Reversal Effects in Charitable Behaviors across Cultures:The Role of Sentence Voice

Incom e per capita was previously found to be sometimes negatively, and sometimes positively, associated with helping across cultures. These opposite effects may result from a type of reversal of helping across the perceived contexts of cultures. The possibility in a language to highlight actors’ responsibility in events through antipassive sentences (cf. “I will help”) predicted higher, and the possibility to decrease the responsibility through passive sentences (“The needy will be helped”) predicted lower levels of charitable behaviors in countries with lower levels of power distance and uncertainty avoidance, and higher levels of income, a set of societal characteristics, which were previously found to positively impact helping. The results highlight the role of sentence voice presumably linked with perceived autonomy in contexts of helping across cultures, ultimately reversing the cultural effects on charitable behaviors.


Highlights:
• Passive voice in languages predicts decreases in countries' charitable behaviors.• Income predicts less charity in societies of languages with passives.
• Power distance predicts more charity in countries of languages with passives.
• Uncertainty avoidance predicts more charity in cultures of languages with passives.
The wealth of a country may not always predict charitable donations of its citizens as one of the poorest countries in the world, Myanmar, with a yearly income of about 1,244 USD per capita occupied the top rank in 2014 with 92% of its people donating money to a charity (Charities Aid Foundation, 2015).In fact, as the number of wealthy people increases in a country, charitable behaviors Corresponding author: ibrahimsenay@sehir.edu.tr may sometimes decrease (Levine, Norenzayan, & Philbrick, 2001).The fact that the per capita income in a society sometimes predicts increased money donations (Aknin, Barrington-Leigh, Dunn, Helliwell, Burns, Biswas-Diener et al., 2013;Smith, 2015), however, begs the question why and how society-level variations may sometimes have opposite effects on helping.

Charitable Behaviors and Perceived Autonomy
Helping actors' perceived autonomy may play an important role in the reversal of helping across different types of contexts.For this reason, it can be used to shed light on why income per capita and possibly other society-level variations may reduce motivation for helping in different times and contexts.According to the theory of psychological reactance when individuals' freedom to choose and control their own behavior is threatened, they experience a state of psychological reactance or reactance arousal.As a result, they get motivated to act in the opposite direction from how they would originally behave (Brehnm & Brehm, 1981).In a series of studies on helping behaviors reviewed by Brehm and Brehm (1981), the initial motivation for helping was found to reverse when the needy is highly dependent on the agents of helping, responsible for the need and likely to continue making future requests of help.All these contexts in which the reversal effects are observed can be characterized as having a high level of threat to the freedom of helping.Specifically, people who are otherwise motivated to help others become relatively less likely to help in contexts in which they experience threats to their freedom of choosing whether or not to help.Although people in more individualistic cultures are more likely to experience psychological reactance as compared to members of more collectivistic cultures (Jonas, Graupmann, Niesta-Kayser, Zanna, Traut-Mattausch, & Frey, 2009), psychological reactance has been shown to be valid and relevant for explaining behaviors and choices in collectivistic cultures (see e.g., Quick & Kim, 2009).As such, it can be potentially helpful in shedding light on cross-cultural patterns of behaviors.
Presupposing a role for threats to perceived autonomy underlying helping may help us better understand the role of wealth in predicting charitable behaviors across societies.Increased helping predicted by increased income or purchasing power in a society was observed in contexts of helping that involved money spending (Aknin et al., 2013;Smith, 2015), while decreased helping linked with this variable was found to be present in contexts that did not involve money spending (Levine, Martinez, Brase, & Sorenson, 1994;Levine et al., 2001).Money and the implied thoughts of power and control can decrease psychological reactance (Sittenthaler, Jonas, & Steindl, 2011).In one study, restricting people's choices increased their heart rates as an index of psychological reactance.However, this effect was observed only when participants previously counted pieces of paper as opposed to money, showing that being exposed to money cues decreased the feelings of threat.Monetary cues and the associated thoughts of power and control were found to increase the general self-efficacy and control (Mukherjee, Manjaly, & Nargundkar, 2013; for a review see e.g., Vohs, 2015).In another study these same variables were found to be negatively associated with standard self-report measures of psychological reactance (Sittenthaler, Traut-Mattausch, Steindl, & Jonas, 2015).Although money donations do not necessarily involve exposure to money cues as one can engage, for example, in such acts with credit cards, which were found to be relatively less effective in creating a feeling of power and control (for a review see e.g., Vohs, 2015), charity in the form of money donations can still be thought to involve exposure to money cues to a greater extent than the other types of helping contexts such as those involving volunteering time and helping a stranger.Therefore, it could be the case that income is positively associated with helping in contexts of low psychological reactance, such as those involving money spending, while predicting relatively decreased levels of helping in other contexts.
Although the effects of income seem to suggest the presence of threats to perceived autonomy in contexts of helping across cultures, a systematic investigation of such effects has been lacking.Language is discussed below as a possible tool that can be linked with the possible role of perceived autonomy for helping across societies.

Language and Perceived Autonomy
Subtle differences in the way in which verbal messages are framed seem to be important in changing perceived autonomy, hence possibly the experience of psychological reactance in contexts of helping.In one study, feelings of participants about helping were verbally represented in slightly different forms (Goodstadt, 1971).The experimenter made either no statement about the participants' feelings for helping, or a mildly controlling statement such as "Apparently you like helping" or a strongly controlling statement such as "It is very clear you like helping".In the no-statement condition, the participant helped the experimenter to the extent the two shared similar attitudes or the experimenter was nice to the participant.However, in the statement conditions the effect was reversed such that helping was greater when the experimenter and the participant held different attitudes or the experimenter was cold to the participant.The mild statement condition was more similar to the no-statement condition than the strong-statement condition.
If language can play a role in changing perceived autonomy, then the features of languages spoken in different cultures may be linked with different levels of psychological reactance likely resulting from changes in the way in which responsibility is represented in these languages.Charitable behaviors across cultures, as a result, can be studied as a function of language and culture to reveal a possible effect of freedom threats underlying charitable behaviors across different cultures.One linguistic feature that shows sufficient variation across different cultures and, hence, may be consequential for charitable behaviors, is the active vs. the passive sentence voice.(e.g., I help the needy vs.The needy is helped).Among 373 languages of the world linguistically analyzed (Siewierska, 2013), about 44% differentiate active from passive sentences.Importantly, this cross-linguistic variation may impact how people in different societies will potentially represent the role of various entities responsible for changing the unfortunate situations of victims.People describe a rape incident in a passive sentence (A is raped by B) if they hold stereotypical beliefs that favor blaming the victims of rape, the acted-upons, as opposed to the assailants (Bohner, 2001).The actor of a passive sentence such as X in the sentence Y is helped by X is generally rated to have less agency/responsibility than the acted-upon of this sentence, which is Y (Harris, 1978;Johnson, 1967).The helping behavior, thus, can be described in a passive sentence (The needy is helped) to suggest that helping is relatively more under the control of the needy, creating a type of freedom threat and psychological reactance for the actors of helping.Contexts in which the needy is perceived to be controlling helping acts such as seeing the needy to be responsible for the need, to be likely to make possible future demands of help and to be highly dependent on the help were all identified as being among the factors in creating freedom threats to the actors of helping (Brehm & Brehm, 1981).It can, therefore, be argued that in contexts of low psychological reactance, such as when the active sentences are more likely to be used in a language to describe the responsibility of actors in events involving helping, cross-cultural attitudes predicting increased helping can bring about an increase in helping behaviors as expected.However, when the perceived autonomy to help is low, hence it may increase psychological reactance, such as when passive sentences are used to describe events in a language, the same cultural factors may motivate a decrease, rather than an increase, in helping.

The Present Study
It may be difficult to determine the role of the active as different from the passive sentences by simply comparing the cultures of languages that have vs. those that do not have passive sentences.That is because when passive sentences are present in a language, active sentences are, by default, co-present with them.However, other types of sentences that exist in some languages other than English such as antipassive sentences can help us tell apart the role of the active from that of the passive sentences.As the complete opposite of passive sentences, antipassive sentences can be used as a proxy for the relative dominance of the active-over the passive-voice mindset in describing events across languages.The meaning change created by antipassives is likened by Polinsky (2013) to that of the meaning distinction in English between sentences like I shot the bear vs.I shot at the bear.In the latter sentence it is not clear whether the bear is affected by the action at all, thus, in such alterations the cause of the action is more likely to be the attempts of the sentence subject rather than the condition of the sentence object.Polinsky gives a specific example from Ukranian as an antipassive language (Polinsky, 2017).The active sentence Did sxopyv ripk-u in Ukranian can be translated as Grandfather grabbed the turnip.The antipassive variant of this active sentence becomes Did sxopyv-sja za ripk-u where the sentence verb becomes intransitive and reflexive by means of the inflection -sja, hence, does not take any direct object.Instead the preposition za (at) becomes necessary to use in the sentence to preserve the logical meaning of the sentence object in the accusative case (ripk-u).The antipassive meaning becomes similar to that of the sentence Grandfather grabbed at the turnip.
In sum, antipassives as compared to passives can be seen as highlighting the role of actors or sentence subjects in reporting events by hiding the role of acted-upons or sentence objects.According to The World Atlas of Language Structures database (Dryer & Haspelmath, 2013) languages with antipassives are distributed over a diverse group of geographical locations ranging from Guatemala (3 out of 4 languages analyzed from that country) to Australia (9 out of 17 languages) to Russia (9 out of 19) and to USA (9 out of 28).Among the countries that do not have antipassives among the analyzed languages are China (0 out of 10) and Papua New Guinea (0 out of 14).In sum, the presence of passive sentences can be analyzed together with the presence of antipassive sentences in a language to see how the features of languages may differentially impact charitable behaviors.
Passive sentences in a language may potentially facilitate speakers' experience of psychological reactance in contexts of helping.In passive sentences (e.g., The needy will be helped) the act of helping is perceived to be under the control of the acted-upon or the needy to a relatively greater extent.This restricted control over helping, in turn, may lead actors of helping to experience psychological reactance (Brehm & Brehm, 1981).One way to measure whether the presence of passive sentences in a language may take the control of helping away from the actors, the actors' perceived personal responsibility to help may be compared across different country languages.Perceived personal responsibility increases actual helping (Bandura, Barbaranelli, Caprara, & Pastorelli, 1996;Darley & Latane, 1968;Krebs, 1970).Therefore, the presence of passive sentences in a language is expected to predict decreased helping in countries where this language is spoken.Similarly, the presence of antipassives can be hypothesized to predict increases in helping behaviors: 1a) The presence of passives in a language should predict relative decreases in charitable behaviors among the members of cultures speaking this language; 1b) The presence of antipassives in a language should predict relative increases in charitable behaviors among the members of cultures speaking this language.
To specifically show whether a reactance experience underlie the effects of society-level variations on helping, the known effects of cultural dimensions on helping may be tested to see if these effects of cultural factors will change direction as a result of the presence of passive vs. antipassive sentences in a language.Power distance, a cultural dimension described as the degree to which the unequal distribution of power in a society is accepted (Hofstede, Hofstede, & Minkov, 2010), was shown to decrease the perceived personal responsibility to help others as well as the actual helping in the form of charitable giving (Winterich & Zhang, 2014).Long term orientation (Winterich & Zhang, 2014) and uncertainty avoidance (Smith, 2015) as cultural dimensions were also reported to be associated with decreases in charitable behaviors.It is, thus, possible that the negative effects of all these three cultural dimensions (power distance, long term orientation, uncertainty avoidance) will be observed in contexts of relatively smaller freedom threats such as when societies speak languages that have antipasssives.However, the negative effects of these cultural dimensions should get reversed in societies of languages with passive sentences due likely to a psychological reactance experience.In addition, as discussed above, income seems to predict increased helping in contexts that are relatively less likely to create psychological reactance such as when helping involves money spending.The positive effect of income on helping should, thus, be observable in contexts of relatively smaller freedom threats such as in societies of languages with antipassives.The positive effect of country wealth should, thus, get reversed in societies of languages that include passive sentences.Overall, the following hypotheses can be proposed: 2a) Power distance, long-term orientation and uncertainty avoidance should predict decreases while income should predict increases in charitable behaviors in cultures of languages that have antipassive sentences; 2b) Power distance, long-term orientation and uncertainty avoidance should predict increases while income should predict decreases in charitable behaviors in cultures of languages that have passive sentences.
Other major cultural dimensions such as individualism-collectivism and masculinity-femininity may not affect helping in a way that can be easily characterized by a psychological reactance motive.Both individualism (Kemmelmeier, Jambor, & Letner, 2006) and collectivism (Moorman & Blakely, 1994;Lampridis & Papastylianou, 2014) can facilitate helping (Miller & Bersoff, 1994;Mullen Skitka, 2009).Similarly, both masculinity and femininity can increase the responsibility to help others under different contexts and among different gender groups (Nelson, Brunel, Supphellen, & Manchanda, 2006).It is therefore difficult to associate a particular increase or decrease in helping with either ends of these cultural dimensions, hence, make a prediction about how an effect can get reversed as a result of a reactance experience.
Finally, as discussed above, helping contexts that involve money donations decrease the experience of psychological reactance (Mukherjee et al., 2013;Sittenthaler et al., 2011, as cited in Mühlberger et al., 2015).Because antipassives are hypothesized to predict decreases in psychological reactance, the effect of antipassives should be relatively stronger on charitable behaviors that involve money spending.Reversely, the effect of passives should be weaker on charitable behaviors that involve money donations: 3) The effects specified in prediction 2a should be stronger, or the effects specified in prediction 2b should be weaker for, charitable behaviors that involve money donations as compared to the other types of charity.
Because all predictions above are about different societies where the objective conditions of the need for help as well as the potential to help differ, human inequality coefficient, which is based on the inequality in life expectancy/ health, education attainment and income, and income per capita were used as two control covariates in testing all the study predictions.

Method Measures
The World Atlas of Language Structures database (Dryer & Haspelmath, 2013) was used to characterize world languages.This database categorizes all languages of the world based on published linguistic evidence according to whether they have passive sentences (Siewierska, 2013) or antipassive sentences (Polinsky, 2013) as well as whether they may have a range of many other linguistic features.To locate countries where each language identified by its ISO 639-3 code is spoken (About the Ethnologue, n.d.), a comprehensive reference work cataloguing the world's known living languages, Ethnologue: Languages of the World, was used.Next, Hofstede's country data (Hofstede, Hofstede, & Minkov, 2010) were used to identify where a country lies on each of the following cultural dimensions: individualism (i.e., the degree of independence maintained among the members of a society), masculinity (i.e., wanting to be the best vs. liking what one does), uncertainty avoidance (i.e., the degree of being threatened by unknown situations), power distance (i.e., the degree of accepting the unequal distribution of power in a society), long-term orientation (i.e., changing vs. keeping cultural traditions) and indulgence (i.e., having low vs. high control over one's impulses and desires).
The data for the World Giving Index (WGI) and/or income per capita were missing for different countries in different years.To minimize the amount of missing data, the last three years of the available data were averaged.The latest income per capita data were available for the year 2015 (UN Statistics Division, 2016).Therefore, the income per capita, World Giving Index (Charities Aid Foundation, 2014, 2015, 2016), as well as human inequality coefficient (UN Human Development Report, 2016) were averaged over the years of 2013 through 2015.Because of the highly skewed nature of the income data (z = 30.82,p < .00001), the raw income data were replaced with the natural logarithm of income per capita.In the World Giving Index report, 140 countries were given an index calculated based on the information from more than 150,000 participants aged 15 and older, who were recruited through random sampling techniques for each of the years of 2013, 2014, and 2015.World Giving Index is a continuous variable, which is based on the average of three country percentages concerning the charitable behaviors of the citizens of a country, which took place in the month prior to the survey: the percentage of people who reported (1) donating money to a charity, (2) volunteering time to an organization, and (3) helping a stranger.However, because of over-or under-reporting of certain components of WGI across different countries, the standardized z scores of each of the three components of WGI was first calculated and then averaged to be used as the standardized overall WGI for a country.The human inequality coefficient (UN Human Development Report, 2016) was based on the inequality in life expectancy, educational attainment and income in countries and was used as a covariate to control for the objective level of the need for help in a country.

Results
Because countries in which a particular language is spoken may be similar to one another, mixed linear models were used to evaluate the effects of languages and cultures on charitable behaviors.The assumption that countries speaking a particular language are correlated with a compound symmetry covariance matrix produced a better model fit in all analyses than the assumption that countries speaking a particular language were uncorrelated (average of fit indices = 2989 vs. 3047 in a smaller-is-better form for the compound symmetry and the scaled identity model, respectively).Therefore, the compound symmetry model assuming that countries speaking a particular language are correlated is used throughout the analyses of study predictions.Satterthwaite approximation was used to calculate the non-integer, denominator degrees of freedom that are adjusted for the correlated scores of countries speaking a particular language.The standardized World Giving Index and its three subcomponents (i.e., the percentage of people (1) helping a stranger, (2) volunteering time, and (3) donating money) were used as a dependent variable in separate models.Including six cultural dimensions and income per capita in the same model or including more than one interaction term in the same model increased multi-collinearity beyond the acceptable threshold (VIF > 5).Therefore, the interaction between the presence of passives or antipassives and each cultural dimension as well as income was tested in a separate model.
First, a series of bivariate correlations among study variables were calculated.As presented in Table 1, the presence of antipassives in a language correlated positively and significantly with the World Giving Index while the presence of passives correlated negatively and significantly with it.The same pattern was present for each component of the World Giving Index.Antipassives only had a positive and significant correlation with income while showing no association with the three cultural factors of interest including power distance, uncertainty avoidance, and long-term orientation.The passives, on the other hand, did not correlate with income while showing a positive and significant association with uncertainty avoidance and long-term orientation.Passives and antipassives.The effects of passives and of antipassives were tested in separate models with the same control variables (per capita income and human inequality coefficient).In the model with the presence of antipassives, income per capita, and human inequality as predictors and world giving index (WGI) as the criterion, the presence of antipassives predicted marginal increases in WGI, b = .25,SE = .14,F(1, 170.18) = 3.065, p = .082,ω 2 = .02.In the model predicting money donations, antipassives had no effect, b = .75,SE = 3.19, F(1, 171.25) = .056,p > .8.The presence of antipassives predicted significant increases in the percentage of people helping a stranger and volunteering time after controlling for income and the human inequality in the country, b = 3.98, SE = 1.91,F(1, 184.57) = 4.337, p = .039,ω 2 < .01 and b = 4.15, SE = 1.90, F(1, 185.66) = 4.759, p = .030,ω 2 = .04,respectively.
Per capita income.In the model with the term of interaction between the presence of antipassives and income per capita, the interaction term predicted a marginal increase in WGI adjusted for the effects of the presence of antipassives alone, income per capita alone and human inequality coefficient, b = .15,SE = .09,F(1, 278.12) = 3.052, p = .082,ω 2 = .01.This pattern was marginally significant in predicting helping a stranger, b = 2.06, SE = 1.23,F(1, 2798.07)= 2.822, p = .094,ω 2 < .01.The effect was not significant in predicting money donations and volunteering time, b = 27.83,SE = 18.50, F(1, 279.27) = 2.456, p > .1, and b = 1.86,SE = 1.19,F(1, 287.22) = 2.434, p > .1,respectively.Thus, increased income can predict marginal increases in WGI and helping a stranger in countries of languages with antipassives.These results partially confirm prediction 2a about income in that income, as shown for the back-transformed values of WGI and income in Figure 1, can predict marginally significant increases in a country's WGI, as well as the number of people helping a stranger, specifically when people in this country speak a language that has antipassives.F(1, 522.12) = 9.135, p = .003,ω 2 = .02,respectively.These results show that the increased income predicts on average a 3% decrease in WGI, 4% decrease in the number of people helping a stranger and 2% decrease in the number of people volunteering time specifically in countries speaking a language with passives.These results confirm prediction 2b about income in showing that income, as shown in Figure 1 for the back-transformed values of WGI and income, predicts decreases in WGI as well as in the value of two of its components in countries having languages with passives.
These results largely confirm prediction 3 about income, which states that either the antipassives-income interaction should be stronger, or the passives-income interaction should be weaker, in predicting money donations as compared to the other types of charity.The anitpassives-income interaction was not stronger, but the passives-income interaction was significantly weaker as expected, in predicting money donations as compared to helping a stranger.Furthermore, the antipassives-income interaction was equally significant, but the passives-income interaction was weaker as expected, in predicting money donations as compared to volunteering time.
Power distance.The antipassives-power distance interaction predicted significant decreases in WGI after controlling for the effects of power distance alone, antipassives alone, income and human inequality, b = -.01,SE = .005,F(1, 249.51) = 6.334, p = .012,ω 2 = .02.The effect was present for money donations and helping a stranger, b = -.30,SE = .11,F(1, 246.64) = 7.792, p = .006,ω 2 = .03and b = -.20,SE = .08,F(1, 266.26) = 6.634, p = .011,ω 2 = .02,respectively.In other words, power distance predicted average decreases in WGI by 2%, in the number of people donating money to a charity by 3% and in the number of people helping a stranger by 2% specifically in countries of languages with antipassives.These results confirm prediction 2a about power distance in that power distance, as shown in Figure 2 for backtransformed values of WGI, predicts significant decreases in WGI as well as in the value of two of its components specifically in countries of languages with antipassives.The term of interaction between the presence of passives and power distance predicted significant increases in WGI after controlling for the effects of passives alone, power distance alone, income and human inequality, b = .01,SE = .004,F(1, 416.34) = 10.073,p = .002,ω 2 = .02.This effect was present for helping a stranger, b = .31,SE = .06,F(1, 416.87) = 31.213,p < .001,ω 2 = .07.The effect was not present for money donations and volunteering time, b = .12,SE = .09,F(1, 415.59) = 2.100, p > .1 and b = .04,SE = .05,F(1, 416.21) = .653,p > .4,respectively.These findings show that the increased power distance predicts average increases in WGI by 2% and in the number of people helping a stranger by 7% specifically in countries of languages with passives.These results generally support prediction 2b in that power distance, as shown in Figure 2, can predict increases in the overall charitable behaviors in countries of languages with passives.
These results confirm prediction 3 about power distance, which states that either the antipassives-power distance interaction should be stronger or the passives-power distance interaction should be weaker in predicting money donations than in predicting the other types of charity.The antipassivespower distance interaction was equally significant, but the passives-power distance interaction was weaker as expected, in predicting money donations as compared to helping a stranger.In addition, the antipassives-power distance interaction was equally significant, but the passives-power distance interaction was weaker as expected, in predicting money donations as compared to volunteering time.
These results partially support prediction 3 according to which either the antipassives-uncertainty avoidance interaction should be stronger or the passives-uncertainty interaction should be weaker in predicting money donations as compared to the other types of charity.The antipassives-uncertainty avoidance interaction was stronger as expected, while the passives-uncertainty avoidance interaction was equally significant, in predicting money donations as compared to volunteering time.However, the passives-uncertainty avoidance interaction was stronger as contrary to prediction 3, while antipassives-uncertainty avoidance interaction was equally significant in predicting money donations as compared to helping a stranger.In sum, for uncertainty avoidance prediction 3 was only confirmed when money donations were compared to volunteering time.
The results partially support prediction 3 about LTO, which requires that either the antipassives-LTO interaction to be stronger or the passives-LTO interaction to be weaker in predicting money donation than in predicting the other types of charitable behaviors.The antipassives-LTO interaction was stronger as expected, but the passives-LTO interaction was neither weaker nor in the expected direction, in predicting money donations as compared to the other types of charitable behaviors.
Other effects.The interaction between individualism and the presence of antipassives did not predict changes in WGI after controlling for the effects of individualism alone, the presence of antipassives, income and human inequality, b = .01,SE = .01,F(1, 214.94) = 2.020, p > .1.Furthermore, the interaction between individualism and the presence of passives did not predict a significant change in WGI after controlling for the effects of individualism, the presence of passives, income and human inequality coefficient, b < .001,SE = .003,F(1, 395.77) = .002,p > .9.Additionally, the antipassives-indulgence interaction and the passives-indulgence interaction were not significant in predicting WGI adjusted for the appropriate control variables, b = .01,SE = .01,F(1, 235.12) = 1.155, p > .2, and b = .004,SE = .004,F(1, 291.03) = 1.009, p > .3,respectively.

Discussion
Overall, the results largely support the study hypotheses.After controlling for income and human inequality in a country, the presence of antipassives in a language predicted marginal increases in the World Giving Index as well as significant increases in two of its components in countries speaking this language (prediction 1a).The passives, on the other hand, predicted significant decreases in the World Giving Index as well as in the values of all of its components except for helping a stranger, for which there was a marginally significant decrease (prediction 1b).Previous studies have shown that the more frequent a linguistic feature is in a language such as certain types of verbs, the more often the speakers of this language will use these as opposed to the other similar types of verbs to describe actions or events (Gennari, Sloman, Malt, & Fitch, 2002;Slobin, 2003).Thus, the presence of passives vs. antipassives in a language as measured in the present study is likely to relate to the use of these constructions to describe events among people speaking this language.The use of passive sentences can function to decrease the perceived control of actors and increase the perceived control of acted-upons in events (e.g., The needy will be helped) (Bohner, 2001;Harris, 1978;Johnson, 1967).The use of antipassive sentences (e.g., I will help), on the other hand, can reverse this pattern (Polinsky, 2013).Because the personal responsibility to help others leads to actual helping (Bandura et al., 1996;Darley & Latane, 1968;Krebs, 1970), the presence of antipassive vs. passive sentences in a language of a society is likely to predict the personal responsibility to help others.
The presence of passives and antipassives further interacted with cultural dimensions including power distance and uncertainty avoidance which were previously found to predict decreases in charitable behaviors across different studies (Smith, 2015;Winterich & Zhang, 2014).Power distance predicted decreases in the World Giving Index as well as in the value of two of its components (money donations and helping a stranger) specifically in cultures with languages that had antipassives while predicting increases in world giving index specifically in cultures of languages that had passives.Similarly, uncertainty avoidance predicted decreases in world giving index and in the value of two of its components (money donations and helping a stranger) specifically in cultures of languages with antipassives, while predicting increases in world giving index as well as in the value of two of its components (money donation and volunteering time) specifically in cultures of languages with passives.These results are in line with prediction 2a and 2b in showing that in cultures of languages with antipassives no reversal in default cultural tendencies to engage in charitable behaviors was observed whereas in cultures of languages with passives there was such a reversal.
Furthermore, either the antipassive-power distance interaction was stronger, or the passives-power distance interaction was weaker, in predicting money donations as compared to the other types of charity (prediciton 3).This effect was also partially present in the interactions of language with uncertainty avoidance.Exposure to money cues in counting money decreases the reactance experience by increasing the feelings of power and control (Sittenthaler et al., 2011).Similarly, exposure to money cues increases self-efficacy (Mukherjee et al., 2013; for a comprehensive review se e.g., Vohs, 2015), which is shown to decrease the reactance experience (Sittenthaler, Traut-Mattausch, Steindl, & Jonas, 2015).It is therefore, plausible to think that money donations differ from the other type of charity in predicting a relatively greater level of reactance experience.
If money donations may involve relatively lower levels of reactance, then, this can help us understand why income is sometimes associated with the decreased helping in some studies (Levine, Martinez, Brase, & Sorenson, 1994;Levine et al., 2001) while correlating positively with prosocial spending and charitable behaviors in others (Aknin et al., 2013;Smith, 2015).The positive association between income and helping was observed in contexts involving money donations while the negative association was observed in the other types of contexts.In the present study, income predicted marginally significant increases in charitable giving in societies of languages that had antipassives while predicting significant decreases in the World Giving Index as well as in two of its components (volunteering time and helping a stranger) specifically in societies with languages that had passives (prediction 2).Moreover, the negative effect of income presumably associated with a type of psychological reactance in cultures of languages with passives was more likely to be observed in the context of donations that did not involve money (prediction 3).Because the effects obtained with income paralleled those obtained with power distance and uncertainty avoidance, and to a certain extent with long-term-orientation, the positive impact of income on helping may be thought to be getting reversed in contexts of helping that do not involve money spending.
One theory that can help us understand this reversal effects is psychological reactance (Brehm & Brehm, 1982).Passive vs. antipassive sentences may be creating these reversal effects because they function to decrease vs. increase the perceived responsibility of actors who try to change events such as those that involve changing the unfortunate situation of the needy.The findings, therefore, suggest that the negative effects observed in previous studies of income on helping may be seen as a reversal of the default positive effect of income on helping.However, because psychological reactance is not explicitly measured in the current study it may be also possible to utilize other theories to explain the obtained results.One such account that can be used to explain some of the results reported above is the cognitive dissonance theory (Cooper, 2007;Festinger, 1957).According to the theory, the perception of an inconsistency between people's behaviors and attitudes produces cognitive dissonance, which, in turn, drives people to reduce this discrepancy by either changing their behavior or their attitude.To feel cognitive dissonance, it is essential that people freely engage in behaviors.Because antipassives vs. passives are expected to respectively increase vs. decrease the perceived autonomy and freedom, people speaking languages with antipassives as opposed to passives might have experienced relatively greater levels of cognitive dissonance.To reduce this aversive state, people normally change their personal attitudes rather than their behavior.However, unlike personal ones, cultural attitudes are more stable and may have forced people to change their behavior rather than their cultural attitudes, ultimately producing culturally consistent patterns of behaviors to a relatively greater extent.That may be why people speaking languages with antipassives as opposed to passives acted more in line with how power distance, uncertainty avoidance and long-term orientation levels of their cultures would predict their behaviors.
Although the experience of a type of cognitive dissonance can help explain some of the results reported in the present study, predicted by prediction 2a and 2b, other results predicted by prediction 3 are difficult to explain in the same way.Cognitive dissonance is shown to create undifferentiated physiological arousal (see e.g., Elliot & Devine, 1994).Money or associated thoughts and feelings of power have been shown to reduce the physiological arousal measured by heart rates (Scheepers, de Wit, Ellemers, & Sassenberg, 2012;Sittenthaler et al., 2011).Thus, in contexts of money donations where the exposure to money cues is expected to be greater as compared to the other contexts, the cognitive dissonance is expected to decrease rather than increase.If, as explained in the previous paragraph, the strength of the effects of language-culture interactions is positively associated with the increased experience of cognitive dissonance in predicting charitable behaviors, then in contexts of money donations, the language-culture interactions should have been weaker rather than stronger.However, the latter was more often the case in the data although it was sometimes not observed such as in the interactions of language with uncertainty avoidance and long-term orientation.Therefore, although the results from the present study are in line with the psychological reactance to a greater extent than the cognitive dissonance theory, they are not sufficiently conclusive to completely rule out cognitive dissonance as a possible explanation.Future studies are needed to more directly measure the effects of psychological reactance vs. cognitive dissonance on charitable behaviors across cultures.
Another related future area of research concerns the levels of description associated with cognitive dissonance and psychological reactance that can be used to explain the reversal effects on cross-cultural behaviors obtained in the present study.Because both theories are mostly used to explain human behaviors at an inter-individual level within cultures, the processes described by the theories may operate differently at higher level of cross-cultural factors.As shown in this study, language, for example, might be a mid-level medium through which interindividual vs. cross-cultural factors might meaningfully interact.Studying the role of language in social contexts that cut across cultures, thus, may be one way in which to merge the gap between the processes operating at inter-individual vs. cross-cultural levels of explanations for human behavior.
There was an unexpected, significant effect for the antipassive-masculinity interaction in predicting the World Giving Index as well as two of its components (money donations and helping a stranger) and significant effect for the passivesmasculinity interaction in predicting helping a stranger in the data.The opposite ends of the masculinity-femininity continuum can facilitate helping for different reasons (Nelson et al., 2006).Females in masculine cultures and males in feminine cultures help more after considering the impact of their helping acts on themselves whereas the other members of the masculine and feminine cultures help more after considering the impact of their helping on others (Nelson et al., 2006).It was the positive effect of masculinity, rather than femininity, that was impacted by the possible reactance experience in the present study.The reason why the positive effect of masculinity, rather than the positive effect of femininity, may have been impacted by the possible reactance experience may be explained by considering the findings in Study 2 of Nelson et al. (2006).The members of more masculine cultures were found in that study to feel more personally responsible to help others after reading donation appeals that highlighted the impact of their donations on others rather than on themselves.Perceiving the helping act to have a greater likelihood of impacting the needy, which is similar to the other-focus condition in Nelson et al. (2006) was described by Brehm and Brehm (1981) to have the potential to create "greater reactance arousal than the low likelihood condition" (p.175).Therefore, masculine cultures may be more likely to trigger psychological reactance as compared to more feminine cultures.Future studies can clarify whether and/or how the positive effect of femininity may be reversed as a result of a reactance or dissonance experience and whether the positive effect of femininity can be understood as a reversal of the positive effect of masculinity.One important difference regarding the role of culture in helping may, therefore, be about how the members of different cultures differ in their perceptions of which type of contexts threaten their freedom to help.
The hypotheses about long-term orientation were not confirmed by the data in the present study.Little research has been done on long-term orientation as it has been one of the last cross-cultural dimensions identified by Hofstede.In one study where the negative effect of this variable on helping was reported, the effect was unexpected and no mechanism was proposed to explain why long-term orientation may have had an effect on charitable behaviors (Winterich & Zhang, 2014).In line with the present study findings, it can be said that the effect of long-term orientation on helping likely involves mechanisms that are different from those operating in the cases of power distance and uncertainty avoidance.Given that power distance has been shown to be about the perceived responsibility to help others (Winterich & Zhang, 2014) and that uncertainty avoidance is affected by the perceived autonomy variation in this study in the same way in which power distance has been affected, the reported effects of long-term orientation on helping (Winterich & Zhang, 2014) may be due to factors that do not involve a type of personal responsibility to help others.Given that long-term orientation refers to the willingness to change vs. continue traditional structures in a society, future studies are needed to see whether effect of this dimension on charitable behaviors may have to do with the general characteristics of more vs.less traditional societies in which helping may have become more vs.less of a habitual act rather than a complex cognitive response.
It should be noted that in the present study a specific relationship is not assumed to be present between language and culture.Some studies show that language is capable of impacting thinking, perception and behavior directly (see e.g., Borodistky, 2001;Roberson, Pak, & Hanley, 2008) while others show that different types of using language result from culture (see e.g., Gopnik, Choi, & Baumberger, 1996;Kashima & Kashima, 1998).In the present study, it is only assumed that culture and language interact to change thinking.This latter conclusion is supported by studies showing that the effects of language on thinking are present to the extent that people internalize their culture (Lucy & Gaskins, 2003), and by studies showing that the effect of language on thought is mediated by the use of language (Gennari et al., 2002;Slobin, 2003), which, in turn, is moderated by culture (Roberson et al., 2008).Furthermore, it is not assumed in the present study that cultural dimensions such as power distance can be solely seen as decreasing the perceived responsibility to help.Because there could be different forms of individualism (see e.g., Vignoles, Owe, Becker, Smith, Easterbrook, Brown et al., 2016) that can lead some high power-distance, or even uncertainty-avoidant cultures to facilitate the feeling of agency, these cultural dimensions may also be expected to trigger a reactance motive.In fact, although both masculinity and femininity were previously shown to increase helping (Nelson et al., 2006), it was the positive effect of masculinity that was reversed by language in the present study.Therefore, a reversal effect can be observed even when a cultural dimension does not have a clear net effect on helping.To the extent that power distance or uncertainty avoidance predict decreases in helping as shown in previous studies (Smith, 2015;Winterich & Zhang, 2014), these dimensions will interact with the presence of passives vs. antipassives, which predict decreases vs. increases in helping (hypotheses 1b and 1b), ultimately creating a reversal effect in charitable giving.In sum, the language-culture interaction effects reported in the present study are independent of the specific main effects of language alone and culture alone on helping as well as the specific bidirectional influences between language and culture.
In concluding, significant sentence-voice effects in charitable giving were obtained in the present study for income, power distance, uncertainty avoidance, and to a certain extent for, long-term orientation and masculinity, as the societylevel variations that were previously found to impact helping in specific ways.Overall, in considering the effects of societal and cultural variables on helping it becomes essential to pay attention to not only cultural variations per se but also the linguistic perception of the contexts of helping in different societies.Such perceptions can restrict the helping agents' responsibility likely creating freedom threats that can ultimately moderate the effects of the society-level and cultural variables on charitable behaviors.

Figure 1 .
Figure 1.The predicted increase in world giving index (WGI) as a function of income per capita in cultures of languages with vs. without antipassive sentences and with vs. without passive sentences overlaid on one graph.

Figure 2 .
Figure 2. The predicted increase in world giving index (WGI) as a function of power distance across cultures of languages with vs. without antipassive sentences and with vs. without passive sentences overlaid on one graph.

Figure 3 .
Figure 3.The predicted increase in world giving index (WGI) as a function of uncertainty avoidance across cultures of languages with vs. without antipassive sentences and with vs. without passive sentences overlaid on one graph.