Interactions between donor Agreeableness and recipient characteristics in predicting charitable donation and positive social evaluation

Participants

A nation-wide sample (n = 387) was recruited through Amazon’s Mechanical Turk service, 284 of who completed all parts of the study and provided usable data (183 women [64%]; 208 white [73%], 24 Asian or Pacific Islander [8%], 22 black [8%], 13 Hispanic [5%], 17 other or unknown [6%]; mean age = 33.5, range = 18–81, sd = 12.3). Previous studies have demonstrated that Mechanical Turk samples are more diverse than, and produce comparably reliable results to, traditional undergraduate samples (Buhrmester, Kwang & Gosling, 2011; Paolacci, Chandler & Ipeirotis, 2010). To maintain high data quality, we restricted participation (via MTurk’s built-in selection tools) to current US residents who had previously had at least 90 of their tasks approved. All participants read an informed consent sheet describing the study and all associated risks and benefits, and were required to click on a button that said “I Agree” before they were able to continue. Each participant was paid $1 for participation plus an additional $1 endowment to be kept or donated as desired. Participants who lacked variance in the outcome variable of interest for each analysis (i.e., who gave the same response on every trial) were excluded; thus the effective sample size for each analysis ranged from n = 200 to n = 284, depending on outcome measure (see below). Non-overlapping analyses of these participants’ data have been reported elsewhere (Y Ashar et al., 2015, unpublished data). All procedures were approved by the University of Colorado Boulder IRB (application #10-0210).

Personality measure

Personality was assessed using the Analog to Multiple Broadband Inventories (AMBI)—a 181-item public domain instrument that accurately recaptures scores on over 200 scales from 8 different broadband inventories—including the NEO-PI-R, HEXACO-PI , TCI-R, and several other widely used measures (Yarkoni, 2010). The AMBI was developed by applying a genetic algorithm to a large dataset drawn from the Eugene-Springfield community sample acquired by Goldberg and colleagues (Goldberg et al., 2006). Extensive convergent and discriminant validation analyses reported in Yarkoni (2010) demonstrated that the AMBI recaptures scores in 8 different broadband measures with extremely high fidelity despite the large reduction in length (a maximum of 5 items per scale). To minimize the number of comparisons performed in the present study, we focused exclusively on the AMBI version of the NEO-PI-R (Costa Jr& McCrae, 1992)—a widely used broadband inventory that provides scores for five major domains of personality (the so-called “Big Five” dimensions of Neuroticism, Extraversion, Openness, Agreeableness, and Conscientiousness) as well as 30 lower-order facet scales.

Procedure

Participants recruited through Mechanical Turk were directed to a custom online experiment written by the authors using the Ruby on Rails web framework. After providing consent and filling out basic demographic information (age, gender, and race), each participant was presented with 16 unique, realistic four-sentence biographies describing an individual in need along with a photograph and the name of an associated charity. We dynamically generated unique biographies in order to elicit a broader range of evaluative responses, relative to a small, static set of stimuli. Each biography was randomly constructed from a pool of 256 candidate sentences focused on one of 16 ‘hardships’ (e.g., suffering from cancer, having a substance abuse problem, etc.), and was randomly paired with one of 40 photographs and 24 names (photographs were paired with gender-consistent names). In total, over 4,000 unique randomly generated biography-photograph configurations were presented to participants during the course of this experiment, out of a total possible 280,000 biography-photograph configurations. A sample biography is: “Daniel has lung cancer. He only smoked cigarettes on weekends when he went out in high school and college. He attends church every Sunday. He has a reputation as somewhat of a gossip.” Additional samples are provided in Table S1.

Participants rated each of the 16 targets on 22 items (listed in Table S2) chosen to assess 7 different attitudes toward recipients, including: (1) feelings of tenderness, (2) personal distress over the other’s plight, (3) perceived neediness, (4) blaming the other for their suffering, (5) likeability of the other, (6) self-similarity to the target, ¹ and (7) overall intent to help. These dimensions were chosen because of prior literature demonstrating their importance in prosocial responding to needy others (Batson et al., 2005; Batson, 2011; Eisenberg, Fabes & Spinrad, 2007; Greitemeyer & Rudolph, 2003; Vollhardt & Staub, 2011).

Participants additionally made a donation to each biography from their endowment (between $0 and $1, in 10-cent increments). To encourage participants to make independent choices on each trial, they were instructed that only one donation trial would be randomly selected, subtracted from their endowment, and donated.

Normative recipient ratings

Because every single biography presented to participants was dynamically generated and unique across the entire study, we generated normative attitudes and donation amounts for each biography using a predictive model. For each biography, we predicted ratings on the 6 attitudes (i.e., tenderness, distress, neediness, blame, likeability, intent to help) as well as charitable donation amounts from the explicitly manipulated properties of the stimuli (the 16 hardships, 256 sentences, 40 faces, and 24 names used to construct the biographies—representing 336 indicators in total). We used ridge regression—an L2-penalized variant of least-squares regression that is robust in the face of large numbers of correlated predictors.

The resulting estimates represent our best prediction as to the ratings and donation choices that the average participant would make when presented with each uniquely generated biography. Importantly, these estimates were based on the full dataset rather than on the research team’s subjective judgment or the results of a much smaller pilot study. In a five-fold cross-validated analysis (to mitigate overfitting) the model recaptured a substantial portion of the variance for all dimensions (Table S3).

Multilevel model

For each of the 7 outcome variables, we fit a multi-level model using the lme4 package for R (Baayen, Davidson & Bates, 2008) in order to quantify the relative contributions of normative attitudes/behavior toward recipients, participant Agreeableness, and their interaction. Formally, the model can be expressed as: ${\displaystyle Y_{ts}={\gamma }_{00}+{\gamma }_{01}\left({\mathrm{Agree}}_s\right)+{\gamma }_{10}\left({\mathrm{Norm}}_{ts}\right)+{\gamma }_{11}\left({\mathrm{Agree}}_s\ast {\mathrm{Norm}}_{ts}\right)+u_{0s}+u_{1s}\left({\mathrm{Norm}}_{ts}\right)+e_{ts}.}$

Here, Y_ts is the observed rating of recipient t made by subject s; Norm_ts is the normative or expected rating for that recipient (produced by the ridge regression model described in the previous section); Agree_s is Agreeableness score for subject s; and e_ts is the residual error. The fixed effects γ respectively model the grand intercept, the subject-level effect of Agreeableness, the trial-level effect of normative rating, and the cross-level interaction between normative rating and Agreeableness. The random terms, denoted by u, represent the subject-level intercepts and subject-specific effects of the normative ratings.

Person-by-situation interaction

Table 1 summarizes the fixed-effects results from the multilevel model—that is, the main effects of normative attitudes or behavior towards recipients, of Agreeableness, and of their interaction—for each of the 6 attitudes plus donation amount. ² For all recipient characteristics, Agreeableness consistently influenced both the main effect and interaction. Main effects were in the expected direction in all cases—i.e., highly Agreeable people were more likely, on average, to perceive recipients as more likeable, less responsible for their troubles, etc. To facilitate interpretation of the interaction between participant Agreeableness and normative attitudes toward recipients, Fig. 2 displays subject-level regression lines alongside summary lines for high- (+1 SD) and low (−1 SD)-agreeable subjects for the dimensions of perceived likeability and amount donated (for other dimensions, see Fig. S1). For all outcomes, Agreeable participants were disproportionately more charitable and prosocial toward targets normatively considered more deserving of donation and prosocial—supporting the hypothesis illustrated in Fig. 1C—rather than being more forgiving of normatively disliked targets, or indiscriminately being more charitable towards all recipients (cf. the competing hypotheses depicted in Figs. 1A and 1B).

Comparison with other personality traits

Although our hypotheses pertained specifically to Agreeableness, for exploratory purposes, we repeated the above analyses for the other four Big Five domains (Extraversion, Neuroticism, Openness, and Conscientiousness). Results revealed systematic relationships between each trait and behavior (Table S3). For example, Extraversion was positively associated with mean ratings of tenderness and likeability, and Openness showed robust main effects similar to Agreeableness on most attitudes (i.e., highly Open people rated targets more favorably), but no interactions with normative attitudes. Interestingly, Conscientiousness interacted with recipient characteristics in a manner similar to Agreeableness, but showed no main effect (i.e., highly Conscientious people were no more likely to rate recipients as more needy or likeable on average, but their responses were generally more extreme at the low and high end than low-Conscientiousness participants). Importantly, including all other Big Five dimensions, as well as gender and age, as subject-level covariates in our multilevel models did not appreciably change any of the Agreeableness results reported above (Table S4).

Facet-level analysis of Agreeableness

Agreeableness is a broad dimension of personality that reflects a wide range of cooperative and affiliative behaviors; in the NEO-PI-R representation of personality, for example, Agreeableness comprises 6 narrower ‘facets’ that some researchers conceptualize as distinct aspects of Politeness and Compassion (DeYoung, Quilty & Peterson, 2007). We hypothesized that the trait and trait-by-situation effects we observed were likely to be driven primarily by the facets related to compassion for other people (Trust, Altruism, and Tender-Mindedness) rather than the facets associated with acquiescence and politeness (Straightforwardness, Compliance, and Modesty). Facet-level analyses provided modest support for this hypothesis: main effects and trait-by-situation interactions were somewhat more robust for the former facets—and particularly for Tender-Mindedness and Altruism—though they were present to some extent for the latter as well (Table S5).