The evolution of preferences and charitable giving: a panel study of the university years

Abstract

Economic preferences are often taken as given, yet evidence shows that preferences respond to life events and change over time. We examine the evolution of other-regarding preferences for a cohort of university students over 5 years, starting before they matriculate and extending one year beyond graduation. Using survey and incentivized measures of preferences, we show that altruism declines over the university years. This decline is reflected in changes in charitable giving over three donation opportunities. We rule out several alternative explanations for the observed change, including cohort differences, perceptions of the charities, and experience with experiments. We show evidence of a ‘giving type’ in charitable giving, with consistency in behavior across giving opportunities. Methodologically, we also show that the incentivized and survey measures are similar at predicting giving types. We conclude that preferences reflect common tendencies over time, while simultaneously showing an overall decline in generosity during the university years.

1. Introduction

A critical part of social development occurs during the 4 years when young adults attend university. In the transition to adulthood, college students are shaped by their teachers and peers and develop life-long values and habits of behavior. Among these habits is charitable giving. In this study, we examine the evolution of altruistic preferences and charitable giving through the 4 years of attending a university and in the years following graduation. The sample is a unique a panel of Rice University students who matriculated in 2016.

A number of scholars have focused on the relationship between age and charitable giving. Using samples of adult subjects of varying education and age, List (2004) shows that, across several different settings, there is a positive correlation between age and other-regarding behavior. Older subjects (>49 years) are more generous than their younger (<19 years) or middle-aged counterparts. As well, the correlation between age and altruism has been explored among children and young adults. For school-age children (6–14 years), Harbaugh and Krause (2000) show only a very small positive effect of age. Similar results are shown in Benenson et al. (2007) (age 6–9 years) and Fehr et al. (2008) (age 3–8 years); both show an increase in dictator game giving with age. In addition, Fehr et al. (2013) show an increase in altruism between ages 8 and 17 years. In a meta-analysis, Engel also shows a positive effect of age on giving. All these studies broadly agree on a positive correlation between age and altruism.¹

While the association between age and altruism has been studied extensively, less attention has been given to the evolution of altruistic preferences among young adults attending a university. In contrast to the results for younger children or older adults, Eckel et al. (2011) show substantially higher levels of giving in a dictator game for high school, when compared with university, students, suggesting that giving declines from high school to college. However, because this study uses two independent samples, selection may play an important role, and it is not possible to infer that preferences change over this time for a given cohort. A handful of studies have investigated giving behavior with student samples focusing on the external validity of lab measures (Benz and Meier, 2008; Galizzi and Navarro-Martinez, 2019). These authors do not examine preference evolution among students over time. It remains largely unknown whether and how the altruistic preferences of young adults change in response to common experiences associated with the university years.

In this study, we explore the stability of altruistic preferences of undergraduate students through their university years. Our rationale for investigating this issue is fourfold. First, college students undergo important academic and nonacademic experiences during the time they spend in their college years. Four years of residential college experience and constant interactions with networks of friends and peers have the potential to shape the otherwise less-crystalized preferences of young adults. Second, most young adults live outside of their parents’ homes for the first time when they start college. While living away from parents means independence, it also brings responsibility. College students make everyday decisions on their own, such as what classes to take, which activities to participate in, etc., all of which contribute to experiential learning that can lead to changes in preferences. Third, college students often face a hard budget constraint when making day to day economic decisions. The experience of making ends meet through university years likely impacts their marginal utility from a dollar, and that in turn should be reflected in all economic decisions made by university students. Finally, we want to know whether the evolution of altruistic preferences translate into charitable giving decisions.

We exploit a unique dataset from a study that followed a cohort of at Rice University between 2016 and 2021 to investigate the evolution of other-regarding preferences through the university years. Our data are especially suited to explore preference evolution comprehensively for several reasons, summarized here and explained in more detail later:

• We collect measures of altruistic preferences for the entering class of 2020 prior to their arrival on the Rice University campus in the Fall of 2016, at the end of their university years in March 2020, and we repeat the same measures in July 2021, more than a year after they have graduated. This allows us to draw a before–after comparison for their time in college and allows us to observe the stability of those preferences once they leave campus. We adopt an inclusive approach to data collection in that we collect both a survey measure and an incentivized measure of altruism (standard dictator game) which allows us to compare preference change across measures.

• In addition to the class of 2020, we elicit the altruistic preferences of the entering classes of 2021, 2022, and 2023. Each of these cohorts are also followed up in 2020 and 2021. This helps us to verify that the class of 2020 is not systematically different from any other entering class at Rice University. Data from additional cohorts also allow us to check for any general changes in other-regarding preferences over time.

• We excluded a segment of the entering class of 2020 from the recruitment pool of the ‘before’ elicitation study conducted in 2016. This excluded or untouched sample (who are seniors in 2020) complete a near replication of our 2016 study at the end of the college term in 2020, allowing us to control for any experiment participation effect on the behavior of our panel. As with the other cohorts, we retest this group again in July 2021.

• During the study period (2016–2021), our panelists had the opportunity to make donations to three important causes: Australian bush fire relief in February 2020; US COVID-19 relief in April 2020; and COVID-19 relief in India in July 2021. We also have somewhat different, all-or-nothing donation decision from January 2021 that we use as part of a robustness check in a later section. The data from these real donation experiments allow us to explore the stability of altruistic preference over time and across domains.

• Our design lets us perform several robustness checks on our primary findings.

Using our data, we first explore if other-regarding preferences of undergraduate students change through the university years. Second, we test if the altruistic preferences of students are stable across measures and contexts and ask whether there is a ‘giving type’. Third, we test whether survey or incentivized measures are better at predicting giving behavior.

Our findings are as follows. First, our data suggest that college students become less other-regarding during the university years. Students in our study transfer lower amounts in a dictator game to their counterparts at the end of their university experience in 2020 than at the beginning in 2016. The weakening of other-regarding preferences is also observed in our participants’ self-reported survey measure of altruism. A key criticism of survey measures is that they are not incentive compatible and therefore, are likely to be inaccurate. Notwithstanding the criticism, we find that both the survey and incentivized measures change in the same direction. A comparison with other cohorts shows that our student panelists are not different from other cohorts. Using the untouched subsample, we also show that regular participation in economic experiments negatively impacts measures of other-regarding preferences. While that finding does moderate the extent of change in other-regarding preferences through college years, our major result of weaker preferences by the end remains.

Second, we examine the stability of preferences across measures. We find positive correlations for a given measure across time, and across measures. We also provide evidence of a giving type. Using factor analysis over donation decisions we sort participants by their degree of generosity and conduct additional analysis.

Third, we examine the relationship between the two altruism measures—survey and dictator transfer—and real donations. We find that both other-regarding preference measures from the end of the university years are correlated more strongly with the charitable contributions made for Australian bush fire relief and both forms of COVID-19 relief by the students through the study period than what is measured at the beginning in 2016, prior to their campus experience. This is an indication that preferences may have changed in a way that affects charitable giving. Finally, while the incentivized dictator-game measure is consistently stronger in predicting real donations, we question whether the additional cost of conducting these measures is worth the gain in accuracy.

Our robustness checks use an all-or-nothing charitable giving task on a subsample of our panel. We find similar results to what is found with the full sample. We also use a different sample of subjects, separate from our panel, who faced the same set of decisions, but not over the same span of time. Again, we find results similar to those from our panel.

Our findings have significance for experimentalists as well as fundraising agencies. For example, our finding that measures of other-regarding preference elicited from freshmen is different from those of seniors indicate that preferences are malleable, and that preferences elicited at one point in time may not correlate strongly with later decisions. Preferences are likely to change during major life events in ways that impact charitable giving. However, we see considerable stability beyond graduation, as students go on to get jobs and earn income, an indication that preferences formed in the university years may persist. Another key finding that change in the survey measure of other-regarding preference does almost as well in predicting charitable giving as the incentivized measure, should be comforting for social scientists, since it is not always possible to collect multiple preference measures.

For fundraisers, our result suggests that higher contributions can be expected from freshmen than seniors. The evidence suggests that a giving type emerges during university years, and this extends at least into their first post-university experience. As de Oliveira et al. (2011) show, our results also suggest that fundraisers are correct in using past donors to other charities as fundraising targets, even for recent college graduates.

In what follows, we present the experimental design and results. Section 2 presents the design and procedures of our experiment. Section 3 presents the results and Section 4 concludes.

2. Experimental design and procedures

The research reported here is part of a panel study examining the evolution of preferences (altruism, risk aversion, patience, competitiveness, loss aversion, ingroup favoritism) across the 4 years that a student is enrolled in a residential university. A cohort consisting of two-third of the entering class of 2020 was recruited and participated in 14 different study waves from 2016 to 2021.² In this article, we explore the evolution and stability of other-regarding preferences using a subset of the data collected during this time. We briefly highlight each of the waves below (also see Supplementary Appendix Section I, Table A1).

2.1 Wave 1 (July 2016)

Subjects were recruited to participate in the panel study in July–August 2016, after they were admitted to Rice University but before they arrived on campus. In the online study, subjects were told that it would take about 25 min, and that they would be compensated USD $5 for completing a survey and could earn additional money depending on their decisions in set of six incentivized tasks. Tasks were presented in a fixed order. Two of the tasks were selected at random for payment. No feedback was provided on the tasks not chosen for payment. A total of 661 subjects were contacted via email and 553 (83.7% response rate) completed this wave of the study. On average, subjects earned $26.79 for their participation. Excluding extreme outliers, it took subjects an average of 20 min to complete. This wave focused on eliciting economic preferences (altruism, risk-aversion, time preference, competitiveness, etc.).

2.2 Wave 2 (February 2020)

A total of 385 panel members participated in this wave of the study. Subjects were told there were two parts to the study. Part 1 was similar to Wave 1 in that there was a survey plus seven incentivized tasks, and it was also conducted online. These tasks were in a fixed order. Part 2 (not used here) was a network elicitation.³ Subjects were told that the study would take approximately 15 min. Excluding outliers, subjects spent slightly over 11 min in the study. On average, subjects earned $33.29 for both parts of the study.

2.3 Wave 3 (March 2020)

This study was designed to document changes in students’ preferences over the college years using the same measures as in Wave 1 (2016). A total of 406 panelists were in this study, which was conducted online. Study subjects were told that the study would take approximately 25 min. Excluding outliers, subjects spent an average of 23 min on the study. The study took place between 17 March and 11 April 2020. On 8 March, Rice University cancelled classes for a week in advance of Spring Break. On 12 March, the University announced that following the end of Spring Break (23 March) all classes would be taught remotely and students were instructed to leave campus by 25 March. Consequently, subjects were involved in this study during a very uncertain time and were not on campus.

2.4 Wave 4 (April 2020)

This portion of the study focused on how subjects were coping with the COVID-19 pandemic. A total of 397 panelists completed this wave.⁴ Subjects were told the study would take approximately 25 min and that there were three components of compensation. First, everyone was compensated $10 for completing a survey embedded in the study. Second, subjects would face five incentivized decision tasks from which one would be randomly selected for payment. Lastly, five participants from the study would be chosen randomly and paid an additional $100. (This latter payment was an attempt to enhance panel retention.) Excluding outliers, panelists spent just over 22 min in the experiment and earned an average of USD $17.06. This does not include the $2,457.84 given to charity.

2.5 Wave 5 (January 2021)

A total of 232 individuals from the original panel completed this wave. Subjects were compensated $5 for completing the survey, which included an embedded experiment concerning vaccinations. The survey was short, lasting 5 min. At the conclusion of the survey, respondents were asked whether they wished to donate their $5 earnings to the local Food Bank. They were presented with a binary choice—either donating or retaining their earnings. Because this donation task is quite different from the others, and not embedded in a larger study, we use it for a robustness check.

2.6 Wave 6 (July 2021)

A total of 218 individuals from the original panel (Wave 1) completed this wave. Subjects were compensated $10 for completing a survey and faced eight incentivized tasks, three of which are of interest to us in this article. Subjects were paid for one of the eight tasks selected at random. Embedded in the survey was the altruism survey measure. Subjects also participated in two dictator games as the dictator (the games were randomly ordered with a task separating them). The first game asked the subject to allocate $20 between themselves and a newly arriving freshman at Rice. This was designed to mimic the setting the panelists experienced in 2016 when they matriculated. The second dictator game asked them to allocate $20 between themselves and someone from the class of 2020 who is also a participant in the study. This mimics the dictator game played both in Wave 1 and Wave 3 in that the recipient was someone from their class. (Instructions for both versions of the dictator game are given in Supplementary Appendix Figures A14–A19.) Finally, subjects faced a real donation dictator game where they were asked to allocate $20 between themselves and a charity (UNICEF USA) providing relief for COVID-19 that was then raging in India. Subjects spent just under 27 min on the study and earned an average of $21.93. This does not include the $722.00 that was sent to the charity by subjects for whom this task was randomly selected for payment.

The primary measures used in the current study consist of a survey, incentivized dictator games with student recipients, and real-donation dictator games with charity recipients. The survey measure was created by the authors, based on the structure used in the 2004 wave of the German Socio-Economic Panel. Subjects were asked to respond on a scale of 0–10 to the following question: ‘How do you see yourself? Do you generally donate a lot of time and money to help others or do you focus primarily on taking care of yourself and family?’⁵ The instructions read, ‘Please use the scale from 0 to 10, where 0 means “Donate a lot to others” and 10 means “Focus on self and family.”’ The scale of this item is then reversed such that a higher value implies higher altruism toward others. This question is repeated in multiple waves of the study with the same wording. These measures are taken from Waves 1, 2, and 6.

The incentivized measure is a standard dictator game with an anonymous partner (Forsythe et al., 1994). The instructions asked subjects to allocate a fixed amount of money ($20) between themselves and another recipient, who was also a student at Rice University and a participant in the study. The participants made their choices among 11 discrete alternatives in $2 increments ($0 for themselves and $20 for the counterpart, $2 for themselves and $18 for the counterpart, and so on). If the task was selected for payment, then the role of decision maker or recipient was determined randomly.⁶ This game was repeated in several waves of the study, and a final time after subjects have left the university, with two different partners: a freshman at Rice, and a participant in the study from the same cohort. Instructions are included in the Supplementary Appendix. These measures are taken from Waves 1, 2, and 6.

The third measure is a real-donation dictator game (Eckel and Grossman, 1996). Here, subjects are given the opportunity to share an amount of money with a charitable organization. This measure has been widely used to study charitable giving.⁷ The organization differs across waves of the study. In all cases, we selected an emergency-related cause that was highly salient at the time of the wave and chose one or more charitable organizations addressing that cause. Thus, we selected the Australian bush fires in early 2020, COVID-19 equipment shortages in the USA later in 2020, and COVID-19 equipment shortages in India in 2021 (See Supplementary Appendix Figures A3–A5 for the distributions). These causes obviously differ, but each was selected because it was in the news at the time. To check the robustness of our results, we use an all-or-nothing charity choice for the local Food bank. These measures are taken from Waves 2, 4, 5, and 6.

3. Results

In the subsequent analysis, we focus on the 198 panelists who participated in all of the waves noted above.⁸ In additional analyses, we include smaller samples of students who matriculated in 2017, 2018, and 2019, as discussed below. These individuals were given the same survey and set of tasks noted in experiment 1 above. We also include the untouched seniors from the class of 2020 (approximately one-third of entering class) who participated in the same design as experiment 3, also in March 2020.

3.1 Changing altruism measures over time

Figures 1 and 2 present the distribution of the survey measure of altruism and dictator transfers before subjects arrived on campus (2016), at the end (2020) of the university period, and 1 year later (2021). The survey measure assesses subjects’ degree of altruism (other-regarding preferences) on a scale of 0–10. Using box-and-whiskers plots we observe (Figure 1) that there is a shift to the left in our panel for 2020 and 2021 compared with when they first arrived on campus in 2016. In other words, more students report a lower degree of altruism at the end of the college term than at the beginning. The same is true with the distribution of dictator transfers (Figure 2) from 2016 to 2021. Prior to arriving on campus subjects were more likely to send something to another entering student. The variance is lower in 2016 and slightly over 60% of participants sent half of their $20 endowment to their counterpart, who was another participant in the study.⁹ Five years later only 37% of the subjects in our panel sent half and 60% transferred less than $10. This result is more similar to typical findings for college students. Further, at the outset of their college career (2016) 10.1% of the panelists sent nothing. In 2020 and 2021, this was 23.2% and 29.9%, respectively. To control for the fact that subjects in 2016 were giving to freshmen like themselves, we included in the July 2021 two dictator game treatments, one where the transfer was to a member of the incoming class of Freshmen, and one where the transfer was to another participant in the study who was a member their own graduating class (the order of these two decisions was randomized). We find no difference in what was sent between these two types of recipients (t = 1.05, df = 197, p=0.29).

Table 1 presents summary statistics from the survey and incentivized measures of altruism. Both measures indicate a higher degree of altruism prior to arriving on college campus than at the end of the college term. Statistically, there is always a difference between the measures in 2016 and those in 2020 and 2021.¹⁰ We also find that within subject correlations are weakest between actions taken prior to arriving on campus and after leaving campus. The correlations are quite high in 2020 and 2021 between items that measure either attitudes or behavior, but not both. These correlations are given in Table 2.

3.2 Possible alternative explanations

The observed decline in altruistic preference of our sample could be due to factors other than their experiences through the university years. In this section, we consider several possibilities. For instance, it may be that there is something unique about the entering class of 2020. Second, it could be that the environment changed over time and that subsequent classes differ accordingly. Third, it is possible that, because our panel has participated in several economic experiments through the 4 years of university, this experience has led them to become more self-regarding. Finally, we cannot rule out the possibility that the onset of the COVID-19 pandemic, which coincided with our 2020 survey, shaped our subjects’ preferences.

First, we evaluate whether the class of 2020 was different from other entering classes. We test this by comparing the dictator transfers of the class of 2020 prior to arriving on campus with that of the entering classes of 2021, 2022, and 2023 at the same time. These latter classes, like the class of 2020, completed a set of preference elicitation tasks including a dictator game, prior to matriculating. We find no evidence that the class of 2020 is any more or less generous than the other classes at the beginning of their college term. The lower triangle of Table 3 provides pairwise t-tests of the different cohorts prior to arrival on campus. Aside from a significant difference between the class of 2022 and 2023, we find no differences in their initial allocations. The upper triangle compares each of the classes in their allocations in the last experiment carried out in July 2021. Again, we find no differences across the different entering classes. As with those in our panel from the class of 2020, we find less is given in 2021 than prior to their arrival on campus.¹¹ From these findings, we can conclude that the class of 2020 was not an outlier. We can also conclude that there is no effect of time for when the cohort matriculated. We observe exactly the same pattern in other classes.

Second, to investigate the role of repeated participation in economics experiments on subjects’ altruistic preferences, we compare both the survey response and dictator allocation for two samples of subjects from the same class: the panel, and the excluded sample. Recall that a portion of the sample from the class of 2020 was set aside at the beginning of the study to afford such comparisons. In March 2020, students from this excluded group (now seniors) were recruited and completed the same set of tasks as noted in the previous section for Wave 4. Initially, 257 individuals participated, and this dropped to 98 by the end of the study in July 2021. We report only those who participated in all tasks in 2020 and 2021.¹² As shown in Table 4, the mean survey measure of altruism for the excluded seniors is nearly the same as the panel, indicating that the decline in self-reported altruism seen in the panel is not due to repeated exposure to experimental procedures. As with the panel subjects, there is a good deal of stability in attitudes between 2020 and 2021.

We also compare the dictator allocation of our panel to that of the excluded seniors. If there is a weakening of other-regarding preferences through the university years unrelated to repeated exposure of our panel to economic experiments, then we should observe a higher proportion of the excluded seniors choosing favorable transfers for themselves than our panel at the beginning of the college term. Table 4 shows that the untouched seniors send more on average to classmates than the panelists, though the amount they send is still substantially below the amount sent by the panel cohort as freshmen in 2016 ($8.44). Notably, by the end of the study in July 2021 the excluded seniors give even less than the panelists ($5.88 vs. $6.20). This may be evidence in favor of a ‘participation-effect’. It is possible that with regular participation, earnings from experiments become more salient, mitigating the ‘house-money’ effect of experimental endowments. Several studies suggest that participants make more generous dictator transfers when endowments are more abstract (presented on computer screens) or windfall than when they are given in cash or is earned prior to playing the dictator game (Cherry et al., 2002; Reinstein and Riener, 2012). We argue that with regular participation in economic experiments, earnings are likely to become more tangible which partially explains the less generous dictator transfers by both the panel and the untouched seniors in 2021.

Finally, it may be that the pandemic changed altruistic preferences and behavior. We might expect that increased fear about COVID-19 would lead people to be less altruistic, conserving resources for themselves. On the other hand, a perception of greater need during the pandemic might lead to an increase in altruism, as donors respond to appeals for COVID-19-related causes. A number of studies have addressed the impact of the pandemic on altruism, and most have found an increase in altruistic preferences and/or behavior. For example, Fridman et al. (2022) use two longitudinal data sets to show an increase in altruistic behavior and preferences in the presence of COVID-19 threat in their geographic area in the USA. They use charity navigator data (2016–2020) to show an increase in giving to charitable organizations during the pandemic whose magnitude is positively correlated with higher levels of COVID-19 within a county. Their parallel dictator game study followed subjects over six months early in the pandemic (March–December 2022). This study shows an increase in dictator-game giving when the COVID-19 threat manifests in their county. Lotti and Pethiyagoda (2022) follow over 1,200 US citizens for 8 weeks during the pandemic (March–May 2021), and show that hypothetical dictator giving to relatives, neighbors, strangers, and to government all increase over time, with larger increases associated with more self-reported concern about COVID-19. Shachat et al. (2021) also show small increases in dictator game giving in repeated cross-section studies using student subjects in Wuhan, China as the pandemic progressed in early 2020. In an online experiment, Grimalda et al. (2021) find no effect of local pandemic intensity on real-donation dictator giving to a charity involved in COVID-19 relief. However, personal exposure to COVID-19 increased giving. This effect was the strongest at the local level when compared with national or international levels. Adena and Harke (2022) conduct an online real-donation experiment with 4,200 participants in the UK and find that mentioning COVID-19 affected giving to charitable organizations. Greater severity and media coverage at the local level also increased giving, but individual experiences with the disease decreased giving, in contrast to Grimalda et al. (2021).

On the other hand, a negative result was found in one study. Brañas-Garza et al. (2022) examine giving by about 1,000 participants during 6-day period in early 2020 in Andalusia, when COVID-19 was increasing rapidly and lockdown measures were being implemented. They use a snowball sample, with earlier participants recruiting later participants. The study used several different tasks with participants earning ‘points’ consisting of lottery tickets for a 100 Euro prize. Participants could choose to allocate their prize for participating to ‘an NGO’ which was not named; two participants would be selected to win the prize. The study compares early with late participants over the 6 days, and finds that later participants contributed less to the organization. However, late participants might differ from early ones in a number of ways, given the recruitment method.

These studies differ from each other in location, presence and size of stakes, recruitment method, sample, recipient identity, (individual or a variety of charities), measure (games vs. surveys), and procedural details, making it difficult to compare them directly. But most of the evidence and the overall pattern is one in which COVID-19 leads, if anything, to an increase in giving, both to other individuals and to charitable organizations, especially those organizations involved in COVID-19 relief. This is in stark contrast to our study, which finds a decline in giving over a period that includes the pandemic.

While our study was not designed to capture the immediate effect of COVID-19 on generosity, we can test for differences in giving over time as due to concern with COVID-19 using survey questions. In Wave 3 (March 2020), as the disease was emerging in the USA, we asked a series of questions about how worried subjects were about COVID-19 and whether they were worried about infecting someone else. The items used a Likert scale, and asked whether subjects agreed or disagreed with a series of statements about COVID-19.¹³ The items do not scale together (Kronbach’s alpha = 0.34) so we treat them as separate dimensions of concern. To test for impact on changes in dictator giving, we use two measures of change in giving. First, we subtract what an individual sent in Wave 3 (2020) from what was sent in Wave 1 (2016). A second measure subtracts what an individual sent in Wave 6 (2021) from what was sent in Wave 1 (2016). This captures a longer interval of exposure to the pandemic. We then regress the changes in contributions on the two items. In addition, we take items from Wave 6 measuring their actions with respect to COVID-19—their views about vaccinations or avoidance behaviors.¹⁴ If concern with COVID-19 is the source of reduced average donations in our sample, then we ought to expect less to be sent as subjects express more concern using any of our survey items. This is not the case. The coefficients for our estimates carry negative signs for the ‘worried’ item, vaccinations, avoidance behavior, and positive signs for the ‘infect others’ item (see Supplementary Appendix Table A4). However, all the items have a small effect and are statistically insignificant. Thus, we conclude that the changes in giving are unlikely to be caused by reactions to the COVID-19 pandemic. We take this as further evidence that the change in altruistic preferences is something that happens over the course of subjects’ college experiences.

3.3 Charitable giving

We now turn to charitable giving. Our panelists had the opportunity to make donations for three different charitable causes between 2020 and 2021. In Wave 2, students could donate toward bush fire relief in Australia in a real donation dictator game. In Wave 4, students were asked whether they would like to donate a part or all their earnings from a survey toward COVID-19 relief. In Wave 6, subjects could donate to COVID-19 relief in India using a real donation dictator game. In the Supplementary Appendix (Figures A3–A5), we plot the separate distributions for charitable giving for all three targets. We then ask whether subjects give consistently across opportunities for charitable giving. Once we do so, we examine whether survey-based or behaviorally-based measures best predict charitable giving.

We look at three charity choices made by our respondents in 2020 and 2021. Following de Oliveira et al. (2011), we test for the presence of a ‘giving type’ in our student population. We use confirmatory factor analysis of subjects’ contributions to uncover a single dimension. Using the factor structure, we calculate factor scores. These are plotted as a histogram in Figure 3. Negative values are individuals who consistently give little to nothing to charities. Individuals clustering around zero indicate mixed giving. Finally, individuals with scores toward +1 indicate those who consistently give a great deal.

These findings indicate considerable heterogeneity, from those who consistently give little or nothing to those who consistently give larger amounts, with modes at either end. The fact that a single dimension is recovered when including all the charity items in the factor analysis provides evidence that there is stability in preferences for giving.¹⁵ If the items were unstable, more than a single factor would be recovered. This ‘giving type’ factor score is used in our analysis below.

3.4 Correlations among giving behaviors

In Table 5, we present the Pearson correlation coefficients among the giving behaviors of our study participants and their dictator game giving. We exploit the panel nature of our data to show how well the behavior correlates with our incentivized measure of other-regarding preference taken both at the beginning of the panel (2016) and after their university years (2021).

We find positive and statistically significant correlations among our students’ giving behavior for all four instances of charitable giving. Moreover, the dictator transfer measures from 2020 and 2021 have stronger correlations with donation choices than the measure from 2016. Correlations among the charitable donations are consistent at around 0.5. Supplementary Appendix Table A2 shows similar patterns of correlations for the survey-based altruism measures and charitable giving.

3.5 What predicts giving?

Because of the panel structure of our data, we can examine predictors of charitable contributions over time. Our approach is to first examine the relative performance of the survey and dictator measures of altruism in predicting the factor score from the charitable giving decisions. Table 6 contains regression coefficients from separate regressions for 2016, 2020, and 2021 altruism measures. For all regressions, the dependent variable is the individual factor score for charitable giving. The coefficients show the relationship between the two altruism measures and the factor score. When both measures are included in the regressions, they have similar impacts on charitable giving. (Supplementary Appendix Table A5 contains coefficients for these regressions run with only one measure included at a time and shows a similar pattern.) This suggests that both measures can predict charitable giving. More recent measures are also better at predicting the giving factor score, compared with the 2016 measures, suggesting that preferences have changed over the university years, with the latter measures more closely related to contemporaneous charitable giving.

Table 7 takes a different approach. Here, we construct a panel data set for all donation decisions, dictator transfers and real donations. Thus, for each subject there are six observations. The table contains estimates from a random-effects panel regression. The dependent variable is the giving amount, and the independent variables shown in the table are: a dummy variable equal to 1 for real donation decisions; a time variable; and an interaction between these two. Model 1 shows that subjects give more to charities than to other student recipients, and that donations decline over time. Model 2 adds the interaction term, which shows that charitable donations decline faster over time. Model 3 includes controls for sex, citizenship (non-US = 1), personality inventory (Big 5), and residential college.¹⁶ These controls leave the main result unchanged.

We next address the question of whether the charities are so different that that might be the cause of the reduction in giving over time. In order to test this, we use the answers to five five-point Likert-scale questions that were posed immediately after the donation decisions assessing subjects’ perceptions of the organization. The questions used for the Australian Bush Fire decisions reference that event, while the second and third charitable decisions both ask about COVID-19 relief. The questions are as follows:

1. To what extent do you agree or disagree that supporting (Bush Fire relief and prevention; COVID-19 assistance) is an important cause? (1 = Strongly Disagree; 5 = Strongly Agree).

2. To what extent do you agree or disagree that to provide (Bush fire relief and prevention; COVID-19 assistance) is the responsibility of the following organization? (1 = Strongly Disagree; 5 = Strongly Agree).

3. How much do you trust the following organizations in providing (Bush fire relief and prevention services) COVID-19 assistance? (1 = Strongly Distrust; 5 = Strongly Trust).

4. Please evaluate the quality of the work done by the following organizations in (supporting Bush fire relief and prevention) providing COVID-19 relief (1 = Poor; 5 = Excellent).

5. How confident are you that donations to the following organizations (for Bush fire relief and prevention; for COVID-19 relief) will be used efficiently?’ (1 = Not at all confident; 5 = Very confident).

In Table 8, we report the results of regressions that include only the three charitable giving decisions. The dependent variable is the amount donated. On the right-hand side, we control for three main variables. Important Cause is the answer to the first item, and Responsibility is the answer to the second. The third variable, Organizational Quality, constructs a scale derived from the remaining three questions, which measure various aspects of organizational quality. Factor analysis was used to confirm that the items load all on a single dimension and predicted values (factor scores) are used.

If the subject believes that the cause supported by the charity is important, then they contribute about $2.50 more on average. Perceived responsibility does not affect donations. The quality of the organization is also positively related to donations. Clearly, there is a decline over time (the variable ‘study’ is a time counter).

In sum, we can show that the preferences measures are consistently related to charitable giving, and that both dictator transfers and charitable donations decline over time during the college years.

3.6 Robustness

As a robustness check, we use a somewhat different charitable giving task carried out in January 2021. Recall that study participants were paid a fixed amount ($5) to answer a short survey about vaccinations. At the end of the study, respondents were asked whether they wish to donate their earnings to the food bank. This was a binary choice—giving $0 or $5. Only 162 members of the panel participated in this task and of that subgroup, 49.4% donated their earnings. In Table 9 we replicate Table 6, predicting giving with the survey and dictator measures used previously. Because the binary nature of the dependent variable we use logit for our estimates.

As in Table 6, we find that early measures (prior to matriculating) have little predictive power. Unlike Table 6, the survey measures have weak predictive power. Later measures using dictator giving (early in 2020 and later in 2021) have stronger predictive power and carry statistically significant coefficients predicting whether a respondent would donate their earnings to the food bank. This is consistent with our findings from Table 6.

A further robustness check uses a sample different from our panel. Recall that the longitudinal study included smaller samples of incoming students from 2017 to 2019 and the untouched sample of students from the class of 2020. Here, we focus on these respondents, treating them as a different sample. We have a total of 216 respondents who initially made earlier and later dictator transfers and survey responses, and who also completed the July 2021 study. Table 10 replicates Table 6 using this sample. We measure initial dictator and survey responses as the first instance in which the respondent was assessed. For someone who matriculated in 2017, this was their response prior to coming to Rice. For the sample of untouched seniors, this would have occurred in March of 2020, as seniors. For the charity score measure we again use factor scores derived from the charitable giving decisions to US COVID-19 relief in Wave 4 (April 2020) and charitable giving to COVID-19 relief in India in Wave 5 (July 2021). We use ordinary least squares regression to predict factor scores from our altruism measures.

Table 10 is similar to Table 6. The initial measures predict charitable giving. The survey measure and the dictator behavior are positive and statistically significant. In Table 6, only the dictator giving behavior was statistically significant. For the 2021 measures, both coefficients are again statistically significant and in the same direction and a similar magnitude as the coefficients from Table 6. In large part this analysis is consistent with what we reported earlier, and further supports the robustness of our results.

4. Conclusion

We investigate changes in other-regarding preference of students through their university experience by exploiting data collected between 2016 and 2021 using the same student sample—the undergraduate class of 2020 at Rice University. Students in our sample play the traditional dictator game three times—first prior to matriculation, again as they near graduation, and again one year after graduation. They also complete a survey measure of altruism. Finally, they are given three different opportunities to donate to charities. Using repeated measures allows us to explore any change in preferences over time as individuals age or are exposed to certain real-world shocks. Our objective for repeating the dictator game in this study is to understand if and how other-regarding preferences of college students change as they become more independent and make everyday decisions under tight budget constraints. We find that students become less other-regarding or more self-regarding by the end of college term, over both altruism measures and with respect to charitable giving.

We document our student participants’ donation behavior for three different disaster-related donation opportunities—Australian bush fire relief in February, 2020 and COVID-19 relief in the USA in April, 2020 and in India in July, 2021. We elicit donations in each of the three cases under varying decision-making environments. Our results suggest that donation behavior is stable across time (2020–2021) and over contexts, in the sense that we can identify a ‘giving type’. Subjects behave consistently (though not identically) across contexts. The weakening of other-regarding preference, however, is reflected in the donation behavior of students, in that charitable giving also declines over time. (Differences across contexts limit our ability to say this with absolute confidence.)

In the context of prior research, we add to the evidence that preferences change over the life cycle. Several studies document the socialization of children to be more other-regarding in early school years (Harbaugh and Krause, 2000; Fehr et al., 2013). Others have shown that older cohorts, over age 50 years, are more altruistic (e.g. List, 2004). Our sample of college students enter the university with the norms of typical high school students (Eckel et al., 2011), and we see that, through their university years, they become less other-regarding. Thus, it appears that something about young adulthood leads individuals to become more self-regarding and less other-regarding.

Another issue we can address is the relative efficacy of survey and incentivized measures of altruism. Our survey measure is strongly correlated with donation decisions, especially for measures taken later in the university years. However, the incentivized measure is stronger. Like Benz and Meier (2008), we find a high degree of correlation between dictator transfers and charitable giving. Nevertheless, given that the survey question does a pretty good job of predicting donations, perhaps the superior accuracy of the dictator game is not worth its cost in terms of incentives and logistics. Our study provides evidence that both are worthwhile, although dictator games are slightly better at predicting behavior.

On balance our results show evidence of a giving type, with a high degree of consistency across domains and over time in charitable giving, despite an overall decline in giving over time. This suggests that charitable organizations do well to identify ‘givers’ and target them for fundraising campaigns. While young adults may be at a point in their lifecycle where giving levels are low, nevertheless, there is considerable heterogeneity in generosity, which can be identified with altruism measures.

Footnotes

Supplementary material

Supplementary material is available on the OUP website. These are the data and replication files and the online appendix. The data used in this paper are also available through Open Science Framework: DOI 10.17605/OSF.IO/AQ9CY.

Funding

This work was supported by the US National Science Foundation [SES-1534403, SES-2027556 to R.W.; SES-2027548, SES-1534411 to C.E.].

Acknowledgements

We thank research assistants Andy Cao, Liam Coolican, Allegra Hernandez, Carly Mayes, Nanyin Yang, and Sora Youn. This work was improved substantially by comments from Erin Krupka, Tanya Rosenblat, and Oluwagbemiga Ojumu. This work was also supported by our lab managers: Economic Research Lab, Texas A&M University, David Cabrera; Behavioral Research Lab, Rice University, Michelle Harris and Annie Pham. This study received IRB approval [Texas A&M University IRB Numbers: IRB2015-0471D, IRB2020-0379D. Rice University IRB Numbers: IRB—777122–1, IRB-FY2020-278, IRB-FY2021-114].

References