Социальный статус и передача информации в экспериментальных играх тема диссертации и автореферата по ВАК РФ 08.00.01, кандидат наук Бондаренко Оксана Юрьевна

• Бондаренко Оксана Юрьевна
• кандидат наук
• 2021

Оглавление диссертации кандидат наук Бондаренко Оксана Юрьевна

Введение

Основные результаты

Список публикаций

Список литературы

Приложение 1. Статья «Влиятельные индивиды: подходы к

моделированию»

Приложение 2. Статья «Измерение социального статуса в экспериментальных играх»

Приложение 3. Статья «Social status and social

learning»

Введение

Введение диссертации (часть автореферата) на тему «Социальный статус и передача информации в экспериментальных играх»

Актуальность исследования

Когда люди принимают решения в повседневной жизни, они опираются не только на собственное мнение, но и на опыт своего круга общения: друзей, знакомых и коллег, экспертов и лидеров мнения. Часть таких социальных взаимодействий могут быть несимметричными -происходить между людьми с разным социальным статусом. Обладание высоким социальным статусом и доступом к ресурсам издавна ценилось в обществе [Henrich, J., Gil-White, F. J., 2001]. Стремление повысить свой статус - один из наиболее часто встречающихся видов человеческой мотивации [Anderson et al., 2015].

Под социальным статусом могут пониматься уважение и восхищение, которыми индивиды обладают в глазах других членов группы [Anderson et al., 2006], или влияние, которое они оказывают на других людей [Ridgeway, Correll, 2006]. В данном исследовании используются два показателя социального статуса: субъективный социальный статус (subjective social status, SSS) и объективный, или социоэкономический статус (socioeconomic status, SES). Под субъективным социальным статусом понимается восприятие индивидом своего места в общественной иерархии. Социоэкономический статус ближе к понятию социального класса и измеряется на основе образования, дохода, престижа работы и других характеристик [Diemer et al, 2013].

Важным вопросом для исследования является влияние статуса на социальное обучение (social learning): на то, как индивид использует информацию из разных источников при принятии решения, насколько важнее будут для него или нее собственное мнение или внешняя информация. Рассмотрим в качестве примера вакцинацию от SARS-CoV-2, которая в 2020 году началась в разных странах. Человек должен принять решение о том, сделать ли прививку, на основе нескольких источников информации. Изначально он или она имеет мнение относительно прививок в

целом и относительно прививки от SARS-CoV-2 - например, не доверяет прививкам в целом. Также человек слушает выступления экспертов в данной области - врачей и вирусологов. Кроме того, он или она наблюдает, как делает прививку значимая фигура, за которой следят тысячи и миллионы людей, например, известный актер. Наконец, человек общается с социальной сетью - друзьями и знакомыми, и узнает от них, кто сделал привику, а кто нет. Какая информация повлияет на итоговое решение человека, какая стратегия действий «перевесит», и повлияют ли на это какие-либо личные характеристики?

В теоретических моделях социального обучения присутствует некоторое число агентов, объединенных в социальную сеть. Агенты получают личные информационные сигналы о неизвестной переменной (состояние мира), а также наблюдают информационные сигналы или действия других агентов. Используя и собственную информацию, и информацию, полученную в ходе наблюдения за другими, агенты обучаются - обновляют убеждения в соответствии с правилом Байеса или некоторым простым фиксированным правилом [DeGroot, 1974; DeMarzo et al, 2003; Golub, Jackson, 2010; Acemoglu et al, 2010; обзор литературы по моделям социального обучения Mobius, Rosenblat, 2014]. При хорошем исходе все агенты, т.е. общество в целом, должны узнать истинное состояние мира. Такой процесс называется агрегированием информации (information aggregation). Однако некоторые индивиды могут оказывать более сильное влияние на своих соседей по сети. Присутствие влиятельных агентов в экономической среде может нарушить или затруднить агрегирование информации, например, способствовать распространению стадного поведения, что показано в [Banerjee, 1992; Bikhchandani et al, 1992; Acemoglu et al, 2010; Golub, Jackson, 2010]. Высокий социальный статус может служить обоснованием такой ассимметрии.

Согласно психологическим и экономическим исследованиям, социальный статус и социальное обучение могут быть связаны между собой.

Низкостатусные индивиды демонстрировали более высокую эмпатию и в большей степени подстраивались под других людей, чем высокостатусные индивиды [Stellar et al., 2012; Varnum et al., 2015; Kraus et al, 2010]. Более низкий социальный класс ассоциировался с более сильным откликом на боль и страдания [Stellar et al., 2012; Varnum et al., 2015], более выраженной реакцией на стресс и отвергание [Muscatell et al., 2016], на двигательную активность других [Varnum et al., 2016]. Индивиды, обладающие более высоким социоэкономическим статусом, хуже определяли эмоции других людей [Kraus et al, 2010].

Возможное объяснение для таких различий предложили Kraus et al, 2012 с помощью концепции «локус контроля» (locus of control). На жизнь людей с низким социальным статусом, по их мнению, влияют обстоятельства, которые они не могут контролировать. В результате такие люди склонны адаптироваться к окружающей среде и реагировать на внешние ограничения и угрозы. Наоборот, люди с высоким социальным статусом более независимы и убеждены в том, что они сами контролируют свою жизнь, и более сконцентрированы на себе, а не на других [Piff et al, 2018]. В рамках другой теории, теории приближения/торможения (approach/inhibition theory) [Keltner et al, 2003], индивиды, наделенные большей силой, демонстрируют мотивацию, связанную с приближением и позитивным аффектом. Для индивидов, наделенных меньшей силой, характерны проявления, связанные с торможением и негативным аффектом.

Цель и задачи исследования

В данном исследовании изучается, как социальный статус влияет на процесс социального обучения.

Для реализации поставленной цели были решены следующие задачи: 1. Изучена и систематизирована теоретическая и эмпирическая литература, анализирующая социальные взаимодействия, в том числе асимметричные взаимодействия между влиятельными агентами и прочими агентами.

2. Выдвинут ряд гипотез относительно социального статуса у игроков, играющих разные роли в экспериментальных играх.

3. Исследован посредством лабораторного эксперимента способ установления асимметричного социального статуса у агентов с помощью экспериментальной игры.

4. Разработана игровая модель для проведения экспериментальной игры на социальное обучение.

5. Разработан способ измерения субъективного социального статуса на основе опроса.

6. Разработаны показатели для измерения лидерских качеств и социального капитала.

7. Выдвинут ряд гипотез относительно связи собственного социального статуса и веса, который игрок придает личному сигналу и действиям другого игрока;

8. Построена эконометрическая модель зависимости действий игрока от личного сигнала, действий партнера и их интеракции со статусом, социоэкономическими и когнитивными и некогнитивными личными характеристиками агентов.

Данные задачи решаются в трех статьях, представляющих диссертацию. В рамках первой статьи «Влиятельные индивиды: подходы к моделированию» представлен обзор подходов, принятых в настоящее время для моделирования социальных взаимодействий между агентами в теоретической и экспериментальной экономической литературе. В рамках второй статьи «Измерение социального статуса в экспериментальных играх» исследуется, каким образом формируется субъективное восприятие субъективного социального статуса индивида в процессе экспериментальной экономической игры, и выбирается экспериментальная игра, меняющая восприятие собственного статуса у людей, играющих разные роли. В рамках третьей статьи «Social status and social learning» экспериментальная игра, рассмотренная в предыдущей статье, используется в качестве

экспериментального условия (priming), и измеряется влияние субъективного статуса и других когнитивных и некогнитивных характеристик на процесс социального обучения.

Степень разработанности темы исследования

Влияние отдельных личных характеристик на социальное обучение исследовалось в прошлых экспериментальных исследованиях. Так, [Duffy et al, 2017] рассматривали влияние возраста и когнитивных способностей на социальное обучение, а [Berger et al, 2018] - влияние идентификации индивида, т.е. собственной позиции по какому-либо вопросу и того, отличалась ли она от позиции других членов группы. Дизайн игры на социальное обучение, использованной в диссертационном исследовании, основан на работах [Anderson, Holt, 1997; Celen, Kariv, 2004; Corazzinni et al, 2012], а также на полевом эксперименте [Burztyn et al, 2014], которые показали, что инвестиционные решения профессионалов финансового рынка подвержены двум каналам влияния: социальному обучению на примере чужого опыта и имитации чужих действий. Эффект от социального обучения выше, если участник обладает более развитой финансовой грамотностью.

Также существует обширная область экспериментальных работ, в которых исследуется влияние статуса на альтруизм, доверие, кооперативное поведение, неэтичное поведение, готовность делиться ресурсами и другие явления. Статус участников устанавливается разными способами: случайным образом присваивается ведущим эксперимента, «зарабатывается» игроками (например, по результатам выполнения задания) или предполагается, что он основан на реальных показателях, например, принадлежности к более престижной школе или более высокой касте. Так, в [D'adda, 2012] социальный статус основан на реальных характеристиках. Участники взаимодействуют в парах, и информация о вкладе лидеров в общественное благо (участников с более высоким статусом) влияет на поведение остальных участников. В [Brooks et al, 2015] в полевом эксперименте в Индии мужчины из низких и высоких каст играют в координационную игру. Пары из людей

из более высоких каст координируются между собой (достигают более высоких выигрышей) менее эффективно, чем пары из людей из низких каст. Смешанные пары показывают средние результаты. Авторы связывают это с культурными особенностями более высокой касты.

В серии работ Ball, Eckel в соавторстве с другими исследователями рассматривают влияние статуса на поведение в экспериментальных играх. В [Ball, Eckel, 1996; 1998] статус присваивался по результатам выполнения заданий (квиза). Результаты заданий суммировались, и участникам, чьи результаты находились выше медианы, присваивался высокий статус, а остальным - низкий статус. Высокостатусные участники эксперимента получали золотые звезды. В игре «ультиматум» как высокостатусные, и низкостатусные участники предлагали высокостатусным участникам больше, чем низкостатусным участникам. В [Ball et al, 2001] статус индуцировался двумя способами: на основе заданий и случайным образом. Участники эксперимента участвовали в игре, моделирующей аукцион с продавцами и покупателями. Либо продавцы, либо покупатели обладали более высоким статусом. Цены были в среднем выше, когда высокостатусные продавцы соревновались с низкостатусными покупателями, и ниже, когда статус покупателей был выше, чем продавцов. Эти результаты сохранялись для обоих видов индуцированного статуса.

Информационную базу исследования составляют данные, полученные на сессиях двух экспериментов, которые проводились в Международной экспериментальной экономики НИУ ВШЭ в Москве. В 2016 году было проведено 6 экспериментальных сессий с 68 участниками для первого исследования «Измерение социального статуса в экспериментальных играх». В 2016-2018 гг. было проведено 14 сессий с 184 участниками для второго исследования «Social status and social learning». Участники всех экспериментальных сессий были набраны через электронную базу лаборатории, и большинство из них были студентами НИУ ВШЭ. Эксперименты проводились в программе Z-tree [Fischbacher, 2007].

Экспериментальная сессия в каждом эксперименте длилась 1,5 часа. Инструментарий, используемый при анализе данных, описан в каждой из статей.

Основные результаты

Основные результаты диссертационного исследования можно разделить на два направления: исследование статуса в экспериментальных играх и исследование влияния статуса на социальное обучение.

Социальный статус в экспериментальных играх

Дизайн эксперимента

В начале эксперимента участники были случайным образом распределены по парам и последовательно играли в игры с разной степенью симметричности: «диктатор», «доверие» и «трудовой контракт». В разных сессиях порядок игр менялся. В игре «диктатор», наиболее ассимметричной из всех игр, распределителя просят поделить фиксированный бюджет между собой и получателем. Роль получателя в отличие от роли распределителя пассивна - он или она принимает выбор распределителя и не может влиять ни на свой, ни на чужой выигрыш. В игре «доверие» один из участников, «доверитель», принимает решение о том, как поделить фиксированный бюджет между собой и другим участником («поверенным»). Сумма, отданная поверенному, умножается экспериментатором на три, и поверенный может вернуть доверителю любую часть суммы. В игре «трудовой контракт» менеджер имеет капитал, из которого он выдает зарплату работнику. Работник выбирает уровень усилий, которому соответствуют разные уровни издержек. Более высокий уровень усилий соответствует более высокой прибыли менеджера, но менее высокой зарплате работника. После пяти раундов каждой игры пары менялись.

После каждой игры участники заполняли анкету, с помощью которой измерялся собственный субъективный социальный статус, а затем субъективный статус партнера по игре (второго игрока). Показатель субъективного статуса рассчитывался на основе двух групп вопросов. Первая

группа вопросов состояла из семи семантических дифференциальных шкал [Ridgeway et al, 1998]. Участник эксперимента должен был поставить себя ближе к тому или другому полюсу, на которых были представлены противоположные по смыслу качества, ассоциирующиеся с положением в социальной иерархии. Вторая группа вопросов состояла из вопроса о положении на воображаемой социальной лестнице, широко используюемого в социологии [The MacArthur Scale of Subjective Social Status, Adler et al, 2000; РМЭЗ, 2015].

После измерения статуса участники эксперимента отвечали на вопросы анкеты для измерения социально-демографических и личных характеристик. Для измерения лидерских качеств были использованы вопросы о том, насколько часто участникам приходилось принимать участие в ситуациях, связанных с принятием ответственности, проявлением личной инициативы, не подчинения авторитету и пр. [Международный сборник личностных шкал (IPIP); Kuhn, Weinberger, 2002]. Также были заданы вопросы о материальном благосостоянии, измерялся позитивный/негативный аффект, социальные нормы и доверие, участие в деятельности разных организаций и другие вопросы.

Результаты

Оценивалась модель, в которой зависимой переменной являлся субъективный статус: собственный статус, статус второго игрока и показатель разницы между собственным субъективным статусом и статусом партнера - относительный статус, а зависимыми переменными - фиктивная переменная, обозначающая роль в игре, и личные характеристики индивидов.

Для игры «диктатор» показатель относительного статуса значимо выше для распределителей, чем для получателей. В играх «доверие» и «трудовой контракт» такой разницы не наблюдалось, и этот результат сохранялся независимо от порядка игр.

Личные характеристики участников также влияют как на абсолютный, так и относительный показатель статуса. Участники с более высоким

материальным положением имеют более высокий субъективный социальный статус, как и участники с более развитыми лидерскими качествами и высоким уровенем экстраверсии (позитивного аффекта). Субъективный статус мужчин выше, чем у женщин. Структура семьи также имеет значение: если участник является младшим ребенком в семье, его субъективный статус и относительный статус ниже. Кроме того, у участников эксперимента, работающих полный рабочий день или частично занятых, а также у индивидов, принадлежащих к семьям с высоким доходом, наблюдался более высокий субъективный статус.

Влияние статуса на социальное обучение

Дизайн эксперимента

В начале эксперимента участники заполняли анкету. Затем они случайным образом были поставлены в пары и на протяжении 5 раундов играли в игру «диктатор», дизайн которой был таким же, как в предыдущем эксперименте. Внизу экрана каждого игрока во время всего эксперимента стояла надпись «Вы - распределитель/получатель».

После игры «диктатор» участники в тех же парах играли 10 раундов в игру на социальное обучение. В начале каждого раунда компьютер случайным образом выбирает состояние мира, представляющее собой целое число из дискретного равномерного распределения в интервале от -7 до 7. Участник получает личный зашумленный сигнал о состоянии мира, где шум также является числом из того же распределения, что и состояние мира. У участника есть две попытки угадать число. Чем точнее угадано число, тем выше выигрыш, а чем дальше решение от истинного состояния мира, тем выше штраф. При первой попытке игрок опирается только на личный сигнал. Затем он видит решение второго игрока при первой попытке и получает вторую попытку угадать состояние мира. После этого игроки узнают состояние мира, свой выигрыш и переходят к следующему раунду игры.

Дизайн игры на социальное обучение основан на прошлых работах, в которых социальное обучение исследовалось с помощью лабораторных

экспериментов. Дизайн, в котором состояние мира представляет собой число, сгенерированное компьютером, используется, например, в [Corazzinni et al, 2012]. Участники должны угадать средний сигнал нескольких игроков, которые соединены с ними в одну сеть. Чем точнее угадан сигнал, тем выше выигрыш. Состояние мира, представленное в виде числа в данном исследовании, позволяет численно измерить, в какой степени участник пересмотрит свое решение при второй попытке угадать состояние мира.

После игры на социальное обучение участники отвечали на вопросы о собственном субъективном статусе и статусе партнера, аналогично шкалам, описанному в предыдущем эксперименте. На следующем этапе участники играли в стандартную лотерею, измеряющую отношение к риску [Holt, Laury, 2002]. После лотереи участники заполняли анкету: стандартные социо-демографические вопросы, вопросы на измерение социоэкономического статуса, вопросы на измерение лидерских качеств, социального капитала и когнитивных навыков, позитивный и негативный аффект. Лидерские качества измерялись по расширенному перечню вопросов по сравнению с предыдущим экспериментом. Для измерения социального капитала была использована методология [Glaeser et al, 2000], согласно которой социальный капитал представляет собой число социальных связей и размер круга общения, популярность и влияние среди сверстников. Когнитивные навыки измерялись с помощью трех стандартных задач без вознаграждения [Frederick, 2005]. Также в анкету были включены вопросы о том, насколько участники склонны доверять людям.

Результаты

Была оценена регрессионная модель, в которой зависимой переменной было решение индивида в игре на социальное обучение во втором периоде, а независимыми переменными - решение индивида в игре на социальное обучение в первом периоде, решение второго игрока в первом периоде, субъективный социальный статус, роль в игре «диктатор» и другие

характеристики, а также их интеракции с решениями игрока и его/ее партнера в первом периоде.

В данном эксперименте показатель субъективного статуса для индивидов, игравших роль распределителей в игре «диктатор», был выше, чем для индивидов, игравших роль получателей, однако разница не являлась значимой. Роль распределителя также не влияла на вес, который участники придавали личному сигналу или действиям партнера. При этом индивиды с более высоким субъективным статусом придавали больший вес личному сигналу и меньший вес действиям партнера в первом периоде.

Было также проанализировано влияние личных характеристик на социальное обучение: лидерских качеств, склонности к риску, социального капитала, эмоционального состояния и других характеристик. Индивиды, не склонные к риску, придавали больший вес действиям партнера в первом периоде и меньший вес личному сигналу, при этом эффект был больше выражен среди мужчин. Индивиды, обладающие лидерскими качествами, меньше полагались на действия партнера при принятии решения. При этом разные показатели социального капитала, например, уровень доверия, негативно коррелировали с социальным обучением. Индивиды с более развитыми когнитивными навыками придавали больший вес действиям партнера в первом периоде, чем личному сигналу. Показатели объективного социального статуса (доход семьи, образование родителей, занятость) не влияли на процесс социального обучения.

Научная новизна работы заключается в следующем: 1. Была проанализирована взаимосвязь между социальным статусом, измеренным в процессе лабораторного эксперимента, и социальным обучением. Было показано, что люди с более высоким субъективным социальным статусом при принятии решения в большей степени опираются на личную информацию, чем на общественную информацию. Таким образом, были подтверждены закономерности, наблюдавшиеся в прошлых исследованиях.

2. Был разработан и опробован новый показатель субъективного статуса на основе вопросов анкеты. В прошлых исследованиях использовался только показатель субъективного социального статуса, измеряемый с помощью «социальной лестницы».

3. Было проанализировано влияние когнитивных и некогнитивных личных характеристик на социальное обучение. В прошлых работах не рассматривалась взаимосвязь этих характеристик и социального обучения. Несклонность к риску ассоциируется с меньшим весом, который присваивается личному сигналу, и большим весом, который присваивается действиям другого игрока. В прошлых исследованиях не рассматривалось влияние отношения к риску на социальное обучение. Лидерские качества ассоциируется с меньшим весом, который присваивается действиям другого игрока. Когнитивные навыки ассоциируются с большим весом, который присваивается действиям другого игрока.

4. Было исследовано влияние экспериментального поведения и роли участника в экспериментальной игре на субъективный статус. Было показано, что с помощью игры «диктатор» у участников можно создать асимметричный субъективный социальный статус - более высокий статус у распределителей и более низкий статус у получателей. При этом данный эффект не подтвердился в эксперименте, в котором участники сначала играли в «диктатор», а потом в игру на социальное обучение. Роль распределителя в игре «диктатор» не влияла на вес, который участники придавали личному сигналу или действиям партнера.

Результаты, полученные в данной работе, вносят вклад в теоретическую и эмпирическую литературу, анализирующую распространение и агрегирование информации в социальных сетях. Индивиды, обладающие высоким социальным статусом, могут хуже реагировать на информационные сигналы от других агентов. Из-за этого с большей вероятностью могут происходить информационные каскады, снижающие общественное благосостояние.

В работе проанализировано, какие личные качества влияют на использование собственной и общественной информации при принятии решения. Это позволяет прогнозировать поведение агентов в социальных сетях, например, в части распространения технологий или изменения социальных норм.

Можно выделить несколько направлений для дальнейших исследований. Во-первых, это поиск теоретического механизма для объяснения того, почему люди с более высоким статусом придают больший вес личному сигналу, чем общественной информации. Например, высокостатусные индивиды могут считать, что другие индивиды менее рациональны и чаще ошибаются, и придавать меньший вес их действиям и решениям, чем это оптимально. Индивиды с более низким статусом могут быть более конформными и получать более высокую полезность, если их действия ближе к действиям других членов социальной сети. Во-вторых, можно анализировать другие характеристики участников, которые могут повлиять на социальное обучение, например, их оценки в университете и реальную популярность среди сверстников, личные качества, измеренные по шкале «большой пятерки», расширенные шкалы измерения когнитивных способностей, а также насколько участники склонны к подчинению авторитету или насколько они считают себя хуже других (шкала личной относительной депривации). В-третьих, можно более подробно проанализировать связь статуса и избыточной уверенности в себе (overconfidence). В-четвертых, можно проверить, сохраняется ли взаимосвязь между социальным статусом и социальным обучением на другой выборке -для людей, имеющих разное образование и разный доход, т.е. более проявленный социоэкономический статус. В-пятых, можно объединять участников игры на социальное обучение не в диады, а в более крупные группы, а также увеличить число попыток угадать состояние мира, чтобы проанализировать, как разный статус повлияет на агрегирование информации и информационные каскады. В-шестых, можно провести эксперимент в

реальной иерархической структуре, например, в фирме с начальниками и подчиненными. Наконец, можно использовать другой вид экспериментального условия (прайминга), например, основанный на воспоминаниях людей о реальной ситуации, когда они обладали властью над другими людьми или когда у кого-то другого была власть над ними.

Список публикаций

Основные результаты диссертационного исследования опубликованы в трех работах общим объемом 6,82 п.л.; личный вклад автора составляет 4,17 п.л.:

— Alexei Zakharov, Oxana Bondarenko (2021) Social status and social learning. Journal of Behavioral and Experimental Economics, 90. - 2,1 п.л. (личный вклад -1 п.л.)

Список литературы диссертационного исследования кандидат наук Бондаренко Оксана Юрьевна, 2021 год

ЛИТЕРАТУРА

Осин Е. (2012). Измерение позитивных и негативных эмоций: разработка русскоязычного аналога методики PANAS // Психология. Журнал Высшей школы экономики. Т. 9 (4). С. 91—110.

Adler N., Stewart J. (2007). The MacArthur Scale of Subjective Social Status.

[Электронный ресурс] Режим доступа: http://www.macses.ucsf.edu/ research/psychosocial/subjective.php, свободный. Загл. с экрана. Яз. англ. (дата обращения: март 2018 г.).

Adler N.E., Epel E.S., Castellazzo G., Ickovics J.R. (2000). Relationship of Subjective and Objective Social Status with Psychological and Physiological Functioning: Preliminary Data in Healthy, White Women // Health psychology. Vol. 19 (6). P. 586—592.

Anderson C., Hildreth J., Howland L. (2015). Is the Desire for Status a Fundamental Human Motive? A Review of the Empirical Literature. [Электронный

ресурс] // Psychological Bulletin. Online publication. Vol. 141. No. 3. Режим доступа: http://psycnet.apa.org/record/2015-11715-001, свободный. Загл. с экрана. Яз. англ. (дата обращения: март 2018 г.).

Anderson C., John O.P., Keltner D., Kring A.M. (2001). Who Attains Social Status?

Effects of Personality and Physical Attractiveness in Social Groups // Journal of Personality and Social Psychology. Vol. 81 (1). P. 116—132.

Andreoni J., Vesterlund L. (2001). Which Is the Fair Sex? Gender Differences in Altruism // The Quarterly Journal of Economics. Vol. 116 (1). P. 293—312.

Ball S.B., Eckel C.C. (1996). Buying Status: Experimental Evidence on Status in Negotiation // Psychology and Marketing. Vol. 13 (4). P. 381—405.

Ball S.B., Eckel C.C. (1998). Stars upon Thars: Status and Discrimination in Ultimatum Games. Working paper. Blacksburg: Virginia Tech.

Berg J., Dickhaut J., McCabe K. (1995). Trust, Reciprocity, and Social History // Games and economic behavior. Vol. 10 (1). P. 122—142.

Berger J., Zelditch M. (eds) (1985). Status, Rewards, and Influence. San Francisco: Jossey-Bass Inc. Pub.

Berger J., Conner T.L., Fisek M.H. (1982). Expectation States Theory: A Theoretical Research Program. Lanham: University Press of America.

Berger J., Fi§ek M.H. (2006). Diffuse Status Characteristics and the Spread of Status Value: A Formal Theory 1 // American Journal of Sociology. Vol. 111 (4). P. 1038—1079.

Bleidorn W., Arslan R.C., Denissen J.J., Rentfrow P.J., Gebauer J.E., Potter J., Gosling S.D. (2016). Age and Gender Differences in Self-Esteem — A Cross-Cultural Window // Journal of Personality and Social Psychology. Vol. 111 (3). P. 396—410.

Bothner M., Godart F., Lee W. (2009). What Is Social Status? Comparisons and Contrasts with Cognate Concepts. University of Chicago Working Paper.

Bucciol A., Cavasso B., Zarri L. (2015). Social Status And personality Traits // Journal of Economic Psychology. Vol. 51. P. 245—260.

Chen Z., Liu R. (2014). Comparing Adolescent Only Children with Those Who Have Siblings on Academic Related Outcomes and Psychosocial Adjustment // Child Development Research.. Vol. 2014. Article ID 578289.

Childers T.L., Rao A.R. (1992). The Influence of Familial and Peer-Based Reference Groups on Consumer Decisions // Journal of Consumer Research. Vol. 19 (2). P. 198—211.

Chudek M., Heller S., Birch S., Henrich J. (2012). Prestige-Biased Cultural Learning: Bystander's Differential Attention to Potential Models Influences Children's Learning // Evolution and Human Behavior. Vol. 33 (1). P. 46—56.

D'Adda G. (2011). Social Status and Influence: Evidence from an Artefactual Field Experiment on Local Public Good Provision. Verein für Socialpolitik, Research Committee Development Economics. Working paper No. 22.

Diemer M.A., Mistry R.S., Wadsworth M.E., Lopez I., Reimers F. (2013). Best Practices in Conceptualizing and Measuring Social Class in Psychological Research // Analyses of Social Issues and Public Policy. Vol. 13 (1). P. 77—113.

Engel C. (2011). Dictator Games: A Meta Study // Experimental Economics. Vol. 14 (4). P. 583-610.

Fehr E., Kirchsteiger G., Riedl A. (1993). Does Fairness Prevent Market Clearing? An Experimental Investigation // The Quarterly Journal of Economics. Vol. 108 (2). P. 437-459.

Festinger L. (1954). A Theory of Social Comparison Processes // Human Relations. Vol. 7. P. 117-140.

Fischbacher U. (2007). Z-Tree: Zurich Toolbox for Ready-Made Economic Experiments // Experimental Economics. Vol. 10 (2). P. 171—178.

Gil-White F., Henrich J. (2001). The Evolution of Prestige: Freely Conferred Deference as a Mechanism for Enhancing the Benefits of Cultural Transmission // Evolution and Human Behavior. Vol. 22 (3). P. 165—196.

Glaeser E.L., Laibson D.I., Scheinkman J.A., Soutter C.L. (2000). Measuring Trust // The Quarterly Journal of Economics. Vol. 115 (3). P. 811—846.

Handgraaf M.J., Van Dijk E., Vermunt R.C., Wilke H.A., Dreu C.K. de (2008). Less Power or Powerless? Egocentric Empathy Gaps and the Irony of Having Little Versus No Power in Social Decision Making // Journal of Personality and Social Psychology. Vol. 95 (5). P. 1136.

Heffetz O., Frank R. (2008) Preferences for Status: Evidence and Economic Implications // Handbook of Social Economiccs. Vol. 1A. The Netherlands: North-Holland. P. 69—91.

Hyman H.H. (1942). The Psychology of Status. Archives of Psychology. New York: Columbia University.

Jasso G. (2001). Studying Status: An Integrated Framework // American Sociological Review. P. 96—124.

Keister L. (2003). Religion and Wealth: The Role of Religious Affiliation and Participation in Early Adult Asset Accumulation // Social Forces. Vol. 82 (1). P. 175—207.

Levitt S., List J. (2007). What Do Laboratory Experiments Measuring Social Preferences Reveal about the Real World? // The Journal of Economic Perspectives. Vol. 21 (2). P. 153—174.

Magee J., Galinsky A. (2008). Social Hierarchy: The Self-Reinforcing Nature of Power and Status // The Academy of Management Annals. Vol. 2 (1). P. 351—398.

Ridgeway C. (1991). The Social Construction of Status Value: Gender and other Nominal Characteristics // Social Forces. P. 367—386.

Ridgeway C.L., Balkwell J.W. (1997). Group Processes and the Diffusion of Status Beliefs // Social Psychology Quarterly. P. 14—31.

Ridgeway C.L., Boyle E.H., Kuipers K.J., Robinson D.T. (1998). How Do Status Beliefs Develop? The Role of Resources and Interactional Experience // American Sociological Review. P. 331—350.

Ridgeway C.L., Erickson K.G. (2000). Creating and Spreading Status Beliefs 1 // American Journal of Sociology. Vol. 106 (3). P. 579—615.

Smillie L., DeYoung C., Hall P. (2015). Clarifying the Relation between Extraversion and Positive Affect // Journal of Personality. Vol. 83 (5). P. 564—574.

Sulloway F.J. (1996). Born to Rebel: Birth Order, Family Dynamics, and Creative Lives. New York: Pantheon Books.

Tajfel H., Billig R.P., Flament C. (1971). Social Categorization and Intergroup Behavior // European Journal of Social Psychology. Vol. 1. P. 149—178.

Thye S. (2000). A Status Value Theory of Power in Exchange Relations // American Sociological Review. P. 407—432.

Watson D., Clark L., Tellegen A. (1988). Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales // Journal of Personality and Social Psychology. Vol. 54 (6). P. 1063—1070.

Weber M. (1978). The Distribution of Power within the Political Community: Class, Status, Party // Economy and Society. Vol. 2. P. 926—940.

Webster Jr.M., Hysom S.J. (1998). Creating Status Characteristics // American Sociological Review. P. 351—378.

Williams K.D. (2010). Dyads Can Be Groups (and Often Are) // Small Group Research. Vol. 41 (2). P. 268—274.

Поступила в редакцию 17 мая 2018 г.

REFERENCES (with English translation or transliteration)

Adler N., Stewart J. (2007). The MacArthur Scale of Subjective Social Status. Available at: http://www.macses.ucsf.edu/research/psychosocial/subjective.php (accessed: March 2018).

Adler N.E., Epel E.S., Castellazzo G., Ickovics J.R. (2000). Relationship of Subjective and Objective Social Status with Psychological and Physiological Functioning: Preliminary Data in Healthy, White Women. Health psychology, 19 (6), 586—592.

Anderson C., Hildreth J., Howland L. (2015). Is the Desire for Status a Fundamental Human Motive? A Review of the Empirical Literature. Psychological Bulletin, 141 (3), 574—601.

Anderson C., John O.P., Keltner D., Kring A.M. (2001). Who Attains Social Status?

Effects of Personality and Physical Attractiveness in Social Groups. Journal of Personality and Social Psychology, 81 (1), 116—134.

Andreoni J., Vesterlund L. (2001). Which Is the Fair Sex? Gender Differences in Altruism. The Quarterly Journal of Economics, 116 (1), 293—312.

Ball S.B., Eckel C.C. (1996). Buying Status: Experimental Evidence on Status in Negotiation. Psychology and Marketing, 13 (4), 381—405.

Ball S.B., Eckel C.C. (1998). Stars upon Thars: Status and Discrimination in Ultimatum Games. Working paper. Blacksburg: Virginia Tech.

Berg J., Dickhaut J., McCabe K. (1995). Trust, Reciprocity, and Social History. Games and Economic Behavior, 10 (1), 122—142.

Berger J., Zelditch M. (eds) (1985). Status, Rewards, and Influence. San Francisco: Jossey-Bass Inc. Pub.

Berger J., Conner T.L., Fisek M.H. (1982). Expectation States Theory: A Theoretical Research Program. Lanham: University Press of America.

Berger J., Fi§ek M.H. (2006). Diffuse Status Characteristics and the Spread of Status Value: A Formal Theory 1. American Journal ofSociology, 111 (4), 1038—1079.

Bleidorn W., Arslan R.C., Denissen J.J., Rentfrow P.J., Gebauer J.E., Potter J., Gos-

ling S.D. (2016). Age and Gender Differences in Self-Esteem — A Cross-Cultural Window. Journal of Personality and Social Psychology, 111 (3), 396—410.

Bothner M., Godart F., Lee W. (2009). What Is Social Status? Comparisons and Contrasts with Cognate Concepts. University of Chicago Working Paper.

Bucciol A., Cavasso B., Zarri L. (2015). Social Status and Personality Traits. Journal of Economic Psychology, 51, 245—260.

Chen Z., Liu R. (2014). Comparing Adolescent Only Children with Those Who Have Siblings on Academic Related Outcomes and Psychosocial Adjustment. Child Development Research, 2014, ID 578289.

Childers T.L., Rao A.R. (1992). The Influence of Familial and Peer-Based Reference Groups on Consumer Decisions. Journal of Consumer Research, 19 (2), 198—211.

Chudek M., Heller S., Birch S., Henrich J. (2012). Prestige-Biased Cultural Learning: Bystander's Differential Attention to Potential Models Influences Children's Learning. Evolution and Human Behavior, 33 (1), 46—56.

D'Adda G. (2011). Social Status and Influence: Evidence from an Artefactual Field Experiment on Local Public Good Provision. Verein für Socialpolitik, Research Committee Development Economics. Working paper No. 22.

Diemer M.A., Mistry R.S., Wadsworth M.E., Lopez I., Reimers F. (2013). Best Practices in Conceptualizing and Measuring Social Class in Psychological Research. Analyses of Social Issues and Public Policy, 13 (1), 77—113.

Engel C. (2011). Dictator Games: A Meta Study. Experimental Economics, 14 (4), 583—610.

Fehr E., Kirchsteiger G., Riedl A. (1993). Does Fairness Prevent Market Clearing?

An Experimental Investigation. The Quarterly Journal of Economics, 108 (2), 437—459.

Festinger L. (1954). A Theory of Social Comparison Processes. Human Relations, 7, 117—140.

Fischbacher U. (2007). Z-Tree: Zurich Toolbox for Ready-Made Economic Experiments. Experimental Economics, 10 (2), 171—178.

Gil-White F., Henrich J. (2001). The Evolution of Prestige: Freely Conferred Deference As a Mechanism for Enhancing the Benefits of Cultural Transmission. Evolution and Human Behavior, 22 (3), 165—196.

Glaeser E.L., Laibson D.I., Scheinkman J.A., Soutter C.L. (2000). Measuring Trust. The Quarterly Journal of Economics, 115 (3), 811—846.

Handgraaf M.J., Van Dijk E., Vermunt R.C., Wilke H.A., Dreu C.K. de (2008). Less Power or Powerless? Egocentric Empathy Gaps and the Irony of Having Little Versus No Power in Social Decision Making. Journal of Personality and Social Psychology, 95 (5), 1136.

Heffetz O., Frank R. (2008) Preferences for Status: Evidence and Economic Implications. Handbook of Social Economiccs. Vol. 1A. The Netherlands: North-Holland, 69—91.

Hyman H.H. (1942). The Psychology of Status. Archives of Psychology. New York: Columbia University.

Jasso G. (2001). Studying Status: An Integrated Framework. American Sociological Review, 96—124.

O.ro. EoHgapeHKO, A.B. 3axapoB

Keister L. (2003) Religion and Wealth: The Role of Religious Affiliation and Participation in Early Adult Asset Accumulation. Social Forces, 82 (1), 175—207.

Levitt S., List J. (2007). What Do Laboratory Experiments Measuring Social Preferences Reveal about the Real World? TheJournal of Economic Perspectives, 21 (2), 153—174.

Magee J., Galinsky A. (2008). Social Hierarchy: The Self-Reinforcing Nature of Power and Status. The Academy of Management Annals, 2 (1), 351—398.

Osin E. (2012). Measurement of Positive and Negative Emotions: Development of Russian Version of PANAS. Psychology. Journal of the Higher School of Economics, 9 (4), 91—110 (in Russian).

Ridgeway C. (1991). The Social Construction of Status Value: Gender and other Nominal Characteristics. Social Forces, 367—386.

Ridgeway C.L., Balkwell J.W. (1997). Group Processes and the Diffusion of Status Beliefs. Social Psychology Quarterly, 14—31.

Ridgeway C.L., Boyle E.H., Kuipers K.J., Robinson D.T. (1998). How Do Status Beliefs Develop? The Role of Resources and Interactional Experience. American Sociological Review, 331—350.

Ridgeway C.L., Erickson K.G. (2000). Creating and Spreading Status Beliefs 1. American Journal of Sociology, 106 (3), 579—615.

Russian Longitudinal Monitoring Survey — HSE, Questionnaire for adults (2015).

Available at: https://www.hse.ru/rlms/ (accessed: March 2018, in Russian).

Smillie L., DeYoung C., Hall P. (2015). Clarifying the Relation between Extraversion and Positive Affect. Journal of Personality, 83 (5), 564—574.

Sulloway F.J. (1996). Born to Rebel: Birth Order, Family Dynamics, and Creative Lives. New York: Pantheon Books.

Tajfel H., Billig R.P., Flament C. (1971). Social Categorization and Intergroup Behavior. European Journal of Social Psychology, 1, 149—178.

Thye S. (2000). A Status Value Theory of Power in Exchange Relations. American Sociological Review, 407—432.

Watson D., Clark L., Tellegen A. (1988). Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales. Journal of Personality and Social Psychology, 54 (6), 1063—1070.

Weber M. (1978). The Distribution of Power within the Political Community: Class, Status, Party. Economy and Society, 2, 926—940.

Webster Jr.M., Hysom S.J. (1998). Creating Status Characteristics. American Sociological Review, 351—378.

Williams K.D. (2010). Dyads Can Be Groups (and often Are). Small Group Research, 41 (2), 268—274.

Received 17.05.2018

O.Yu. Bondarenko

National Research University Higher School of Economics,

Moscow, Russia

A.V. Zakharov

National Research University Higher School of Economics,

Moscow, Russia

Measurement of Social Status in Experimental Games

Abstract. The majority of social and economic interactions take place between people of different social status. Age, position, income and other factors affect the way people evaluate their position in the society. We investigate how self-estimation of the social status is formed when an individual participates in an economic experimental game. In our experiment subjects are set in pairs and play consequently the dictator game, the trust game and the labor market (contract) game. After each game we measure their subjective socioeconomic status using two different scales. We show that participation in the dictator game affects the perception of one's social status to the greatest extent: the status of dictators is higher than the status of recipients. Prescription of roles in other games does not have such an effect. Active behavior, gender, income, etc. also affect the subjective status.

Keywords: social status, game theory, dictator game, group behavior.

JEL Classification: C72, C91, D83, Z13.

Приложение 3. Статья «Social status and social learning»

Alexei Zakharov, Oxana Bondarenko (2021) Social status and social learning // Journal of Behavioral and Experimental Economics, 90. https://doi.org/10.1016/j.socec.2020.101647

Разрешение на копирование:

Согласно https://www.elsevier.com/about/policies/sharing, автор статьи может использовать полную журнальную версию статьи в своей диссертации при условии, что указан DOI статьи.

© 2021. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/

Journal of Behavioral and Experimental Economics 90 (2021) 101647

Contents lists available at ScienceDirect

Journal of Behavioral and Experimental Economics

journal homepage: www.elsevier.com/locate/jbee

Social status and social learning^

Alexei Zakharova, Oxana Bondarenkob

a Higher School of Economics, Moscow, Russia

b The Center for the Study of Customs Tariff and Non-Tariff Regulation

H)

Check for updates

ARTICLE INFO ABSTRACT

Keywords: We studied the effect of social status on social learning and other personal characteristics in an experimental

Social st:atus game where individuals in a dyad made repeated attempts to guess the underlying state of the world. Several sets

Sodal leammg of survey questions were deployed to control for socioeconomic status, the subjective perception of social status,

EcoDomlc expenments and leadership traits, as well as quality and quantity of social interactions, and cognitive reflection. Risk aversion

was measured using an incentivized task. We also induced social status in each pair of subjects using a dictator game. We found that people with high subjective social status relied less on observed choices of other subjects and put more weight on private information. Subjects who were less risk-averse and showed more leadership traits, were also less likely to learn from the actions of others. Some effects were gender-specific. Our finding that social learning is stronger in low-status individuals can imply higher likelihood of information cascades in hierarchical networks.

1. Introduction

Much of human interaction takes place between people of different social status — a property that has been defined as "the prominence, respect, and influence individuals enjoy in the eyes of other group members" (Anderson et al., 2006), admiration from other members of the group (Magee & Galinsky, 2008) or influence exerted on other people (Ridgeway & Correll, 2006). Id the workplace, there are bosses and rank-and-file employees, in the military — subordinates and commanding officers, while in more traditional societies women have lower status than men. Other sources of social status may include wealth, education, or occupation prestige (Diemer, Mistry, Wadsworth, Lopez, & Reimers, 2013), social popularity (Glaeser, Laibson, Scheinkman, & Soutter, 2000), and even one's ranking in an online computer game (Evers, de Ven, & Weeda, 2015). Social status hierarchies are also ubiquitous in the animal kingdom (Chiao, 2010) and are central to social behavior of many species.

In many economic contexts it is important to know the degree to

which one's social status affects social learning, or how different sources of information (such as friends and contacts, role models, mass media, or the Internet) are consulted to update belief and arrive at a decision. Social learning and peer effects shape personal decisions in such areas as public good provision (d'Adda, 2012), financial planning (Bursztyn, Ederer, Ferman, & Yuchtman, 2014), physician prescriptions (Nair, Manchanda, & Bhatia, 2010), political behavior and persuasion (Weeks et al., 2017), or academic achievement (Van Ewijk & Sleegers, 2010).

A key theoretical and empirical question regarding social learning is whether information aggregation takes place: if individuals receive noisy signals about the underlying state of the world, does the society eventually learn that state (Acemoglu, Dahleh, Lobel, & Ozdaglar, 2011)? Asymmetries in learning due to social status may potentially make information aggregation more difficult, as the individuals with high social status will have a disproportionately large influence, as their private signals will carry more weight. Especially, this will likely be the case if individuals with high social status are also more connected and listened to by a larger number of peers and/or the network is

☆ The authors thank Anastasia Antsygina, Alexei Belianin, Lubomir Cingl, Heike Hennig-Schmidt, Ole Jann, Andrea Mattozzi, Tatiana Mayskaya, David Myatt, Roman Zakharenko, and seminar participants at HSE and Nuffield College for their very helpful feedback on this project. We also like to thank Stepan Aleksenko and Roman Solntsev for their excellent research assistance. Declaralion of interest: Done. All supplementary and replication material can be found at https://github. com/morhellis/Status-information. We wish to acknowledge the support of NES Center for the Study of Diversity and Social Interactions which received state support according to the Government Decree of April 9, 2010 No 220 and the Contract providing the grant No 14.U04.31.002 of June 26, 2013. The article was prepared within the framework of the Basic Research Program at the National Research University Higher School of Economics (HSE) and supported within the framework of a subsidy by the Russian Academic Excellence Project '5-100'.

E-mail addresses: [e-mail] (A. Zakharov), [e-mail] (O. Bondarenko).

https://doi.org/10.1016/j.socec.2020.101647

Received 8 May 2020; Received in revised form 8 September 2020; Accepted 17 November 2020

Available online 19 November 2020

2214-8043/© 2020 Elsevier Inc. All rights reserved.

hierarchical, with higher-status people being closer to the top. Theoretical models predict that the presence of influential agents can make learning in networks more difficult (Acemoglu, Ozdaglar, & Para-ndehGheibi, 2010; Golub & Jackson, 2010).

Empirical research in psychology, sociology and economics provides behavioral foundation for the relationship between social status and social learning. In the course of natural selection people learned that it is more effective both from the informational and psychological perspective to copy high-rank group members (Henrich & Gil-White, 2001). Low-status subjects show more empathy and interdependency, and are more attuned to others than their high-status counterparts (see Stellar et al., 2012; Varnum, Blais, Hampton, and Brewer (2015); Kraus, Cote, and Keltner (2010)). Lower social class is associated with stronger neural, physiological or emotional responses to others' pain and suffering (Stellar et al., 2012; Varnum et al., 2015), to others' motor movements (Varnum, Blais, & Brewer, 2016), and to social stress and rejection (Muscatell et al., 2016). Individuals of higher socioeconomic status, defined in terms of educational attainment or occupational prestige, are worse at judging the emotions of others (Kraus et al., 2010), show fewer socially engaged emotions (such as feeling friendly or guilty) and more disengaged emotions, such as feeling proud or angry (Na et al., 2010), and are less likely to feel close to other people in their social networks.

A possible explanation proposed by Kraus et al. (2012) is that the life circumstances of low-status individuals are more likely to be influenced, or be perceived to be influenced, by forces outside their immediate control. This gives rise to the culturally ingrained contextualist social cognitive tendencies or "an external orientation to the environment motivated by managing external constraints, outside threats, and other individuals". High status entails relative material and social freedom, leading to a different cognitive mindset — one characterized by greater perceptions of control and self-sufficiency, tendency to explain behavior as caused primarily by personal influence, and greater attendance to own (vs. others) mental state (Piff, Kraus, & Keltner, 2018).

Although social status and learning are correlated, establishing a causal link between social status and social learning is complicated for several reasons. First, people differ in the size of their reference groups. Some may value the opinion, or react to the actions of only a few people (such as friends or role models), while others consult much larger reference groups when forming an opinion. Hence, individual A is more likely to learn from individual B, than vice versa, if the reference group of A is smaller. At the same time, the size of one' s reference group may be linked to social status. For example, those who are more extroverted and open to contact with other people are also more likely to be perceived as leaders (Judge & Bono, 2000).

Second, social contacts can be asymmetric, and this asymmetry is likely to be correlated with social status. A high-status individual A is more likely to belong to the reference group of a low-status individual B, than vice versa. Politicians, public intellectuals, or celebrities can be listened to or serve as a role model for thousands of people, without even personally knowing most of them.

Finally, people tend to have different levels of knowledge. When making a decision, individual B is more likely to learn from prior words or actions of individual A if the latter is more knowledgeable in that specific area. At the same time, either actual or perceived knowledge is likely to be correlated with social status (Paulhus & Morgan, 1997), so correlation between status and learning can arise even if people are fully Bayesian rational.

Our goal is to study how social status affects social learning in a laboratory experiment where subjects were divided into pairs. Each pair of subjects played 10 rounds of a social learning game. In every round of the game, each subject received a noisy signal about the underlying state of the world, and was given two attempts to guess that state. On the first attempt, only the private signals were observed. Before the second attempt, the subjects observed the first attempts of their peers. We then analyzed the extent to which these two sources of information

contributed to the second attempted guesses of the subjects. By design, the signals received by both subjects in a pair had identical distributions, so both subjects in a pair were equally well informed about the underlying state of the world.

We inferred subjective social status using a survey-based measure administered at the end of the experiment, composed of the McArthur 10-step ladder and several other questions. We also used several sets of questions to capture the socioeconomic status or the person's status-related characteristics observable to others. Other questions were used to infer the frequency with which one assumes leadership roles, and the frequency/quality of his or her social interactions.

Prior to the social learning game, the same pair of subjects played five rounds of the dictator game with fixed roles. In the previous work (Bondarenko & Zakharov, 2018) the authors found that playing as a dictator substantially increased the player' s subjective social status (these results are described in Appendix C). The goal of the dictator game stage was the experimental induction of social status.

The specific contribution of our research is that we were able to isolate the effects of social status from those of asymmetry in knowledge and the number/direction of social contacts. Our study is the first to identify the effect of either social status or risk aversion on social learning in an incentivized experiment. Previously, several other factors were reported to have an effect on social learning in experimental settings, such as age and cognitive ability (Duffy et al., 2017) or shared identity (Berger et al., 2018). The interaction of social status, peer effects, and social learning was also studied in several field experiments. Bursztyn et al. (2014) found that investment decisions were subject to peer effects, with the utility arising both from using the information provided by the peers, and from imitating their decisions, with the utility from imitating the peer being stronger if the peer's decision is considered to be more informed. The approach we take is different — everybody has information of the same quality, and instead we look at the individual-level characteristics that affect social learning.

We did not find that social learning was significantly affected by the assignment of subjects into dictator or recipient roles. However, we find that survey-based measures of social status predict the degree to which private signals and the actions of others are used during the second guess attempt. Higher self-perceived social status leads to less weight put on the peers decision, and greater weight on one's private signal. Reporting having taken leadership roles (such as organizing events, being an entrepreneur, speaking publicly, or persuading others to change opinion) during the past year also leads to underweighing the peer' s decision. These results persist when controlling for risk preferences and cognitive reflection, which are also correlated with social learning in our experiment.

In our experimental design, the players in a dyad were aware that their status was not equal (unlike in Chandrasekhar, Larreguy, and Xandri (2015) or Grimm and Mengel (2020)), and social status (including its subjective measures) was measured using a survey — something that was not done in some earlier works (e.g. Mobius, Phan, & Szeidl (2015)).1

Our work is related to the literature on coordination games. In fact, the design of our experiment was similar to Cornand and Heinemann (2014) and Shapiro, Shi, and Zillante (2014), with two important exceptions. First, in these works each subject observed his or her own signal and a common public signal, while in our case each subject in a pair acted on a private signal, observed the action of the other player, and acted again. Thus, we were able to focus on the asymmetry in the dissemination of information, while keeping fixed the number of social contacts (the subjects were arranged in pairs) and the knowledge of the subjects (it was common knowledge that everyone received signals of

1 Interestingly, in Grimm and Mengel (2020) the payoff of the subjects increases with emotional intelligence, although the authors do not report whether it affected the decision weights of the subjects.

the same precision). Second, in our experiment the subject was not explicitly rewarded for conforming to the action of his or her peer; instead, we were looking for individual-level correlates of the weights put on private signals and actions of peers.2

Our work is also related to the literature on conformism in social networks (e.g. Liu, Patacchini, & Zenou (2014)). Low-status individuals may have a preference for conforming to their high-status peers. Thus the low-status player takes the choices of the high-status player as a social norm, and deviating from that norm is costly for him or her. At the same time, the high-status player has no such concern for the choices made by the low-status player. Our results imply that the social norm that determines the strength of strategic complementarity is not necessarily uniform.

At a further distance from our work is the experimental literature looking at the effect of merit-based status on the willingness to share resources (Schurter & Wilson, 2009; Ball & Eckel, 1996; Bracha, Heffetz, & Vesterlund, 2009), or the effect of status on unethical behavior (Schurr & Ritov, 2016; Gill, Prowse, & Vlassopoulos, 2013). Our work is also related to the literature investigating the effect of power on advice taking where individuals primed with high power discount advice from others (Tost, Gino, & Larrick, 2012).

The rest of this paper will be structured as follows. In Section 2 we describe the design of the experiment. Section 3 provides the results of the experiment. Section 4 concludes.

2. Experiment design

We implemented 14 experimental sessions with a total of 184 participants at the Laboratory for Experimental and Behavioral Economics of the Higher School of Economics in Moscow, Russia. The experiment was computer-based, using the Z-tree program (Fischbacher, 2007). The median age was 21 years, while 38% of the subjects were men. Almost all subjects were students of Higher School of Economics, recruited via online announcement3.

The total list of sessions is given in Table B.1. Each experimental session lasted approximately 1 h and 30 minutes, including decisions and payment. The subjects were paid in private, with the total earnings of each subject written on a sheet of paper that was shown to each subject. The payoff at the end of the experiment was equal to the show-up fee of 200 Russian Rubles (RUR), plus the payoff from three stages of the experiment: the dictator game, the social learning game, and the risk elicitation task. The mean payoff was 799 RUR or $10, the minimum payoff was 340 RUR or $4.3, and the maximum payoff was 1215 RUR or $15.2.

2.1. Inducement of social status

In the first stage of the experiment the subjects played 5 rounds of dictator game in fixed pairs and in fixed roles, with roles in each pair allocated at random in the beginning of the experiment. We preferred to induce social status using random allocation of roles, because a merit-based allocation — either based on performance in an experimental task (Ball et al., 2001; Ball and Eckel, 1996; Charness et al., 2011; Eckel

2 Social influence was also studied in a sequential dictator game (Cason & Mui, 1998).

3 Id our experiment we use the student subject pool. Many experimental

studies prove external validity and generalizablity of results obtained on a standard sample. Similar patterns of experimental behavior in bargaining, trust, bidding and other areas are found in students and professionals. For review of such studies see Falk and Heckman (2009); Camerer (2011); Frechette (2011). Examples of more recent studies include comparison of cheating between students and users of public transport (Dai, Galeotti, & Villeval, 2018), students and real government officials (Hanna and Wang, 2017), students and actual taxpayers (Alm, Bloomquist, & McKee, 2015).

and Wilson, 2007), or on some real-life characteristic (such as subjects' grades (Schurter & Wilson, 2009), high/low-profile school/caste (Liebe & Tutic, 2010; Brooks, Hoff, & Pandey, 2015), social popularity (Glaeser et al., 2000), or morality and respect (d'Adda, 2012)) — will produce an allocation of status that is correlated with competence and better access to information. That, in turn, will bias the results, as in a social learning game it could be optimal to imitate the behavior of the referent if his or her status is linked to perceived competency. In the dictator game, the dictator and the recipient are in the asymmetrical positions of power: the dictator is entitled with the budget and has control over both his or her and recipients payoff while the recipient has nothing and cannot control the outcome. Previously, Bondarenko and Zakharov (2018) found that, in the dictator game, dictators scored higher than recipients on several measures of subjective social status, while in two other types of games there was no difference between participants of different roles. These results are briefly reported in Appendix C.

At the beginning of each round of the dictator game, the dictator in each pair was given a budget of 100 ECU that he or she could share with the recipient (the exchange rate was equal to 2.5 Rubles per 1 ECU at the time of the experiment), while the recipient took no action. At the end of each round, subjects were informed of their earnings for that round. The earnings from that part of the experiment were equal to the earnings from a randomly selected round of the dictator game. At the instruction stage and during the game, the dictators were referred to as "allocators" and the recipients as "receivers" in order to avoid the experimenter demand effect; the same wording appeared in the instructions for the dictator game stage.

In some experimental sessions, dictators belonged to one of the two types. Dictators of the first type could give the recipient any amount between 0 and 50 ECU, with the recipient receiving twice the amount that was given by the dictator. Dictators of the second type could give any amount from 0 to 100 ECU, with the amount received equal to the amount given. There was a 50% change that a dictator would be of any of the two types, the types of the dictators remained fixed throughout the five rounds of the dictator game, and the recipients did not know the type of the dictator they were paired with.4

2.2. The social learning game

After the dictator game, the subjects, in pairs, were assigned to a task where each subject observed an imperfect signal about the unobserved state of the world, and made two guesses about the state, relying on two inputs: the private signal, and (for the second attempt) on the observed guess of his or her peer. There were 10 rounds of the social learning game; the pairing of the subjects was retained from the dictator game, and did not change between rounds. Our goal was to see what determined the importance of the two inputs to the subject's decision.

Prior to the first round of this stage of the experiment, the subjects completed a small quiz to test their understanding of the rules and how the payoffs were calculated.

In the beginning of each round and for each pair, the computer generated a number X, drawn from a discrete uniform distribution on { — 7, — 6,..., 7}. The goal of subjects in each round was to guess the value of X. For each subject i = 1, 2 in the pair, the computer generated an integer Yi, which was also drawn from a discrete uniform distribution on { — 7, — 6,..., 7}. Initially, each subject observed the private signal Zi = X + Yi (which could be an integer between -14 to 14). In the instructions,

4 Id sessions 1-5 the dictator belonged to only one type. The recipient received the amount given to him by the dictator and the dictators payoff was 100 - the amount donated. Id sessions 6-14 the dictator belonged to the first or the second type described above. Two types of dictators were used to obtain larger variation in donations so that we could check whether larger donation received by the recipient affected her subjective social status (see Section 3).

the subjects were informed that X and all the numbers Yi are statistically independent.

After observing X + Yi, the subject was given the first attempt to guess the value of X. After the first attempt, the subject was informed about the attempted guess of the subject that he or she was paired with, and was given the second attempt to guess X. The payoff of the subject in each round was 120 ECUs, minus any deductions made for not guessing the value of X correctly. For each of the two guess attempts, the subject was deducted the amount of ECUs equal to the minimum of 50, and 10 times the absolute difference between X and the subject's guess. During every guess attempt, the subjects were reminded about their roles in the dictator game. At the bottom of the screen, the subject read either "You are an allocator" of "You are a receiver", depending on his or her role in the dictator game.

If we assume that there are no intrinsic costs or benefits (e.g. preference for conformity), the Bayesian equilibrium in this game is straightforward to calculate. The second-period payoffs are not affected by his or her first-period decision. Thus, the first-period decision should minimize the expected first-period penalty, and is given by Table B.2. As the penalty function is linear, there are multiple optimal responses to some values of signals, and the second-period actions depend on how one randomizes over first-period best responses; Table B.3 gives the on-equilibrium-path second-period responses for the case where that randomization is uniform.

Using weighted OLS, we can then calculate the expected value of coefficients if we were to regress the second-period decision on the subject' s private information and on the first-period decision of his or her peer, assuming that everyone plays strategies given in Tables B.2 and B.3. We estimated the first coefficient to be equal to a1 = 0.3266, and

the second coefficient to be equal to b = 0.6513. Our goal is to determine whether such coefficients estimated from the actual data depend on the status of the individual, and how they relate to the benchmark values given above.

2.3. Risk preference elicitation

The social learning game was followed by a risk lottery task, where each subject had to make 10 decisions (this design has been first used in Holt & Laury (2002)). Each decision was a choice between a safe lottery that offered 50 RUR with some probability p and 40 RUR with probability 1 p, and a risky lottery that offered 96.25 RUR with probability p, and 2.5 RUR with probability 1 — p. The values of p varied from 0.1 to 1 in 0.1 increments. The subjects were informed that, at the end of the experiment, one pair of lotteries (corresponding to some p) would be selected at random, and the lottery chosen by the subject would be used to determine his or her payoff in that part of the experiment. Higher willingness to take risks should correspond to a higher proportion of risky lotteries.

2.4. Measuring social status and other personal characteristics

Subjects completed a survey that contained socio-demographic questions, such as gender, age, parental education, siblings, and income. Other questions were designed to elicit the subject's social status, sociability, cognitive ability, and a number of other characteristics.

First, we measured the individual's subjective social status or the perception of one's relative standing in the society. We asked "Which of the following best describes you?" and presented the subject with 7 scales related to status, power, and confidence (a similar set of scales to measure subjective social status was used by Ridgeway, Boyle, Kuipers, & Robinson (1998)). The eighth question, known as the McArthur 10-step ladder (Adler, Epel, Castellazzo, & Ickovics, 2000), was of the following form: "In our society, there are people who stand at high positions and people who stand at lower positions. Please state where you are on the 10-step ladder, where 1 is the lowest step, and 10 is the

highest step". By taking the first principal component of the first eight questions, we construct the subjective status index (the Cronbach's a for the eight questions was 0.8263, while the eigenvalue for the first component was equal to 3.8172; see Table B.4). The same eight questions were then asked to measure the subjective perception of the peer's social status. The subjective assessment of peer status index was constructed from similar eight questions where the subject was asked to evaluate his or her peer (Cronbach's a = 0.8200, eigenvalue for the first component was 3.6857; see Table B.5). For every single component of the subjective status index, assessment of own status was more favorable that of the peer's status, with the difference significant at p < .001 (paired sample t-test) in each case; thus the subjects exhibited self-serving bias. In Appendix E we look at the correlates of subjective assessment of own and peer social status; in males in particular, the former was higher in those who were dictators in the first stage of the experiment.5

The second set of questions measured socioeconomic status or more objective characteristics related to power, prestige, and access to resources: family income, past andanticipated future change in family income, parental education, and whether the individual had younger or older siblings.6

Third, we looked at the non-cognitive skills that are correlates of socioeconomic status and labor market outcomes (Weinberger, 2014; Deming, 2017). A set of questions measured leadership skills: we asked how often (on a 1-10 scale) during the past year the subject took part in activities associated with responsibility, initiative, or not yielding to group pressure and authority. In particular, we asked whether the subject organized meetings/events, led a voluntary association, was an entrepreneur, moderated a group in social networks, managed a large sum of money, spoke publicly, convinced someone to change one' s opinion, expressed an opinion different from that of the majority, and was ranked in the top 5% of her class. Taking the normalized first principal component of these questions, we construct the index of leadership skills (the Cronbach's a for the nine questions was 0.7290, while the eigenvalue for the first component was equal to 3.0586; see Table B.6). Separately, we asked questions about participation sports and in political/professional organizations or clubs. A measure of sociability was constructed with questions about how many friends the subject has, how often he or she meets with friends and is invited to parties, how often he or she meets new people, how often people turn to the person for advice, how active he or she is in social networks, and whether the subject is dating someone. Taking the normalized first principal component of these questions we construct the sociability

5 Although students usually do not have either an established occupation or income, it is common practice to measure their socioeconomic status by asking about parents' education and family income in social science research. Research that supports using the McArthur ladder question on adolescent subjects, relating subjective status to stress, pessimism, coping style (Goodman et al., 2001) and sense of control (Goodman et al., 2001; Kraus, Piff, & Keltner, 2009) — results mirrored in the studies of adults (Adler et al., 2000). Diemer et al. (2013) review a variety of indicators measuring socioeconomic status. Family income, parents' education and McArthur's ladder are recommended for usage in all samples including adolescents and students while such measures as occupational prestige are more justified for adult samples. In papers on education studies cited by the literature review on socioeconomic status and behavior of students in college (Walpole, 2003), students' socioeconomic status is measured by family income and parents' education and is considered to be a valid measure.

6 There are several reasons why the number of siblings can be relevant to one's status: parents having to share a limited amount of cognitive or material resources between the children, greater parental attachment to firstborn or only children, and the dilution of intellectual resources in a large family (Chen & Liu, 2014).

Table 1

Summary statistics.

mean sd obs

Male 0.38 0.48 1800

Age 20.68 2.46 1800

Both parents have higher education 0.67 0.47 1800

Has older sibling(s) 0.27 0.45 1800

Has younger sibling(s) 0.33 0.47 1800

Only child 0.45 0.50 1800

Risk aversion 0.54 0.20 1800

Cognitive reflection 0.47 0.41 1800

Positive affect 0.00 1.00 1800

Negative affect 0.00 1.00 1800

Subjective status 0.00 1.00 1800

Subjective status, partner 0.00 1.00 1800

Leadership index 0.00 1.00 1580

Civicness index 0.00 1.00 1720

Socialization index 0.00 1.00 1570

Participates in organisations 0.23 0.42 1800

Has sports category 0.15 0.36 1800

Works part or full time 0.49 0.50 1800

Subjective health (1-10) 7.61 2.11 1800

Family income (1-5) 4.04 1.10 1720

Change in well-being (1-5) 2.99 0.84 1800

Change in expected well-being (1-5) 2.41 0.73 1800

People can be trusted 0.28 0.45 1800

index (Cronbach's a = 0.5835, eigenvalue for the first component was 2.0637; see Table B.7).7

Other questions were used to produce control variables that are potential correlates of subjective and socioeconomic status. Cognitive reflection was measured with three non-incentivized questions, using wording from Frederick (2005); cognitive ability was found to be related to risk (Dohmen, Falk, Huffman, & Sunde, 2010) and to predict some forms of strategic behavior in experiments (Branas-Garza et al., 2019; Hoppe and Kusterer, 2011). Subjective health (which, together with income, is correlated with higher status (Adler et al., 2000; Diemer et al., 2013)) was measured on a 1-10 scale. A measure of civicness — a concept related to social capital, see Algan, Cahuc, and Sangnier (2016) — was calculated as the normalized first principle component based on five survey questions regarding the justifiability of certain types of unethical behaviors, such as not paying for public transport (Cronbach's a = 0.7281, eigenvalue for the first component was 2.4110; Table B.8 has specific question wording). The survey also included the binary measure of interpersonal trust and a question on whether the subject is employed.

Finally, we measured subjects' emotional state. We calculated the positive and negative affect using the PANAS questionnaire (Watson et al., 1988). These two scales are used to measure the positive and negative emotions experienced by the person. Positive affect has been linked to extroversion (Smillie, DeYoung, & Hall, 2015) which in turn, is correlated with higher status (Bucciol, Cavasso, & Zarri, 2015). Negative affect is related to neuroticism (Watson, Wiese, Vaidya, & Tellegen, 1999) which, in turn, can be correlated with risk aversion and uncertainty (Rustichini et al., 2012).

The survey consisted of two parts. For a random subset of subjects, Part I (comprising questions on subjective health, income, interpersonal trust, and the civicness measure) was asked at the beginning of the

Importance of private signals and observed actions

by DG role by status quartile by safe choices

Recepient

Q1 Q2 Q3 Q4 5 or fewer over 5

Private signal

Peer action

Fig. 1. The weights of private signal and observed peer action for the second-period decision.

experiment, while Part II followed the risk aversion task. For other subjects, both parts of the survey followed the risk aversion task. Each subject had a 50% probability of being included in the first group; the effect of this treatment is reported in Appendix D.8 Summary statistics for all variables are given in Table 1.9

3. Results

Our goal was to analyze how subjects weight their private signals and the observed actions of other subjects when trying to guess the state of the world, and to see whether the social status of the individual (and one' s perception of the peer' s status) affected these weights.

3.1. First-period actions

We begin by looking at how the first-period actions of subjects depend on their signals. After observing private signal z,, individual ican be certain that the true state of the world lies in the set Si(z,) = { — 7,., 7} n {%i — 7, ...,Zi + 7}. In 98.9% of observations, the first-period decisions lie within these intervals. In 76.4% of observations, the first-period decisions were of the same sign as and less extreme than the signals, belonging to the sets S2(Zi) = Si(zt) n {min{Zi,0},...,max{Zi,0}}. In a smaller fraction of observations, 38.5%, the individuals chose a guess that was equal to one half of their signals, rounded upward or

downward: xn e j ^J, |~§j |. Finally, in as much as 56.8% of cases the

subjects chose a value that minimized the expected first-period penalty, given by Table B.2.10

To investigate the effect of private signals on their first-period actions, we estimated the following reduced-form model:

X1it = a Zi, + a2WiZi, + ß1 + ß2W + eu,

(1)

where Xnt is the first guess of individual i in round t, Zit is the signal received by individual i in round t, and Wi are controls that may include

7 We treat "decline to answer/don't know" answers as missing, and use a 8 The survey was divided into two parts to test for the priming of questions

multiple imputation algorithm to fill these observations. order on social learning, see Appendix D.

9 The number of observations was calculated as the number of subjects multiplied by the number of rounds in the social learning game. Four subjects were excluded from the dataset due to a Z-tree failure. Questions used for constructing the leadership index and the socialization index were not included in the first two sessions.

10 When guesses are chosen at random, on average, the expected first-period penalty is minimized in 21.8% of cases.

Table 2

The effects of treatment, subjective status, and objective status on second-period action.

Dictator Subj-own Subj-other Income Inc. (exp) Inc (retr) Parental ed. Yo. sib. Old. sib. Only child

Priv. sig. X [Var.] 0.0157 0.0267*** 0.00450 -0.00369 -0.0107 0.00606 -0.0196 0.0230 0.00647 -0.00975

(0.0176) (0.00731) (0.0117) (0.00869) (0.0117) (0.0119) (0.0205) (0.0182) (0.0202) (0.0175)

Part. act. X [Var.] 0.000310 -0.0802*** -0.000207 0.00447 0.0390 0.00547 0.0434 -0.0308 -0.0596 0.0322

(0.0437) (0.0180) (0.0258) (0.0209) (0.0274) (0.0277) (0.0501) (0.0466) (0.0492) (0.0430)

r2 0.675 0.681 0.675 0.672 0.675 0.676 0.675 0.675 0.676 0.675

N 1800 1800 1800 1720 1800 1800 1800 1800 1800 1800

OLS regressions. Dependent variable is individual's second-period action. Standard errors clustered by subject. Other covariates not shown. [Var.] is as follows. Column 1: Dictator (0 or 1); Column 2: Subjective status index; Column3: Peer's subjective status index; Column 4: Income category (1-6); Column 5: Expected change in well-being (1-5); Column 6: Retrospective change in well-being (1-5); Column 7: Both parents have higher education (0 or 1); Column 8: has an older sibling (0 or 1); Column 9: Has a younger sibling (0 or 1);

Column 10: In only child (0 or 1)

* p < 0.10, ** p < 0.05, *** p < 0.01

Table 3

The effects of other covariates on second-period action.

Risk Lead Active Social Health Trust Civic Male Employed Sports

Priv. sig. X [Var.] -0.0999** 0.00447 0.0108 0.0108 0.00351 0.0403** 0.00562 -0.0171 -0.0235 0.00210

(0.0454) (0.00894) (0.0201) (0.0101) (0.00495) (0.0201) (0.00850) (0.0190) (0.0174) (0.0287)

Part. act. X [Var.] 0.346*** -0.0553** -0.0907* -0.0304 -0.00753 -0.112** -0.0307 0.00394 0.0143 -0.0531

(0.104) (0.0230) (0.0479) (0.0215) (0.0123) (0.0512) (0.0236) (0.0449) (0.0429) (0.0782)

r2 0.679 0.683 0.677 0.680 0.675 0.678 0.675 0.675 0.675 0.675

N 1800 1580 1800 1570 1800 1800 1720 1800 1800 1800

OLS regressions. Dependent variable is individual's second-period action. Standard errors clustered by subject. Other covariates not shown. [Var.] is as follows. Column 1: Fraction of safe choices on the risk task; Column 2: Leadership skills; Column 3: Active in a sports/environmental/professional organization, labor union, or political party (0 or 1); Column 3: Sociablity index; Column 5: Subjective health (1-10); Column 6: Interpersonal Trust (0 or 1); Column 7: Civicness index; Column 8: Male (0 or 1); Column 9: Employed part-time or

full-time (0 or 1); Column 10: Has a sports degree (0 or 1)

* p < 0.10, ** p < 0.05, *** p < 0.01

variables related to one' s social status.

Estimating model (1) while setting a2 = f>2 = 0 yields the coefficient a1* = 0.477(0.007), with standard error clustered by subjects. This was slightly less than a1 = 0.500 that was predicted for Bayesian rational individuals with no preferences for conformity.

We did not find that status in the dictator game, socioeconomic status, or subjective social status were associated with the weight of the private signal in the first-period decision. Other covariates were not significant as well. In Table B.9 we estimated Model (1) assuming that the effect of private signal on the first-period action is moderated by social status. In Column 1 we checked whether the weight of the private signal was different for dictators and confederates. In Columns 2 and 3, we used the own and peer's subjective social status indices. In Columns 4-9, we used various personal characteristics that may be associated with objective social status: income, expected and past changes in income, parental higher education, and whether the subject had older or younger siblings. None of the coefficients that we reported in the table are significant. In Table B.10 we used additional covariates: risk preferences, leadership skills index, whether one is active in civic or political organizations, socialization index, subjective health, interpersonal trust, civicness, gender, and whether the person is employed. Of all these variables, only trust and civicness were found to have an effect — people who believe that others can be trusted and people with less tolerance for rule-breaking put somewhat less weight on their private signals. These effects were not very large: the weight decreased by 0.0167 for each standard deviation increase in civicness, and were smaller by 0.0381 for those who believed that others could be trusted.

3.2. Second-period actions

We proceeded to analyze second-period decisions. As much as 98.2% of second-period choices were consistent with private information and lay in the Sj (%i) sets. The majority of the choices, 72%, were also located in the sets S4(%i,xj_j) = {min{z;,xj_j},...,max{z;,xj_j}}, between private signals and peer first-period action.

For the second-period action, the following reduced-form model was

estimated:

X2il = aZ + a2WiZil + b1X^i, + b2WtX^ it + c + c2Wt + e2iI (2)

Here, X2it is the second guess of individual i in round t.

Fig. 1 shows the coefficients for the private signal and the peer action when Model (2) was estimated for different groups of subjects, while setting a2 = b2 = c2 = 0. In the left graph, we show the coefficients of the models estimated separately for dictators and recipients. In the middle graph, the subjects are divided into four groups based on their subjective status index quartile. Finally, in the right graph, the subjects are divided into groups based on their risk preferences (the first group contains 94 subjects who made 5 or fewer safe choices in the lottery task11; the second group — 86 subjects who made more than 5 safe choices).

The weight of private information was higher for dictators than for recipients, while the weight of peer action was slightly lower; however, the differences were not statistically significant. At the same time, these weights were linked to subjective social status. The weights of the private signal and peer actions differed between first and fourth (p = .0010 and p = .0013, respectively), second and fourth (p = .0886 and p = .0143) status quartiles, and first and third quartiles (p = .0885 and p = .0295). Risk preferences also mattered: subjects who made over 5 safe choices in the risk aversion task put less weight on the private signal, and more weight on the peer action (p = .0101 and p = .0029).

In Table 2 we estimated Model (2) assuming that the effect of private signal and peer action on the second-period action is moderated by the individual's role in the dictator game, subjective social status, or objective social status. We reported only coefficients a2 and b2. In Table B.11 we repeated the estimation for the first round only.

We found that subjective social status was highly correlated with how an individual uses information to arrive at the second-period decision. An individual with a higher subjective status puts more weight on private information, and less weight on the first-period action of his or

11 A safe choice is one where the lottery offering 40 or 50 RUR is chosen.

Table 4

Determinants of second-period action.

1 2 3 4 5 6 7

Private signal x Status 0.0303*** 0.0269*** 0.0297*** 0.0273*** 0.0266***

(0.00819) (0.00817) (0.00837) (0.00835) (0.00779)

Partner's guess x Status -0.0669*** -0.0732*** -0.0611*** -0.0712*** -0.0750***

(0.0204) (0.0179) (0.0199) (0.0177) (0.0186)

Private signal x Leadership -0.0118 -0.0113

(0.00934) (0.00936)

Partner's guess x Leadership -0.0152 -0.0152

(0.0249) (0.0237)

Private signal x Active -0.0116 -0.0194

(0.0191) (0.0200)

Partner's guess x Active -0.0229 0.000933

(0.0483) (0.0498)

Private signal x Risk -0.120** -0.105** -0.111** -0.0965**

(0.0465) (0.0454) (0.0465) (0.0453)

Partner's guess x Risk 0.341*** 0.345*** 0.309*** 0.320***

(0.0969) (0.0984) (0.0954) (0.0968)

Private signal x Trust 0.0318 0.0349* 0.0310 0.0329*

(0.0211) (0.0199) (0.0210) (0.0196)

Partner's guess x Trust -0.0924* -0.0962* -0.0890* -0.0917**

(0.0532) (0.0489) (0.0498) (0.0456)

Private signal x Pos. affect -0.00178 -0.00314 -0.0102

(0.00880) (0.00831) (0.00786)

Partner's action x Pos. affect 0.00147 0.00450 0.0241

(0.0228) (0.0217) (0.0210)

Private signal x Neg. affect -0.0165* -0.0176** -0.0113

(0.00900) (0.00860) (0.00811)

Partner's action x Neg. affect 0.0412* 0.0430* 0.0260

(0.0233) (0.0222) (0.0214)

Private signal x Cognitive -0.0179 -0.0313 -0.0399** -0.0381**

(0.0190) (0.0190) (0.0193) (0.0186)

Partner's guess x Cognitive 0.123** 0.124** 0.164*** 0.151***

(0.0520) (0.0492) (0.0492) (0.0479)

r2 0.694 0.689 0.696 0.691 0.676 0.681 0.685

N 1580 1800 1580 1800 1800 1800 1800

OLS regressions. Dependent variable is individual's second-period action. Standard errors clustered by subject. Other covariates not shown.

* p < 0.10, ** p < 0.05, *** p < 0.01

her peer; this association was observed over all 10 rounds, as well as during the first round (Column 2, Tables 2 and B.11). The second effect is particularly large: a one standard deviation increase in subjective status will result in a 0.08-0.115 decrease in the coefficient. At the same time, social learning was not affected by the roles in the dictator game.12 It was not associated with most indicators of objective social status as well, and was not related to how the individual perceived the social status of his or her peer. The only significant effect was that of being the only child in the family — in the first round, the onlyborn put greater weight on private information (Column 10, Table B.11).

The association between other covariates and social learning is reported in Table 3, and in Table B.12 for the first round only.

We found that risk aversion was associated with more learning from the actions of peers, and less learning from private information in rounds 1-10 (Column 1, Table 3), but not in the first round.13

Leadership skills (which were defined as the frequency with which the individual took on various leadership roles, such as organizing

12 The mean amount donated in the dictator game was 22.7 (sd of 5-round mean 17.9). Subjective social status of dictators was slightly higher than that of recipients, but the difference was Dot significant (p = .5837, two-tailed t-test). For the recipients, the weights put on private and public information also did not depend on the average donation received in the dictator game.

13 Interaction coefficients for the weights of private and public information did

not change much (to -.082 (.045) and 0.298 (.105), respectively) if we only considered the 158 subjects (or 85.6% of the total amount) who did Dot switch to a less risky lottery on the risk aversion task.

events or speaking in public), or membership in civic or political organizations are associated with less learning from the actions of peers in rounds 1-10 (Columns 2 and 3, Table 3), but, once again, this association was not significant in the first round.

We are not aware of any literature directly claiming the link between risk preferences and attention to the actions of others (something that we detect in our study) and/or empathy. At the same time, risk aversion has been associated with lower socioeconomic status, particularly lower income and education (Donkers, Melenberg, & Van Soest, 2001; Guiso, Sapienza, & Zingales, 2018), behavioral inhibition (or tendency to demonstrate withdrawal in unfamiliar situations) and external locus of control (Beauchamp, Cesarini, & Johannesson, 2017), lower extraversion and higher neuroticism (Becker et al., 2012; Lonnqvist et al., 2015), and lower power (Anderson & Galinsky, 2006). Individuals exhibiting these traits are more highly attuned to how others think of them and evaluate their own actions (Keltner et al., 2003), while low power, status, and a diminished sense of personal control were associated with increased attention to others' thoughts and actions (Kraus et al., 2012). Leadership has also been related to self-confidence (McCormick, 2001), personality, particularly to extraversion and, negatively, to neuroticism (Bono & Judge, 2004), and disinhibition (Keltner et al., 2003).

If anything, we found that various components of social capital are negatively associated with social learning. In rounds 1-10, interpersonal trust was positively correlated with learning from private information, and negatively correlated with learning from publicly observed actions of others (Column 6, Table 3), while lower tolerance for rule-breaking behavior was found to be negatively associated with the weight put on social information in the first round (Column 7, Table B.12). It may seem

counter-intuitive that more trusting people are less reliant on social information and are more likely to learn from the actions of others. However, it is consistent with the hypothesis that social learning is driven by subjective status, as trust implies willingness to accept vulnerability (Hong & Bohnet, 2007), and trusting behavior has been found to be positively correlated with both income and social status (Brehm and Rahn, 1997; Delhey and Newton, 2003; Subramanian et al., 2003). In the first round (but not in all 10 rounds) individuals who were more outgoing, spent more time socializing, and had more social contacts, put greater weight on private signals and less weight on social information (Column 4, Table B.12). Subjects with better self-reported health put more weight on private information in the first round (Column 5, Table B.12).14

In Columns 1 and 2 of Table 4 we combined the significant variables from Tables 2 and 3 (in Table B.13 we report correlations between the explanatory variables), keeping either the leadership index or activity in civic or political organizations. We found that the coefficients produced using subjective status, risk aversion, and interpersonal trust all retained their significance. At the same time, neither the leadership skills nor membership in civic/political groups were any longer associated with learning from private information or from peer action. All coefficients retained their significance when in Columns 3 and 4 we estimated the same models while controlling for cognitive reflection.

In Columns 3 and 4 we repeated the estimation while controlling for cognitive reflection, which was associated with more learning from the actions of peers. This is not surprising, as the individuals (even ones with low subjective social status and high risk aversion) put much less weight on the actions of their peers than Bayesian individuals would. Cognitive reflection was found to correlate with rational behavior in other works. In a social learning game subjects with higher CRT scores chose optimal information more frequently (Duffy, Hopkins, Kornienko, & Ma, 2019); in a repeated Prisoneres Dilemma game individuals with lower cognitive skills played the less profitable defection strategy more often (Cason & Mui, 2019) and exhibited higher error rates (Proto et al., 2020); individuals with lower cognitive skills were less willing to take risks (Dohmen et al., 2010). In Column 5 we looked at the correlation between the person' s emotional state and social learning. We found that the subjects who reported experiencing more negative emotions also put greater weight on social information, and less weight on private information. This relationship was robust when controlling for cognitive refection (Column 6), but disappeared if we also accounted for subjective social status (Column 7). Negative mood can signal new or challenging situations that call for less reliance on preexisting knowledge and greater attendance to social cues, and result in more accurate social

14 Introducing dictator game status in models 2-10 of Table 3 or in models 1-10 in Table B.12 will not yield significant effects on social learning, while the existing coefficients retain statistical significance.

judgements (Forgas, 2013). At the same time, negative affect was positively correlated with neuroticism and negatively — with extroversion (Watson et al., 1999), while both of the latter two traits were related to leadership (Judge, Bono, Ilies, & Gerhardt, 2002) and social status.

In Table B.14 we looked at how social learning is related to both personality traits and life experiences. We found that leadership (Column 1) and participation in organizations (Column 2) were associated with less learning from partner actions, while negative affect was associated with less use of private information, and greater learning from partner actions.15The effects of life experiences and negative affect disappeared when controlling for subjective status (Columns 3 and 4). When controlling for cognitive reflection (Coluns 5 and 6), the effect of participation on organizations was also not present (Column 6). When controlling for either subjective status or cognitive reflection, positive affect was negatively correlated with social learning (Columns 3 and 5); we attribute it to the fact that positive mood and leadership were positively correlated.

We proceeded to test whether the associations that we found between status, other covariates, and social learning, are gender-specific (recall that subjects were not aware of the peers' gender). In Tables B.15 and B.16 we repeat the regressions made in Tables 2 and 3, introducing the full set of interaction terms with gender. We report only the triple interaction terms between gender, private signal/observed action, and the variable of interest. The negative association between the reliance on private signal and risk aversion that we reported previously was manifested in males to a greater degree than in females — in fact, it was present in males only (this followed from estimating the model separately for males and females, which is not reported in this table); gender differences in risk preferences were found in a number of studies (see Croson and Gneezy, 2009, for review of relevant literature). Parental education was one variable related to objective social status where the association with social learning was gender-specific. While there was no effect for males, in females, having both parents with higher education was associated with more learning from the observed action of the peer, and less weight put on the private signal.

Finally, we examine whether our findings hold for subsets of subjects that may be more representative of Russian population at large (recall that our subjects were 180 young students from a selective university). In Table B.17 we repeat the estimation of Table 2 for the 88 subjects who work part or full-time. Individuals with a stable socio-economic position may behave differently from university students, many of whom do not have a stable occupation. Still, we find that our main result holds — subjective social status is negatively associated with the weight put on the peer's decision, and is also positively associated with the weight put on private information. Moreover, expected decrease in income is associated with less weight put on private information and more weight

15 We realize the limitations of that finding, because it is difficult to disentangle the two effects. On the one hand, personality traits have been linked to leadership traits (Hofmann and Jones, 2005; Judge and Bono, 2000), as well as labor market outcomes (Lee & Ohtake, 2012; Wiersma & Kappe, 2017), career decisions such as self-employment (Hamilton et al., 2019), and other outcomes such as the feeling of control over life circumstances (Prevoo & ter Weel, 2015). On the other hand, personality traits in childhood and young adulthood are malleable and affected by parental investment (Borghans, Duckworth, Heck-man, & Ter Weel, 2008), socioeconomic status in childhood Almas, Cappelen, Salvanes, S0rensen, and Tungodden (2016), and group status (Dasgupta et al., 2017, find that students marginally admitted to more selective colleges and thus have a smaller class rank than their peers exhibit lower extraversion and conscientiousness). Other traits are such as risk aversion can be shaped by traumatic events like wars or economic crises (Malmendier & Nagel, 2011; Kim & Lee, 2014). The returns to non-cognitive traits may also depend on socio-economic context such as parental education (Lundberg, 2013). For future research, it would be very interesting to evaluate the social learning effects of random life experiences that occur independent of personality.

период 1/1

Вы - распределитель. Какую сумму от 0 до 100 УЕ вы хотите передать получателю

Вы - распределитель.

Fig. A1. Dictator game, Allocator's decision.

put on peer's action. Remarkably, this effect is present only in subjects who work full or part-time, which shows that own income is a substantial component of the social status.

In Table B.18 we repeat the estimation for the 59 subjects whose mother or father does not have higher education (in the full sample, as much as 67% of subjects have both parents with higher education, meaning that the subjects were drawn from an upper tail of status distribution). We find that, despite of the smaller sample size, subjective social status is still negatively associated with the weight put on peer' s decision. We also find that subjects who have younger siblings put more weight on private information and less weight on peer action. However, to our knowledge, there are no relevant works that could explain such effect by subjects whose parents (at least, one of them) do not have higher education.16

4. Discussion

In this paper we investigated whether an individual' s social status correlates with how likely is he or she to use social or private information to arrive at a decision. We found that high subjective social status is associated with less weight placed on the observed actions of peers and more weight on private information, although the roles in the dictator game did not affect social learning. On the other hand, the components of one's socioeconomic status — in particular, income and parental education — were not correlated with the use of private/social information. A possible factor that can mediate between objective and subjective social status is relative deprivation that is known to correlate with both low status and less prosocial orientation (Callan, Kim, Gheorghiu, & Matthews, 2017). The relevance of different aspects of social status can also depend on the country and cultural context (Park et al., 2013); replicating our results in different countries is a matter of future research. It would also be interesting to replicate our results on a

16 On the contrary, children with siblings may compete for limited material resources of the family (see, for instance, Chen & Liu (2014)).

sample of people from different social stratas.

We found that several other individual characteristics were associated with more weight put on private information, and less weight on social information: greater tolerance for risk taking, leadership skills, interpersonal trust, and sociability. All of these findings are consistent with the theory that low-status individuals are likely to be more vigilant to threats, have lower personal sense of control, and be more attuned to others (Kraus et al., 2012). These results were robust when controlling for cognitive reflection; the latter was associated with a greater weight on social information, putting it closer to the equilibrium value. The effect of risk aversion on social learning was greater in men (previously, women were found to be more risk-averse than men (Croson & Gneezy, 2009)).

Different theoretical frameworks can be used to interpret our results. Individuals may not hold their peers to be fully rational, leading them to underweight social information (Cornand & Heinemann, 2014; Shapiro et al., 2014; Weizsäcker, 2010). Social status can then be related to the level of reasoning in the sense of the level-k model (Stahl and Wilson, 1995). Individuals of high subjective social status may give less consideration to their peers, considering them to be less rational, and underweight their actions as a result. Low-status subjects may have preferences for conformity, e.g. their guesses being closer to the guesses of their peers. Another explanation can be provided by the theory developed by Keltner et al. (2003). They showed that high-power individuals exhibit approach-related motivation and positive affect while low-power ones show inhibition-related tendencies and negative effect. Each of these processes was associated with certain brain activity (Boksem, Smolders, & Cremer, 2012). In either case, finding the underlying rationale for our results is the matter for future research.

Our results have implications for dissemination/aggregation of information in networks. Individuals with high social status are also likely to be the ones who have many followers. The fact that high-status individuals are less likely to react to the signals of others can make information cascades and herding more likely, possibly decreasing social welfare.

16 On the contrary, children with siblings may compete for limited material resources of the family (see, for instance, Chen & Liu (2014)).

период 1/1

Вы - получатель. Подождите, пока распределитель решает какую сумму вам передать.

Вы - получатель.

Fig. A2. Dictator game, Receiver's screen.

Fig. A3. Dictator game, end of round.

Declaration of Competing Interest

None.

Appendix A. Experiment design and procedures

Dictator game, Allocator's screen. You are the Allocator. Which amount from 0 to 100 ECU would you allocate to the Receiver? You are the Allocator (Figure A1).

Dictator game, Receiver's screen. You are the Receiver. Please wait while the Allocator decides which sum to pass to you. You are the Receiver. (Figure A2).

Dictator game, end of round. You are the Allocator. Your payoff is 90 ECU. You are the Allocator. (Figure A3).

период 1/1 часть 1

Компьютер загадал числа X, У1 и У2 Вам известно, что сумма чисел X и У1 равна 2. Вам предоставляется попытка угадать значение Х- Скажите, чему оно равно?

Г-7 Г-6 Г-5 Г -4 Г-З Г -2 Г -1 ГО Г 1 Г 2 Г 3 Г 4 Г 5 Г6 Г 7

Вы - получатель.

Fig. A4. Main part, first guess attempt.

Период 1/1. часть 2.

Второму игроку была известна сумма чисел X и У2. Он пробовал угадать число X и предположил, что оно равно 0.

Напомним, что сумма X и У1 равна 2.

При первой попытке Вы предположили, что число X равно 1

Вам предоставляется вторая попытка угадать значение X. Скажите, чему оно равно?

Г -7 Г-6 Г -5 Г -4 Г-З Г -2 Г-1 ГО Г 1 Г 2 ГЗ Г4 Г 5 Г 6 Г 7

Вы - получатель

Fig. A5. Main part, second guess attempt.

Main part: first guess attempt. Computer randomly assigned numbers X, Y1 and Y2. You know that the sum of X and Y1 equals 2. You've got the first attempt to guess the value of X. Choose one of the options below, please. You are the Receiver (Figure A4).

Main part: second guess attempt The second player knew the sum of X and Y2. He tried to guess X and assumed that it was 0. We remind you that the sum of X and Y1 equals 2. You've got the second attempt to guess the value of X. Choose one of the options below, please. You are the Receiver (Figure A5).

Main part: end of round The true value of X equals -3. At first attempt you assumed that X equals 1. At the second attempt you assumed that X equals 4. At the second attempt the other player named number -4. Your payoff: 30.0 ECU. You are the Receiver (Figure A6).

Status questionnaire Please fill in a short questionnaire. Which of the following best describes [you/the person you interacted with in the previous game]?

Dominant — Subordinate

Unconfident — Confident

High status — Low status

Fig. A6. Main part, end of round.

Fig. A7. Subjective status questionnaire, own.

Leader — Follower

Controls resources — Does not control resources Dependent — Independent Passive — Active

In our society there are people who occupy higher social positions and people who occupy lower social positions. Please state where [you/the person you interacted with in the previous game] stand on the ladder of 10 steps where 1 is the lowest step and 10 is the highest step (Figure A7).

Risk aversion task: intro screen Please make ten decisions that you will see on the screen. One of them will affect your payoff at this stage of the experiment. Press OK as soon as you are ready (Figure A8). Cognitive reflection test

1. A bat and a ball cost $ 1.10 in total. The bat costs $ 1.00 more than the ball. How much does the ball cost?

2. If it takes 5 machines 5 minutes to make 5 widgets, how long would it take 100 machines to make 100 widgets?

Пожалуйста, примите десять предлагаемых вам решений.

Одно из них повлияет на ваш выигрыш в этой части эксперимента. Нажмите ОК как только вы готовы.

Вы - получатель

Fig. A8. Risk aversion task, intra screen.