Which Features of Prosocial Spending Recollections Predict Post-Recall Happiness? A Pre-registered Investigation Free

We conducted a pre-registered examination and used the largest known collection of spending recollections gathered to date from Aknin and colleagues’ (2020) Registered Replication Report examining the emotional benefits of prosocial spending. As part of this previous report, Aknin and colleagues (2020) recruited a sample of 5,199 participants and randomly assigned them to either recall a previous prosocial or personal spending experience. Participants reported both their baseline and post-recall levels of happiness and were also asked to rate how happy they were at the time of purchase. Over 2,500 participants described a recent prosocial spending memory. We coded each prosocial spending memory for information on whom participants spent money on and what they purchased. Informed by Self-Determination Theory (Ryan & Deci, 2000), we also coded each spending memory for the extent to which the description contained cues of affiliation, volition and impact. Moreover, we also coded for authenticity and detail. We pre-registered our hypotheses that higher levels of each of these features would be associated with higher levels of post-recall happiness as reported by spenders while controlling for the spenders’ baseline well-being. Taken together, these data offer greater insight into how people spend their money on others, and what features of prosocial spending recollections are associated with greater well-being. Our pre-registration, data, and syntax can be found on the Open Science Framework (OSF) at the following link: https://osf.io/s4hqf/.

As part of a previous study (Aknin et al., 2020, Study 3), a total of 5,199 participants were recruited on Amazon’s MTurk in exchange for a small financial payment. Participants reported their baseline well-being and were then randomly assigned to recall a time they spent approximately $20 on either themselves (personal spending; n = 2,614) or on someone else (prosocial spending; n = 2,585). Afterward, participants reported their momentary well-being on several measures described in more detail below. We focus on responses from participants in the prosocial spending condition. We use a final sample from 2,572 participants (M_age = 38.89, SD = 11.8; 58.1% female, 41.5% male, .3% other) who were randomly assigned to the prosocial spending condition. In line with our pre-registered exclusion criteria, this final sample excluded 12 participants who were assigned to the prosocial spending condition but reported a personal spending experience and one participant who did not provide a spending memory at all.

As noted in Aknin et al. (2020) and the accompanying pre-registration (https://osf.io/d6ymu/), participants were first asked to report their baseline happiness on two items assessing state (“Do you feel happy right now?”; 1–Not at all; 5–Extremely) and trait happiness (“In general, I consider myself…”; 1–Not a very happy person; 7–A very happy person; Lyubomirsky & Lepper, 1999). These items were standardized and averaged to create a measure of baseline happiness.

Participants were then asked to recall and describe a recent spending experience. Specifically, participants were randomly assigned to see one of the two following prompts:

Participants were required to use at least 150 characters in their spending description to encourage at least a minimal level of task engagement. The average word count (n=2572) was 59.45 (SD = 31.83) with a minimum of 14 words and a maximum of 413 words.

Following the recollection, participants were prompted to report their current positive affect on the Positive and Negative Affect Schedule (PANAS; Watson et al., 1988), along with the addition of the word “happy.” Consistent with Aknin and colleagues (2020) and as noted in our pre-registration, we use the average of the 11 positive items (10 from the PANAS and “happy”) as our main outcome variable. It is worth noting, however, that participants also reported both their current positive emotion and their recalled positive emotion at the time of purchase on the Scale of Positive and Negative Experience (SPANE; Diener et al., 2009).

Each spending memory was coded along several dimensions (see Table 1).

Several details of each spending memory were coded for exploratory investigation. For instance, we coded the spending target (on whom participants spent money) and spending content (what participants bought) because we thought it would be useful to provide future researchers with descriptive information and insight into how people spend their money on others, on average.

A team of three independent coders, blind to hypotheses and participants’ self-reported happiness, indicated whether each spending memory referred to any of the following spending targets: (1) oneself, (2) a parent, (3) a sibling, (4) a child, (5) a significant other/romantic partner, (6) a friend, (7) a colleague, (8) extended family, and/or (9) someone that the participant did not know. We also included an additional coding option for “someone else” to flag purchases made for another person with an unknown or undisclosed relationship with the participant, but reliability for this dimension was low (a = .55). Coders were asked to check all that apply (1 = yes, 0 = no) and only mark “yes” when there was clear evidence for a particular target relationship category. For example, if a spending memory stated, “I bought a coffee for Sam,” coders were asked not to guess Sam’s relationship to the participant (i.e., a friend, sibling, colleague), but rather select “someone else” on the coding guide.

Similarly, coders indicated whether each spending memory referred to any of 12 key types of purchase content, including: (1) personal necessities, (2) food, (3) clothing, (4) medical/health related items or costs, (5) gifts, (6) personal resources, (7) tickets to something, (8) an activity/experience, (9) toys/games, (10) bills, (11) money, or (12) sporting equipment. Once again, coders were asked to check all that apply (1 = yes, 0 = no) and to only code if there was clear evidence for a particular content category included in the memory.

Considering past research and theory (Aknin & Whillans, 2021; Dunn et al., 2014; Lok & Dunn, 2020; Ryan & Deci, 2000), we predicted that spending recollections with higher levels of affiliation, volition, impact, authenticity, and detail would be associated with greater happiness. As a result, we pre-registered our intention to use directional, (i.e., one-tailed tests) when investigating these relationships.

Affiliation, authenticity, and level of detail were each coded by a computer text analysis software called the Linguistic Inquiry and Word Count Program (LIWC; Pennebaker et al., 2015). LIWC is a text analysis program that quantifies the degree to which various constructs are found within a text. For affiliation, LIWC calculates the number of times the text refers to another person with terms like “ally” or “friend”, and divides this number by the length of text. Similarly, LIWC codes authenticity through the disclosure of additional information and by counting the number of times that the text refers to words previously shown to reflect personal detail and honesty, and then divides by the length of the text. Finally, level of detail is indexed by the number of interrogatives within the text, including who, what, when, where, and how.

The other two dimensions –volition and impact – were rated by a team of three independent coders, blind to hypotheses and participants’ self-reported happiness. For volition, coders rated the extent to which the purchase was made out of obligation (obligatory; required) or made by choice (volitional; freely chosen) on a 7-point Likert scale (1 – Very much obligatory; 7 – Very much volitional). For impact, coders noted whether or not (1 = yes, 0 = no) the participant indicated that the purchase helped someone in need.

Because this data set offers the largest collection of prosocial spending recollections to date, we coded whom individuals spent their money on (spending target) and what they bought (spending content). As shown in Table 1, descriptions of participants’ last prosocial purchases revealed that people were most likely to spend money on a significant other (19.32%), friend (19.33%), or child (19.16%) and least likely to spend on a stranger they did not know (1.94%). Meanwhile, participants were most likely to buy gifts (64.59%) and food (38.07%). Interestingly, individuals in this sample were less likely to spend money on other “necessities” to help others, such as bills (.94%) and medical/health related costs (1.43%).

We followed our pre-registered analytic plan posted on the OSF (https://osf.io/s4hqf/) with two notable deviations. First, we implemented a Bonferroni correction to control the family-wise error rate. With five pre-registered directional (one-tailed) tests, we use an α = .02 as our threshold for statistical significance. Second, we handled missing values using pairwise deletion to retain as many cases as possible and thus maximize statistical power across our analyses.

As stated in our pre-registration, we predicted that spending recollections with higher levels of affiliation would be associated with higher levels of post-recall happiness. To test this hypothesis, we examined whether spending recollections containing higher levels of affiliation, as coded by LIWC, were associated with higher levels of post-recall happiness on the PANAS while controlling for baseline well-being using a partial correlation. In contrast to our prediction, analyses revealed no association between affiliation and post-spending positive affect, r(2515) = -.00, CI₉₈ [-.05, ∞], p = .581, one-tailed (see Table 2). Thus, participants expressing higher levels of affiliation in their spending recollections did not appear to report higher happiness afterward.

We pre-registered the prediction that spending recollections which contained higher levels of volition would be correlated with higher self-reported happiness. To test this hypothesis, we used a partial correlation to examine whether spending memories coded to have higher levels of volition were associated with greater post-recall happiness on the PANAS while controlling for baseline well-being. As predicted, we found a small positive correlation, r(2564) = .07, CI₉₈ [.03, ∞], p < .001, one-tailed (see Table 2). These results indicate that participants expressing higher levels of volition in their spending recollections reported slightly higher happiness after recalling their act of prosocial spending.

As noted in the pre-registration, we predicted that spending recollections reflecting higher levels of impact would be associated with higher levels of post-recall happiness. To test this hypothesis, we used a partial correlation to examine whether spending memories with higher levels of impact as coded by three independent coders, were associated with higher levels of post-recall happiness while controlling for baseline well-being. In line with our prediction, analyses revealed a small association between impact and post-recall positive affect, r(2565) = .05 CI₉₈ [.01, ∞], p = .003, one-tailed (see Table 2). Thus, participants expressing higher levels of impact in their spending recollections did appear to report higher happiness afterward recalling their act of prosocial spending.

As stated within the pre-registration, we predicted that prosocial spending recollections containing higher levels of authenticity would be associated with higher levels of post-recall happiness. We used a partial correlation to examine whether spending recollections containing higher levels of authenticity, as coded by LIWC, were associated with higher levels of post-recall happiness while controlling for baseline well-being. In contrast to our prediction, analyses revealed a non-significant negative association between authenticity and post-recall positive affect, r(2515) = -.05, CI₉₈ [-.09, ∞], p = .992, one-tailed (see Table 2).

Finally, as noted in the pre-registration, we predicted that spending recollections containing higher levels of detail (interrogatives) would be associated with higher levels of post-recall happiness. To test this hypothesis, we used a partial correlation to examine whether spending memories containing higher levels of detail, as coded by LIWC, were associated with higher levels of post-recall happiness on the PANAS while controlling for baseline well-being. In contrast to our prediction, analyses revealed no association between level of detail and post-spending positive affect, r(2515) = .04, CI₉₈ [-.00, ∞], p = .030, one-tailed (see Table 2). Thus, participants who provided greater detail in their spending recollections did not report higher happiness afterward.

One potential alternative explanation for the above results may be that people who are simply more engaged with the task are also more likely to report both greater post-recall happiness and greater details about the volition and impact of their prosocial spending experience. Thus, we repeated our five primary tests and additionally controlled for word count to examine whether our findings remain robust to this potential third variable. As above, we used a Bonferroni correction to control the family-wise error rate in this set of exploratory, post-hoc, directional tests; we use an α = .02 for our threshold of statistical significance. Similarly, we handled missing values using pairwise deletion to maximize statistical power across our analyses.

Overall, controlling for word count did not meaningfully alter the conclusions of most of pre-registered analyses (see Table 2). Indeed, when word count was included as an additional predictor, greater levels of volition continued to predict greater post-recall happiness, while levels of affiliation, authenticity, and detail did not predict post-recall happiness. However, the association between impact and post-recall happiness weakened somewhat when controlling for word count. Thus, engagement may in part explain the relationship between impact and post-recall happiness.

We coded over 2,500 prosocial spending recollections to learn more about how people engage in everyday acts of generosity and whether various theoretically motivated dimensions conveyed in the memories predicted spenders’ post-recall happiness. As hypothesized, we found that participants expressing higher levels of volition and impact in their recollections also reported higher levels of happiness after recalling their generous acts. Meanwhile, levels of affiliation, detail, and authenticity expressed in spending recollections did not predict higher levels of post-recall happiness.

Importantly, the positive associations we observed between happiness, volition, and impact emerged when we controlled for baseline happiness reported by the participant before recalling their spending memory. Including baseline happiness as a control variable helps to reduce concerns of reverse causality – the possibility that happier people are more likely to engage in generous acts that were freely chosen or impactful. Thus, while our analyses are correlational in nature and reveal small effect sizes that require cautious interpretation, it seems possible that acts of generosity that are recalled to be more volitional and possibly more impactful could lead to greater happiness upon reflection.

With this possibility, the present findings converge with past experimental research demonstrating the importance of volition and impact for reaping emotional rewards from prosocial action. As noted above, previous research has shown that people randomly assigned to give in ways that are freely chosen or have a clear positive impact on others report higher levels of happiness afterward than people randomly assigned to give in ways that are pre-determined or have unclear benefits for others (e.g., Aknin et al., 2013; Weinstein & Ryan, 2010). Similarly, these findings align with previous recollection experiments in which participants report greater happiness after recalling a prosocial act that was freely chosen or clearly helped others then when recalling a prosocial act that was required or ineffective (Lok & Dunn, 2020). Here, we extend upon these findings by demonstrating that prosocial recollections naturally containing higher levels of the same key factors – volition and impact – are also associated with higher levels of happiness upon reflection. In doing so, this work captures the various ways that people can engage in prosocial spending in real life, which allows for a full spectrum observation of these dimensions.

Counter to predictions and past work (Lok & Dunn, 2020; Titova & Sheldon, 2021), we did not find that spending recollections containing more cues of affiliation, authenticity, and detail were associated with greater post-recall happiness. One reason for these null effects could be that participants’ spending recollections were quite brief (M = 59 words) and some dimensions, such as concepts of affiliation and authenticity, could be hard to convey in this limited space. It is also possible that using LIWC to code affiliation by counting the number of times that other people were mentioned in a recollection may not be a sensitive measure of relatedness. Indeed, relatedness is typically measured through self-reports of how connected and attuned participants feel to others. Thus, it is possible that human coding is a more sensitive instrument to assess overall feelings of connection that spenders feel towards the recipients. Overall, low, or imprecise estimates of these dimensions may help to explain the lack of association.

In addition to coding several theoretically motivated dimensions of prosocial spending, we also considered what people bought and for whom, providing greater insight into how people engage in small, common forms of generous spending. This exploratory investigation revealed some intriguing patterns within the findings. For instance, this large sample reported spending almost equally (~19%) on their children, a significant other/partner, and their friends (see Table 1). In addition, many people discussed occasions on which they bought food for others (approx. 38%). Notably, this rate is strikingly similar to that observed in previous research by Aknin and colleagues (2013) who coded spending memories from participants in Uganda and Canada and found that 34% and 47% of prosocial spending memories, respectively, mentioned purchasing food for others. These converging results suggest that food sharing may be a relatively common form of generous spending in diverse contexts around the globe.

Other features of the descriptive data align with past work and raise new questions. For instance, our coding of over 2,500 prosocial spending memories indicates that 65% reference a gift while only 4% mention providing a personal necessity. Considering past research suggesting that givers systematically misestimate what recipients truly want, such as by overlooking practical and requested gifts in favor of the unfeasible and unrequested items (Galak et al., 2016; Gino & Flynn, 2011), these data may shed light on the frequency of this mismatch.

Another notable feature of the spending memories coded here was that 52% referenced spending money to help a target of an unknown status. This finding suggests that unless specifically instructed, many participants who recall their past prosocial spending memories may not regularly describe the target, including the target’s relationship to the participant. As such, if future researchers utilize the recall paradigm to examine the emotional consequences of giving to specific targets, such as romantic partners, friends, and colleagues, they may need to directly instruct the spender to talk about the nature of their relationship to the target, or include a separate question to collect this information. Alternatively, researchers may wish to randomly assign spenders to describe spending on particular recipients. Taken together, these and other features of our data raise interesting questions to be explored both within this dataset and in future research. We hope that by posting our data and code on the Open Science Framework (https://osf.io/s4hqf/), we will inspire others to use and add to this valuable dataset and this body of research.

Perhaps one of the most fascinating features of the present work is that human coders and a textual analysis program were able to reliably glean insight into several dimensions of everyday forms of generous spending from minimal text. Indeed, human coders trained in <10 hours coded concepts, such as volition and impact, with moderate levels of agreement (ICC(2,3) = .64 and .75 respectively). In turn, coder ratings of these brief snippets of text were able to predict the spenders’ post-recall happiness. These initial analyses offer exciting promise. As psychology begins to consider the possibilities of textual analysis, researchers may be able to derive new insights into the human experience (Jackson et al., 2021). We hope this work inspires others to use various technologies and human powered coding to glean new forms of information.

This project is not without limitations. First, the findings we present here are correlational in nature and should therefore be interpreted with caution. Second, because we focused exclusively on coding prosocial recollections, we were unable to compare if and how these same dimensions of interest predict happiness after recalling an act of self-directed spending. While a substantial body of previous research on Self Determination Theory (e.g., Ryan & Deci, 2000) suggests that many of these dimensions, such as volition, may predict greater well-being in most domains, it would be worthwhile examining whether these factors are more (or less) valuable in the realm of generosity compared to spending in other domains.

Third, as mentioned above, participants only provided brief spending recollections of the last time they spent $20 on someone else. Descriptions of $20 purchases may provide a limited window into everyday acts of generosity because they are tied to a particular amount and simply reflect the most recent (or easily recalled) experiences. That said, we suspect that $20 spending memories are valuable for several reasons. One reason is that this dollar value appears to be a familiar, accessible, and meaningful amount to many people. Nearly all participants were able to provide a response to this prompt, suggesting that $20 prosocial spending choices occur and can be recalled. As a result, recent and memorable accounts of generous actions help to index the ways in which people help others. In fact, these relatively mundane experiences may provide a conservative test of theory because more detailed accounts of extraordinary generosity would likely reveal stronger associations (e.g., Brethel-Haurwitz & Marsh, 2014). Another reason that $20 spending memories may be useful is that this dollar value is consistent with previous research on prosocial spending (e.g., Aknin et al., 2012, 2013, 2020). Therefore, the information offered by the current investigation offers greater insight into the spending targets and content of familiar forms of spending examined to date, as well as when these memories may be associated with more/less happiness.

Finally, while two of the five predictors (impact and volition) examined here were significantly associated with higher levels of post-recall happiness, the correlations were small (rs < .07, with lower bound estimates reaching .03). That said, we think the correlations reported here are meaningful because they align with past research, the relationships are consistent with pre-registered hypotheses, and the findings emerge when controlling for baseline well-being, thereby minimizing concerns of reverse causality. More broadly, our detection of small effects is consistent with recent theorizing arguing that many complex facets of human behaviour are multiply determined, leaving any single predictor unlikely to account for a large degree of variance (Götz et al., 2021). The low correlations reported here suggest that other variables may play a much larger role in predicting post-recall happiness.

While the correlations reported here are undoubtedly small, the frequency with which many people engage in prosocial spending provides an opportunity for small associations to accrue over time (Aknin et al., 2020; Dunn et al., 2008). Indeed, as Funder and Ozer (2019) recently explained, a correlation on the magnitude of r = .05 offers little explanatory value in a single, solitary event. However, when this association is considered at scale – through numerous prosocial spending memories accumulated in a year or lifetime – details regarding the volition and impact of one’s generous spending may have greater emotional relevance.

What features of everyday generosity lead givers to feel good? We coded recollections of generous purchases from over 2,500 people and along five key dimensions (affiliation, volition, impact, authenticity, and level of detail) to see which were associated with greater happiness. As predicted, people reported greater happiness after recalling generous purchases that were rated as high in volition and impact. Meanwhile, and in contrast to our predictions, people did not report higher levels of happiness when their spending memories contained more cues of affiliation, authenticity, and detail. These data provide insight into how people spend money on others and what types of spending memories are associated with more happiness. More broadly, these data offer researchers a simple first look at everyday human action – a valuable but often overlooked feature of meaningful and progressive science (Rozin, 2001).

The authors thank Armaghan Aliabadi, Devan Gill, Ira Rishi, Rachel Pham, and Zohra Kantawala for their invaluable assistance!

The Authors have no competing interests.

All the stimuli, presentation materials, participant data, and analysis scripts can be found on this paper’s project page on OSF at the following link: https://osf.io/s4hqf/

Contributed to conception and design: TAC, LBA

Contributed to acquisition of data: TAC, LBA

Contributed to analysis and interpretation of data: TAC, LBA, JDEP

Drafted and/or revised the article: TAC, LBA, JDEP

Approved the submitted version for publication: TAC, LBA, JDEP