HAVING, GIVING & TAKING
BIG DATA ON THE RELATIONSHIP BETWEEN SOCIAL CLASS AND
PROSOCIAL BEHAVIOR
In the current research it is examined how social class is related to prosocial behavior. Whereas previous research has found inconsistent results using experimental lab settings and survey approaches, we analysed actual lending and borrowing behavior in a natural setting. Study 1 (N=16.251) found that there is no meaningful relation between social class and lending. Study 2 (N=98), combined behavioral data with survey data and found, contrastingly, that higher social class is associated with more lending. Higher social class was also associated with more borrowing. We conclude that the theoretical field on prosocial behavior highly benefits from a more natural research approach. Statistical challenges regarding analysing big data are discussed.
Student name: Mayra Kapteyn
Student number: 10002758
Supervisor: Gerben van Kleef
Participative societies thrive on prosocial behavior. People help each other out by sharing, volunteering and many other kinds of behavior that are intended to benefit
another. They do this either because it brings them some kind of reward in exchange and/or because they feel an empathetic and compassionate response to someone else’s need (Batson & Shaw, 1991). Helping somebody brings prosperity for the other, but how does one’s own prosperity relate to the willingness to help the other? Do people who have more, give more, or is it the other way around?
In this research project we assessed how social class is related to prosocial behavior. Social class can be measured by material resources, such as education (Snibbe & Markus, 2005), income (Kraus & Keltner, 2009) and occupational status (Oakes & Rossi, 2003), as well as social class rank, which is a subjective perception of rank in comparison to others (Kraus, Piff & Keltner, 2009). Different social classes experience different levels of opportunity, which shapes the way they think (Johnson & Kreuger, 2006). We will explain how these cognitions differ and how that might impact prosocial behavior.
Two viewpoints suggest contrasting hypotheses regarding social class and prosocial behavior. First, the negative relation hypothesis stems from the social cognitive theory of class (Kraus, Piff, Mendoza-‐Denton, Rheinschmidt & Keltner, 2012), which suggests that lower class’ act more prosocial than upper class individuals because they are more attuned to their environment (Piff, Kraus, Côté, Cheng & Keltner, 2010). Contrastingly, a positive relation hypothesis stems from the concept of Noblesse Oblige. This viewpoint predicts that prosocial behavior is higher among upper class individuals because their relative cost of helping is lower in comparison to lower class individuals (Dovidio, Piliavin, Schroeder & Penner, 2006; Batson & Shaw, 1991).
Negative relation hypothesis: Social class and Contextualism
Recent research has suggested that there are cognitive differences between lower and upper class individuals that influence people’s prosocial behavior (Piff et al., 2010). Lower class individuals are suggested to have a more contextual social cognition, which means they draw on external forces to explain personal, social and political events (Kraus et al., 2012). This contextual thinking implies paying a lot of attention to other people’s
social cognition generate behavior that is highly influenced by other people. Upper class individuals, on the other hand, tend to have what is called a solipsistic relation to the outside world (Kraus et al., 2012). This refers to an individualistic orientation to the environment, motivated by internal states, goals and emotions. It implies a higher sense of personal control over one’s life outcomes.
Indeed, in comparison to their upper class counterparts, lower class individuals are more dependent on their external world for their personal outcomes (Argygle, 1994), and experience less control over their lives (Johnson & Krueger, 2006; Lachman & Weaver, 1998). This lack of personal control drives lower class individuals to explain success to situational factors, while upper class individuals attribute success to internal traits (Kraus et al., 2009).
This reduced sense of personal control and dependency on other people (Kraus et al., 2009) may entice lower class people to engage more with one another. In personal
interactions, lower class individuals show a more socially engaged non-‐verbal style than upper class individuals, who show relatively more impolite behaviors such as self-‐grooming (Kraus & Keltner, 2009). It is theorised that due to lower class’ reduced sense of personal control, they think in a contextual way, leading them to be socially engaged.
Consequentially, their contextual cognition may result in them being more helpful towards other people (Piff et al., 2010; Kraus et al., 2012).
The question remains if lower class’ contextual focus leads them to be more prosocial than upper class individuals. Research on prosocial behavior has shown that lower class individuals are better at judging other people’s emotions, signifying more empathic accuracy (Kraus, Côté & Keltner, 2010). They also report higher levels of compassion in response to seeing someone else suffering, which is also reflected in their decreased heart rate, a symptom associated with feeling compassion (Stellar, Manzo, Kraus & Keltner, 2012). This enhanced empathetic accuracy and feelings of compassion may lead lower class individuals to act more prosocially than upper class individuals.
To examine this hypothesis, Piff et al. (2010) conducted four studies and found evidence that social class is negatively related to displays of prosocial behavior. However,
behavior. The measures of prosocial behavior are either attitude measures (Study 2) or experimental measures in a lab setting (Study 1, 3 and 4). We will address each study and explain how these measures lack external validity.
In Study 1, Piff et al. (2010) found that subjective social rank is related to decreased generosity in the Dictator Game. The Dictator Game is an adequate, highly controllable measure of prosocial behavior. It is however a very simplified reconstruction of reality, because the situation depicted the Dictator Game –having to distribute points between oneself and a stranger-‐ is one that does not (often) present itself often in real life. Therefore, additional methods using measures closely related to real life is needed.
Second, they found that manipulated social rank and income negatively predict attitudes on the amount of money people should donate to charity. This attitudinal measure is flawed because it does not control for social desirability bias (Randall & Fernandes, 1991). This is a serious problem to validity because different social classes may be more or less triggered to respond socially desirable. Especially lower class individual’s contextual
cognition (Kraus et al., 2012) may make them more susceptible to social desired responding because they may be more attuned to leaving a good impression with others. The second problem with the attitudinal measure is that attitudes generally don’t predict behavior very well: only when the timing, context, action and target of the attitude measure and the behavior are similar (Ajzen & Fishbein, 1977). In Study 2, prosocial behavior is measured by the question “what portion of one’s salary should be allocated to charitable donations”. Because there are no real costs involved, answering this question in a prosocial manner is much easier said than done, so the action in the attitude measure does not resemble the actual action. Therefore, the measured attitudes on donations arguably do not predict class-‐ driven behavior well.
Third, Piff et al. (2010) reported a negative relationship between social class and prosocial behavior, mediated by egalitarian values. However, they assessed prosocial behavior using the Trust Game, which is not a valid measure of prosocial behavior. The participant namely allocates points to another participant, while the other participant has the chance to return the favour with increased value of the points. This task does not measure prosocial behavior; rather, it measures whether participants choose a risky,
cooperative but potentially rewarding strategy, or a safe individualist strategy, with less potential rewards. Prosocial behavior is defined as behavior intended to benefit the other (Brief & Motowidlo, 1986), but in this case, the ultimate intention of the participant may just be to receive the maximum points for themselves. Upper class’ solipsistic cognitions may lead them to choose a more individualistic strategy, but that does not mean they’re less prosocial. Therefore, this is not an accurate measure of prosocial behavior.
Fourth and finally, it was reported that compassion moderates the negative
relationship between social class and helping behavior. This experiment was the only explicit behavioral measure used in this research. The measure of prosocial behavior was the time the participant took to help a female confederate who arrived late to do her task. This measure is biased by social norm rigidity, because the help recipient’s distress (and thus, need for help) is a direct consequence of her own lack of punctuality. Lack of punctuality is something people can disapprove strongly of, especially towards women (Kanekar & Vaz, 1993). Bowles and Gelfand (2010) found that when a low-‐status individual (operationalised as “lacking a high-‐status track record”, as is the case with the confederate) violates a norm, upper class individuals punish more heavily than lower class individuals. In a subsequent study, they found that men are more eager to punish female norm violators than male norm violators. These findings seriously question the conclusion drawn by Piff et al. (2010),
because the supposedly direct effect of social class on prosocial behavior may be confounded by norm rigidity towards the female, norm-‐violating confederate.
Thus, although Piff et al. (2010) may have a solid theoretical background to
hypothesize that social class inhibits prosocial behavior, their measures of prosocial behavior lack external validity. The Dictator Game in Study 1 is accurate but simplified, the measure in Study 2 is merely an attitude measure, study 3 measures strategy instead of helping and the measure in study 4 is confounded by norm rigidity. These flawed measures of prosocial behavior imply that these experimental results may not be valid in the real world. Therefore, we cannot conclude on a negative relation between social class and prosocial behavior.
Positive relation hypothesis: Noblesse Oblige
being prosocial is higher. The higher the cost compared to rewards, the lower the probability that somebody will help (Dovidio et al., 2006; Batson & Shaw, 1991). So simply because lower class individuals have relatively less to give, they may act less prosocial than upper class individuals.
Korndörfer, Egloff and Schmukle (2015) tested whether there is a positive or a negative relation between social class and prosocial behavior. They conducted eight survey studies and reported mixed results. For example, Study 1, which was conducted in Germany, reported no significant relationship between social class and relative amount of money spent on charity among donating households. Contrastingly, Study 2, which was conducted in the U.S., reported a negative relationship between social class and relative amount of money spent on charity among donating households. Then Study 3 –also using U.S. data on donating-‐ reported a positive relationship. Studies 4 (conducted in Germany) and 5
(conducted in the US) also reported positive relationships between social class and reported volunteering. The results from Study 6 reported a meaningless (b=.06) but significant
positive association between social class and volunteering, among 37.000+ participants internationally. Study 7 found a significant positive association between objective social class and everyday helping, but no significant association between subjective social class and everyday helping. Study 8 found a significant positive relationship between social class and allocated points in the Trust Game. These results are not consistent, but seem to point in the direction that there is a positive relationship between social class and prosocial behavior, contrary to the findings from Piff et al. (2010).
However, the methodology used in this study also lacks external validity. Seven out of eight studies conducted were survey measures. Survey measures on past behavior (such as volunteering and donating behavior) are biased by socially desirable responding and recall bias (Coughlin, 1990; Randal & Fernandes, 1991). Especially prosocial behavior is sensitive to social desirability bias because prosocial behavior is very socially desired in definition. Therefore, survey measures are not adequate measures of prosocial behavior.
The only behavioral measure used in this research is the Trust Game in Study 8, which (as described above) measures behavior intended to benefit the self, not the other. So in eight studies, none of the measures of prosocial behavior contain actual behavior.
Therefore, these results don’t provide sufficient evidence that there is a positive relation between social class and prosocial behavior, or to state that the negative relation hypothesis from Piff et al. (2010) is invalidated. There is behavioral research needed in a natural setting in order to assess a valid relation between social class and prosocial behavior.
Current Research
The current literature on social class and prosocial behavior has used only
experimental or survey data, but no behavioral data in a natural setting, and therefore it may not be valid in the real world. Therefore, we tested whether social class is positively or negatively related to prosocial behavior with actual behavioral measures in a natural setting. We tested if participant’s social class is associated with lending household items through an online sharing platform called Peerby. Hypothesis 1a is that social class is negatively
associated with lending and hypothesis 1b is that social class is positively associated with lending.
Balancing Giving and Taking
Aside from the lack of behavioral measures, another missing element in the current literature regarding social class and prosocial behavior is the balance between giving and receiving help. Thus far, we have a unilateral understanding of prosocial behavior, because we only have information on how much people give, and none on how much people take. The social cognitive theory on social class (Kraus et al., 2012) predicts that upper class individuals experience more personal control and are therefore less attuned to their
environment. This may have different implications for requesting help. Hypothesis 2a is that upper class’ elevated sense of personal control (Johnson & Krueger, 2006; Kraus et al., 2009) reduces their tendency to ask for help because of elevated levels of individualism (see Kraus et al., 2012). Contrastingly, and alternative explanation is that upper class’ elevated sense of personal control triggers assertiveness – not being afraid to ask. Following this reasoning, hypothesis 2b is that higher social class is associated with asking for help more often.
By combining the measures between giving and receiving help, we can get a sense whether there is truly a relation between social class and prosociality, or that there is only a relation between social class and activity on Peerby. Namely, if upper or lower social class
individuals would both borrow and lend more than the other, it wouldn’t necessarily mean that they are more prosocial – it could also just mean that they are more participative on Peerby. Therefore, we tested the relation between social class and a prosociality ratio of lending minus borrowing. Hypothesis 3a is that the relationship between social class and prosocial behavior upholds when subtracting borrowing behavior from lending behavior. This would signal strong differences in prosociality among social classes. Contrastingly, hypothesis 3b is that there is no relation between social class and prosociality when subtracting borrowing behavior from lending behavior. This would signal that a difference between social classes may be due to different levels of activity on the Peerby platform, and not due to a difference in prosociality.
In the second study, in order to validate that self-‐report measures are indeed
inadequate measures of prosocial behavior, we also assessed how the self-‐report measures on borrowing and lending relate to the behavioral measures on borrowing and lending. Therefore, hypothesis 4 is that there is a low correlation between self-‐report lending and actual lending, and hypothesis 5 is that that there is a low correlation between self-‐report borrowing and actual borrowing.
Study 1 uses a large dataset (N=16.251) in order to assess robust general findings on the relations between social class, borrowing, lending and prosociality ratio. We measured social class by combining average street income and average house value of the participant’s street. In Study 2, we enriched the street level social class measures and behavioral data from Peerby with survey data on income, age, gender and self-‐reported borrowing -‐and lending.
Study 1
In study 1, big data is used to assess if social class is positively or negatively related to lending (hypothesis 1) and borrowing (hypothesis 2) on the Peerby platform. Third, we tested if the relation between social class and lending would uphold when subtracting borrowing from the lending score.
The behavioral data is gathered from Peerby, an online sharing platform where neighbors lend each other household items for free. Peerby saves button clicks on the
website and app onto their database. The measure of lending is somewhat determined by what objects people have in their homes, so it is important that these are not luxury items that only upper class individuals possess. The most requested items on Peerby are: (1) drill, (2) ladder, (3) standing tables, (4) bike, (5) trailer and (6) car. There’s no way of knowing what items participants exactly have in their home, but the top items clearly aren’t exclusive to higher social classes. Therefore, it should not confound the measure of prosocial
behavior.
Method Participants
A dataset containing 92.679 participants was provided by Peerby. 49.182 participants were excluded because of missing values for all of the social class measures, namely a) they did not provide their full 6-‐digit zip code area or b) there was no data on income or house value available for their specific zip code area. To account for the high number of people who just signed up for Peerby to ‘take a look around’, we selected members who at least lent out once. After excluding inactive members, 16.167 active members were left in the dataset. These participants were members for 510.36 days on average (SD = 273.67). Procedure
The Peerby platform works demand based: when someone needs something, they send out a request to their neighbors with a personal message, see Figure 1. The receiver of the message can then click ‘Help neighbor X’; ‘Not now’ or ‘I don’t have it’. If someone clicks “Help neighbor X”, the two neighbors enter a chat page where they can make arrangements to pick up the item. As there are often more than one neighbor offering the item, the
requesting neighbor chooses one of the offering neighbors, then picks the item up at their home address and returns it after use.
Measures
Lending. Lending was measured by the total number of ‘Help’ clicks per member, see Figure 1. This is the total number of help clicks on both the app and the web platform (M = 4.20, SD = 7.37). The recipient sees the date, a photograph of the requester, the distance to the recipient, their name, the item that they need and a personal message they provide. They can choose to click ‘Help’; ‘Not now’ or ‘I don’t have it’. There is also a flag button, in case the request is inappropriate or unwanted.
Borrowing. Requested help was measured by the total amount of requests the participant has placed on the website, see Figure 2. In order to request an item on the Peerby platform, the participant describes the item that they need and a short story to describe what they need it for (see Figure 2). Then Peerby sends the request to max. 250 of the participant’s neighbors. It is communicated that people receive an offer from their neighbors in thirty minutes on average. The average total number of borrowing is 1.42 times (SD = 2.27).
FIGURE 2. A PEERBY REQUEST FORM
Prosociality index. A prosociality index was computed by standardizing borrowing and lending, and then subtracting borrowing from lending. Thus the higher the score, the more prosocial the behavior on Peerby.
Social Class. Social class was measured by standardizing the average income and the average house value in the participant’s street. These two variables are highly correlated, r(7550)=.68, p<.001, thus predict social class reliably. If one of the two values was missing, only the other variable was used as proxy for social class. There was more data available on income (N=16167) than on house value (N=7550).
This data is obtained from the Dutch national databank (CBS, 2012), who published it as customized data by the request of Sinfore and the Jan van Es Institute. House value is only published when there are at least 20 venues in the 6-‐digit zip area, rounded off and reported in units of thousand. The average income is only published when there are at least 10
average monthly income was below the minimal (€500) or above the maximum (€10.000) value, the minimal (€500) or maximum (€10.000) value was reported.
Results
The first step was to assess how social class relates to prosocial behavior. Hypothesis 1a was that higher social class is associated with less lending (see Piff et al., 2010), while hypothesis 1b predicted a positive relation between social class and lending (see Korndörfer et al., 2015).
Due to the non-‐parametric distribution of the data, we ranked the values and computed Spearman’s Rho correlation. Table 1 shows that, supporting the positive relation hypothesis, a very weak but significant positive relationship was found between social class and prosocial behavior. Hypothesis 2a was that social class would be negatively associated with social class, while hypothesis 2b predicted a positive association. We found that higher social class was associated with less borrowing, therefore confirming hypothesis 2a. We thus confirmed hypothesis 3a, which suggested that the found relationship between social class and prosocial behavior is indeed caused by higher prosocial behavior among upper class individuals, and not by increased Peerby activity overall.
Table 1
Spearman’s rho Correlations between measures, (N)
Measure Lending Borrowing Prosociality ratio
Social Class .03* (16251) -‐.04* (16251) .05* (16251) Note. *p<.001
Although these results suggest significant relations, the coefficients are extremely weak. The explained variance of social class on lending is only ρ2=.001, and the explained
variance of social class on prosociality is ρ2=.003. Therefore, we cannot conclude there is a meaningful relationship between social class and prosocial behavior.
In Study 1, we did not find a meaningful relationship between social class and prosocial behavior. This indicates that either the assumed relationship between social class and prosocial behavior is non-‐existent in a natural setting, or that the relationship is more complex than could be measured in this research design.
In the current study, it was impossible to account for demographic data such as gender and age. Moreover, we could not assess participant’s social class directly, but used proxy data from the street the participant lived in. Additional research is needed to verify the quality of the social class measure and to control for age and gender. That way we can draw conclusions on the relation between social class and prosocial behavior.
Study 2
Our second study investigated the relationship between social class and borrowing and lending among a smaller sample of Peerby members, with direct measures of income, education level, age and gender. However, unfortunately, we weren’t able to control for age and gender because the data did not meet parametric assumptions.
Taking the results of Study 1 in account, we predicted that, following hypothesis 1b, higher social class would be associated with more lending and borrowing. The findings in Study one also set direction for the second hypothesis, namely that higher social class would be associated with less borrowing. Third, we expected that following hypothesis 3a, the relation between social class and lending would uphold when accounting for borrowing behavior. Fourth, we hypothesized that self-‐report measures on lending would weakly correlate with actual lending. Similar to that account, our fifth hypothesis was that self-‐ report measures on borrowing would be weakly correlated with actual borrowing.
Method
Participants. Participants are members of Peerby who responded to a survey regarding participation on sharing platforms, conducted by Stipo (N= 180). Stipo is a consultancy firm that published a report on participative Internet Platforms (Stipo, 2015). Participants were contacted through e-‐mail and asked to complete a 58 item survey on their Peerby behavior. A €50 voucher was allotted among the participants. 71 participants were excluded because of missing data. We also excluded 11 participants because their lending
count was above 40, so they were clear outliers who impacted results disproportionately. Two of these participants were employees of Peerby or Stipo. This left 99 participants in the final analysis (57 female, 42 male) who were members for 539.85 days on average (SD = 262.62). Participants ranged in age from 25 to 65 years old (M = 44.09, SD = 11.35). 97% of the respondents reported a Dutch nationality.
Measures
Self-‐report income measure. The self report measure inquired “what is your income level on a yearly basis” using six categories: (1) <€20.000, (2) €20.000-‐€30.000, (3) €30.000-‐ €50.000, (4) €50.000-‐€70.000, (5) €70.000-‐€100.000, or (6) >€100.000.
This manner of questioning is suboptimal because it remained unclear whether it referred to personal or household income, and before or after tax deductions. However, because it is not likely that there is an effect of social class on the way people respond to this question; we assume that the inaccuracy is distributed equally among social classes. In order to calculate correlations using income as a scale, participant’s income levels were recoded into numerical scale variables (1) €18.000*, (2) €25.000, (3) €40.000 (4) €60.000, (5) €85.000, and (6) €120.000. Participants reported a median income level of €40.000.
Street income. The street income measure was (similar to Study 1) obtained from the CBS data regarding mean income per month in the participant’s street of residence (CBS, 2012). The mean monthly street income was €2840, with a standard deviation of €989.
In order to assess the validity of the income proxy measure, we calculated Pearson correlation between the and the self-‐report income measure. The correlation was moderate, r(74)=.37, p<.01. So even though one measure is direct and the other indirect, they are moderately associated with each other.
Street house value. The average street house value was obtained from CBS data regarding average house values per street, following the same procedure to study 1. The average house value was €212.450 (SD = €87.468).
*
Social class. A composite measure of social class was computed by standardizing the self-‐report income measure, the street income measure and the street house value
measure, then averaging the scores on these three measures. If one of the measures was missing, the average of the other measures was taken. A Crohnbach’s Alpha (using
standardized values) of α=.73 showed this composite of social class was a reliable predictor. Education level. Education level was assessed using six categories: (1) lbo (lower craft education), (2) vmbo (high school), (3) mavo (high school), (4) havo (high school), (5) vwo (high school), (6) mbo (college), (7) hbo (college), and (8) university. The median education level was HBO (college).
Interestingly, education level and self-‐reported income were not correlated, r(96)=.17, p=.10. Therefore, education level was not combined with the other social class measures to represent social class. This lack of reliability in using education as a measure of social class may be explained by the egalitarian education model in the Netherlands. Higher education is government funded with additional funding for lower class individuals. Thus, even though education has previously been used as a measure of social class (Snibbe & Markus, 2006), this measure may not be valid in highly egalitarian educational systems.
Lending behavior. Lending behavior was, similar to Study 1, tracked by Peerby. The measure represents the total number of times the participant has clicked “Help neighbor X”, in order to lend something to another member of Peerby, see Figure 1. Participants lent 9.37 times on average (SD = 8.5).
Self-‐reported lending. Participants self-‐reported how often they lent things per year using six categories: (1) once per week, (2) twice or more per week, (3) once a month, (4) a few times per month, (5) a few times per year, or (6) once a year. In order to calculate correlations, these scores were recoded into scale values of lending frequency per year: (1) once per week into 52, (2) twice or more per week into 104, (3) once a month into 12, (4) a few times per month into 24, (5) a few times per year into 2, and (6) once a year into 1.
Participants reported an average lending of 4.64 times per year (SD = 5.95). The discrepancy between mean of the self-‐report data and the behavioural measure can be
explained by the fact that offering to lend does not always result in actual lending. Often, help requesters receive multiple lending offers and choose one neighbor to borrow it from.
Borrowing behavior. Borrowing behavior was also tracked by Peerby, just like done in Study 1 (M = 2.75, SD = 3.39). The count number represents the total number of times the participant has requested to borrow something from the other members during their
membership.
Self-‐reported borrowing. In the survey, participants were asked how often they borrowed things per year, using six categories: (1) once per week, (2) twice or more per week, (3) once a month, (4) a few times per month, (5) a few times per year, or (6) once a year. This scores were recoded into scale values of lending frequency per year: (1) once per week into 52, (2) twice or more per week into 104, (3) once a month into 12, (4) a few times per month into 24, (5) a few times per year into 2, and (6) once a year into 1. The average frequency was 3.21 times per year (SD = 4.77).
Prosociality ratio. The prosociality ratio was computed by subtracting the standardized borrowing score from the standardized lending score. This represents the balance between providing for others and receiving help. The higher the score, the more lending in comparison to borrowing.
Results
In this study, we assessed if higher social class is related to more prosocial behavior. We used behavioral data on lending and borrowing to measure prosocial behavior, in combination with proxy social class measures and survey measures on income, gender, age and self-‐reported lending and borrowing. Because the behavioral data does not meet parametric assumptions, we ranked the data and computed Spearman’s rho. Table 2 shows a summary of the correlations between the social class and prosocial behavior, age and gender.
Hypothesis 1 was that higher social class is associated with more lending. Indeed, Spearman’s rho reported that there is a moderate positive association between social class and lending, r(96)=.23, p<.05. This indicated that higher social class is associated with more
Hypothesis 2 was that higher social class is associated with less borrowing. In contrast to that prediction, borrowing was marginally significant associated with social class in a positive direction, r(98)=.16, p=.06. This means that upper class individuals seem to borrow more than lower class individuals.
So upper class participants both lend and borrow more, relative to lower class participants. Hypothesis 3 was that the relation between social class and prosociality would uphold when accounting for both borrowing and lending. Because of the unexpected positive relation between social class and borrowing, there was no significant relation between social class and prosociality ratio, r(98)=-‐.03, p=.40. This suggests that the
difference in lending activity found among social classes may be explained by different levels of participation on Peerby in general, and not by different levels of prosociality. However, in order to conclude this, we must conduct mediational analysis, which is not possible using Spearman’s rho.
As for hypothesis 4, we predicted that self-‐reported lending would be weakly
associated with actual lending. We found that indeed, self-‐reported lending was moderately correlated with actual lending, r(80)=.22, p<.05. This indicates that what participants report on lending is not accurate in representing actual behavior. Similarly, hypothesis 5 predicted that self-‐reported borrowing would also be weakly associated with actual borrowing. Our results showed that this correlation is indeed weak and only borderline significant, r(58)=.22, p=.051. So in borrowing too, participants’ self reports are not accurate. Due to this
inaccuracy, as can be seen in Table 2, self-‐reported lending is not significantly associated with social class, while actual lending is significantly higher among upper class participants, compared to lower class participants. This informs us that self-‐report measures don’t accurately represent people’s actual behavior.
Finally, we found that age is moderately correlated with lending, self-‐reported lending and prosociality ratio. This implies that age may be a factor that can explain the positive relation between social class and lending. But, similar to the other possible confounds, there is mediational analysis needed before we can conclude that the relation between social class and prosocial behavior is indeed mediated by participant’s age.
Correlation Matrix between measures of Prosocial behavior, Social Class and Demographic Measures (N)
Measure Lending Self-‐reported Lending Borrowing Self-‐reported Borrowing Prosociality ratio Social Class .23 (98)** .06 (79) .16 (98)* -‐.12 (57) -‐.03 (98) Age .29 (99)*** .19 (80)** -‐.15 (99) -‐.12 (58) .32*** (99) Gender .11 (99) .10 (80) .02 (99) -‐.05 (58) .08 (99) Note. *p<.10, **p<.05, ***p<.01 Discussion
The results of Study 2 show that, confirming hypothesis 1b, there is indeed a positive relation between social class and prosocial behavior. This clearly falsifies the negative
relation hypothesis stipulated by Piff et al. (2010), who suggests that due to lower class’ highly contextual cognition, they are more prosocial than lower class people. In this study, were able to connect survey data to actual behavioral data gathered in a natural setting. This gives our findings great external validity, while also ensuring internal validity on our social class measure, because we were able to measure income directly.
In contrast to the predicted hypothesis 2a, there was also a (borderline significant) positive relation found between social class and requesting help. So upper class participants were more likely to both lend and borrow on Peerby. When subtracting borrowing from lending in one prosociality ratio (hypothesis 3), there was no relationship found between social class and prosociality ratio.
These results signal the possibility that the relation between social class and prosocial behavior may not be be caused by upper class’ enhanced prosociality, but on their increased tendency to participate on Peerby in general. Due to statistical limitations in analysing the non-‐parametric dataset, we could not conduct mediation or control analyses for this alternative explanation. Our results do however send a clear signal that prosocial behavior needs to be put in perspective of help giving as well as giving requesting.
Also concerning correct measuring of prosocial behavior, we assessed the validity of self-‐report data. Confirming hypothesis 4, we found a weak correlation between self-‐
reported lending and actual lending behavior. Similarly, confirming hypothesis 5, we found a (borderline significant) weak relationship between self-‐reported borrowing and actual borrowing requests. We do need to be careful interpreting these results. Apart from to the response errors such as recall bias and socially desired responding (Coughlin, 1990; Randall & Fernandes, 1991), this low correlation may be caused by the rigid categories participants had to answer to. For example, 65% of participants reported borrowing “a few times per year”. This categorical way of asking is necessary because people aren’t able provide an exact number how often they did something. However, this does lead categorical data to lose much of the variance that natural data has, especially when events are not frequent and spread out over a long period of time (Coughlin, 1990). So, our results show that categorical self-‐report measures specifically don’t predict real-‐world behavior accurately.
Finally, an alternative hypothesis could be that the relation between social class and lending is confounded by participant’s age. We found a moderate correlation between age and lending on Peerby. Because the data was not distributed normally, it was impossible to control for age. The found relationship between social class and lending may be caused by the fact that older people have more material resources and lend more things. However, the positive relation between social class and borrowing can not be explained by age, because older people do not seem to borrow more than younger people. Future research using parametric datasets (or new statistical methods) are needed to assess if age mediates the relation between social class and prosocial behavior.
General Discussion
The current research investigated whether having resources is related to giving resources to somebody else. Our two studies found mixed results. Study 1 found an
extremely weak correlation between social class and prosocial behavior, signalling that there is no meaningful relation between how much people have and how much they give. Study 2 found a moderate positive relationship between social class and lending. We also found a marginally significant positive relation between social class and borrowing.
The two studies signalled very different strengths regarding how higher social class is related to more prosocial behavior. This discrepancy might be caused by the large number of unengaged members of Peerby in Study 1, whereas the sample who responds to a survey (Study 2) consists of people who are more committed to the platform. The Study 2 sample indeed had an average lending count of 9.37 times, which is more than double in
comparison to the average of 4.20 times in Study 1. An alternative explanation could be that the relation between social class and lending only arises among very active members. We tested this possibility by only selecting very active members (lending > 10) in Study 1. This did not result in a significant correlation. So the lack of result in Study 1 can’t be explained by the large group of members who are only sporadically active. Most probably, the data in Study 1 was simply too complex to be correctly analysed with the current statistical programs. The dataset of Study 1 contains a lot of participants who are in some way
different than the participants of Study 2, but we don’t know exactly in what way. This is the challenge that psychological research has facing the opportunities of big data.
The use of big data in the current study does enable this research to be the first on social class and prosocial behavior to use behavioral data in a natural setting. This has the advantage that, whereas previous research on social class and prosocial behavior lacked external validity (Piff et al., 2010; Korndörfer et al., 2015), there is no risk of experiment constructs, socially desired responding or other responding errors (Randall & Fernandes, 1991; Coughlin, 1990). We found evidence that indeed, what people report on how much they lend and borrow, does not reflect what how much they actually lend and borrow. So using natural behavior data, these results robustly contradict the negative relation
hypothesis postulated by Piff et al. (2010). Balancing Giving and Receiving Help
While upper class people may be more likely to help in comparison to lower class people, they’re also more likely to request help. This high level of activity among upper class individuals may be explained by the possibility that due to their elevated personal control (Johnson & Krueger, 2006; Kraus et al., 2009), upper class individuals adopt innovations (such as Peerby; see Rogers, 2010) sooner than lower class individuals, and are therefore more active on Peerby both in borrowing as well as in lending. Additional research must
assess whether personal control and innovation adoption mediate the positive relationship of social class with borrowing and lending.
Another way to experimentally assess how giving and receiving help is related, is to reverse the Dictator Game. A Reversed Dictator Game could communicate to participants that an experiment partner has received 10 points and is free to decide how to distribute it. The participant would get the possibility to ask their experiment partner for a portion of the points. Requesting help would then be measured with by the amount of points asked by the participant.
Economic Inequality as Moderator of Upper Class Prosociality
Apart from methodological problems with the evidence provided by Piff et al. (2010), there is a new theoretical explanation for the contrasting evidence in the literature so far. The contrasting evidence between the current study and those published by Piff et al. (2010) may be explained by the amount of economic inequality in the country where the research took place. While the data analysis of the current research was in progress, Côté et al. (2015) published findings reporting that economical inequality moderates the effect of income on generosity in the Dictator Game.
They found that the negative effect of social class on prosocial behavior only emerges when there is great economic inequality, specifically, when the Gini coefficient (representing economic inequality) is .485 or higher. Contrastingly, when the Gini coefficient is .454 or lower, they found that social class is associated with equal or more prosocial behavior. This pattern also arises when participants are experimentally induced to think there is high vs. low economic inequality in their home state. This explains why Piff et al. (2010), who conducted their studies in California (one of the most unequal US states; Wilkinson & Pickett, 2009), found a negative relationship between social class and prosocial behavior, while the current research found neutral and positive relationships, conducting research in the Netherlands. Here, the Gini coefficient is .251, signifying high economic equality.
Therefore, in the current study, the model proposed by Côté et al. (2015) holds up. However, in order to conclude that the positive relationship between social class and prosocial
