) famous book title, ­Women, Fire, and Dangerous ­Things. A Language-­Opinion Connection?

We can look at examples closer to home. Consider, again, nuances in grammatical gender. Whereas En­glish, a Germanic tongue, obliges speakers to distinguish between he and she, speakers of Romance languages—­e.g., French, Italian, Portuguese, Spanish, and ­others—­are additionally required to designate the gender of all nouns, making Romance languages even more gendered than En­glish. Then t­ here’s a language like Rus­sian, which obliges its speakers to do what speakers of En­glish and Romance tongues already do

Los t i n Tra n s l ati o n ? 5

grammatically, but also requires inflecting verbs for gender in the past tense. In contrast, speakers of Estonian, Finnish, Hindi, Hungarian, Indonesian, Turkish, and Viet­nam­ese make use of absolutely no grammatical gender markers at all. They are, in a word, genderless. The use of tenses i­ sn’t any more consistent. While En­glish uses a f­ uture tense form to talk about tomorrow (e.g., “it w ­ ill rain”), Finnish lacks grammatical means of marking the ­future and, consequently, Finnish speakers talk about ­future using pre­sent tense (Dahl and Velupillai 2011). Temporal meta­phors vary as well. In En­glish, we think about time in terms of distance: “it was a long night,” “they have had a long friendship,” “another long meeting!” The Greek language, however, construes time in a manner akin to a quantity. Thus, a Greek speaker would say “it was a big night,” “they have had a big friendship,” and “oh no, another meeting that lasts much!” (see Casasanto et al. 2004). Languages also differ in the extent to which they use active versus passive voice. En­glish speakers use an active voice when describing events and are taught in school to prefer it. Hence, an En­glish speaker w ­ ill use transitive sentences, such as Jeremy spilled the coffee, even when this act was unintentional. Yet Japa­nese and Spanish speakers prefer not to mention the agent when describing accidents and would instead say that the coffee got spilled. Minor differences? Yes. Trivial consequences? No. Alas, how we remember events and assign causality can affect high-­stakes outcomes, like descriptions of eyewitness accounts (Fausey and Boroditsky 2011). Clearly, then, considerable diversity exists between languages and what they grammatically oblige their speakers to do. This is in­ter­est­ing for two po­liti­cal scientists like us b ­ ecause language is a valuable currency in politics. In mass publics throughout the globe, citizens draw on their own words to debate, deliberate, and ultimately choose what they believe is in their best interest. Indeed, p ­ eople use language to voice support for or opposition to vari­ous policies ranging from the mundane issues of, say, local trash collection, to the more central issues of in­equality, poverty, conflict, environmental sustainability, and more. ­These opinions, expressed through language, can have far-­reaching consequences when they influence ­whether we increase or decrease equality, advance or hinder development, prepare for or ignore the ­future, or even start or end wars. And we know that many times, public opinion systematically affects the course of politics (Stimson 1999, 2004; Stimson, Mackuen, and Erikson 1995). Does this mean that language is a fundamental force b ­ ehind mass opinion, influencing the shape of po­liti­cal attitudes, beliefs, and outlooks expressed by individuals?

6 I ntro d u c ti o n

You would think this is the case. If the language we speak can affect ­ hether we construe events as accidents or foul play (Fausey and Boroditsky w 2011), then it seems plausible it can also affect how p ­ eople interpret corruption, fraud, or poor government performance—­and even influence w ­ hether individuals are willing to hold public officials to account for ­these outcomes (Healy and Lenz 2014; Huber, Hill, and Lenz 2012; Lenz 2012; Malhotra and Kuo 2008). Moreover, if language influences w ­ hether we perceive the ­future as being very dif­fer­ent from ­today (Chen 2013), then it stands to reason that language can shape public support for future-­oriented policies, such as environmental protection or social security reforms (Shaw and Mysiewicz 2004; Winter 2006; Yeager et al. 2011). Indeed, inasmuch as language affects ­whether we express prejudice (Danziger and Ward 2010), it might also play a role in (de)escalating intergroup conflict (Horo­witz 1985; Kinder and Dale-­Riddle 2012; Tesler and Sears 2010). Similarly, insofar as language impacts ­whether we pay attention to gender distinctions (Boroditsky et al. 2003; Flaherty 2001), it might also shape our understanding of gender equality and ac­cep­tance of nonbinary individuals. Fi­nally, if language can impact our moral reasoning (Keysar, Hayakawa, and An 2012), it may shape our understanding of what is right and wrong in politics and policy, as well as how willing we are to compromise on our beliefs (Haidt 2012; Ryan 2017, 2019). Each of ­these examples drives home an intuitive point: that it is highly plausible for language to shape po­liti­cal opinions. And what we have provided above is just a short list of language effects with pos­si­ble po­liti­cal consequences. The ways in which linguistic differences might ­matter for politics are numerous. ­After all, language is the lifeblood of politics. Po­liti­cal scientists, therefore, should not turn a blind eye to the role that language might play in shaping po­liti­cal thinking and choices. Furthermore, if the language we speak can shape how we construct real­ity, then exploring how linguistic differences might shape mass po­liti­cal choices could push forward research in vari­ous subfields. For example, better understanding how language shapes thought can advance research in po­liti­cal psy­chol­ogy by broadening our understandings of attitude formation and change. In addition, greater knowledge of language effects on p ­ eople’s attitudes, be­hav­ior, and preferences can shed new light on cross-­country variation in policy choices and outcomes, relations between ethnic groups, international interactions, and many other phenomena. Our goal with this book is to start mapping the effects of language on po­liti­cally relevant attitudes by building a comprehensive framework that

Los t i n Tra n s l ati o n ? 7

explains ­whether, why, and how language affects mass opinion across several domains. We are not the first to display this curiosity. Po­liti­cal scientists, like the eminent David Laitin, have previously examined the link between language and politics (Laitin 1977). But the primary focus of prior research has been the macro-­level ­causes and consequences of language choice and policies (cf. Laitin 1992, 1998; Laitin and Ramachandran 2016; Liu 2015; Liu and Pizzi 2018), rather than how structural differences between tongues impact individual decision-­making. Indeed, very few studies that we are aware of have examined the in­de­pen­dent effects that language might have on po­liti­cal outcomes at the individual level (Laitin 1977). And, when this focus exists, many times the goal of scholars has been to treat language as an indicator of another variable of interest, such as ethnic differences or diversity, not linguistic differences per se (e.g., Chandra 2012; Ferree 2012; Horo­w itz 1985). For instance, prior research has studied language use to better understand ethnicity and ethnic relations (Adida et al. 2016; Garcia Bedolla 2005; Laitin 1998; Laitin, Moortgat, and Robinson 2012) or explored how language skills and exposure to foreign languages impact attitudes ­toward immigrant integration (Hopkins 2014, 2015; Hopkins, Tran, and Williamson 2014; Sobolewska, Galandini, and Lessard-­Phillips 2017). ­These are all impor­tant studies that inform our own work. But ultimately, they are not what we are r­ eally interested in—­namely, the linkages between the language that ­people speak and the po­liti­cal views they express. The goal of this book is to begin mapping ­these effects of language on po­liti­cally relevant attitudes. How hard can this r­ eally be, right? If linguists and cognitive scientists have accumulated evidence of language effects on ­human thought, then it seems a ­simple m ­ atter to just graft ­these basic insights onto the study of po­liti­cal opinion. Yet, as many po­liti­cal scientists know, the realm of politics is a peculiar one, where many of the qualities that research on language effects takes for granted—­heightened attentiveness, strong engagement, and considerable effort—­are relatively weak, if not entirely absent among individuals (Delli Carpini and Keeter 1996; Lodge and Taber 2013; Pérez 2016a; Zaller 1992). Hence, explaining how language impacts po­liti­cal thinking demands a theoretical leap not yet taken, one connecting thinking for speaking to the cognitive and affective pro­cesses under­lying individual opinions (Tourangeau, Rips, and Rasinski 2000). As we ­will argue in the chapters to come, language influences public opinion, not by determining what ­people think about or by providing access to a distinct po­liti­cal real­ity

8 I ntro d u c ti o n

(cf. Sapir 1958; Whorf 1940). Instead, language systematically aids in the interpretation of politics and the expression of po­liti­cal opinions. —­—­— But even if questions about theory are answered, another clarion call beckons: how to establish, empirically, that language impacts public opinion? Again, ­there is more than meets the eye when looking at previous work in linguistics and cognitive science. Th ­ ere is, for example, no getting around the fact that prior work has done an impressive job of establishing that language can cause shifts in p ­ eople’s thinking in several domains, including individual judgments of space, time, and objects (see Boroditsky 2001 and Pérez 2015 for overviews). But t­ here is also no getting around two other facts, which are inextricably tied to each other—­and which ­matter im­mensely for po­liti­cal science and other data-­driven disciplines. The first is that, notwithstanding the breathtaking range of domains in which researchers have uncovered language effects, the one that is most impor­tant to us—­public opinion—is the one that is glaringly missing from this accumulated inventory. Only one study that we are aware of has directly assessed w ­ hether differences between languages can causally influence what individuals opine about politics (Laitin 1977). The second uncomfortable fact is that, while evidence about the generic causal connection between language and ­human cognition is strong, evidence of this pattern’s breadth throughout the world is surprisingly weak. Prior work has mostly unearthed language effects in carefully controlled lab settings, with small samples of con­ve­nience, a l­ imited range of language treatments, and a lack of outcomes that directly speak to politics—­hardly the kind of stuff that ­will convince other social scientists that language influences mass opinion. This tradeoff between internal and external validity (Campbell and Stanley 1963)—­between pinning down the causal effect of language and establishing its robustness—is a real one in po­liti­cal science, where evidence that generalizes to the rough-­and-­tumble world of politics has greater appeal. The advantage, as we see it, is that in convincing po­liti­cal scientists about the merits of language effects, we also extend and fortify what language researchers have already done before us. To move ­these boulders forward, then, a more convincing set of research designs is needed to make the case that language shapes mass opinion.

Los t i n Tra n s l ati o n ? 9

­Toward a Theory of Language-­Opinion Effects

With a now clearer sense of the theoretical and empirical stakes involved, let us turn h ­ ere t­ oward a basic sketch of our theoretical argument, which we develop in greater depth in chapter 1. We use this framework to evaluate language-­opinion effects in several po­liti­cal domains, including mass opinion t­ oward gender equality, environmental policy, ethnic relations, and candidate evaluation. This approach—of studying language effects across multiple domains—­lets us test several observable implications of our argument, allows us to isolate varied circumstances ­under which language-­opinion effects wax and wane, and enables us to distinguish language effects from the impact of p ­ eople’s cultures. In ­these ways, we chart new territory in our understanding and appraisal of language effects, thus helping to advance language effects research in both theoretical and methodological terms. Our framework rests on two major pillars. The first is supplied to us by Dan Slobin, a famed linguist who birthed the notion of thinking for speaking (Slobin 1996, 2000). The gist of Slobin’s s­ imple but power­ful idea is that languages vary in the degree to which they grammatically force their speakers to attend to and encode certain aspects of their environments (Boroditsky 2001; Fuhrman et al. 2011; Ogunnaike, Dunham, and Banaji 2010; Slobin 1996, 2003). Think, for example, about basic grammatical differences between languages, such as the nuances between a gendered and genderless tongue. If one wishes to say the sun in Spanish (a gendered language), one would need to say el sól, with the definite article el denoting that the sun is masculine in that tongue. Yet a genderless language like Estonian would not require a speaker to distinguish the sun as male or female—in fact, as we explained before, even the word for he and she is the same in this tongue. Thinking for speaking is therefore qualitatively dif­fer­ent for Spanish and Estonian speakers. This implies that dif­f er­ent language speakers can be biased ­toward focusing on ­those aspects of the world that their tongue demands. If speaking a language requires one to make certain distinctions between objects, such as colors, gender, and time orderings, then speakers take for granted that ­these categories actually exist in the world and are relevant to a judgment, task, or opinion (cf. Boroditsky 2001; Boroditsky et al. 2003; Cubelli et al. 2011; Fuhrman et al. 2011; Vigliocco et al. 2005). Thus, language can structure thought by making some distinctions more salient (Cubelli et al. 2011; Hunt and Agnoli 1991) and certain ­mental associations and categories more accessible than ­others (Danziger and Ward 2010; Ogunnaike et al. 2010).

10 I ntro d u c ti o n

Our second pillar, belief-­sampling, comes to us from John Zaller, Roger Tourangeau, Milton Lodge, Charles Taber, Norbert Schwarz, and other social scientists invested in understanding how, exactly, ­people form and express their opinions about politics. In par­tic­u­lar, the work of ­these individuals and their colleagues teaches us that the average citizen approaches public affairs with low levels of attention, information, and effort (Lodge and Taber 2013; Pérez 2016a; Schacter 1999; Simon 1985; Zaller 1992). Instead of having ready-­made opinions on all ­matters, ­people construct opinions based on considerations that are salient when a topic is broached (Lodge and Taber 2013; Schwarz 2007; Tourangeau et al. 2000; Zaller 1992)—­considerations that may have been made more salient by the features of the language that they speak. Merging thinking for speaking with belief-­sampling yields our language-­ opinion hypothesis. This is the falsifiable notion that language affects most aspects of opinion formation and expression in a mea­sur­able way: from how ­people utilize information, to what considerations they retrieve from memory, to how t­ hese considerations inform one’s opinions. R ­ unning through this claim is a basic pro­cess. Language makes some ­mental content more accessible in ­people’s minds (see Pérez 2015 for a review). That is, the language one speaks makes some associations, beliefs, knowledge, and values more prominent in ­people’s memory, thus heightening their accessibility. This is what nudges opinions in predictable directions. Our argument is versatile in more ways than one. It can explain ­whether grammatical features of a language w ­ ill impact the po­liti­cal opinions p ­ eople express. It can also explain how language shapes po­liti­cal opinions, focusing on the influence of language on the learning and recall of relevant information for opinion formation. It is also able to anticipate when language ­will ­matter for mass opinion, and just as impor­tant, when it ­will not. Fi­nally, our framework can clarify why language can affect mass opinion across a diverse set of seemingly unrelated po­liti­cal domains that include gender and LGBTQ equality, environmental policy, ethnic relations, and candidate evaluation and choice. For example, as we have already explained, languages vary in the degree to which they grammatically oblige speakers to attend to and encode gender as part of their everyday experience. Our language-­opinion hypothesis suggests that speaking a genderless language should promote greater perceived equity between men and w ­ omen. By neglecting to formally distinguish between male and female objects and individuals, speakers of

Los t i n Tra n s l ati o n ? 11

t­ hese languages are not as sensitized to pay attention to gender, including traditional gender roles and categories. With gender not as mentally salient, speakers of genderless tongues should express more gender-­blind attitudes, both in terms of gender equality and ac­cep­tance of nonbinary individuals. Similarly, languages across the world vary by w ­ hether they require speakers to attend to and encode time. Futured tongues, like French, oblige speakers to use specific verbs to differentiate between temporal tenses, which alerts speakers that the ­future is distinct and distant from the pre­sent. In contrast, futureless tongues, like Finnish, do not oblige speakers to grammatically distinguish between ­these tenses, making tomorrow seem temporally closer and more similar to ­today. Drawing on our language-­opinion hypothesis, we would expect that speakers of futureless languages ­will treat pending rewards as less distant and more pressing than speakers of futured tongues, leading them to support efforts and practice be­hav­iors that are more present-­oriented. Beyond grammar and lexicon, language can affect po­liti­cal opinions ­because it plays a role in the encoding and recall of po­liti­cal information—­a crucial aspect of the opinion-­formation pro­cess. For example, in many nations throughout the world, the language of an ethnic majority coexists—­ sometimes peacefully, but often in tension—­with the language of a minority. Our language-­opinion hypothesis predicts that majority and minority languages w ­ ill prime dif­f er­ent considerations in memory, thus directly impacting what individuals attend to when thinking about politics. ­Here, a minority language should draw speakers’ attention to the presence of ethnic divisions in society, given the status of that tongue in society. When speaking a minority language, ­people are therefore more likely to notice and prioritize ethnic divisions when expressing po­liti­cal opinions. That nudge is absent when speaking the majority language in the same society. Yet in other instances, speakers of a minority tongue may develop slightly dif­f er­ent sets of ideological associations and beliefs simply ­because their languages expose them to varied sources of information and, as a result, greater access to specific content in memory. In the United States, for example, Latino individuals might have m ­ ental access to dif­fer­ent types of information depending on the language (En­glish or Spanish) they use when they learn and retrieve this content. Hence, who p ­ eople are and what they believe about politics might depend, in some mea­sure, on the language they use to navigate the po­liti­cal world and form opinions about it.

12 I ntro d u c ti o n

­ oward a Methodological Assessment T of Language-­Opinion Effects

Our primary method for drawing reliable inferences is experimentation. Most of our empirical studies rely on experiments with bilinguals who live in the same cultural context. In t­ hese experiments, we randomly assign the language in which respondents express their po­liti­cal opinions. This ­simple, yet power­ful, design allows us to effectively hold constant all other (un)observed differences between bilinguals besides language and clearly identify any linguistic effects. This is worth stressing. We do not simply compare groups of individuals who speak dif­fer­ent languages—­we randomly assign the language in order to identify its effect on opinions. We conduct our experiments in large and heterogeneous samples of bilingual adults in nations like the United States, Sweden, and Estonia. This diversity of national contexts allows us to exploit several naturally occurring linguistic features (e.g., the grammatical nuances between a futured and futureless tongue or the availability of pronouns to denote ­people as masculine, feminine, or in a nonbinary way) that shed new light on the language-­opinion connection. That is, we did not pick ­these countries at random or out of con­ve­nience, but focused on them b ­ ecause they provided the language contexts that best allowed us to identify the vari­ous language effects. For example, Sweden is the only country in the world that has formally ­adopted a gender-­neutral pronoun, permitting us to design realistic experiments where we randomize gendered versus genderless pronoun use. Similarly, Estonia is a relatively unique context where a sizeable bilingual population speaks languages that differ on multiple relevant dimensions, including ­whether they are gendered, futured, or have a minority status. Randomizing language assignment in this context allows us to identify the effect of grammatical and nongrammatical features of a language on vari­ ous outcomes of interest. We discuss the advantages as well as potential limitations offered by t­ hese research sites in greater detail in the respective chapters. We then pair t­ hese experiments, whenever pos­si­ble, with systematic analyses of cross-­national survey data to more fully and convincingly show that language shapes public opinion ­toward gender equality, LGBTQ tolerance, environmental conservation, ethnic conflict, and po­liti­cal candidates beyond the immediate context of many of the experiments we conducted. In this way, our research strategy significantly enhances the external validity of language effects research (Kinder 2011; McDermott 2011), which to date

Los t i n Tra n s l ati o n ? 13

has largely been confined to small-­scale lab experiments, often with con­ve­ nience samples of undergraduate students. Methodologically, we also illuminate how language effects on public opinion manifest themselves—­that is, we address the question of mechanisms (Baron and Kenny 1986; Elster 1989). Prior research on language effects has mainly focused on establishing that language c­ auses shifts in ­people’s thinking. Scholars have spent less time, however, on isolating the precise mechanisms through which this influence occurs: the how and the when. To this end, we craft a series of experiments that lead to four types of innovation in terms of mechanisms. First, our experiments demonstrate that language effects are generated not only by the grammatical quirks that certain tongues make salient—­such as the obligatory gendering of objects—or the precise words employed by language speakers in everyday situations, but also by priming certain po­liti­ cally relevant associations. While the grammatical and lexical effects have been part of prior work on language effects, the priming effects we report have not. Yet from the point of view of attitude formation and change (e.g., Taber and Young 2013; Valentino and Nardis 2013), or even from the point of view of multilingual survey research (e.g., Pérez 2009, 2011), understanding such priming effects is crucial to grasping language’s influence on mass opinion, since so many patterns in what p ­ eople think about politics are generated, often mechanically, via priming (Lodge and Taber 2013; Tourangeau et al. 2000; Zaller 1992; Zaller and Feldman 1992). Second, some of our experiments specifically pinpoint how language shapes opinions via thinking for speaking. To this end, our experiments uncover the power­ful influence of language in making certain m ­ ental associations and categories more cognitively accessible, which allows individuals to draw on them to form and express their opinions about politics more easily. This evidence allows us to document, in a clearer way than before, a fuller chain of causation from linguistic features to ­mental salience and accessibility of certain associations to expression of attitudes reflecting ­those associations. In this way, our experiments expose the “black box” of language effects to new light, helping to clarify with added precision how the tongue one speaks impacts the attitudes and outlooks about politics that one reports. Third, we have designed some of our experiments to explore when language effects weaken or dissipate. This allows us to establish that the influence of language on mass opinion is variable and can be minimized by information that ­people have at their disposal. Specifically, we establish the influence of social norms, i.e., a clear sense that certain opinions are

14 I ntro d u c ti o n

(de)valued by o ­ thers in the mass public, in attenuating language effects. Establishing when language does and does not ­matter in relation to other stimuli offers significant theoretical advancement over existing work, which has single-­mindedly focused on establishing language effects. It also has significant practical relevance—­a constant concern for po­liti­cal scientists—­ because it offers directions on how to overcome language effects that may be socially deleterious. Fi­nally, our methodological approach wrestles, systematically and profoundly, with one of the more per­sis­tent bugbears in research on language and thinking: the role of culture. Prior work has been frustratingly unable to fully distinguish the effect of language from that of culture. We do this in chapter 3 by taking advantage of three unique experiments with adults who share the same cultural milieu, speak the same language, but who use dif­fer­ent words to express po­liti­cal views. And, as our data permit, in other chapters we control statistically for specific manifestations of culture, including national and ethnic identities, ideological worldviews, and specific value dispositions. Combined, this collection of methodological interventions provides robust and consistent evidence that language—­independently of innumerous other forces—­systematically nudges public opinion in meaningful directions. A Look Ahead

The path from idea, to theory, to research design, and data analy­sis is a long and winding one, with several obstacles to surmount along the way. We therefore start our journey in chapter 1 by engaging a smoldering debate around what is known as the Sapir-­Whorf hypothesis: the broad notion that differences between tongues determine how ­humans interpret their world. Critics have quashed this hypothesis as unfalsifiable, noting that it is stated so broadly that it is impossible to know when it is wrong. Moreover, empirical support for this hypothesis has rested more on scintillating anecdotes, rather than hard scientific evidence proper. Th ­ ese two features alone would have done in any hypothesis. But we ­will learn that, instead of burying this idea, a cadre of linguists that includes Dan Slobin, Lera Boroditsky, and o ­ thers has refashioned this claim into a more constructive conversation about when, why, and how language can, generically, shape ­human cognition. We ­will therefore spend some time familiarizing ourselves with the basic cognitive princi­ples of language effects laid out by t­ hese researchers, while also interrogating the evidence supporting ­these tenets.

Los t i n Tra n s l ati o n ? 15

Based on ­these efforts, we then formulate our own theoretical argument to explain language effects on mass opinion throughout the globe—­and across diverse po­liti­cal domains that include gender equality, environmental policy, ethnic relations, and candidate evaluation. This new framework creates synergy between two heretofore isolated mechanisms. The first of ­these is thinking for speaking—­the idea that languages vary in their grammatical organ­ization, which obliges speakers to focus on certain aspects of their world (Slobin 1996, 2000). The second mechanism is belief-­sampling—­the notion that instead of having ready-­made opinions on all m ­ atters, ­people construct opinions based on considerations that are salient when a topic is broached (Lodge and Taber 2013; Tourangeau et al. 2000; Zaller 1992). Blending ­these components into a unique alloy yields our language-­opinion hypothesis: the falsifiable claim that language shapes opinion formation and expression by making some ­mental content more accessible, which nudges opinions in predictable directions. Together, this framework and general hypothesis guide our empirical hunt for language effects on public opinion in the chapters that follow. Having clarified our theoretical stakes, chapter 2 begins a steady campaign to empirically appraise our language-­opinion hypothesis. To this end, we focus on the realm of gender in­equality, exploring ­whether the presence (absence) of grammatical gender affects how p ­ eople think about gender equality. We draw on a trio of studies to accomplish this goal. The first of ­these is a large-­scale survey experiment in Estonia, where we randomly assigned bilingual adults to complete their interview in Rus­sian (a language that uses grammatical gender) or Estonian (a language that does not use grammatical gender at all). We find evidence that strongly aligns with our language-­ opinion hypothesis. That is, in comparison to individuals who interview in Rus­sian, ­those who interview in Estonian express more liberalized attitudes ­toward gender and gender relations, including weaker ste­reo­typical views about ­women and greater support for females in vis­i­ble and meaningful po­liti­cal roles and offices. We then show, using rich cross-­national survey data and sophisticated statistical modeling techniques, that this empirical regularity is not a narrow function of Estonian-­Russian bilinguals or the national setting in which they find themselves. Indeed, across a variety of mass publics throughout the world, individuals who report speaking a genderless tongue at home express similarly liberalized attitudes t­ oward the increased presence of ­women in society, the economy, and politics. We then cement t­ hese results with a second experiment that replicates our core finding, while also illuminating when language effects weaken. Specifically, we

16 I ntro d u c ti o n

show that language effects on mass opinion about gender relations dissipate when individuals can avail themselves of clear information about public (un) popularity of mea­sures tackling gender disparities. In chapter 3, we continue our assessment of language effects by solving a nagging challenge that even linguists have had a hard time addressing: namely, the extent to which language effects are driven by structural differences between tongues, rather than cultural differences between individuals. Imagine, for example, asking a Spanish speaker and an En­glish speaker their opinions about traditional gender roles and finding that the former individual expresses more conservative views than the latter. Does this pattern emerge ­because Spanish is a more gendered tongue than English—or b ­ ecause Spanish speakers inhabit a paternalistic culture characterized by machismo? To gain leverage over ­these points, we travel to Sweden to study its citizenry’s use of gender-­neutral pronouns and how this might affect their views about gender equality and LGBTQ rights. In 2015, Sweden a­ dopted the use of a new gender-­neutral pronoun, hen. This means that as part of their grammatical toolkit, Swedes now seamlessly use hen alongside the explic­itly gendered hon (she) and han (he). We have ­here, then, a nation that introduces a new pronoun, where ­people—­all sharing a single national culture—­rehearse this pronoun in their everyday lives. What do we find? Three major patterns. First, we discover that the words that ­people use— in this case, pronouns—­matter in terms of nudging ­people’s views about gender and LGBTQ equality in a liberalizing direction. Second, ­these language effects operate through a mechanism that puts our language-­opinion hypothesis into even sharper relief. Specifically, we demonstrate that gender-­ neutral pronouns decrease the m ­ ental salience of males, which then has the downstream effect of liberalizing ­people’s views about gender and LGBTQ equality. Last, since all participants inhabit the same cultural milieu but vary in terms of their pronoun use, we can say more definitively than before that ­these effects are due to linguistic nuances, rather than cultural effects in disguise. Chapter 4 broadens our empirical scope by delving into the question of language effects on p ­ eople’s construal of time. Similar to grammatical gender, languages throughout the world vary in the degree to which they require speakers to attend to and encode temporal nuances. For example, some languages make use of a separate ­future tense (we call ­these futured languages), while ­others use pre­sent tense to talk about the ­future (i.e., futureless tongues). We seize this grammatical feature to assess w ­ hether language systematically impacts mass opinion about future-­oriented policies.

Los t i n Tra n s l ati o n ? 17

Accordingly, we undertake two new studies, each yielding more evidence in ­favor of our language-­opinion hypothesis. The first of ­these is another original experiment where we manipulate interview language in a public opinion survey of bilingual adults who speak a futured (Rus­sian) and futureless (Estonian) tongue. Among other tantalizing results, we find that individuals who express their opinions in a futureless tongue are significantly more supportive of imposing a “green” gasoline tax in order to help protect the environment. In other words, the nature of the language that ­people interview in nudges them to make po­liti­cal choices that address long-­term policy goals. We then avail ourselves one more time of rich cross-­national data to show that this basic empirical regularity exists more widely across mass publics beyond the one in which we undertook our experiment. Up to this point, our empirical chapters focus on the role grammatical features of tongues play in shaping public opinion. Chapters 5 and 6 shift gears by grappling with some of the ways in which language subtly, but indelibly, leaves its stamp on the attitudes, beliefs, and outlooks that ­people report when asked about their views. In par­tic­u­lar, we zero in on the interplay between language and priming. In the study of public opinion, priming is a work­horse mechanism whereby ­mental content is made more accessible through features that are extraneous to the domain about which one’s views are mea­sured. This includes the wording and order of questions, the response options that items offer, the presence or absence of interviewers—­ and, as we aim to demonstrate—­the language that one interviews in. To begin establishing this point, chapter 5 examines ­whether the ­simple act of interviewing in a minority versus majority language can heighten the ­mental salience of ethnic diversity and divisions. In other words, does the language that p ­ eople interview in impact how much they notice and take into account ethnic divisions when expressing po­liti­cal opinions? Drawing on our theoretical framework, we reason that interviewing in a minority language should prime ethnic divisions, leading ­people to take these divisions into account when judging politics. We test this claim with two survey experiments that assign ­people to interview in ­either a minority (Rus­sian) or majority (Estonian) language. Consistent with this priming-­via-­language mechanism, we discover that ­people who are assigned to report opinions in a minority language rank ethnic relations as a more impor­tant po­liti­cal issue and are more likely to correctly identify anti-­minority actors in politics. Language, we find, makes ethnic diversity more mentally accessible, which leads ­people to directly integrate this consideration into their views about politics. Building on this unique finding, we then show, in a now familiar fashion,

18 I ntro d u c ti o n

that the empirical regularity captured by this experimental result emerges in nations beyond the one where we undertook our experiment, suggesting that the degree to which p ­ eople notice and approve of ethnic diversity in society depends on ­whether or not they interview in a minority language. Chapter 6 then considers w ­ hether language shapes opinions through linguistic tags, i.e., the way information is stored and or­ga­nized in memory. Working in a US context, where civic affairs are overwhelmingly conducted and enshrined in English—­the majority language—we argue that information in this domain is stored and or­ga­nized (i.e., tagged) in En­glish, even if some new information in this domain is learned in a dif­f er­ent language, such as Spanish. To show this, we undertake two original experiments with Latino bilingual adults. This pair of studies is unique in that they assign bilinguals to learn new po­liti­cal information (about a po­liti­cal candidate) and report their po­liti­cal judgments in one or both of their tongues. As a result, we can observe individuals who learn new po­liti­cal information in En­glish and express a po­liti­cal opinion also in En­glish, as well as individuals who learn the same new information in Spanish but report a judgment in En­glish. This unique design allows us to show that language affects the encoding and retrieval of po­liti­cal information. That is, language directs ­people to integrate new po­liti­cal information (e.g., details about a candidate) with other related considerations (i.e., other po­liti­cal information) they have already learned in En­glish. We also show that language affects how this information is recalled and affects opinions: interviewing in a majority tongue (En­glish) as opposed to a minority language (Spanish), both facilitates recall of this new information and facilitates making connections to other po­liti­cal information, which are tagged in En­glish in memory. Language, it appears, also shapes how po­liti­cal information is learned and retrieved. In our concluding chapter, we review t­ hese findings, taking special care to tease out and discuss their implications for the study of po­liti­cal attitudes. In par­tic­u­lar, we discuss how a better understanding of the language-­opinion connection deepens our grasp of po­liti­cal decision-­making in mass publics. We also delve into the vari­ous insights that our findings provide for some major po­liti­cal science questions of our time. But first, let us start with the basics by turning to chapter 1 and learning about the promise and pitfalls of studying language effects from the perspective of linguists.

1 Explaining the Language-­ Opinion Connection Language shapes the way we think, and determines what we can think about. —­B EN JA M I N W H O R F

The h ­ uman tongue is one of our species’ most venerable tools, honed across millennia to satisfy a host of personal and social needs. Yet the scientific study of language and ­human cognition is of fairly recent vintage. The notion that dif­fer­ent language speakers perceive distinct realities has been around since at least the late 1700s, when German phi­los­o­phers like Johan Georg Hamann, Johann Gottfried, and Wilhelm von Humboldt engaged in erudite discussions about the topic. But it was in the early twentieth c­ entury that Edward Sapir gave this provisional idea its scientific imprimatur. Working at the crossroads of anthropology and linguistics, Professor Sapir studied Amerindian languages like Navajo and Southern Paiute, using his vast knowledge to plant the seeds of a princi­ple that would ­later dominate research in ­these fields. In his view: ­ uman beings . . . ​are . . . ​at the mercy of the par­tic­u­lar language which H has become the medium of expression for their society. . . . ​[T]he “real world” is . . . ​built upon the language habits of the group. No two languages are ever sufficiently similar to be considered as representing the same social real­ity. (Sapir 1958: 69) 19

20 C H A P TE R 1

But if Sapir enshrined ­these basic tenets for anthropologists, linguists, and other scientists to follow, it was ­really his protégé, Benjamin Lee Whorf, who promoted their veneration in the acad­emy and in the popu­lar mind. By day, Whorf worked as a fire prevention inspector for the Hartford Fire Insurance Com­pany. By night (or whenever ­else he made time) he was an amateur linguist who was also engrossed with studying aboriginal tongues. Very l­ ittle about Whorf ’s efforts and insights was amateurish, however. He was a voracious reader, an astute observer, and a gifted communicator—­and he harnessed ­these skills to pursue self-­directed research into indigenous languages, such as Náhuatl. In 1931, he enrolled at Yale University as a part-­ time gradu­ate student u ­ nder Sapir’s tutelage, where he refined his mentor’s thinking about language; published prodigious amounts on his own efforts in leading linguistic journals; and heralded his unique insights to scientific and lay audiences alike. During his heyday, Mr. Whorf was known for many t­ hings—­his penchant for studying esoteric languages, his interminable memory for linguistic anecdotes, and his voluminous capacity for writing about languages. Yet nuance was not one of his strengths. In Whorf ’s doctrinaire view, all h ­ uman thought was determined by the unique worldview laid out by one’s tongue. This notion, now known as the Sapir-­Whorf (SW) hypothesis, implies that languages are distinct from each other, to the point of being incomparable (i.e., linguistic relativity); and that nuances between tongues determine how h ­ umans interpret their world (i.e., linguistic determinism). Whorf ’s intellectual activities emboldened this notion and spread it throughout the acad­emy and beyond, culminating in a towering edifice of knowledge on language and thinking. But it was a ­house of cards that caved with scrutiny. Despite the force with which Whorf expressed his views throughout intellectual circles, skeptics found linguistic determinism to be, well, indeterminate. The first swipe came from ­those judging Whorf ’s claim of linguistic relativity as too ham-­ fisted. According to this rebuke, if we are interested in studying the effects of two or more languages on ­human thought, but ­these languages are dissimilar, then what are we r­ eally d ­ oing? The answer, said skeptics, is comparing “apples” with “oranges.” Yet in order to eke out empirical evidence that language impacts thinking, it is imperative that we compare “apples” to “apples.” ­People must perceive the same t­ hing before they can claim t­ here is linguistic variation in their construal of it. For example, if I reason that language shapes personal attitudes, then ­these attitudes must be similar before I can say t­ here are any differences in their expressed levels that are induced

Th e La n g uag e - Opi n i o n Co n n ec ti o n 21

by language. It is a subtle point, but a crucial one to allow. For without it, the SW hypothesis becomes a self-­defeating exercise in identifying nuances between languages without actually being able to contrast them. The SW hypothesis was also landed a punch by skeptics who considered it unfalsifiable. This was deemed especially true of the assertion that language determines thought. In hindsight, it was perhaps a poor choice of words by Sapir and Whorf. But the prob­lem was ultimately theoretical, not semantic. A hypothesis anticipating that language determines thought rules out any mitigating circumstances to ­these anticipated effects—­circumstances that might prove theoretically illuminating. For example, are ­there conditions ­under which language effects are less likely? Are language effects more prevalent among some types of individuals? The SW hypothesis snuffed out ­these possibilities by fiat. As a result, some scholars found it challenging to know when this claim was wrong and when it was actually right. For this ­gaggle of doubters, the SW hypothesis was tantalizing on the surface, but vacuous at its core. A third blow to the SW hypothesis was struck by critics who decried its weak empirical basis. This notion initially may have been parsimoniously elegant, ­these critics acknowledged. But in the end, it was mostly supported by direct observation and anecdote—­hardly convincing, scientific evidence. What is more, when some researchers modified this hypothesis in order to submit it to rigorous examination, it buckled ­under the pressure. Case in point: well-­crafted research showed that non-­English speakers could, in fact, learn En­glish color categories, even if their own native tongue possessed few words for color (Heider 1972; Rosch 1975). From where ­these critics stood, the SW hypothesis was still in desperate search of air-­tight evidence. This sustained succession of jabs put Sapir and Whorf ’s thinking about language on the ropes. But it was the changing winds of science in the late 1950s that fi­nally secured its downfall, with the linguist, Noam Chomsky, his students, and their followers burying it u ­ nder an avalanche of new theorizing and evidence on linguistic universalism. Chomsky’s idea of Universal Grammar was the diametric opposite of SW, suggesting that all h ­ umans come equipped with a shared, hardwired capability to learn language. H ­ uman languages are thus mainly defined by their commonalities, not by their trivial differences. It was a paradigmatic shift that would prevail for de­cades in the scientific study of language. When the tides of skepticism about linguistic determinism receded, scholars found themselves beached on the shores of ­whether language could affect h ­ uman thought at all. It would take a new cadre of researchers de­cades, beginning in the late 1980s, to fully rehabilitate

22 C H A P TE R 1

this lit­er­a­ture into a more constructive conversation about how language might affect thought. In the rest of this chapter, we walk readers through t­ hese developments, sensitizing them to what we now know about the connections between language and thought. We w ­ ill learn that scholars have replaced the e­ arlier braggadocio of the SW hypothesis with a more mea­sured understanding of how, exactly, language influences ­human thought. We ­will also learn about the ways in which this new generation of scholars has added empirical bite to theoretical claims about language effects through clever and rigorous research designs. But we w ­ ill also learn that, for all of this new energy, insight, and evidence about language effects, the exact ways in which language shapes po­liti­cal thinking—­the focus of our book—­remains shrouded in mystery. This blind spot arises ­because po­liti­cal cognition involves more than simply grafting general insights about language effects onto domain-­ specific thinking. Indeed, as we ­will show, po­liti­cal thought is a unique realm where many of the qualities that language effects research takes for granted—­ heightened attentiveness, strong engagement, and considerable effort—­are relatively weak, if not absent (Delli Carpini and Keeter 1996; Zaller 1992). Thus, explaining how language impacts this form of cognition demands a theoretical leap that we ­will take ­toward the end of this chapter. But we are getting too far ahead of ourselves. Let us therefore begin, first, with the fallout of the SW hypothesis and the scholars who have picked up the pieces to reconstitute linguistic determinism into a more respectable idea with a new name, a new look, and a new set of falsifiable expectations. Taming the Big Bad Whorf

Following the implosion of the SW hypothesis, scholars of language effects found debris scattered in three distinct directions, allowing many of them to eventually reconstruct what went wrong with this theoretical framework, while setting the stage for what could be improved about it. The first component ­here, linguistic relativity, involved the notion that each language entailed its own unique worldview, which allegedly led speakers of varied tongues to experience distinct realities. On the surface, this strict sense of linguistic relativity seems uncontroversial enough: if you speak a dif­fer­ent language, then you perceive a distinct real­ity. But as we already hinted, a deeper pondering of this claim reveals a fundamental flaw. Insofar as two or more languages are, by definition, incomparable, our ability to learn anything about how the nuances

Th e La n g uag e - Opi n i o n Co n n ec ti o n 23

between them generate differences in individual thinking is ruled out by fiat. This is perfectly captured by the circular logic inherent in linguistic relativity: language A is structurally dif­fer­ent from language B; therefore, speakers of tongue A see the world distinctly from speakers of tongue B. Why does this perceptual nuance arise? ­Because language A and language B are dif­fer­ent! The prob­lem h ­ ere, as should be clear, is not that we cannot learn anything by making comparisons between tongues. Alas, the entire scientific enterprise is inherently comparative (Geddes 2003; King, Keohane, and Verba 1994). The challenge, instead, is that any comparison between two or more languages demands a common scale or dimension, so that variations in a concept can be understood in a relative and absolute sense. In the case of linguistic relativity, the fundamental question becomes: relative with re­spect to what? Viewed this way, it is easier to appreciate that the apples to oranges perspective laid out by Sapir and Whorf is more logically and conceptually tenable if we transform it into an apples to apples comparison. IN WHORF’S WAKE: THINKING FOR SPEAKING

This is where the linguist, Dan Slobin, comes in. A trained psycholinguist, Slobin shared some of his colleagues’ apprehension about the circularity implied by Sapir and Whorf ’s claims. But he was also unpersuaded that language had no effects on ­human thought. His path out of this morass is what became known as thinking for speaking (Slobin 1987, 1996, 2003). For Slobin, the ham-­fistedness of linguistic relativity could be relaxed and made more tenable if we shifted scholarly focus from uncovering linguistic differences in worldview to unearthing nuances in how ­people harness language in order to interpret their surroundings. In Slobin’s view, ­people encounter similar realities as they navigate the world, which puts all individuals on a comparable scale. What varies with re­spect to language is the degree to which a person’s tongue obliges them, grammatically, to attend to specific aspects of a shared experience. For example, imagine a Spanish speaker and En­glish speaker joining a clutch of strangers in a room. Whereas the En­glish speaker might seamlessly interact with ­these ­people by informally referring to them as you, our Spanish speaker must grapple with distinguishing the more intimate and informal tú from the more formal and distant usted. This relative nudging, on the basis of language, is what Slobin calls thinking for speaking, and it is what has single-­handedly helped to revive interest in assessing the effects of language on ­human cognition. As Slobin describes it:

24 C H A P TE R 1

­ very utterance represents a se­lection of characteristics, leaving it to the E receiver to fill in details on the basis of ongoing context and background knowledge. Part of the background is knowledge of what is obligatory or typical of the language being used. (Slobin 2003: 2) The shift in focus h ­ ere is both subtle and profound. Unlike Sapir and Whorf, and the intellectuals ­shaped by their thinking, Slobin is not proposing or implying that events, situations, worldviews, or other entities are presented by the world to individuals in static form, where their job is simply to encode them to memory. Rather, individuals encounter similar realities that they filter with the help of language: a pro­cess that helps ­people give meaning to t­ hose realities. Thus, it is precisely in this area of interpretation and meaning-­making that one is more likely to uncover the influence of language, since, according to Slobin, languages vary appreciably in the extent to which they grammatically privilege some aspects of the world over o ­ thers. This notion of thinking for speaking once again reminds us that all science—­very much including linguistics—is incremental and cumulative. Slobin did not discard the SW hypothesis so much as refurbish it with slight tweaks. But his innovations clarify, more brightly than before, how language can affect h ­ uman thought, thus exposing this black box to new light. In ­doing this, Slobin teaches us that language does not shape cognition by granting us access to unique ­mental content that is already formed, unchanging, and ripe for retrieval. Rather, the power of language stems from its ability to nudge ­people in terms of how they sort, filter, construct, and interpret the experiences they encounter. This is an on-­line pro­cess (e.g., Hastie and Park 1986; Lodge, McGraw, and Stroh 1989; Lodge, Steenbergen, and Brau 1995), occurring in real time and in the ser­vice of some ulterior goal, which means that language effects are variable, rather than constant. This on-­line feature also reins in expectations about the magnitude of language effects. Thinking for speaking encourages us to anticipate language effects that are characterized by soft touches, rather than dramatic flair. Such effects might be weaker and less attention-­grabbing than the linguistic determinism of Sapir and Whorf, but they “are not without scientific interest,” Slobin (2003: 2) reminds us. Indeed, what makes t­ hese linguistic nudges tantalizing from a scientific perspective is that, unlike the SW hypothesis, language effects on ­human thinking are more falsifiable than before: our expectations about when this claim is correct are just as clear as when it might be completely wrong.

Th e La n g uag e - Opi n i o n Co n n ec ti o n 25

FROM THINKING FOR SPEAKING TO NEW EVIDENCE

But if Slobin planted the conceptual seeds of a revitalized framework to study language effects, then Lera Boroditsky has seized this fertile ground to harvest new layers of evidence to corroborate anew the effect of language on ­human thinking. A trained cognitive scientist with a breathtaking knowledge of vari­ous languages, Professor Boroditsky has formulated and undertaken a variety of clever experiments that more convincingly establish language effects on ­people’s understanding of time, space, and objects (Boroditsky 2003). Indeed, whereas prior empirical assessments of language effects often focus on comparing monolingual speakers of varied languages, Boroditsky (2001) often compares bilinguals by manipulating dif­f er­ent features of their tongues. Consider her work on language and construals of time (Boroditsky 2001). As she explains, En­glish speakers verbally describe time in horizontal terms, as in “that which is ahead of us.” Mandarin speakers describe time horizontally, but they also sometimes describe it vertically, as in “­earlier events being up” and “­later events being down.” Seizing on this nuance, Boroditsky exposed Mandarin-­English bilinguals (of varying age at which they started acquiring En­glish) to horizontal or vertical cues. Horizontal cues consisted of images with captions like “the black worm is ahead of the white worm.” In turn, vertical cues consisted of photos with captions like “the black ball is above the white ball.” Following such primes, Boroditsky had bilinguals quickly answer questions like “does March precede April?” Consistent with thinking for speaking, she found that the longer the bilinguals had spoken Mandarin, the quicker they w ­ ere to correctly answer that March does, in fact, precede April when primed with vertical cues. This suggests that the presence of the vertical meta­phor for time in Mandarin results in a dif­f er­ent temporal pro­cessing for Mandarin than En­glish speakers. Follow-up studies with further methodological refinement showed even stronger evidence of such language effects (Boroditsky, Fuhrman, and McCormick 2011; Fuhrman et al. 2011). Boroditsky’s methodological prowess extends to thinking for speaking about objects, too. In collaborative work with Lauren Schmidt and Webb Phillips, she has investigated w ­ hether discussing inanimate objects as if they ­were masculine or feminine can lead ­people to construe ­these objects as being gendered (Boroditsky, Schmidt, and Phillips 2003; see also Sera, Berge, and Pintado 1994). To this end, German and Spanish speakers ­were instructed to rate the similarities perceived between photo­graphs of (fe)male

26 C H A P TE R 1

­ eople and photo­graphs of inanimate objects. They found that speakers of p each language appraised grammatically feminine objects as more similar to females and grammatically masculine objects as more similar to males. In addition, Boroditsky and her colleagues reveal that Spanish and German speakers also assign more feminine (masculine) attributes to objects, consistent with their grammatical gender. When asked, for example, to describe a key, which is masculine in German, but feminine in Spanish, German speakers employ adjectives like hard, heavy, jagged, while Spanish speakers use adjectives like intricate, ­little, lovely. When describing a bridge, however, which is feminine in German and masculine in Spanish, the former described it as elegant, fragile, and pretty, while the latter described it as long, strong, and sturdy. ­These are only two examples in a longer caravan of studies by Boroditsky and her many collaborators demonstrating language effects on cognition (e.g., Boroditsky, Fuhrman, and McCormick 2011; Fausey and Boroditsky 2010, 2011; Fedden and Boroditsky 2012; Fuhrman and Boroditsky 2010), which is itself only a slice of a larger train of studies by other scholars (not Boroditsky) establishing comparable effects via thinking for speaking and other routes (e.g., Konishi 1993; Levinson 1996; Li and Gleitman 2002; Lucy and Gaskins 2001; Sera et al. 1994; see Everett 2013 for an extensive review). So, what’s the prob­lem ­here? ­There are three of them, which together leave doubts about the veracity of ­these effects: studies with small samples, drawn from con­ve­nience populations, and conducted in highly contrived research settings. This is a question of external validity (McDermott 2011; Shadish, Cook, and Campbell 2002). To what extent can researchers uncover effects similar to Boroditsky, but in larger samples of more heterogeneous populations and in settings beyond the tightly scripted and controlled confines of laboratories? Enter Keith Chen. An economist by training, Chen has taken thinking for speaking to new heights, demonstrating that speaking a language that grammatically links the pre­sent and ­future is substantively associated with personal engagement in future-­oriented be­hav­iors, such as saving more, retiring with more wealth, smoking less, and practicing safer sex—­a set of patterns he unearths via rigorous statistical modeling techniques (Chen 2013). Remarkably, ­these empirical regularities are uncovered across hundreds of thousands of individuals spread globally throughout more than 75 nations, which vastly extends the work of cognitive scientists like Boroditsky and ­others. In fact, his work has inspired o ­ thers to take a similar approach

Th e La n g uag e - Opi n i o n Co n n ec ti o n 27

in other domains, such as gender equality (Prewitt-­Freilino, Caswell, and Laakso 2012; Santacreu-­Vasut, Shoham, and Gay 2013). So, what’s the challenge h ­ ere? That for all of its technical sophistry, Chen’s findings and ­others like it ­don’t meet the high causal standard set by experimentalists like Boroditsky and ­others. More plainly, despite Chen’s statistical efforts, his work and ­others like it cannot rule out completely that the link between language and thinking is coincidental and correlational, rather than causal.1 This is perhaps best illustrated by the entanglement of language and culture. That is, how much of the results that Chen and ­others find emanate from linguistic, rather than cultural, differences between p ­ eople? In par­tic­ u­lar, it remains unclear how much of p ­ eople’s inclination to save, retire with more wealth, smoke less, and practice safer sex is driven by the linguistic quirks that (do not) strongly c­ ouple the pre­sent and f­ uture tense, rather than the s­ imple fact that individuals inhabit varied cultures that privilege future-­oriented views and practices. Language Effects on Po­liti­cal Cognition?

Our discussion so far implies that the notion of thinking for speaking is a power­ful tool to isolate language effects on ­human thinking. But even with all of the theoretical and empirical assistance from Slobin, Boroditsky, Chen, and o ­ thers, we still fall short on one crucial front. We wish to explain the effects of language on politics, a domain untouched by previous scholars. To complicate ­matters, it is also a realm that is conceptually distinct from the ones in prior work (see Pérez 2011 for an overview). The po­liti­cal realm is complex, and forming opinions about it requires ­mental energy (e.g., Gailliot et al. 2007), with a majority of citizens in most polities simply uninformed about and inattentive to po­liti­cal facts and pro­cesses (Delli Carpini and Keeter 1996). Hence, p ­ eople’s po­liti­cal views are greatly affected by considerations—­beliefs, knowledge, values, identities, etc.—­that happen to be at the top of their head the moment they express their opinions (Lodge and Taber 2013; Tourangeau et al. 2000; Zaller 1992). For this reason, we cannot simply transfer the insights from t­ hese domains and apply them to po­liti­cal opinions in a one-­to-­one fashion. If we are to better grasp language effects on politics, then we ­will have to take the insights of Slobin, Boroditsky, Chen, and o ­ thers and move beyond them in a fresh and revealing way. Specifically, an explanation of how language affects po­liti­cal opinion has to account for the unique ways in which ­people

28 C H A P TE R 1

form and express po­liti­cal attitudes. In order to accomplish this, we ­will start by turning to what is known as belief-­sampling. Belief-­Sampling: How Individuals Form Po­liti­cal Opinions

Our theory-­building campaign starts with a detailed explanation of some fundamentals b ­ ehind belief-­sampling in public opinion formation. This demands a level of depth that is hard to skim over. But we think a grasp of this minutiae is essential to explaining the role of language in the expression of mass opinion. Let us therefore go back to 1964, with the groundbreaking research of the po­liti­cal scientist, Phillip E. Converse. In that year, Converse published his now (in)famous chapter, which documented the ideological innocence of the American mass public. Across a variety of empirical tests, he showed that when individual Americans expressed their po­liti­cal views, ­there was hardly any ideological structure to them. Thus, when professing a liberal or conservative opinion about politics, ­people’s attitudes zig-­zagged in terms of direction and magnitude, with individuals often reporting dif­fer­ent opinions about the same domain at two points in time. Converse (1964) observed, for example, that the same person could, in one month, ­favor more government intervention on one issue, yet only a few months ­later, take a completely opposite view on the same issue. The ensuing de­cades vindicated Converse’s (1964) original insights, showing that ideological innocence is not a function of when opinions w ­ ere appraised and what they w ­ ere assessed about (e.g., Converse 1970; Converse and Markus 1979; Markus 1982), with the most recent addition to this vast lit­er­a­ture provided by Kinder and Kalmoe (2017). Yet the bigger lesson that public opinion scholars drew from this controversy is that mass opinion seemed to have alarmingly weak foundations, leading scholars to won­der how, exactly, ­people actually form their attitudes? Enter the notion of belief-­sampling. In the tidal waves of research that followed Converse’s work, scholars found that addressing mea­sure­ment error, clarifying question wording, and tightening response options often improved ideological coherence—­but only at the margins (Achen 1975; Neijens 2004; van der Veld and Saris 2004). Despite efforts to extenuate the ideological incoherence of the mass public, Converse’s original insight remained as steadfast as ever. Even efforts directed at more substantive explanations fell short of toppling over Converse’s main conclusion. For example, several po­liti­cal scientists turned to core values as a way to uncover structure in ­people’s opinions, suggesting that predispositions such as individualism,

Th e La n g uag e - Opi n i o n Co n n ec ti o n 29

egalitarianism, humanitarianism, and the like undergird ­people’s opinions (Rockeach 1973; Schwarz 1992). Yet a broad and sober look at this accumulated lit­er­a­ture suggests that while values can structure ­people’s opinions, ­these effects seem ­limited to select issue domains (Feldman 1983, 1988, 2003; Feldman and Steenbergen 2001) and often among po­liti­cally aware individuals (Brewer 2003; Goren 2001; Kalmoe 2020; Sniderman, Brody, and Tetlock 1991; Zaller 1992). Hardly the kind of stuff to overturn Converse’s claim. The same applies for published work on heuristics (Popkin 1993; Sniderman et al. 1991). ­Here, scholars theorized that individual opinions are driven, not by the type of higher-­order, ce­re­bral thinking that ideology or values might imply, but rather, by information shortcuts or rules of thumb. This was initially an attractive idea, since every­one, it would seem, can use information shortcuts. For example, individuals could form policy opinions based on how they felt about the groups implied by policy proposals. Yet a close look ­here also revealed that effective heuristic use required the type of po­liti­cal engagement that only a few select individuals within a mass public actually display (Delli Carpini and Keeter 1996; Luskin and Bullock 2011). Alas, it was not u ­ ntil the late 1980s and early 1990s that psychologists and po­liti­cal scientists in other quarters began coming around to the idea that perhaps—­just perhaps—­people did not have ready-­made po­liti­cal attitudes to report. By this view, the lack of structure would seem to emanate from the absence of fully formed attitudes that p ­ eople could retrieve from memory in order to report. In psy­chol­ogy, for example, the work of Roger Tourangeau, Lance Rips, and John Rasinski (2000; see also Tourangeau 1984, 1987, 1992; Tourangeau and Rasinski 1988; Tourangeau, Rasinski, and D’Andrade 1991) established that mass opinion is highly variable and hard to anticipate ­because the question-­answering pro­cess is more subtly complicated than other scholars admit (see also Schwarz 1997). Consider that any person answering a survey question has to decipher the question, identify relevant content in memory to answer it, and then match that content to the response options provided by a question. In po­liti­cal science, John Zaller and Stanley Feldman (1992) showed, in a similar vein, that ­people’s opinions ­were buoyed by a web of interconnected considerations—­beliefs, knowledge, identities, etc.—­that ­were made mentally accessible by the content of questions (Feldman 1989). Drawing on ­these and other insights, Zaller (1992) assembled what became the model to study and predict public opinion, one where p ­ eople’s attitudes w ­ ere characterized as “top-­of-­the-­head” responses driven by considerations that w ­ ere primed by survey questions (Zaller and Feldman 1992). The reason for this

30 C H A P TE R 1

priming effect is the architecture of long-­term memory (LTM)—­the area in p ­ eople’s minds where considerations are actually stored. According to the work of Collins and Loftus (1975) and o ­ thers (e.g., Lodge and Taber 2013), ­people’s considerations are or­ga­nized in LTM in a lattice-­like network where numerous concepts are linked to each other, to varied degrees, in an associative fashion. This structure implies that activation of a concept w ­ ill call forth related ones via spreading activation. This last point is critical, as it helps to explain how ­people recruit considerations from their LTM into their working memory (WM)—­the “top of their head”—in order to report their opinions (Tourangeau et al. 2000; Zaller 1992). Connecting Belief-­Sampling to Thinking for Speaking

The notion of belief-­sampling encourages us to consider individual opinions as fluid and amenable to a host of forces that impact the content of ­people’s beliefs. Accordingly, the sample of considerations that p ­ eople bring to bear on their opinions is predictably sensitive to a variety of situational cues that heighten the accessibility of content in memory. One of ­these forces, we theorize, is language. More precisely, we claim that language shapes belief-­sampling through thinking for speaking. Recall that at its most fundamental level, thinking for speaking implies that certain categories in memory ­will be made salient by virtue of the language that one speaks (Boroditsky 2001; Boroditsky et al. 2003; Cubelli et al. 2011; Fuhrman et al. 2011; Slobin 2003; Vigliocco et al. 2005). If the vocabulary, grammar, or another linguistic feature obliges speakers to make certain distinctions or associations between objects (e.g., based on color, gender, time, ­etc.), then ­these nuances become salient for speakers. ­Because of one’s tongue, ­people must attend to certain aspects of the world, which brings them to the “top of their head.” This merging of thinking for speaking with belief-­sampling is the basis for our language-­opinion hypothesis. Together, they provide a mechanism by which language affects opinion formation and expression: from what information ­people use and how, to what considerations they retrieve from memory, to how t­ hese considerations inform their opinions. This mechanism relies on a basic pro­cess, namely, that language makes some ­mental content more accessible, at least temporarily, in ­people’s minds, which makes it more likely that this content w ­ ill enter their sample of considerations when expressing an opinion. This yields a testable claim: specific features of a language indicate which associations, beliefs, knowledge, or values are

Th e La n g uag e - Opi n i o n Co n n ec ti o n 31

likely to become more prominent in ­people’s memory, thus heightening their accessibility. In this way, our language-­opinion hypothesis enables us to predict the direction in which such ­mental accessibility is likely to affect opinions. It also enables us to falsify our claims about a linguistic influence on ­people’s attitudes. If, for example, the concern is that language primes cultural values or national identities—­rather than a grammatical or associative feature of a tongue—we can yield concrete alternative explanations and test them against our linguistic claim. Fi­nally, although the connection between po­liti­cal opinions and po­liti­cal be­hav­ior is hardly straightforward (e.g., Van Zomeren et al. 2012; Van Zomeren et al. 2004; see also Kinder 1998), our proposed mechanism implies that chronic priming of linguistic features through public discourse might hold one key to when attitudes guide actions. In subsequent chapters, we ­will elaborate on how our hypothesis can be tested on dif­f er­ent features of a tongue and in varied domains of public opinion. Yet our main contention is versatile in other ways as well. For example, it can be used to explain ­whether grammatical features of a language ­will impact the po­liti­cal opinions ­people express. And, it can also explain how language shapes po­liti­cal opinions by focusing on the links between language and the learning and recall of information for opinion formation. This positions us to better anticipate when language ­will ­matter for mass opinion, and just as crucially, when it ­will not. All of this, we ­will ­later demonstrate, flows from this initial theoretical spark that we label our language-­opinion hypothesis. But just how original and new is this basic hypothesis? It’s a fair question. Although po­liti­cal scientists have examined the influence of language on politics, their efforts have been quite a few degrees removed from our specific focus on language and mass opinion across the globe. For example, some po­liti­cal scientists have studied the vari­ous links between language policy and socioeconomic and po­liti­cal development (Laitin 1992, 1998; Laitin and Ramachandran 2016). Moreover, other po­liti­cal scientists have employed language as an indicator of ethnic groups in studies of intergroup conflict (Horo­witz 1985; Laitin and Posner 2001; Posner 2005). And still other po­liti­cal scientists use language to signal the presence of outgroups to members of an ingroup (Hopkins 2014; Hopkins, Tran, and Williamson 2014; Newman, Hartman, and Taber 2012, 2014). Th ­ ese are impor­tant studies in their own right. But they do not speak directly to the language-­opinion connection that is the center of our attention ­here. Indeed, on that front, we are aware of only two studies that engage this exact phenomenon.

32 C H A P TE R 1

The first is a late-1970s experiment by David Laitin, which examined the impact of language on cultural and po­liti­cal opinions in a small sample of Somali students. The second is a series of analyses appraising the association between interview language and po­liti­cal attitudes, beliefs, and knowledge in samples of US Latinos (Lee and Pérez 2014). While Laitin’s (1977) study demonstrates that language can causally impact individual attitudes, its small sample and lab-­like setting limit the external validity of this result. In turn, while the research of Lee and Pérez uncovers a reliable and robust link between language and attitudes in surveys of Latino public opinion, it does not establish a direct causal relationship on a wide scale (Pérez 2016a). Fi­nally, both of ­these studies fall short of providing a full theoretical explanation as to why language—­and language alone—­should influence mass po­liti­ cal opinion. We have provided the pillars of that framework through our fusion of belief-­sampling and thinking for speaking. Our next obstacle, then, is to begin assembling evidence in support of this major claim. Summary and Next Steps

This chapter has outlined the changing trajectory of scientific work on the connection between language and h ­ uman thought. The arc of this endeavor is one that can be accurately described as being driven by conceptual refinement and theoretical innovation. Indeed, from its origins as a forcefully blunt claim that language determines thought, to the now more nuanced contention that, ­under some circumstances, language might shape thinking, the basic hypothesis that p ­ eople’s tongues influence how they think about objects, time, and space has under­gone serious extenuation, rehabilitation, and now, more rigorous empirical evaluation (Boroditsky 2001). But for all of this pro­gress in clarifying what language does and when and where it does it, the tides of ­these innovations have not fully reached the shores of po­liti­cal cognition. We argue that this is a theoretical leap yet to be taken ­because, unlike the many domains that linguists and cognitive scientists study, po­liti­cal cognition is peculiar, insofar as p ­ eople often lack high levels of po­liti­cal awareness, are averse to investing much effort in po­liti­cal thinking, and—on top of all this—­lack ready-­made opinions about politics to report. To begin resolving this impasse, we put forth an original framework whose primary contention is our own language-­opinion hypothesis. This framework, we explained, relies on a unique alloy of two basic princi­ples: thinking for speaking and belief-­sampling. The primary prediction of this fusion is that the grammatical strictures of our tongues influence some of

Th e La n g uag e - Opi n i o n Co n n ec ti o n 33

the considerations that float to the top of our head, thus affecting our views about politics in many domains. That, at least, is our contention. But ­whether it is empirically valid is a dif­fer­ent question, one requiring evidence from not just one domain, but several. To this end, in the next chapter we begin our empirical assault on the language-­opinion hypothesis by examining the pos­si­ble links between gendered languages, genderless tongues, and the opinions that p ­ eople express about how to remedy gender in­equality. Let us turn to that effort now.

2 ­Battle of the Sexes

The only way to solve the prob­lem of ­women’s subordination is to change ­people’s mindset and to plant the idea of gender equality into ­every mind. —­Q I NG RONG M A

We live in a world that is characterized by neck-­b reaking changes in technology, scientific advances, the economy, and more, pushing many nation-­states ­toward new heights in their development. But t­ here is one con­spic­u­ous area that still trails b ­ ehind this juggernaut of modernization. When it comes to relations between men and ­women, the latter still lag ­behind the former across vari­ous domains—­especially in politics, where they are woefully under-­represented and under-­placed in higher echelons of power (Butler, Tavits, and Hadzic 2022; Cheng and Tavits 2011; Hadzic and Tavits 2019, 2021; Hinojosa 2012; O’Brien 2015; O’Brien and Rickne 2016; Paxton, Hughes, and Painter 2010). What explains ­these nagging disparities? One of the more persuasive answers to this question pertains to the mindset that Qingrong Ma alludes to in her epigraph. According to her view, and the views of many o ­ thers (Epstein 2007; Inglehart and Norris 2003; Iversen and Rosenbluth 2010), the prevalence of patriarchal attitudes and beliefs promotes and maintains ­women’s unequal status, serving to negatively impact w ­ omen’s po­liti­cal repre­s en­ta­tion (Huddy

34

Bat tle o f th e S e xe s 35

and Terkildsen 1993; Sanbonmatsu 2002), their economic participation (Fortin 2005), and their social opportunities, even in highly developed socie­ties (OECD 2012). This is an intuitive and compact explanation. But it is also one that is incomplete. For inasmuch as gendered attitudes are symptomatic of continued gender inequalities, the root ­causes of ­these beliefs still remain opaque, thereby underlining the challenge in explaining where ­these opinions arise from in the first place and why they persist. Armed with our language-­opinion hypothesis, this chapter revisits the question of gender in­equality and assesses the degree to which language—­ and language alone—­causes p ­ eople to think differently about gender relations and politics. We zero in on one of the more time-­tested patterns in ­human communication: the varied genderedness of tongues. As we explained in the introduction, all ­human languages vary in the degree to which they require speakers to attend to and encode gender (Boroditsky, Schmidt, and Phillips 2003; Corbett 1991; Cubelli et al. 2011; Vigliocco et al. 2005). Genderless languages, such as Estonian or Finnish, do not require speakers to designate the gender of objects or even individuals—­the word for “he” and “she” is the same in ­these tongues. In contrast, gendered languages, like Spanish and Rus­sian, require speakers to differentiate genders and assign it to objects.1 Spanish speakers, for instance, must mark the object “moon” as feminine by using the definite article la, as in la luna. This chapter’s mission is twofold. First, it aims to explain how our language-­opinion hypothesis connects the presence (absence) of grammatical gender to public attitudes ­toward gender equality. Second, it appraises this hypothesis and its implications through tightly controlled experiments and sophisticated analyses of global survey data. ­These efforts shed light on several theoretical questions: Does speaking a genderless, rather than gendered, language cause ­people to express more gender-­equal attitudes? If so, do ­these effects occur ­because genderless tongues pay less attention to gender or ­because speakers of genderless languages are biased ­toward females? How widely replicable are ­these language effects—­are they confined to only a handful of nation-­states? And, fi­nally, ­under what circumstances do ­these language effects dissipate? Let us therefore stop asking questions and instead start answering them by stipulating the necessary links between language, grammatical gender, and individual opinions about gender equality. This is the task we turn to next.

36 C H A P TE R 2

How Language Impacts Attitudes ­Toward Gender Equality

In the introduction, we called attention to linguists’ notion of thinking for speaking. The basic premise of this idea, you may recall, is that all languages vary in their grammatical organ­ization, which obliges speakers to focus on dif­fer­ent aspects of their experience when using a specific tongue (Boroditsky 2001; Boroditsky et al. 2003; Fuhrman et al. 2011; Slobin 1996). If speaking a language requires one to make certain distinctions between objects (e.g., dif­fer­ent colors, time orderings), then a speaker may take for granted that ­these categories exist in the world and are relevant to judgment (Boroditsky 2001; Boroditsky et al. 2003; Boroditsky and Gaby 2010; Danziger and Ward 2010; Fuhrman et al. 2011; Hunt and Agnoli 1991). Applied to grammatical distinctions by gender, thinking for speaking implies that language should affect how salient (or not) gender is for vari­ous language speakers. Indeed, carefully conducted research reveals that speakers of gendered tongues are more keenly aware of gender differences: they are more likely to categorize the world in feminine and masculine terms and to proj­ect gender features onto objects and individuals (Boroditsky et al. 2003; Cubelli et al. 2011; Konishi 1993; Phillips and Boroditsky 2003; Sera, Berge, and Pintado 1994). For example, young Spanish speakers generally rated object photos as masculine/feminine according to their grammatical gender (Sera et al. 1994). Relatedly, when Rus­sian speakers ­were asked to personify days of the week, they generally personified grammatically masculine days as males and grammatically feminine days as females ( Jakobson 1966). This was likely automatic and not controlled ­because l­ ater, when asked, they ­were not able to explic­itly say why they had done so. Con­temporary empirical research further confirms that the influence of grammatical gender on ­mental repre­sen­ ta­tion of words is automatic (e.g., Maciuszek, Polak, and Swiatkoska 2019). Gendered language speakers are also more likely to attain their own gender identity sooner than speakers of less gendered tongues. For example, Guiora et al. (1982) studied c­ hildren, ages 16–42 months, who spoke tongues varying by their genderedness: Hebrew (highly gendered), En­glish (medium gendered), and Finnish (genderless). By 28–30 months of age, 50% of Hebrew speakers expressed gender identification compared to 21% of En­glish speakers and 0% of Finnish speakers. Prior work also reveals that language effects like ­these arise from structural (i.e., grammatical) differences between tongues. For example,

Bat tle o f th e S e xe s 37

researchers have taught individuals fictitious languages that are completely stripped of any cultural or other contextual information and found that ­these tongues still affect individual be­hav­ior in predicted ways (Boroditsky et al. 2003). Furthermore, ­these effects emerge on both linguistic and non-­ linguistic tasks, the latter using photo­graphs as stimuli and button pressing as response without requiring subjects to produce or pro­cess any language (Fausey and Boroditsky 2011; Fuhrman et al. 2011; Phillips and Boroditsky 2003). This indicates that language impacts cognition with l­ittle to no verbalization. Thinking for speaking, then, implies that languages can set a frame of mind for how ­people think about dif­fer­ent realms. But we are interested in how language affects what p ­ eople think about politics: a domain where few ­people have ready-­made attitudes to report. H ­ ere is where the second pillar of our theory—­belief-­sampling—­comes in. Recall that belief-­sampling is an idea that originates from survey research and suggests that p ­ eople construct their po­liti­cal opinions on the basis of ­those considerations in memory that are more mentally accessible (Lodge and Taber 2013; Tourangeau, Rips, and Rasinski 2000; Zaller 1992). Our language-­opinion hypothesis fuses the ideas of thinking for speaking and belief-­sampling and sees language as a power­ful force b ­ ehind po­liti­cal attitudes ­because it affects which considerations become mentally accessible and influential on subsequent judgments. Our hypothesis leads to the following expectations about language, gender, and public opinion. It suggests that speakers of gendered languages are, on average, more grammatically sensitized to the feminine or masculine qualities of individuals or objects ­because language makes ­these distinctions mentally salient and accessible for them. This ­mental accessibility, in turn, makes it more likely that t­ hese considerations affect po­liti­cal judgment and opinions about gender roles: a pro­cess that po­liti­cal scientists ­will recognize as priming—or the heavier weighting of a par­tic­u­lar consideration on a person’s opinion, judgment, or evaluation (Valentino, Hutchings, and White 2002; Valentino and Nardis 2013). In the case at hand, this suggests that speakers of a gendered language ­will be more attuned to the distinctions between gender categories and the unequal roles assigned to them. We expect the opposite among speakers of genderless tongues. Th ­ ose languages, our hypothesis suggests, do not prime gender as strongly ­because they do not require speakers to make distinctions on this basis. This blurs the formal distinction between (fe)male objects—in effect, changing “­people’s mindset,” as Qingrong Ma recommends. Therefore, speakers of genderless

38 C H A P TE R 2

languages are less likely to perceive gender differences and more likely to see the roles of men and w ­ omen as similar. This logic suggests, not that some ­people fail to distinguish between men and ­women, but rather, that some individuals find it harder to perceive a “natu­ral” asymmetry between the sexes. This blurred distinction between (fe)males, we claim, nudges some ­people’s views in a more egalitarian direction. Our main prediction, then, is that speakers of genderless tongues ­will express more egalitarian outlooks about ­women’s place in politics and society, and more support for public efforts seeking to rectify gender imbalances.2 How far off the mark are we about this? The Leap from Hypothesis to Evidence

To say that differences in the structural features of languages cause ­people to construe the same real­ity differently is perhaps controversial enough (Boroditsky 2003; Pérez 2015). But to empirically assess this claim is a challenge all upon itself. The rub of the ­matter is this. We wish to see ­whether ­people who speak dif­fer­ent languages interpret gender relations distinctly. S ­ imple enough, right? Just investigate how varied language speakers form their views about w ­ omen, men, and their respective roles in society and politics. But the challenge h ­ ere is that two or more individuals can express dif­fer­ ent opinions about ­these ­matters based on nonlinguistic ­factors that are observed and unobserved. What is needed, then, is a research design that can hold constant ­these (un)observed differences between varied language speakers, while zeroing in on the influence of language itself. This is where experimentation can come in handy (Shadish, Cook, and Campbell 2002). By randomly assigning speakers of dif­fer­ent tongues to express their opinions, we can effectively hold constant, by design, the manifold differences between them, besides language (Campbell and Stanley 1963). But this solution now poses a prob­lem of its own. Most individuals cannot simply be assigned to speak just any old language, for they generally speak only one tongue—­the one they are fluent and competent in—­which makes random assignment very difficult. What’s to be done about this? Our proposed solution is to study bilingual individuals (Boroditsky 2001; Boroditsky, Fuhrman, and McCormick 2011; Pérez 2015, 2016a). In essence, we ­will study individuals who can speak, read, and understand two languages; and preferably, two languages that vary systematically in the structural nuance that is supposed to influence their thinking.

Bat tle o f th e S e xe s 39

For this intricate task, we travel to Estonia. We visit this small corner of the Baltic Sea, not as a way to be methodologically exotic, but in an effort to be conceptually rigorous. Estonia is an ideal setting for testing our hypothesis for several reasons. First, it possesses a sizeable population that is equally proficient in a gendered (Rus­sian) and a genderless (Estonian) language: about 61% of the population identify Estonian and 29% Rus­sian as their first language. Roughly 44% of the former group and 36% of the latter speak the other language well enough to qualify as bilingual according to our definition. Estonia’s current official language is Estonian. Yet the country was annexed into the USSR ­until 1991, which made Rus­sian a prominent tongue, with most Estonian speakers acquiring at least working knowledge of Rus­sian and many becoming proficient. Large-­scale immigration of Rus­ sians before 1991 also created a sizeable Russian-­speaking population, some of whom acquired proficiency in Estonian. Integration programs, intermarriages, and schools offering general education (equivalent to K-12) in ­either Estonian or Rus­sian further increased this bilingual population. Second, prior research shows that in terms of po­liti­cal opinions and values, Estonians and Rus­sians in Estonia have more in common with each other than with any other group outside Estonia (Lauristin and Vihalemm 1997; Maimone 2004). ­These studies also suggest that Rus­sians in Estonia do not express more traditional or conservative values than Estonians do. This makes Estonia a uniquely ideal setting to isolate language effects (from ­those of values or ideology) on public opinion. Taking advantage of ­these features, we conducted two experiments in Estonia. Study 1 is a survey experiment we administered from May 26 to June 12, 2014. Twelve hundred (N = 1,200) Estonian-­Russian bilingual adults ­were randomly assigned to interview in Estonian or Rus­sian. Study 2 is a smaller experiment (N = 262) using the same design, with the aim of replicating and extending Study 1’s findings. We conducted Study 2 from March 22 to April 10, 2016. This pair of experiments provide a compelling and straightforward between-­subjects design for identifying language effects (Dunning 2016; Green 2004). We then supplement t­ hese studies with additional data that take us beyond Estonia. Study 3 is a cross-­national analy­sis of the World Values Survey (WVS), which appraises the external validity of our experimental results and mitigates concerns over the generalizability of our Estonian findings. In sum, we corroborate our language-­opinion hypothesis in the realm of gender equality by marshaling evidence from three studies, each varying in

40 C H A P TE R 2

the amount of control we have over language, the type of po­liti­cal settings we operate in, the types of ­people we study, and the range of po­liti­cal opinions about gender relations that we focus on (Campbell and Stanley 1963). While any single analy­sis that we pre­sent, or even a single study alone, may not be convincing, we hope that combined, the totality of the evidence is able to persuade even the most skeptical reader that the way we speak affects our views on gender equality. Let us therefore begin prosecuting our empirical case against this language-­ opinion hypothesis by g­ oing to Estonia and appraising w ­ hether use of genderless language ­causes ­people to express more gender-­equal attitudes. Does Language Causally Impact Gender-­Related Attitudes?

Study 1 randomly assigned twelve hundred (1,200) Estonian-­Russian bilingual adults to interview in Estonian (a genderless tongue) or Rus­s ian (a gendered tongue). This large sample reflects our endeavor’s novelty. Our sample (N = 1,200) allows us to unearth small, but meaningful, language effects in ­either direction, if ­these exist.3 We provide further details about this study’s design in appendix 2.1. We conducted Study 1 via telephone in partnership with TNS Emor, a leading Estonian survey firm with extensive experience conducting survey research for public-­and private-­sector clients, including the Eurobarometer Survey (since 2004). The universe for this study included residents of Estonia, ages 18–74, who live in private h ­ ouse­holds and can speak Estonian and Rus­sian. The sampling frame for this survey consisted of randomly drawn landline and mobile phone numbers in Estonia. TNS Emor identified eligible bilinguals with two items tapping self-­ rated skill in Estonian and Rus­sian, respectively. ­After first reporting their native language, potential respondents w ­ ere asked about their fluency in that tongue. For example, if a person reported Estonian as their native language, they ­were asked “In your opinion, how well do you know Estonian?” where the response options ­were: (1) Do not know the language at all; (2) Can understand a ­little, but cannot speak; (3) Can understand and can speak a ­little; (4) Can understand, speak, and write; and (5) Fluent. The same question was also asked of their second language, Rus­sian. Our respondents are bilinguals who answered “4” or “5” on both items. ­These participants w ­ ere then randomly assigned to interview in Estonian or Rus­sian.4 Our language manipulation was designed to set a distinct

Bat tle o f th e S e xe s 41

linguistic milieu by informing bilinguals that all subsequent instructions and questions would be in their assigned tongue (cf. Marian and Neisser 2000).5 Specifically, our language treatment was worded as follows: Based on your answers to some of the previous questions, it appears that you are fluent in both Estonian and Rus­sian. Therefore, we ­will let the computer program randomly select which language we continue this interview in [SHORT PAUSE]. [Estonian/Rus­sian] was selected. This means that ­after this point, the rest of the interview w ­ ill take place in [Estonian/Rus­sian]. This is not a language test. We are simply interested in your opinions as an [Estonian/ Rus­sian] speaker.6 Thirty-­eight p ­ ercent (38%) of our sample consists of bilinguals whose first language is Rus­sian and who, on average, learned Estonian at the age of 15. In turn, ­those bilinguals in our sample (62%) whose first language is Estonian learned Rus­sian, on average, at the age of 14. Fi­nally, 61% of our respondents are female, with a median age of 52 and secondary level of education. As we discuss in appendix 2.2, ­these and a few other characteristics are balanced across both experimental conditions, allowing us to make full use of randomization for our inferences about language’s causal effects.7 Following the manipulation, respondents answered several items related to perceptions of ­women and their role in society and politics, which we describe below (full item wording is provided in appendix 2.1). We constructed ­these items on the basis of prior research on attitudes about gender, and several of them are adapted from well-­established national or international surveys such as the General Social Survey and the Amer­ic­ as Barometer. Our items are designed to tap attitudes about gender roles and gender bias in several ways, including: (a) ­those expressed in gender ste­ reo­types, which foster unequal perceptions of men and ­women (e.g., Bauer 2015; Dolan 2014; Koch 2002; Schneider and Bos 2014); (b) preferences over ­women’s participation in politics and po­liti­cal leadership roles, where females are substantially underrepresented (e.g., Banda and Cassese 2021; Cheng and Tavits 2011; O’Brien 2015; O’Brien and Rickne 2016); and (c) support for policies aimed at changing norms that promote gender equality (e.g., Cassese and Barnes 2019; Jung and Tavits 2021). For the policy items, we chose issues that are relevant and topical in Estonian public discourse. More specifically, the survey included the following questions about w ­ omen and politics, which serve as our dependent variables.

42 C H A P TE R 2

Emotional ­women and Emotional men are seven-­point scale ratings of how emotional (keyed as 7) versus rational (keyed as 1) bilinguals believe men and ­women to be, with the item order randomized. We use ­these ratings in two ways. First, we analyze them individually and in their original format (variable names Emotional ­women: single rating and Emotional men: single rating). Second, we take the difference between t­ hese ratings to create a mea­sure ranging from − 6 to 6, where positive values indicate greater ste­reo­typical belief in w ­ omen as emotional (relative to men) (variable name Emotional ­women: relative rating) (cf. Kinder and Kam 2009). Paternity leave queried bilinguals about ­whether they agreed (keyed as “1”) or disagreed (keyed as “0”) with a proposed change in f­ amily leave policy that would allow a ­father to stay home, while the ­mother can return to work as soon as she is able to (the law at the time allowed f­ athers to stay home only ­after babies are at least 70 days old). Female Defense Minister asked bilinguals “If the party that you normally like nominated a generally well-­qualified ­woman to be Minister of Defense, would you support that choice?,” with support coded as “1” and opposition as “0.” Female po­liti­cal recruitment asked ­whether one strongly disagreed (4), somewhat disagreed (3), somewhat agreed (2), or strongly agreed (1) that ­women should be recruited to “top-­level government positions.” We recode this item so that higher values reflect greater agreement with the statement. Speaking Genderless Language ­Causes More Gender-­Equal Attitudes

We first examine ­whether interviewing in Estonian affects how much asymmetry ­people perceive between men and ­women in terms of gender ste­reo­ types by focusing on Emotional ­women: relative rating. ­Here, higher values reflect individual views of ­women as more emotional than men, with the scale midpoint indicating no sensed difference between men and w ­ omen on this stereotypic trait. We find that interviewing in Estonian significantly reduces how emotional respondents see w ­ omen relative to men. The mean value for respondents interviewing in Rus­sian is 1.34 (95% confidence interval (CI): 1.17, 1.52) and in Estonian 1.14 (95% CI: 0.98, 1.30), for a difference in means: t = 0.20, p = 0.09, two-­tailed test. ­These numbers tell an in­ter­est­ing story. The first piece of information we learn is that survey respondents in Estonia do not express strong gender ste­reo­types in general. Recall that the maximum

Bat tle o f th e S e xe s 43

value on this variable is 6, i.e., that would be the score if a respondent saw ­women completely ste­reo­typically as emotional (and not rational) and men ste­reo­typically as rational (and not emotional). If a respondent saw men exactly the same way as men in terms of how emotional versus rational they are, then the score would be 0. We see that the mean scores for both Rus­sian and Estonian speakers are closer to 0 than to 6, indicating some ste­reo­typical belief in ­women as being more emotional than men. The second piece of information we learn from ­these numbers—­and one that is relevant for testing our language-­opinion hypothesis—is that language m ­ atters. While the scores are low for both the Estonian and Rus­sian speakers, they ­aren’t the same. As the difference in means test shows, the mean score for t­ hose respondents who w ­ ere assigned to speak Rus­sian is significantly higher than the average score for the respondents assigned to speak Estonian. That is, w ­ omen and men are seen in less ste­reo­typical ways when speaking genderless Estonian than when speaking gendered Rus­sian. The same finding is reflected in the regression analy­sis presented in t­ able 2.1 (column 1). In short, a­ fter using randomization to hold constant all other (un)observed differences between respondents interviewing in Estonian versus Rus­sian, the former are still less likely to perceive w ­ omen as more ste­reo­typically emotional than men. This is in line with what our language-­ opinion hypothesis suggests. Having found evidence that affirms the cognitive validity of our proposed mechanism, we now look at ­whether it generates our anticipated po­liti­cal effects on ­people’s opinions about gender equality. We do this by estimating the effect of interviewing in Estonian on our other dependent variables: Paternity leave, Female Defense Minister, and Female po­liti­cal recruitment. If our language-­opinion hypothesis is correct, we should observe Estonian interviewees consistently reporting more gender-­equal opinions than Rus­ sian interviewees. Do they, in fact, manifest this pattern? Columns 2–5 in t­ able 2.1 suggest they do, with the last column displaying the effect on an additive scale of all three gender-­equal policies.8 Since most of ­these individual estimates are based on nonlinear models, we translate the raw coefficients into predicted probabilities, as presented in figure 2.1. Accordingly, panel A shows the shift in the probability of supporting changes in paternity leave policy. Among respondents assigned to interview in Rus­ sian, the probability of supporting this policy change is 35%. But if a person is assigned to interview in Estonian, the probability of endorsing this proposal climbs reliably by 8 points to 43% (first difference (FD) = 0.08, 95% CI: 0.02 to 0.14). Thus, simply by interviewing in a genderless tongue,

44 C H A P TE R 2

­TABLE 2.1.

The Effect of Genderless Language on Opinions ­Toward Gender Equality, Study 1 Emotional ­women: relative rating (OLS)

Paternity Leave (Probit)

Female Defense Minister (Probit)

Female po­liti­cal recruitment (Ordered Probit)

Gender-­ equal policy scale (OLS)

Estonian interview

−0.202^ (0.120)

0.206* (0.075)

0.221* (0.078)

0.140* (0.063)

0.246* (0.053)

Constant

1.344* (0.086)

−0.379* (0.054)

0.402* (0.054)

—­

1.603* (0.038)

1,153

1,140

1,156

1,154

1,108

N

Note: Table entries are OLS, probit, or ordered probit coefficients with standard errors in parentheses. Dependent variables are indicated in column headings. *p