Warming up cool cooperators

Explaining why someone repeats high-cost cooperation towards non-reciprocating strangers is difficult. Warm glow offers an explanation. We argue that warm glow, as a mechanism to sustain long-term cooperation, cools off over time but can be warmed up with a simple intervention message. We tested our predictions in the context of repeat voluntary blood donation (high-cost helping of a non-reciprocating stranger) across 6 studies: a field-based experiment (n = 5,821) comparing warm-glow and impure-altruism messages; an implementation study comparing a 3-yr pre-implementation period among all first-time donors in Australia (N = 270,353) with a 2-yr post-implementation period (N = 170, 317); and 4 studies (n = 716, 1,124, 932, 1,592) exploring mechanisms. We show that there are relatively warm and cool cooperators, not cooling cooperators. Cooperation among cool cooperators is enhanced by a warm-glow-plus-identity message. Furthermore, the behavioural facilitation of future cooperation, by booking an appointment, is associated with being a warm cooperator. Societal implications are discussed.

Explaining why someone repeats high-cost cooperation towards non-reciprocating strangers is difficult.Warm glow offers an explanation.We argue that warm glow, as a mechanism to sustain long-term cooperation, cools off over time but can be warmed up with a simple intervention message.We tested our predictions in the context of repeat voluntary blood donation (high-cost helping of a non-reciprocating stranger) across 6 studies: a field-based experiment (n = 5,821) comparing warm-glow and impure-altruism messages; an implementation study comparing a 3-yr pre-implementation period among all first-time donors in Australia (N = 270,353) with a 2-yr post-implementation period (N = 170, 317); and 4 studies (n = 716, 1,124, 932, 1,592) exploring mechanisms.We show that there are relatively warm and cool cooperators, not cooling cooperators.Cooperation among cool cooperators is enhanced by a warm-glow-plus-identity message.Furthermore, the behavioural facilitation of future cooperation, by booking an appointment, is associated with being a warm cooperator.Societal implications are discussed.
Human cooperation presents the individual with the dilemma of deciding between what is good for them versus what is good for others 1 .This dilemma has been defined in the literature in a variety of ways ranging from paying a personal cost to benefit others 1 to engaging in behaviours with mutual benefits [2][3][4] that do not necessarily require coordinated action but may include reciprocal interaction 2,3,5 .Indeed, there is a clear clustering of cooperative behaviours, ranging from altruism, in terms of the simple division of resources without direct reciprocity (for example, dictator game), to more complex reciprocal interactions (for example, public good and trust games), that define the human cooperative phenotype 5 .Human society is characterized by a remarkably high degree of cooperation 1 and for such high levels of cooperation to survive, definable benefits for the helper (and recipient) need to outweigh costs 1,4,[6][7][8] .Indeed, several benefits (for example, reputation building) have been identified 1,[7][8][9][10][11][12][13][14][15][16][17][18][19][20][21] .Here we examined one aspect of the cooperative phenotype-altruism-whereby the donor bears a cost to benefit a non-reciprocating stranger in the absence of any other mechanisms to support altruism (for example, costly punishment) 1,8,14,15,22 .Under such circumstances, we argue that warm glow sustains high-cost altruistic cooperation directly through reinforcement and indirectly by facilitating a behavioural commitment to cooperation.We tested these hypotheses using voluntary non-remunerated blood donation (VNRBD) as a real-world model of cooperation.
VNRBD follows the structure of the anonymous dictator and public goods games used to demonstrate warm glow in the lab 20,21,[23][24][25][26] and presents cooperation that incorporates both altruism [27][28][29] and Article https://doi.org/10.1038/s41562-023-01687-6less likely.Therefore, we propose that an intervention based on warm glow (χ) will be more effective for these cool cooperators by eliciting feelings of warm glow.Further, as donor identity (π) strongly predicts repeat donation, a warm-glow message that also emphasizes donor identity will be more effective at encouraging cool cooperators to donate again.Therefore, a message that highlights both warm glow and donor identity should be especially effective for cool cooperators who can be identified behaviourally as those who do not book their next appointment to donate.
However, the reinforcing effect of warm glow (λ) may diminish over time, especially if the interval between cooperative acts is long, resulting in affective decay 60,61 and unintentional free-riding 62,63 .Thus, we may have cooling cooperators.Research based on affective chronometry (the study of mood change over time) suggests that affect decays 60,61 .However, there is evidence that for positive affect augmenters, such as our warm cooperators, positive affect decays at a slower rate than for negative affect augmenters, such as cool cooperators 64 .Therefore, for cool cooperators (ω i ), this decay may be more rapid than for warm cooperators (ω ii ).Understanding the temporal dynamics of warm glow should help explain why warm-glow messages are more effective for cool cooperators, for example, if their warm glow decays more quickly 51 .
So far, two field-based experiments have exogenously manipulated warm-glow messages to test their impact on real-world cooperation 52,53 .Both experiments involved donating financial resources.The first study compared the effects of warm-glow messages (for example, "Please donate your refund -think of the good feeling in helping others.")against norm-based messages (for example, "Many of our customers from this store regularly donate their refunds…") displayed at supermarkets on the likelihood of donating bottle return refunds to charity 52 .Significantly higher donation rates were observed in supermarkets displaying warm-glow messages than supermarkets displaying norm-based messages 52 .However, as messages were randomly allocated by supermarket rather than by consumer, it cannot be ruled out that these results were due to differences in consumer populations between supermarkets 52 .
In the second field-based experiment, participants were randomly allocated to one of three message conditions: warm glow ('… Warm your heart'), pure altruism ("… Make Alaska better for everyone") and a no-treatment control (no message).The likelihood of participants donating to a chosen charity following a financial windfall and the amount donated were higher with exposure to the warm-glow than to the other messages 53 .
We further developed this work on the effectiveness of exogenous warm-glow messages.While previous research compared warm glow with pure altruism 53 , studies have consistently shown that blood donation is unlikely to be an example of pure altruism [27][28][29]50,[65][66][67] . However,impure-altruism may motivate blood donors 27,65,66 .Therefore, we contrast warm-glow with impure-altruism messages; our warm-glow message focuses on the donor's warm glow with no reference to benefits to recipients, whereas our impure-altruism message focuses both on the donor's warm glow and benefits to recipients (Supplementary File 2 and Table 1).Thus, both warm-glow and impure-altruism messages feature warm glow as an active ingredient of the message. However,the impure-altruism message also includes pure altruism as an active ingredient (benefits to the recipient).Thus, we tested whether warm-glow or impure-altruism messages would enhance return donations more, especially for those who had not committed to a repeat donation (cool cooperators).
We further explored whether priming donor identity would enhance the effectiveness of warm-glow and impure-altruism messages.We predicted that (1) messages including a donor-identity prime would be more effective than those that did not prime identity and, (2) messages with a higher number of active elements would be more effective than those with fewer active elements.Therefore, we predicted that the rank contributing to the public good 30,31 .Specifically, blood donors in a VNRBD system pay a cost to donate (for example, time, risk of fainting) to benefit a non-reciprocating stranger they will never meet (the recipient) 29 .Donation in a VNRBD system also contributes to the public good 30 , as donated blood is both non-excludable (freely available to anyone who needs it) and non-rivalry (there is sufficient blood, such that transfusing one person will not mean that another person cannot be transfused) 31 .This sufficiency of supply within a VNRBD system is maintained by transfusion services modelling and managing future supply and demand characteristics 32,33 .
Andreoni 11,20,21 proposed that altruism can be motivated by one of the following: (1) warm glow, (2) pure altruism or (3) impure altruism.Warm glow refers to helping that is motivated solely by the positive feelings experienced as a result of helping and not by achieving a particular charitable outcome.In contrast, pure altruism refers to helping, not to feel good about helping, but solely to achieve a successful charitable outcome.Finally, impure altruism is a combination of warm glow and pure altruism: the giver experiences warm glow and wants to donate sufficient resources to ensure a successful outcome.As such, warm glow can be assessed either as experienced positive emotions arising from giving or behaviourally, in terms of individual contributions to a cause when these contributions result in no material benefit to that cause 23 (Supplementary File 1 for details on assessing warm glow).
The current paper develops this understanding (Fig. 1b).We start from the premise that donors experience varying degrees of experienced warm glow (ω) [26][27][28]50 . Som experience stronger warm glow (ω ii : 'warm cooperators') than others (ω i : 'cool cooperators'), specifically ω ii > ω i .The function ω ii > ω i represents a relative relationship whereby cool cooperators still experience some warm glow but to a lesser extent than warm cooperators.Warm cooperators (ω ii ) are predicted to have a greater probability of repeat cooperation via direct reinforcement effects of warm glow (λ).We also propose a second indirect behavioural mechanism, booking another appointment to donate during the donor's current donation (β), through which stronger experienced warm glow increases the probability of repeat cooperation (ρ).Psychologically, booking will enhance a commitment to cooperate (Ø) 54 and reinforce the donor's self-identity as a donor (π) 54,55 , which in itself is a crucial predictor of prosociality 56,57 .A behavioural commitment to cooperate will also facilitate future cooperation through planning and scheduling (putting the date in the diary), which reduces the likelihood that a further donation will be derailed by competing commitments or forgetting 58 .Thus, booking an appointment will be a strong predictor of subsequent cooperation.Consistent with evidence that positive affect is associated with taking steps to facilitate future prosociality 59 , we predict that making a booking (commitment to cooperation) will be more likely for warm cooperators (ω ii ) than for cool cooperators (ω ii ).
For cool cooperators (ω i ), the reinforcing effects of warm glow (λ) and the likelihood of booking another appointment (β) are both Article https://doi.org/10.1038/s41562-023-01687-6order of message effectiveness would be: 'impure-altruism-plus-identit y', 'warm-glow-plus-identity', 'impure-altruism-only' and finally 'warm-glow-only' (Supplementary File 2).Specifically, people experience variation in warm glow from giving (ω) (see main text for factors that influence the extent of experienced warm glow) and this predicts future donations (ρ) based on the reinforcing effects of warm glow (λ) (see main text).Exogenous manipulations of warm glow (χ) predict repeat donations.b, Model to be tested.We extend a by suggesting an additional behavioural mechanism, booking another appointment to donate (β), which enhances commitment to cooperate (Ø) and identity as a donor (π).
Overall increased feelings of experienced warm glow increase the probability of booking, with this being stronger for warm cooperators (those with higher levels of experienced warm glow) than cool cooperators (those with lower levels of experienced warm glow).Booking (β) predicts return donation (ρ), which is again stronger for warm cooperators.An exogenous manipulation of warm glow/ impure altruism will increase the probability of making another donation if a donor identity prime is present.As cool cooperators are less likely to book and have their donor identity strengthened, the warm-glow/impure-altruism-plusidentity messages should be maximally effective for cool cooperators.Green, behavioural mechanism; peach, psychological mechanism; brown, exogenous manipulation; light blue, overall experienced warm-glow; yellow, effects linked to being a warm-cooperator; dark blue, effects linked to being a cool cooperator; greys, indications of associated behaviours or action.
Overall, the aim of this paper is to extend the existing literature on warm glow and test key predictions outlined in Fig. 1b.Specifically: (1) the effectiveness of warm-glow /impure-altruism messages will be Article https://doi.org/10.1038/s41562-023-01687-6enhanced when presented in conjunction with a donor-identity prime, (2) warm-glow/impure-altruism messages with a donor-identity prime will be most effective for those who have not made a commitment to cooperation, (3) experienced warm glow will have a direct effect on future cooperation, (4) experienced warm glow will have an indirect effect on future cooperation through a commitment to cooperation and (5) experienced warm glow decays over time, with that decay being greater for cool vs warm cooperators.We addressed these aims by testing the hypotheses set out in Table 1 (hypotheses marked with a ~ were pre-registered) across six studies.
Studies 1 and 2 test behavioural hypotheses 1-3 and address the main question: "Compared to a business-as-usual (BAU) control, does a warm-glow/impure-altruism message plus an identity prime encourage repeat donation, specifically among those who have not booked an appointment for a future donation?"Study 1 is a field-based experiment that examined donor return behaviour as a function of booking status (booked, not booked) and four active message arms created by crossing message type (warm-glow, impure-altruism) with the donor identity prime ("… that's when you became a blood donor …": yes, no) (see Fig. 2 and Supplementary File 2 for details of message structure) compared to a BAU control.First-time blood donors (5,821)  were randomly allocated to one of the five treatment conditions.Study 2 tested the most effective message from the field-based experiment (warm-glow-plus-identity) in an implementation trial.This trial compared return behaviour among all first-time Australian blood donors in three 1-yr time windows before the warm-glow-plus-identity message was implemented (total n = 270, 353: 54.6% had not booked) and two 1-yr time windows after implementation (total n = 170,317: 65.6% had not booked).
In studies 3 to 5, we explored the behavioural mechanisms that explain the effectiveness of the 'warm-glow-plus-identity' message (hypotheses 4-7).These address the main question "Does experienced warm glow predict future cooperation directly and indirectly via a commitment-to-cooperation?".In Study 3 (survey of 716 first-time donors), we evaluated whether booking status (booked, not booked) was associated with warm glow or pure altruism.Study 4 (survey of 1,124 donors) compared four groups of donors by crossing donor status (first-time, novice) with donor type (plasma, whole blood) to explore the association between booking status and experienced warm glow and whether experienced warm glow and booking status predicted Behavioural effects H 1 : Commitment-to-cooperate effect: 'Those who book another appointment will be more likely to return than those who do not book.'~✓

✓ ✓ ✓
H 2 : Effect of the warm-glow message: 'Compared to a BAU control, the impure-altruism-plus-identity and warm-glow-plus-identity messages will have the strongest effects on donor return.'~✓✓ H 3 : Message by donor commitment-to-cooperate effect: 'A message that contains an identity component (warm-glow and/or impure-altruism) will have a significant effect on donor return only in those who did not book another appointment.'~✓

Behavioural mechanisms
H 4 : Experienced warm-glow commitment-to-cooperation effect: "Experienced warm glow will be higher in those who have booked another appointment compared to those who have not."

✓ ✓ ✓
H 5 : Direct effect of experienced warm glow: "Increased experienced warm glow predicts retuning to make a repeat donation."

✓ ✓
H 6 : Indirect effect of experienced warm glow: "Experienced warm glow will be indirectly linked to return donor behaviour through making a commitment to cooperate (booking another appointment)."~✓H 7 : Temporal dynamics of warm glow: "Warm glow will decay over time, especially for cool cooperators (those who have not booked another appointment)."

✓
Treatment messages: active ingredient H 8 : Affordance of warm glow: "Both warm-glow and impure-altruism messages should afford more experienced warm glow, compared to BAU message."✓ H 9 : Identity enhances warm glow: "Experienced warm glow should be greater in the presence of an identity prime."~✓H 10 : Warm-glow messages focus on the donor: "The warm-glow compared to the impure-altruism message should focus more on the donor and less on the recipient or both donor and recipient."~✓H 11 : Impure-altruism messages focus on maintaining blood supply: "The impure-altruism compared to the warm-glow message should focus more on maintaining blood supply (the public good)."~✓H 12 : "Focus on the donor mediates the association between exposure to a warm-glow-plus-identity message and experienced warm glow."~✓Studies and design.Study 1. Longitudinal field-based experiment: 5,821 first-time blood donors were randomly allocated to one of five warm-glow and impure-altruism conditions (with or without an identity prime) and a BAU control.Study 2. Pre-, post-implementation study comparing donor return rates in 270,353 first-time donors before the implementation of the warm-glow-plus-identity message with 170,317 donors post-implementation.Study 3. Survey of 716 first-time donors.Study 4. Survey of 1,124 donors, crossing donor status (first-time, novice) with donor type (plasma, blood).Study 5. Longitudinal study with 932 first-time donors by measuring warm glow at two timepoints (baseline).Study 6. Online between-subjects experiment: 1,592 members of the public were randomized to one of six conditions.~Pre-registered prediction.All other behavioural effects and behavioural effects hypotheses were developed in the initial paper and tested formally in the subsequent revisions.Treatment-message hypotheses were developed in response to reviewers' comments.Data from studies 3 to 5 were secondary data analyses to address components of the warm-glow model in the initial submission.Study 6 was designed to test hypotheses developed after the initial submission of the paper.

Article
https://doi.org/10.1038/s41562-023-01687-6return behaviour.Study 5 investigated the temporal dynamics of experienced warm glow (does warm-glow decay and, if so, is the rate of decay higher among those who did not book another appointment) among n = 932 first-time donors by measuring experienced warm glow at two timepoints (baseline, 12-week follow-up).Study 5 also explored the association between experienced warm glow, booking status and return behaviour.Study 4 also explored the indirect effect of warm glow on return behaviour via a commitment to cooperate.Study 6, an online experiment (n = 1,592), explored the active ingredients of the 'warm-glow-plus-identity' message that make it the most effective (hypotheses 8-12).

Results
We first present the results of tests of the behavioural effects (hypotheses 1-3; Studies 1-2), followed by tests of the behavioural mechanisms (hypotheses 4-7; Studies 3-5) and the active elements of the treatment messages (hypotheses 8-12; Study 6).

Study 1 Field-based experiment
Table 2 shows the number of donors by treatment (warm-glow-only, warm-glow-plus-identity, impure-altruism-only, impurealtruism-plus-identity and no-message BAU control), their implied commitment to cooperate (booked or did not book a future appointment) and return rates.As expected, based on the randomization to treatments and the fact that donors were exposed to the treatment messages after their booking decisions, the likelihood of booking did not statistically significantly vary across treatments (χ 2 (4) = 4.198, P = 0.380, φ = .027).Thus, any variability in return rates is due to the treatment messages and cannot be attributed to initial variation in booking status. 2 shows that for each treatment group, making a booking resulted in higher return rates.

Effect of behavioural commitment (H 1 ). The final column of Table
Behavioural effects of warm glow (H 2 and H 3 ).First, we compared whether exposure to an active message, regardless of the message content (combined across the four treatments) influenced return rates.The results comparing return rates for the no-message BAU control to those who were exposed to an active message, regardless of content, are presented in Table 2 (last row).The results show that return rates were not significantly different following exposure to any active message, both for donors who had booked and those who had not.Thus, we found no evidence that simple exposure to any message, regardless of content, influenced return rates.
Next, we tested whether exposure to a specific treatment message (for example, warm-glow-plus-identity or impure-altruismplus-identity) increased return rates more than exposure to the BAU control among those who booked or those who did not book.The 'warm-glow-plus-identity' message was the only significant predictor of return rates among those who had not booked another appointment, and there were no significant effects among those who had booked (Supplementary File 2 and Table 2).
To consolidate the above univariate analyses, we conducted an intention-to-treat analysis with treatment, booking status and the treatment by booking-status interaction predicting returning to donate.These analyses controlled for other salient predictors of return to donate (age, gender and blood type).A logistic regression model (Table 3) indicated that the likelihood of returning increased with age and was higher among O− blood donors than O+ donors.Donors exposed to the 'warm-glow-plus-identity' message were 1.279 times more likely to return than the BAU control.No other treatment messages impacted return significantly relative to the BAU control.A significant main effect of booking status showed that, overall, donors who booked another appointment were 3.242 times more likely to return than those who had not booked.

Study 2 Implementation analysis
The 'warm-glow-plus-identity' message was implemented and sent to all first-time donors in Australia, 6 weeks after their first donation (Supplementary File 3).We compared return rates among those who   (1) = 1.496P = 0.226, φ = 0.020 All tests are two-tailed and no adjustment for multiple comparisons was applied.WG, warm-glow-only; WG + I, warm-glow-plus-identity; IA, impure-altruism-only; IA + I, impure-altruism-plus-identity.An overall comparison of the BAU treatment to the combined message treatments shows that there was no overall effect of simply being exposed to a message (χ 2 (1) = 0.211, P = 0.646, φ = 0.006); results in the last row also show no effect of simply being exposed to a message for those who booked and those who did not book.Results in the last column show that booking enhances return rates across all treatments.a Exact P values for the comparison of return rates of those who booked vs those who did not for ( 1 For those who had booked, there was a 0.88% (95% CI 0.402, 1.358) significant (Z = 3.610, P = 0.0003) increase in returning donors.For those who had not booked there was a 9.06% (95% CI = 8.718, 9.402) significant (Z = 51.963,P < 0.001) increase in returning donors.There was a significant interaction with the percentage increase in return rates for those who had not booked, significantly greater than those who had booked (Z = 27.292,P < 0.001), with a percentage difference of 8.18% (95% CI = 7.593, 8.767).Analyses were conducted using procedures for Z tests for proportions detailed in refs.98,99 and implemented in ZumaStat 4.0.All analyses were two-tailed and no adjustments were made for multiple comparisons.Error bars are 95% CIs.
not book) and two slightly overlapping post-implementation windows (n = 170,317, of which 58,500 booked and 111,817 did not book).Figure 3 plots the aggregate return rates pre-and post-implementation for those who booked and those who did not.For those who had booked, there was a significant increase (0.88%; 95% CI = 0.402, 1.358; Z = 3.610; P < 0.001) in returning donors, whereas, for those who had not booked, the percentage increase was 9.06% (95% CI = 8.718, 9.402), which was significant (Z = 51.963,P < 0.001).There was a significant interaction with the percentage increase in return rates for those who had not booked, significantly greater than those who had booked (Z = 27.292;P < 0.001; % increase = 8.18%; 95% CI = 7.593, 8.768).Thus, the 'warm-glow-plus-identity' message positively affected return behaviour for both those who had booked and those who had not, but the effect was significantly larger for those who had not booked.Due to the slight overlap in the two post-implementation time windows, we also compared the two post-implementation windows separately with the pre-implementation window immediately before the post-implementation period.The pattern of results is identical to the aggregated analyses (see Supplementary File 3 and Fig. 2).We also ran a sensitivity analysis to adjust these aggregate return rates for aggregate age and percentage of women (see Supplementary File 3).The hypothesized positive effect of the warm-glow-plus-identity message for those who had not booked remained.

Warm glow higher in those who book an appointment (H 4 ).
Secondary analysis of data from Studies 3 to 5 allowed us to test the hypothesis that those who book will report higher experienced warm glow (Table 4, panels A-C, respectively).
As Andreoni's model 11,20,21 specifically contrasts warm glow with pure altruism, Study 3 was used to explore the relative effects of warm glow and pure altruism with respect to booking status (see Supplementary File 4 and Table 6 for the derivation of the measure of 'anticipated warm glow').We regressed (logistic model) booking status (0, did not book, 1, booked) on 'anticipated warm glow' and 'pure altruism.'Anticipated warm glow' was significantly associated with booking status (B = 0.0847, s.e.= 0.0423; P = 0.045; 95% CI = 0.0018, 0.1677), such that higher levels of 'anticipated warm glow' were associated with booking another appointment, whereas for pure altruism there was no significant variation (B = −0.0648,s.e.= 0.07503; P = 0.338; 95% CI = −0.2118,0.0822).We explored whether this association remains when other affective states associated with blood donation-'anticipated feelings of calmness' and 'anticipated negative emotions' (see Supplementary File 4 and Table 6 for the derivation of these measures)-are included in the model, along with controlling for donors' age and gender.The resulting logistic regression is reported in Table 4a.The positive effect of anticipated warm glow on booking status remained but was not significant (P = 0.097).Anticipated negative emotions were not significantly associated with booking, as was pure altruism or anticipated calmness (a Heckman selection model (Supplementary File 4 and Table 7) showed no evidence of selection bias).
Study 4 provided a replication of Study 3 with a different measure of warm glow while controlling for donor status (first-time or novice) and donor type (whole blood or plasma).A logistic regression model was specified to test the effects of reported warm glow on booking another appointment, controlling for age, gender, donor status (first-time donors or novice donors) and donation type (whole blood or plasma).Warm glow was significantly and positively associated with booking status, but neither donor status nor donation type was significantly associated with booking (Table 4, panel B; for additional analyses predicting warm glow and robustness checks for sample clustering, see Supplementary File 4 and Tables 8-10).
Study 5 also shows that booking status predicts reported warm glow, with those who book reporting higher warm glow.
Across these analyses, there is clear evidence that making a commitment to cooperate (booking another appointment to donate) is associated with higher levels of warm glow.However, the effect size varies.Thus, using the mean scores for warm glow for those who had and had not booked across studies 3, 4 and 5, we report a simple random effects meta-analysis using Comprehensive Meta-Analysis v.2.The results show that there was an overall significant positive effect across the three studies (standard difference in means = 0.209, s.e.= 0.052; 95%CI = 0.106, 0.312; P < 0.001; Cohen's d = 0.201; r = 0.100).Thus, there is a consistent positive association between reported warm glow and booking status (see Supplementary File 4 for analysis details and Supplementary Fig. 3 for Forest plot).

Warm glow (H 5 ) and booking (H 1 ) predict future donation.
Results from Studies 4 and 5 (Supplementary File 5, and Tables 11 and 12

Experienced warm glow predicts donations via booking (H 6 ).
We proposed a path model (detailed in Supplementary File 5 and Fig. 4) that specified that experienced warm glow positively affected return behaviour via booking.The indirect effect of experienced warm glow on return via booking another appointment was significant (B = 0.022, s.e.= 0.006; P < 0.001; 95% CI = 0.013, 0.031).

Article
https://doi.org/10.1038/s41562-023-01687-6 Warm glow decays quicker in those not booking (H 7 ).We hypothesized that experienced warm glow cools over time and that this cooling would be quicker for those who did not book.We test this prediction using longitudinal data from Study 5, in which experienced warm glow was assessed 1 day (wave 1) and 12 weeks (wave 4) post-donation in those who had booked and those who had not.Table 4c shows the results of a generalized estimating equations model evaluating whether booking status predicted experienced warm glow across time.Results showed that those who booked reported higher average levels of experienced warm glow overall (M = 15.32,s.e.= 0.155) than those who did not (M = 14.68, s.e.= 0.215).There was no significant effect of time on experienced warm-glow levels or significant moderation effect of time by booking status.This indicates that the association of booking status on experienced warm glow was stable over time.Thus, rather than having cooling cooperators, we have cool cooperators (who do not book another appointment) and warm cooperators (who booked another appointment).

Study 6 The active ingredients of warm-glow-plus-identity
It is important to identify the active ingredients of the 'warm-glowplus-identity' message that prompts its effectiveness.To reflect theory, the warm-glow messages used across the studies focused solely on the donor, with no reference to recipient benefits, while the impure-altruism messages focused on both the donor and the recipient 11 .Thus, both warm-glow and impure-altruism messages should afford experienced warm glow more than a BAU control, as both contain active warm-glow ingredients (H 8 ), with afforded warm glow greater when a person's identity as a donor is primed (H 9 ).Furthermore, warm-glow messages, compared with impure-altruism messages, should focus more on the donor than the recipient (H 10 ) and less on donating to maintain the public good (blood supply) (H 11 ).We predict that the warm-glow-plus-identity message is effective because the feelings of warm glow it generates are about the donor.Thus, exposure to a 'warm-glow-plus-identity message' should be indirectly linked to afforded warm glow through an enhanced focus on the donor relative to the recipient or both the donor and the recipient (H 12 ).
We tested these predictions using data from Study 6 in which participants were exposed to one of the four messages used in the field-based experiment.We assessed each message's ability to elicit: (1) 'afforded experienced warm glow' (possible range 2 to 14, where 14 is high warm glow), (2) a focus on the 'donor, the recipient or both' (−50 is focus on the blood donor, 0 is focus equally on both the donor  ).These models specify the effects of messages (X) to predict the outcome (Y) 'afforded warm glow' directly and indirectly via focus ('afforded focus on the donor, recipient or both' (the mediator, M)).These models were specified in PROCESS 4.0, with estimates based on 5,000 bootstraps with a multinominal X (messages) and BAU as the reference (n = 1,521).Age, gender and supply ('afforded focus on donating to maintain blood supply') were specified as confounders of M and Y. Predicting M from X has an R 2 of 0.086, P < 0.001.Predicting Y from both X and M has an R 2 of 0.333, P < 0.001.All analyses were two-tailed and no adjustments were made for multiple comparisons.
and the recipient, and +50 is focus on the recipient) and (3) perceptions about donating to maintain blood supply (1, not at all; to 7, completely).We included two control conditions: a simple reminder BAU control condition and an identity-prime-only condition.Supporting H 8 , the warm-glow and impure-altruism messages were judged to afford more warm glow than BAU (Supplementary File 6 and Fig. 5, and the planned contrasts in File 6).Warm-glow messages were judged to afford slightly lower warm glow than the impure-altruism messages (Supplementary Table 13).Supporting H 9 , exposure to an identity prime resulted in greater afforded warm glow (F (  15).Thus, warm-glow messages are more donor-focused and impure-altruism messages focus on both the donor and the recipient as well as on maintaining blood supply.
A series of mediation models with treatment message (X) predicting afforded warm glow (Y) directly and via an afforded focus on the donor, recipient or both (M) (Fig. 4) explored the hypothesis that the warm-glow-plus-identity message affords feelings of warm glow through a focus on the donor more than the recipient.All the messages, compared with the BAU control, predicted increased warm glow, as did a relative focus on the donor.There was a significant positive indirect effect of exposure to the 'warm-glow-plus-identity' message on levels of afforded warm glow via an afforded relative focus on the donor (B = 0.147, s.e.= 0.044; 95% CI = 0.068, 0.239).In contrast, the significant indirect effect of exposure to an 'impure-altruism-plus-identity' message was negative (B = −0.122,s.e.= 0.043; 95% CI = −0.214,−0.044).That is, exposure to an impure-altruism-plus-identity message is associated with an afforded greater recipient focus, linked to reduced afforded warm glow.There were no other significant indirect effects.

Article
https://doi.org/10.1038/s41562-023-01687-6 Therefore, the effectiveness of the 'warm-glow-plus-identity message' is attributable to experienced warm glow being associated with an increased relative focus on donor identity.

Discussion
We showed that warm glow is an important mechanism for sustaining high-cost cooperation directed towards a non-reciprocating stranger.We predicted that warm glow would decay and that exogenous warm-glow messages would act to remotivate cooling cooperators, especially among those who had not behaviourally committed to future cooperation.However, we found evidence that warm glow is stable over time.Therefore, we propose a revised model, whereby there are warm cooperators and cool (rather than cooling) cooperators, with a behavioural commitment to cooperate (booking another appointment) being associated with a warm cooperator, resulting in an increased likelihood of future cooperation.
Our observation that warm glow is stable is consistent with recent reports that warm glow, following blood donation, remains stable over several days 42,68 .However, further studies are needed to measure warm glow over time, both in the lab and in the field, to assess patterns of stability and decay for different types of helping and what predicts any variation in these patterns.Indeed, levels of experienced warm glow are proportional to the cost of cooperation 23,26 and blood donation is a costly act of cooperation that is therefore likely to generate high levels of warm glow 69 .This may account for why experienced warm glow remains stable, even for cool cooperators.Less costly acts of cooperation may generate less warm glow that may decay 63 .Thus, we identified stable warm and cool cooperators of a high-cost cooperative act and reported evidence that warm glow among warm cooperators facilitated a smooth transition to repeat cooperation 51,59,70 .We show that this facilitation happens through two complementary mechanisms.First, a reinforcing route (λ) 36,49,51 that signals a future reward 23 .Second, warm cooperators take behavioural steps through commitment to cooperate that facilitates the likelihood of returning to donate again 27,57 .In the context of blood donation, this is exemplified by booking the next donation, which minimizes derailment of future cooperation 58 and potentially reinforces the donor's self-image as a committed donor 55 .Thus, in the absence of an exogenous intervention, warm cooperators are more likely to return and cool cooperators drop out.
For any public health intervention to be effective, they have to contain active ingredients 67 .We tested predictions for the active ingredient of 'warm-glow-plus-identity' messaging and showed that all messaging afforded warm glow, which was increased most when this messaging was donor-focused rather than focused on benefit to the recipient.Thus, encouraging other repeat acts of cooperation at a societal level (for example, vaccination, social distancing, mask-wearing) [71][72][73][74][75][76][77][78] may be achieved most effectively via warm-glow messages that focus not only on the feeling of warm glow but also link the warm glow directly to the person's identity as a cooperator.For example, the altruism vaccination hypothesis suggests that people are motivated to get vaccinated to protect others [71][72][73][74] (but see ref. 75).Emphasizing warm glow linked to the person's identity as a good citizen may enhance the effectiveness of such messages (for example, "Be the person who makes a positive contribution to helping our society, get a vaccination to avoid spreading the flu and feel the warm glow from helping.").Volunteer blood donation is sustained by 3-4% of the eligible population donating blood at any one time 29,67 .Many other cooperative acts aimed to help the wider society (for example, helping in a natural disaster or war zone) also rely on a few volunteers.Thus, our findings suggest that warm glow may be critical to 'sustaining' high-cost humanitarian aid in a small group of volunteers.
Improving conversion rates from first to repeat donations is a crucial objective for transfusion services worldwide, as repeat donors lead to savings on recruitment costs and improved donor safety [79][80][81][82] .Warm-glow messages offer a simple, low-cost and effective means to help achieve this.Furthermore, behavioural commitment to cooperation by booking another appointment to donate was a very powerful predictor of future prosocial behaviour, therefore making it easier for people to book another appointment will be beneficial.Taking this further, facilitating commitments to cooperate in other prosocial areas (for example, signing up for a flu vaccination appointment, making a diary commitment for volunteering) would also be likely beneficial.
Although the effect sizes reported in this paper are small for experienced warm glow and exogeneous warm glow, a small effect size should be interpreted as important when the behaviour is hard to change (for example, making a repeated donation) but importantly, the intervention is simple and low cost, and the population effect is large 83 .Indeed, we showed that the 'warm-glow-plus-identity' message resulted in an extra 3,049 donations in the year following implementation (Supplementary File 3).
Our findings show that the effect of warm glow extends to cooperative behaviours beyond donating money to the donation of bodily tissue (blood).This is interesting as these acts differ in substantive ways.Many people donate money and this is often low cost in terms of time, effort and discomfort but can range from low to high cost in terms of finances [84][85][86][87] .However, few donate blood, which is generally high cost in terms of time, effort and potential discomfort 88 .Notably, those who donate only blood or only money differ on several demographic (blood donors are younger, more likely to be childless) and psychological characteristics (blood donors are less interested in politics and are more risk-taking) 87 .Furthermore, as levels of experienced warm glow are proportional to the cost of cooperation 23,26 , blood donation is likely to generate greater warm glow than less costly forms of cooperation 88 .It may be the case that a threshold exists for experienced warm glow, which once passed triggers sustained helping.Indeed, such a threshold model that becomes self-reinforcing has been suggested to sustain blood donor behaviour 67 .
Finally, we acknowledge a number of limitations in the reported studies.First, we used secondary data analyses in 3 studies (studies 3 to 5) to explore the association between experienced warm glow, booking status and behaviour.These studies were not designed to assess these associations specifically and, as such, replication of these findings are needed, especially using more widely ranging assessments of experienced warm glow and emotional reactions to blood donation 23 .Second, we assess the temporal dynamics of experienced warm glow at two timepoints, and it would be informative to assess warm glow across a greater number of timepoints post-donation and into subsequent donations to gain a more detailed understanding of the temporal dynamics of experienced warm glow as a function of booking status and time.Third, we did not assess exogenously manipulated warm glow and experienced warm glow together, hence we could not test directly whether the exogenous manipulation of warm glow enhances experienced warm glow in blood donors.However, the pattern of results across the studies indicates that the manipulation of warm glow, especially when donor identity is primed, does afford experienced warm glow and that this is associated with booking and return behaviour.

Ethics approvals and consent
Ethics approval for studies 1, 3, such, the Australian Red Cross Lifeblood Human Research Ethics Committee did not require additional consent for study 1 as the messages were sent out as part of the normal donor recruitment protocol and posed minimal risk.Study 2 drew on data aggregated at the population level, collected by the Australian Red Cross Lifeblood.As all donors signed a general declaration consenting to assist blood donor research when they attend to donate, no additional ethics approval was required for study 2.

Study 1 Field-based experiment
Participant compensation.Participants did not receive compensation for this study.

Sampling and sample.
A sample of 5,821 (Mean age = 31.94,s.d.= 11.96;% women = 60.7%)new whole-blood donors from across Australia with A−, A+, O− and O+ blood types (based on the Australian Red Cross Lifeblood donor segmentation policy) were recruited.However, A and O positive and negative groups are the most prevalent, accounting for 86% of Australian blood donors.Donors who had made their first whole-blood donation 6 weeks earlier and who had not donated previously in Australia were recruited.For each week of the field-based experiment, the entire eligible population was sampled, with all donors who met the eligibility criteria being selected and randomly allocated to conditions using simple randomization.This process continued until a priori target numbers per condition were achieved.Twenty donors were excluded: 12 due to email bounces and 8 due to being permanently deferred from donating blood after their initial donation, leaving a final sample of n = 5,801 (Mean age = 31.92,s.d.= 11.95;% women = 60.7%;see Supplementary File 7 for detail on sample bias and randomization checks, and Supplementary Fig. 9).This sample is generally representative when compared to the average age (32.5 yr) and percentage of women (56%) for all first-time whole-blood donors in the three pre-implementation windows in Study 2.
Procedure.New A−, A+, O− and O+ donors were recruited to the trial when they made their first donation between 16 April 2018 and 8 July 2018, with the first message issued on 22 May 2018.Donors were randomly allocated (using simple randomization) to one of four active arms and a BAU control arm.The control arm contained no message.All other communications from Lifeblood, including an SMS reminder that the donor could donate whole blood again sent 12 weeks post-donation, were identical across trial arms.Thus, the only difference between conditions was message type (warm-glow, impure-altruism) crossed with the donor identity prime ("… that's when you became a blood donor …": yes, no) (see Fig. 2 and Supplementary File 2 for details of message structure).
Measures.We measured donor characteristics including (1) age, (2) gender (0 = men; 1 = women), (3) blood group (ABO: A−, A+, O−, O+) and (4) whether they had booked to make their next donation immediately after their first donation (scored 1) or had not initially booked (scored 0).Donor return behaviour.Donor return was measured as verified attendance at the donor centre to make a high-cost donation (whole blood or plasma) within 3 months after becoming eligible.These data were collected using eProgesa 5.03.The data analysts who extracted the attendance data were blind to experimental treatments.
Power analysis.Warm glow has a small effect on predicted blood donor attendance 53,55 .To achieve 80% power to detect a small effect with an alpha of 0.05 in a simple regression model with 8 predictors (for example, message type and identity prime conditions, age, gender, blood group, booking status and terms for the interaction of booking status with experimental conditions), 757 donors per arm are needed.Thus, we aimed for 1,000 new donors per arm to allow for ad hoc exploratory analyses.
Pre-registration.The trial was pre-registered on the Open Science Framework (OSF reference: https://osf.io/5m69k).Pre-registered predictions are detailed in Table 1.

Study 2 Implementation study
Participant compensation.Participants did not receive compensation for this study.
Design.The most effective message of those tested in the field-based experiment was selected by Lifeblood to be rolled out nationally on 9 July 2018, with effectiveness measured by marketing click-through data (the highest proportion of click-to-open) from the original recruitment sample.Analysis of attendance data collected in the field-based experiment confirmed the click-rate message choice (see Supplementary File 8 and Fig. 10 for the timeline of the field-based experiment and implementation analysis).The implementation sample included all new donors across Australia who had donated whole blood, excluding donating for autologous and therapeutic reasons.We collated aggregate data on whether they had attended to make a second high-cost donation (whole blood or plasma) within 3 months and whether they had initially booked the next appointment in-centre.We compared the frequency of return in the 3 yr immediately before message roll-out with the frequency of return in two time windows after implementation.The 3-yr pre-implementation period covered three 12 Pre-registration.This study was not pre-registered.Rather, data were collated from the Australian Red Cross Lifeblood (via eProgesa 5.03) on a yearly aggregate level to test the prediction from the main finding of the field-based experiment that a warm-glow-plus-identity message would encourage more second-time donations in those who have not booked.

Study 3 Booking, warm glow and pure altruism
Participant compensation.Five AU$100 gift vouchers were awarded after a random draw of participants.
Rationale.This study was designed primarily to examine blood donors' concerns and attitudes towards blood donation during the coronavirus disease 2019 (COVID-19) pandemic.We used secondary data analysis to explore the association between anticipated warm glow (and other anticipated emotional states), pure altruism and booking status in first-time donors, as these were the focus of the field-based experiment.

Sample.
Seven-hundred and sixteen first-time donors took part (Mean age = 41.20,s.d.= 14.85; % women = 65.5%,n = 469 (see Supplementary File 7 for sampling strategy)); the current sample was older and had a higher percentage of women.
Procedure.Donors were sent a de-identified online survey invitation between 12 August and 24 September 2020.A reminder was sent on 27 Article https://doi.org/10.1038/s41562-023-01687-6 August 2020.Data were extracted on 24 September 2020.The survey was hosted on Qualtrics (https://www.qualtrics.com/).The survey was sent out after donors had made their donations and had the opportunity to book their next donation, but the time between booking and survey completion varied.

Measures. The following variables were used.
'Anticipated warm glow' was indexed using adjectives of the type that have been used to assess warm glow in previous research (for example, refs.24,44).Specifically, donors were asked "What do you think donating blood in the future during the coronavirus pandemic would be like?", to which they responded using 12 adjectives (Unrewarding, Pointless, Displeasing, Negative, Stressful, Unsatisfying, Rewarding, Worthwhile, Pleasing, Positive, Relaxing, Satisfying), each measured on a 4-point scale (1 = Not at all, 2 = Slightly, 3 = Moderately and 4 = Extremely).Of these adjectives, Rewarding, Worthwhile, Pleasing, Positive and Satisfying represent the type of adjective that have been used to assess warm glow in previous research (for example, ref. 24).A principle-axis-factor analysis with oblique rotation (Supplementary File 4 and Table 6) applied to these 12 adjectives indicated a 3-factor solution, with the 'anticipated warm glow' adjectives Rewarding, Worthwhile, Pleasing, Positive and Satisfying (α = 0.77) forming one factor 24 , the adjectives Relaxing and Stressful forming a factor whereby high scores equated to relaxing (termed 'Anticipated calmness'; mean inter-item correlation = 0.34) 89 and the adjectives Pointless, Negative, Unrewarding, Unsatisfying and Displeasing forming the final factor, which was termed 'anticipated negative affect' (α = 0.68) 89 .
'Pure altruism' was assessed with two items concerning the donor's motivation for their last donation ("To help someone in need" and "It's important for society"), each measured on a 5-point scale (1 = Not important at all, 2 = Slightly important, 3 = Moderately important, 4 = Very important, 5 = Extremely important).
Booking status.We extracted data about whether donors had booked another appointment after their last donation from participants' linked but de-identified donor records (n = 294, 41.1% booked).

Additional selection model predictor.
As the study took place during the COVID-19 pandemic, we included a question that assessed whether the donor had had a COVID-19 test ("Have you ever been tested for coronavirus?":Yes, No).Power analysis.There was no data on which to base a power calculation.However, if we assume a small effect size, then we would need 363 donors who booked and 363 who did not (n = 726) with a power of 0.80 and α = .05.In our sample, 294 booked and 422 did not.As such, the overall sample size was sufficient to detect a small effect.
Pre-registration.This study was not pre-registered.

Study 4 Warm glow, booking status, donor status and type
Participant compensation.Participants did not receive compensation for this study.

Rationale.
The study was designed primarily to examine how donors' emotions change across a single donation.The study also included a measure of donors' intrinsic motivation to donate blood 90 .Within the literature, warm glow has been conceptualized as intrinsic motivation 24,44,[91][92][93] , and intrinsic regulation from self-determination theory has been linked to assessing warm glow 93 .Thus, warm glow was assessed using a subscale designed to measure intrinsic regulation for blood donation.In this study, we explored the association between booking and warm glow, controlling for donor status (first-time vs novice) and type (plasma vs whole blood).Procedure.Participating donors were approached by a different researcher in each of three donation centres and prompted to complete questionnaires at several timepoints across their donation.The assessment of warm glow (that is, intrinsic regulation) was completed in the refreshment area after donation.Donors could book during the session.Data were collected between April 2016 and October 2017.

Sample. In
Measures.The following indices were assessed.

Warm glow (intrinsic motivation)
. This was assessed using the intrinsic regulation subscale of the Blood Donor Identity Survey (BDIS) 90 , which comprises 3 items ("I enjoy donating blood", "For me, being a blood donor means more than just donating blood" and "Blood donation is an important part of who I am"), with each item measured on a 7-point scale (1 = Not at all true, 7 = Very true).These items reflect the psychological concept of warm glow 24,44 and were summed to yield a warm-glow score, with higher scores indicating higher subjective warm glow (Mean = 14.66, s.d.= 4.42, α = 0.80).
Booking status.We gained data about whether donors had booked another appointment during the donation session when they completed the survey from participants' linked but de-identified donor records (781, 69.5% booked).Donor return behaviour.Donor return was measured as verified attendance to donate within 6 months of the index donation, collected using eProgesa 5.03 (670 or 59.6% returned to donate).Power analysis.There was no data on which to base a power calculation.However, if we assume a small effect size, then we would need 363 donors who booked and 363 who did not (n = 726) with a power of 0.80 and α = .05.In our sample, 781 booked and 343 did not.As such, the overall sample size was sufficient to detect a small effect.
Pre-registration.This study was not pre-registered.

Study 5 Temporal stability of warm glow following booking
Participant compensation.Participants could opt-in to receive AU$5 compensation for every survey completed.

Rationale.
The study was designed primarily to examine how donors' emotions (reflective and anticipated) changed following their first donation.Warm glow was assessed at two of the four study timepoints: 1 day after donating blood (wave 1) and 12 weeks later (wave 4).We used these data for secondary data analysis to assess: (1) whether booking influenced warm glow, (2) the stability of warm glow over time and (3) whether the stability of warm glow was moderated by booking status. the average age (32.5 yr) and percentage of women (56%) for first-time whole-blood donors in the three pre-implementation windows in Study 2, this sample was slightly older and had a higher percentage of women.

Sample. Nine
Procedure.In a longitudinal design, data were collected in four waves following donors' first donation.The first (wave 1) assessments were completed approximately 1 day after the donor's first donation.Participants who opted-in for follow-up were invited to complete assessments at three subsequent timepoints: 4 (wave 2), 8 (wave 3) and 12 (wave 4) weeks after their initial donation.Warm glow (intrinsic motivation) was measured in waves 1 and 4. All data were collected between January 2017 and September 2017.
Measures.The following measures were assessed.

Warm glow (intrinsic motivation).
This was assessed at waves 1 and 4 using the BDIS 88  Booking status.We gained data about whether donors had booked another appointment after their last donation from participants' linked but de-identified donor records (n = 612, 65.7% booked).

Donor return behaviour.
Return was measured as verified attendance to donate within 6 months of the index donation, collected using eProgesa 5.03 (n = 682, 73.2% returned to donate).

Power analysis.
There was no data on which to base a power calculation.However, if we assume a small effect size, then we would need 363 donors who booked and 363 who did not (n = 726) with a power of 0.80 and α = .05.In our sample, 612 booked and 320 did not.As such, the overall sample size was sufficient to detect a small effect.
Pre-registration.This study was not pre-registered.

Study 6 Validation of warm-glow and impure-altruism messages
Participant compensation.Based on Prolific's rates, participants received £1.25 compensation for completing the study.
Experimental design.The main design was a 2 (warm glow: warm glow, impure altruism) by 2 (identity: present, absent) between-subjects design.This represents the four main messages used in the field-based experiment (Study 1) (warm-glow-only, warm-glow-plus-identity, impure-altruism-only and impure-altruism-plus-identity; Fig. 2).We added two additional conditions to compare warm-glow and impure-altruism messaging with (1) a no-message BAU control ("The good news is in a few weeks you'll be ready to donate."); and (2) an identity-prime-only condition ("Most people can't remember what they were doing 6 weeks ago, but this was the day you became a blood donor.The good news is in a few weeks you'll be ready to donate again.").This was an online experiment hosted on Qualtrics (https://www.qualtrics.com/uk/),with participants randomly allocated using simple randomization to one of these six conditions.All data were collected on 24 March 2021.
Power calculation.A Cohen's d of 0.344 for positive ratings of warm-glow vs altruistic messages for blood donation was previously recorded 66 .Achieving 80% power to detect a small effect with α = 0.05 for a 2 (warm glow: yes, no) × 2 (identity: yes, no) design requires 265 participants per condition.For a series of pairwise comparisons of experimental groups, detecting a small effect with a Cohen's d of 0.344 requires 135 participants per condition.
Outcome measures.The following measures were assessed.
Afforded donor vs recipient focus.This was calculated as the average of 2 items ("To what extent do you think that the message focuses on the blood donor, the recipient of blood or both?" and "To what extent do you think that the message focuses on the emotions of the blood donor, the recipient of blood or both?") measured on a scale from −50 (primarily on the blood donor) to +50 (primarily on the recipient of blood) with a midpoint of 0 (equally on both the donor and recipient).
Afforded warm glow.This was calculated as the sum of 2 items ("The message makes me feel that donating blood is personally rewarding in itself."and "The message makes me feel that donating blood would make me feel like a good person.")measured on a 7-point scale (1 = not at all, to 7 = completely).
Afforded focus on maintaining blood supply.This was assessed with a single item ("The message makes me feel that donating blood would ensure that there is enough blood for all who need it.")measured on a 7-point scale (1 = not at all, to 7 = completely).

Statistical analyses for Studies 1-6
All data were analysed using standard statistical packages (IBM SPSS v.26 and 27, ZumStat, Psychometrica, Stata 17, MPlus 8.4, PROCESS 4.0, Comprehensive Meta-Analysis 2).All tests were two-tailed.All effect sizes for all analyses are reported using Cohen's d.Cohen's d was derived for comparison across multiple-group means using the procedures described in ref. 94 and from Z scores using the procedure described in ref. 95, with both implemented in Psychometrica 96 .ORs were converted to Cohen's d using procedures described in refs.96,97.Comparisons of percentages between groups, including interactions, used procedure detailed in refs.98,99.Mediation models were run in MPlus and PROCESS 100 .Meta-analysis was conducted using Comprehensive Meta-Analysis (v.2) 97 .Our main analyses had dichotomous outcomes and were analysed using logistic models and non-parametric tests.When continuous variables were outcomes, we present boxplots (for example, afforded warm glow) or Q-Q plots (experienced warm glow, Study 5; see Supplementary File 9 and Table 11a,b).In all cases, there was evidence of non-normality.We analysed these data using analysis of variance (ANOVA) and ordinary least squares (OLS) regression models as Monte-Carlo studies show that these techniques are robust to non-normality especially when sample sizes are large (greater than 300 for ANOVA and 1,000 for OLS) 101,102 .All results reported are two-tailed with no adjustments for multiple comparisons made.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

nature research | reporting summary
April 2020

Field-specific reporting
Please select the one below that is the best fit for your research.If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf Behavioural & social sciences study design All studies must disclose on these points even when the disclosure is negative.

Study description
We report results from six studies.
Study 1:We report a field-based experiment where we randomly allocate 5,821 1st time blood donors to one of four active message conditions based on warm-glow or impure altruism messages and a pure no-message control (n 5,801 took part after exclusions).We explored the moderation of messages by booking or return attendance.We collected quantitative data on the number who returned and actually attend to make a donation 3 months later.
Study 2: We confirmed the field-based experimental findings in a large-scale implementation trial where we compared the return rate to attend for a 2nd donation in the population of all new donors in Australia who had made their first donation in one of three one-year time windows prior to the warm-glow message being implemented (Ns = 90,317, 93,430, & 89,606), with the same return rates made by all new donors in Australia in one of two one-year time windows after the warm-glow message had been implemented (Ns = 81,766 & 88,551).We collected quantitative data of the number who returned and actually attend to make a donation 3 months later.
Study 3: A quantitative cross-sectional survey to evaluate whether booking was associated with self-reported warm-glow or purealtruism.(n=716 first-time donors).We used self-reported survey items to assess anticipated warm-glow, negative affect, calmness and pure-altruism, booking status was recorded from donor records.
Study 4: Quantitative survey of n=1,124 donors.We examined 4 groups of donors by crossing donor status (first-time, novice) with donor type (plasma, blood) and explored the association between booking, self-reported warm-glow, and return behaviour.We collected quantitative data of the number who returned and actually attend to make a donation 3 months later, we assessed warmglow by self-reports and booking status was recorded from donor records.
Study 5: Longitudinal survey at two-time points.We investigated the temporal stability of warm-glow among n=932 first-time donors by measuring warm-glow at two-time points (baseline; 12-week follow-up) and the effects of warm-glow and booking on return behaviour.We collected quantitative data of the number who returned and actually attend to make a donation 3 months later, we assessed booking status from donor records.. Study 6 was an online quantitative experiment (n = 1, 592) to validate the warm-glow and impure-altruism messages to be used in the field-based experiment.We collected quantitative data from self-reported data.

Research sample
Study 1: The field-based experiment was conducted on a sample of 5,821 1st time blood donors with 5,801 taking part after deferral criteria were applied.The initial sample of 5,821 (Mage = 31.98,[Sd = 11.96],% female = 60.7%) were new whole-blood donors from across Australia.After twenty donors were excluded (twelve due to email bounces and 8 due to being permanently deferred from donating blood) after their initial donation, the final sample was n=5,801 (Mage = 31.92,[Sd = 11.95],% female = 60.7%).This sample was chosen as our study focus was on increasing the conversion rate to make a second donation in first-time donors.Compared to the average age (32.5 years) and percentage of women (56%) for all 1st time whole blood donors in the three pre-implementation windows in study 2, this sample was generally representative.Sampling was targeted to ensure that the gender balance and age (18-69: donors over 70 were excluded) distribution was representative of the donor population.This sample was chosen as our study focused on exploring the links between self-reported warm-glow and booking status in first-time donors and this sample had items that matched the type of items that had been used previously to assess self-reported warm-glow and also sample first-time donors.In the nearest post-implementation time window (2019-2020) the average age was 34.3 with 57% women.Thus, the current sample was older and had a higher percentage of women.

nature research | reporting summary
April 2020 Study 4: In total, 1,124 donors (Mage = 30.27,Sd = 11.51,% women = 55.2%, n = 621 women) participated.Of these, 401 were firsttime whole-blood donors (Mage = 30.38,Sd = 11.879;% women = 54.6%,n = 219 women), 208 first-time plasma donors (Mage = 30.07,Sd = 11.50, % women 50.5%, n = 106 women), 379 were novice whole-blood donors (Mage = 29.61,Sd 11.49, % women = 60.4%, n = 229 women) and 136 novice plasma donors (Mage = 30.53,Sd = 10.75; % women = 49.3%, n = 67 women).This sample was chosen as our study focused on exploring the links between self-reported warm-glow, booking status and return behaviour and this study contained assessments of all of these.We also wished to explore the effects of first-time vs novice (more experienced donors) and type of donation (whole blood vs plasma) and this study contained data on all of these.The sample was representative in terms of the percentage of women but slightly younger than the average age (32.5 years) for 1st time whole blood donors in the three pre-implementation windows in study 2.
Study 5: Nine-hundred and thirty-two first-time donors who completed the survey provided warm-glow responses at waves 1 and 4 (Mage = 36.36,Sd = 14.02, % women = 70.3%,n = 655 women).This sample was chosen as our study focused on exploring the temporal stability of warm-glow in first-time donors, and in this study, warm-glow was assessed at two-time points.The study also had data on booking status and behaviour and this study also addressed the effects of booking on the stability of warm-glow and explored the prediction of behaviour from warm-glow and booking status.Compared to the average age (32.5 years) and percentage of women (56%) for 1st time whole blood donors in the three pre-implementation windows in study 2, this sample was slightly older and had a higher percentage of women.
Study 6: A total of 1,592 participants were recruited through Prolific (https://www.prolific.co/)(Mage = 36.47,Sd = 13.00,% women = 50.2%,n women = 795).This sample was chosen as our study was designed to explore the validity of the messages used in studies 1 and 2 in the general population.The sample is representative of 50.2% women compared to 51% of the population, but slightly younger at 36.47 years compared to the UK population average of 40.4 years.

Sampling strategy
Study 1: For the field-based experiment a sample of 5,821 new whole blood donors with A-, A+, O-and O+ blood types across Australia, who made their first whole blood donation six weeks previously, were recruited using the following criteria: (1) made their first whole blood donation six weeks ago, and (2) had not donated previously.For each week of the field-based experiment, all donors who met these eligibility criteria were selected and randomly allocated to conditions.The whole population meeting the sampling eligibility criteria were included.This process continued until the pre-determined numbers per condition were achieved.Sample size was determined as follows.As warm-glow has a small effect on predicted blood donor attendance, to achieve 80% power to detect a small effect with an alpha of .05 in a simple regression model with 8 predictors (e.g., message type and identity prime conditions, age, sex, blood group, booking status, and terms for the interaction of booking status with experimental conditions), 757 donors per arm are needed.Thus, we aimed for 1,000 new donors per arm to allow for ad-hoc exploratory analyses.
Study 2 For the implementation study all new donors in Australia across three pre-implementation time windows and 2 postimplementation time windows.We sampled the whole target population of first-time donors in Austrailia across three preimplementation periods and two pots-implementtaion periods.
Study 3: 20,000 donors were contacted with an even split across the 3 groups; (1) first-time donors (first donated after February 2020), (2) existing eligible donors who had donated after February 2020, and (3) existing eligible donors who had not donated since February 2020.Sampling was targeted to ensure that the gender balance and age (18-69: donor over 70 were excluded) distribution was representative of the donor population.Nine-hundred and sixty-six first-time donors were recruited.Of these 250 1st-time donors started the survey but did not complete the majority of measures including the measure of adjectives used to assess warmglow.As 250 did not complete the survey (non-completers) but started it, and 716 did (completers), there is a potential for selection bias.Indeed, completer status is predicted by having had a COVID-19 test (χ2 (1) = 3.373, p = .066,φ = .061)with those who had a COVID-19 test more likely to be completers, and being older (M = 41.20,Sd = 14.85 vs M = 34.14,Sd = 14.33; t (964) = 6.531, p =.000; DCohen = 0.421).Completer status was not predicted by gender (χ2 (1) = 0.020, p = .887,φ = -.005).Rebooking status was not predicted by being a non-completer or completer (χ2 (1) = 1.151, p = .283,φ = .035)nor as a function of having had a COVID-19 test (χ2 (1) = 3.315, p = .069,φ = .061).To account for any potential selection bias we completed a sensitivity analysis using a Heckman Probit Selection model with COVID-19 test, gender and age as predictors f selection bias.There was no evidence of selection bias.This study constituted a secondary analysis of these data.There was no data on which to base a power calculation.However, if we assume a small effect size then we would need 363 donors who booked and 363 who did not (n = 726) with a power of .80 and α = .05.In our sample 398 booked and 329 did not.As such, we had a sufficient sample size to detect a small effect.
Study 4: New (1st donation) and novice (1 or 2 previous donations and no plasma donations for whole blood donors, and 1 or 2 previous plasma donations and less than 3 whole blood donations for plasma donors) donors who were either whole blood or plasma donors were recruited by convenience sampling in donor centres, at three large urban Australian Red Cross Lifeblood (Lifeblood) centres (Brisbane Donor Centre, Sydney Town Hall and Melbourne Mt Waverley) from April 2016 to October 2017.A total sample of 1,153 donors were recruited.Of these, 441 were 1st time whole blood donors, 218 1st time plasma donors, 385 were novice whole blood donors and 136 novice plasma donors.Overall, there were 354 plasma donors and 799 whole blood donors There were missing data on warm-glow (n = 26), age (n = 1) and gender (n = 2).Once these were removed the final sample size was 1,124.Of these, 401 were 1st time whole blood donors, 208 1st time plasma donors, 379 were novice whole blood donors and 136 novice plasma donors.Overall, there were 344 plasma donors and 780 whole blood donors.There was no data on which to base a power calculation.However, if we assume a small effect size then we would need 363 donors who booked and 363 who did not (n = 726) with a power of .80 and α = .05.In our sample 781 booked and 329 did not.As such, we had a sufficient sample size to detect a small effect.
Study 5: All 25,428 first-time whole blood donors were invited to participate in the study between January 2017 and September 2017.4,472 first-time donors were recruited of these data were missing from 500 participants on intrinsic regulation (warm-glow) at wave 1, leaving complete data on 3,972.Of the 4,472, 136 donors were actively deferred from donating for medical and life-style reasons.Those deferred were more likely to be women (χ2 (1) = 18.296, p = .000,φ = .
= -0.329,p =.743; DCohen = -0.021).However, as deferral status influenced the decision to re-book (χ2 (1) = 5.571, p = .016,φ = -.036), with those deferred less likely to rebook, we excluded those who were actively deferred from the analyses.After removal of those actively deferred there were 4336 1st-time donors at wave 1.At wave 4 a total of 932 completed the survey providing data on warm-glow.Those who completed the surveys providing data on warm-glow at waves 1 and 4 were significantly older (M Wave 1 only = 34.88,Sd = 13.39 vs M Waves 1 & 2 = 36.27,Sd = 14.09: t (4334) = -2.863p =.004; DCohen = -0.087)and less likely to be male (χ2 (1) = 8.231, p = .004,φ = -.044)than those who completed only wave 1, however, they did not significantly vary on warm-glow at wave 1 (M Time 1 only = 15.25,Sd = 4.06 vs M time1 & 2 = 14.97,Sd = 4.18: t (3849) = 1.892, p =.059; DCohen = 0.061 ).Furthermore, warm-glow at wave 4, for those who completed both surveys (M = 15.21,Sd = 4.06) was not significantly different from warm-glow at wave 1 (M = 15.25,Sd = 4.06) in those who just completed the survey at time 1 (t (one-sample (931) = -0.320p =.749; DCohen = -0.021).As such, there is no evidence that initial levels of warm-glow influenced who remained in the study from waves 1 to waves 4.However, we controlled for age and gender in these analyses.There was no data on which to base a power calculation.However, if we assume a small effect size then we would need 363 donors who booked and 363 who did not (n = 726) with a power of .80 and α = .05.In our sample 612 booked and 320 did not.As such, we had a sufficient sample size to detect a small effect Study 6: Participants (aged 18 to 80 from the UK) were recruited through Prolific (https://www.prolific.co/)to an experiment hosted on Qualtrics (https://www.qualtrics.com/uk/).A total of 1,592 were recruited.There were 266 participants in the BAU/Control, 273 in the 'identity-only', 263 in the 'warm-glow-only', 256 in the 'warm-glow-plus-identity', 279 in the 'impure-altruism-only', and 255 in the 'impure-altruism-plus-identity' conditions.Ferguson, Farrell, and Lawrence 64 report a Cohen's d of 0.344 for positive ratings of warm-glow vs altruistic messages for blood donation.Achieving 80% power to detect a small effect with an alpha of .05 for a 2 (warm-glow: yes; no) 2 (identity: yes; no) design requires 265 per condition.For a series of pairwise comparisons of experimental groups, detecting a small effect with a Cohen's d of 0.344 requires 135 participants per condition.

Data collection
We collected data on verified donor attendance for studies 1, 2, 4 and 5.This was achieved in a blinded manner, with data analysts who extracted the attendance data blind to the experimental condition that the donor was assigned to.We collected verified data on booking status in studies 1, 2, 3, 4, and 5.
We collected survey data on experienced warm-glow in studies 3, 4, 5, and afforded warm-glow in 6, and pure-altruism in study 3.
In study 3 we assessed 'anticipated warm-glow' using adjectives of the type that have been used to assess warm-glow in previous research.Specifically, donors were asked: "What do you think donating blood in the future during the coronavirus pandemic would be like?" to which donors responded using In studies 4 and 5 experienced warm-glow was assessed using the intrinsic regulation subscale the Blood Donor Identity Survey (BDIS) which comprises 3 items ("I enjoy donating blood", "For me, being a blood donor means more than just donating blood", and "Blood donation is an important part of who I am") with each item measured on a 7-point scale (1 = Not at all True, 7 = Very True).
We collected data, in study 6, on the perception of the messages to be used in the field-based experiment (study 1) with respect to afforded warm-glow, recipient vs donor focus and provision of the public good (blood supply).'Afforded Donor vs Recipient Focus' was calculated as the average of 2 items ("To what extent do you think that the message focuses on the blood donor, the recipient of blood or both?" and "To what extent do you think that the message focuses on the emotions of the blood donor, the recipient of blood or both?") measured on a scale from -50 (primarily on the blood donor) to +50 s (primarily on the recipient of blood) with a midpoint of 0 (equally on both the donor and recipient).'Afforded Warm-Glow' was calculated as the sum of 2 items ("The message makes me feel that donating blood is personally rewarding in itself" and "The message makes me feel that donating blood would make me feel like a good person) measured on a 7-point scale (1 = "not at all" to 7 = "completely").'Afforded Focus on Maintaining the Blood Supply' was assessed with a single item ("The message makes me feel that donating blood would ensure that there is enough blood for all who need it.")measured on a 7-point scale (1 = "not at all" or 7 = "completely.").
Others present when data was collected: For study 1 donors received a message there may have been other present when they read it but we do not have data on this.For study 2 this was aggregated population data on booking status and return behaviour other donors and staff may have been present when donors booked.Study 3 was an online survey (Qualtrics) and it is assumed that donors complete this independently.For study 4 data was collected using an electronic tablet in a donor centre with other blood donors and blood collection staff present.For study 5 data was collected using an online questionnaire (Qualtrics) and it is assumed that donors complete this independently.For study 6 data was collected using an online questionnaire (Qualtrics) and it is assumed that donors complete this independently. Timing

Data exclusions
Study 1: For the field experiment 20 participants were excluded, of which 12 had bounced emails and 8 with permanent deferrals (evidence of blood-borne infection).These were not pre-determined and analyses showed that they were evenly distributed across the study conditions.
Study 2: There were no exclusions in the implementation trial.
Study 3: Nine-hundred and sixty-six first-time donors were recruited.Of these 250 1st-time donors started the survey but did not complete the majority of measures including the measure of adjectives used to assess warm-glow.As 250 did not complete the survey (non-completers) but started it, and 716 did (completers), there is a potential for selection bias.

Non-participation
Study 1: 20 donors were excluded (12 due to bounced emails, and 8 due to permanent deferrals) With respect to the initial randomization of 5,821 donors dropping to 5801 after the 20 exclusions.There was no evidence that exclusion systematically varied across arms or affect distributions of age and sex.Thus, there was no evidence for bias.
Study 2: There were no dropouts.
Study 3: Nine-hundred and sixty-six first-time donors were recruited.Of these 250 first-time donors started the survey but did not complete the majority of measures including the measure of adjectives used to assess warm-glow.As 250 did not complete the survey (non-completers) but started it, and 716 did (completers), there is a potential for selection bias.To account for any potential selection bias we completed a sensitivity analysis using a Heckman Probit Selection model and there was no evidence for bias.
Study 4: A total sample of 1,153 donors were recruited.Of these, 441 were first-time whole blood donors, ,218 first-time plasma donors, 385 were novice whole blood donors and 136 novice plasma donors.There were missing data on warm-glow (n = 26), age (n = 1) and gender (n = 2).Once these were removed the final sample size was 1,124. .Study 5: Of the 4,472 recruited donors, 136 donors were actively deferred from donating for medical and life-style reasons.After removal of those actively deferred there were 4,336 first-time donors at wave 1.At wave 4 a total of 932 completed the survey providing data on warm-glow.The other data were excluded doe to survey non-completion.

Randomization
In study 6 participants were block-randomized by sex to each of the 6 conditions to ensure equal numbers of males and females For the field-based experiment (study 1), this was simple randomization.There was no randomization for the implementation study where we compared pre-and post-implantation time periods or for studies 3, 4 and 5 which were observational.Study 2 ( implementation study) was based on aggregate data and we could not control for age and sex.However, these were consistent across the time window and by booking status so unlikely to have influence the pattern of results observed.

Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies.Here, indicate whether each material, system or method listed is relevant to your study.
931) = -0.320p =.749; DCohen = -0.021).As such, there is no evidence that initial levels of warm-glow influenced who remained in the study from waves 1 to waves 4.However, we controlled for age and gender in these analyses.
Recruitment Study 1: For the field-based experiment the whole population of all 1st time donors who met the inclusion criteria were sampled for each week of the field-based experiment and randomly allocated to conditions.The whole population meeting the sampling eligibility criteria were included.This process continued until the pre-determined numbers per condition were achieved.As such, there is little scope for bias donor were sample and allocated to recieve a message based o the randomization schedule.A sample of 5,821 (Mage = 31.98,[Sd = 11.96],% female = 60.7%)new whole-blood donors from across Australia with A-, A+, O-and O+ blood types (this was based on the Australian Red Cross Lifeblood donor segmentation policy at that time.However, A and O positive and negative groups are the most prevalent, accounting for 86% of Australian blood donors) who had made their first whole-blood donation six weeks prior and had not donated previously in Australia were recruited.Twenty donors were excluded: twelve due to email bounces and 8 due to being permanently deferred from donating blood after their initial donation, leaving a final sample of n=5,801 (Mage = 31.92,[Sd = 11.95],% female = 60.7%.With respect to the initial randomization of 5,821 donors, age did not vary significantly across treatments (F (4, 5816) = 0.725, p = 0.575: DCohen = 0.059), nor did sex ( 2 (4) = 3.542, p = 0.472: DCohen = 0.050).The majority of the sample had blood type O+ (45.9%), followed by A+ (36.9%), and O-(10.7%),A-(6.4%) and O (1%).The distribution of blood type did not vary by treatment ( 2 (16) = 18.16, p = 0.314: DCohen = 0.112).The percentage who booked immediately postdonation was 37.7%, with 62.3% not booking while in the donor centre, and this did not vary by condition ( 2 (4) = 4.163, p = 0.384: DCohen = 0.054).After the 20 donors were excluded (Supplementary Figure 8), donor age remained nonsignificant by condition (F (4, 796) = 0.796, p = 0.527: DCohen = 0.059), as did sex ( 2 (4) = 3.536, p = 0.472: DCohen = 0.054).The majority of the sample had blood group O+ (46%), followed by A+ (36.9%), and O-(10.7%),A-(6.4%).The distribution of blood type did not vary by condition ( 2 (12) = 14.62, p = 0.263; DCohen = 0.100).The percentage who booked was 37.7% with 62.3% not initially booking and this did not vary by condition ( 2 (4) = 4.198, p = 0.380: DCohen = 0.054).Thus, the randomization by arm remained intact after the exclusions, as such there is no evidence of bias due to exclusions.
Study 2: For the implementation study all new donors in Australia across three pre implementation time windows and 2 post implementation time windows were selected.As the whole populations are sampled for the field-based experiment and the implementation study we feel there is little to no bias in the results especially as the findings from the field-based experiment are confirmed in the implementation study.
Study 3: 20,000 Australian donors were contacted with an even split across the 3 groups; (1) first-time donors (first donated after February 2020), (2) existing eligible donors who had donated after February 2020, and (3) existing eligible donors who had not donated since February 2020.Sampling was targeted to ensure that the gender balance and age (18-69: donor over 70 were excluded) distribution was representative of the donor population.A 10-15% response rate was expected and as such some self-selection bias is possible.,however, the attained sample was representative.Donors were sent a deidentified online survey invitation between the 12th of August and the 24th of September 2020.A reminder was sent on the 27th of August 2020.Data were extracted on the 24th of September 2020.The survey was hosted on Qualtrics (https:// www.qualtrics.com/).Nine-hundred and sixty-six first-time donors were recruited (mean age = 39.37 years, Sd = 15.03years; 66% female with 23.7% having had a COVID-19 test).The low number of COVID-19 tests reflect the low rate of COVID-19 infection in Australia at the time.Of these 250 first-time donors started the survey but did not complete the majority of measures including the measure of adjectives used to assess warm-glow.As 250 did not complete the survey (noncompleters) but started it, and 716 did (completers), there is a potential for selection bias.Indeed, completer status is predicted by having had a COVID-19 test (χ2 (1) = 3.373, p = .066,φ = .061)with those who had a COVID-19 test more likely to be completers, and being older (M = 41.20,Sd = 14.85 vs M = 34.14,Sd = 14.33; t (964) = 6.531, p = .000;DCohen = 0.421).
Completer status was not predicted by gender (χ2 (1) = 0.020, p = .887,φ = -.005).Booking status was not predicted by being a non-completer or completer (χ2 (1) = .Thus there is a potential selection bias here.However, the sample was representative.Of these, data were missing from 500 participants on intrinsic regulation (warm-glow) at wave 1, leaving complete data on 3,972 participants.We did not impute these missing data as data were missing on all items pertaining to warm-glow and the missing percentage was high (11%).Of the 4,472 recruited donors, 136 donors were actively deferred from donating for medical and life-style reasons.Those deferred were more likely to be women (χ2 (1) = 18.296, p = .000,φ = .064)but did not vary by age (M deferred = 33.).As such, there is no evidence that initial levels of warm-glow influenced who remained in the study from waves 1 to waves 4.However, we controlled for age and gender in these analyses as these demonstrated systematic variation and potential for bias.

Ethics oversight
Study 1: The field-based experiment was approved by The Australian Red Cross Lifeblood Ethics Committee on 7th May 2018 with the first randomized message sent out on the 22nd of May for those who had made a donation on the 16th of April (Reference: Davison 04052018).
Study 2: The implementation study is based on aggregate data at the population level and as such no individual are identifiable and donors consent for their data to be used for research when they sign up as a blood donor.Note that full information on the approval of the study protocol must also be provided in the manuscript.

Fig. 1 |
Fig.1| Theoretical models linking warm glow to cooperation.a, Current knowledge.Summary of the general findings supporting warm-glow theory.Specifically, people experience variation in warm glow from giving (ω) (see main text for factors that influence the extent of experienced warm glow) and this predicts future donations (ρ) based on the reinforcing effects of warm glow (λ) (see main text).Exogenous manipulations of warm glow (χ) predict repeat donations.b, Model to be tested.We extend a by suggesting an additional behavioural mechanism, booking another appointment to donate (β), which enhances commitment to cooperate (Ø) and identity as a donor (π).Overall increased feelings of experienced warm glow increase the probability of booking, with this being stronger for warm cooperators (those with higher

Fig. 3 |
Fig. 3 | Aggregate return behaviour pre-and post-implementation of the 'warm-glow-plus-identity' message in new donors as a function of booking status (Study 2).For the pre-implementation, the aggregated point estimate for the percentage return rate for those who booked was 61.47 (95% CI = 61.20,61.74; n = 75,414/122,681) and for those who did not book, 21.49 (95% CI = 21.28,21.70; n = 31,738/147,672).For the post-implementation, the aggregated point estimate for the percentage return rate for those who booked was 62.35 (95% CI = 61.96,62.74; n = 36,472/58,500) and for those who did not book, 30.55 (95% CI = 30.28,30.82; n = 34,156/111,817).For those who had booked, there was a 0.88% (95% CI 0.402, 1.358) significant (Z = 3.610, P = 0.0003) increase in returning donors.For those who had not booked there was a 9.06% (95% CI = 8.718, 9.402) significant (Z = 51.963,P < 0.001) increase in returning donors.There was a significant interaction with the percentage increase in return rates for those who had not booked, significantly greater than those who had booked (Z = 27.292,P < 0.001), with a percentage difference of 8.18% (95% CI = 7.593, 8.767).Analyses were conducted using procedures for Z tests for proportions detailed in refs.98,99 and implemented in ZumaStat 4.0.All analyses were two-tailed and no adjustments were made for multiple comparisons.Error bars are 95% CIs.

Study 1 :Study 3 :Study 4 :Study 5 :Study 6 :
The field experiment ran from the 22nd of May 2018 to the 8th of July with the 3 month follow up on attendance completed by the end of Nov 2018.Study 2: For the implementation study the following time windows were analysed.Pre-implementation period covered three time windows consisting of the 16th of April to the 15th of April for years (1) 2015-2016, (2) 2016-2017 and (3) 2017-2018 and the postimplementation period covered two slightly overlapping time windows (1) the 9th of July 2018 to the 8th of July 2019 and (2) the 16th of April 2019 to the 15th of April 2020.Donors were sent a de-identified online survey invitation between the 12th of August and the 24th of September 2020.A reminder was sent on the 27th of August 2020.Data were extracted on the 24th of September 2020 nature research | reporting summary April 2020 The survey took place between April 2016 to October 2017.The survey took place between January 2017 and September 2017.The online survey took place on the 24th of March 2021, and all data were collected on that day.

| Treatment messages used in Studies 1 and 6. The
messages formed when warm glow (warm glow, impure altruism) is crossed with identity (present, absent).
To account for any potential selection bias we completed a sensitivity analysis using a Heckman Probit Selection model with COVID-19 test, gender and age as predictors f selection bias.There was no evidence of selection bias.Study 4: A total sample of 1,153 donors were recruited.There were missing data on warm-glow (n = 26), age (n = 1) and gender (n = 2).Once these were removed the final sample size was 1,124.Of these, 401 were 1st time whole blood donors, 208 1st time plasma donors, 379 were novice whole blood donors and 136 novice plasma donors.Overall, there were 344 plasma donors and 780 whole blood donors.Study 5: Of the 4,472, 136 donors were actively deferred from donating for medical and life-style reasons..After removal of those actively deferred there were 4336 1st-time donors at wave 1.At wave 4 a total of 932 completed the survey providing data on warmglow.There is no evidence that initial levels of warm-glow influenced who remained in the study from waves 1 to waves 4.However, we controlled for age and gender in these analyses.
If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.For the field-based experiment we have the following characteristics.In the control condition we had an n of 1162, with a mean age 32.31 (SD = 12.06) with a %female of 60.5.In the warm-glow-identity condition we had an n of 1164 with a mean age of 31.72 (SD = 11.78) and a %female of 60.8.In the warm-glow condition we had an n of 1164 with a mean age of 32.82 (SD = 12.03) and a %female = 60.4.In the positive affect identity condition we had an n of 1164 with a mean age of 31.92 (SD = 12.12) with a %female of 62.7.In the positive affect condition we had an n of 1167 with a mean age of 31.66 (SD = 11.81) with a %female of 59.0.The majority of the sample had blood type O+ (45.9%), followed by A+ (36.9%), and O-(10.7%),A-(6.4%)andO (1%).Study 2: For the implementation study the % female at each time window was as follows: Pre-implementation time window one (2015-2016: n = 90,317) the mean age was 32.0 (Sd = 13.6) and % female was 55%, time window two (2016-2017: n = 93,430) the mean age was 35.5 (Sd = 13.4) and the %female was 56%, and time window 3 (2017-2018: n = 86,606) the mean age was 33.1 (Sd = 13.0) and the %female was 57%.In the post-implementation period for time windows one (9th of July 2018 to the 8th of July 2019: n = 81,766) the mean age was 34.9 (Sd = 1354) and the %female was 56%, and for time window two (6th of April 2019 to the 15th of April 2020: n = 88.551) the mean age was 34.3 (Sd = 13.2) and the %female was 57%.For the aggregate pre-implementation mean age Ws 35.5 (Sd = 13.3:n= 270,353) and the post-implementation period the mean age was 34.6 (Sd = 13.3:n= 170, 317).As these are aggregate data to examine the effects of age and sex on the aggregate results we re-ran the aggregated analysis on percentage return rates adjusted for aggregate age and sex (percentage women).We obtained estimates of age-and sex-adjusted percentage return rates from a simple general linear model (n = 10).This simple adjustment for age and sex does not alter the pattern of results in terms of the key interaction and the effect of the warm-glow-plus-identity message on return rates for those who do not book.April 2020 impute these missing data as data were missing on all items pertaining to warm-glow and the missing percentage was high (11%).Of the 4,472, 136 donors were actively deferred from donating for medical and life-style reasons.Those deferred were more likely to be women (χ2 (1) = 18.296, p = .000,φ = .064)but did not vary by age (M deferred = 33.88,Sd = 12.83 vs M not deferred = 35.20,Sd = 13.57:t (4470) = 1.119, p =.263; DCohen = 0.033).Those deferred were not significantly different from those not-deferred on warm-glow at wave 1 (M deferred = 15.06,Sd = 4.34 vs M not deferred = 15.18,Sd = 4.09: t (3970) = 0.315, p =.753; DCohen = 0.009) or wave 4 (M deferred = 15.48,Sd = 4.09 vs M not deferred = 15.21,Sd = 4.07: t (955) = -0.329,p =.743; DCohen = -0.021).However, as deferral status influenced the decision to re-book (χ2 (1) = 5.571, p = .016,φ = -.036), with those deferred less likely to rebook, we excluded those who were actively deferred from the analyses.After removal of those actively deferred there were 4336 1st-time donors (M age = 35.20 years, Sd = 15.57years; % female = 67%) at wave 1.At wave 4 a total of 932 completed the survey providing data on warm-glow.Those who completed the surveys providing data on warm-glow at waves 1 and 4 were significantly older (M Wave 1 only = 34.88,Sd = 13.39 vs M Waves 1 & 2 = 36.27,Sd = 14.09: t (4334) = -2.863p =.004; DCohen = -0.087)and less likely to be male (χ2 (1) = 8.231, p = .004,φ = -.044)than those who completed only wave 1, however, they did not significantly vary on warm-glow at wave 1 n = 173 women) and 780 whole blood donors (M age = 30.01,Sd= 11.65,%female = 57.4%,n= 797) who did not differ by age (t (1122) = 1.14, p =.254; DCohen = 0.068 ) but did by gender (χ2 (1) = 4.93, p = .026,φ= -.066).There were 609 1st-time donors (M age = 30.62,Sd= 11.69;%female = 53.4%,n= 325 women) and 515 novice donors (M age = 29.85,Sd= 11.30;%female = 57.5%,n= 296 women) who did not differ by age (t (1122) = 1.11, p =.267; DCohen = 0.066 ) or gender (χ2 (1) = 1.81, p = .178,φ= -.040).Study 5: 4,472 first-time donors were recruited (mean age = 35.16years, SD = 13.55 years; 68% female), of these data were missing from 500 participants on intrinsic regulation (warm-glow) at wave 1, leaving complete data on 3,972.We did not nature research | reporting summary April 20203.315,p = .069,φ = .061).To account for any potential selection bias we completed a sensitivity analysis using a Heckman Probit Selection model with COVID-19 test, gender and age as predictor of selection bias.Study 4: New (1st donation) and novice (1 or 2 previous donations and no plasma donations for whole blood donors, and 1 or 2 previous plasma donations and less than 3 whole blood donations for plasma donors) donors who were either whole blood or plasma donors were recruited by convenience sampling in donor centres, at three large urban Australian Red Cross Lifeblood (Lifeblood) centres (Brisbane Donor Centre, Sydney Town Hall and Melbourne Mt Waverley) from April 2016 to October 2017.There is room for self-selection bias due to the convenience sampling procedure.In total, 1,124 donors (Mage = 30.27,Sd = 11.51,% women = 55.2%, n = 621 women) participated.Of these, 401 were first-time whole-blood donors (Mage = 30.38,Sd = 11.879;% women = 54.6%,n = 219 women), 208 first-time plasma donors (Mage = 30.07,Sd = 11.50, % women 50.5%, n = 106 women), 379 were novice whole-blood donors (Mage = 29.61,Sd 11.49, % women = 60.4%, n = 229 women) and 136 novice plasma donors (Mage = 30.53,Sd = 10.75; % women = 49.3%, n = 67 women).In total, 1,124 donors (Mage = 30.27,Sd = 11.51,% women = 55.2%, n = 621 women) participated.Of these, 401 were first-time whole-blood donors (Mage = 30.38,Sd = 11.879;% women = 54.6%,n = 219 women), 208 first-time plasma donors (Mage = 30.07,Sd = 11.50, % women 50.5%, n = 106 women), 379 were novice whole-blood donors (Mage = 29.61,Sd 11.49, % women = 60.4%, n = 229 women) and 136 novice plasma donors (Mage = 30.53,Sd = 10.75; % women = 49.3%, n = 67 women).Thus there was some variability by age and sex, also the sampling design involved clustering.To account for this we controlled for age and sex and ran the statistical model with both standard error and error clustered within sample region.The results show that the results did not differ for the models with standard errors or clustered errors.Study 5: All 25,428 first-time whole blood donors were invited to participate in the study between January 2017 and September 2017.All eligible donors were sent an email invitation asking them to participate in the study.A reminder email was sent after 5 days.Nine-hundred and thirty-two first-time donors who completed the survey provided warm-glow responses at waves 1 and 4 (Mage = 36.36,Sd = 14.02, % women = 70.3%,n = 655 women).All 25,428 first-time whole blood donors were invited to participate in the study between January 2017 and September 2017.4,472 first-time donors were recruited (M age = 35.16,Sd = 13.55;% female = 68%) 1.151, p = .283,φ = .035)nor as a function of having had a COVID-19 test (χ2 (1) = nature research | reporting summary The study protocol was approved by Lifeblood's Human Research Ethics Committee on 18th September 2015 with reference number 2015#07.Study 5: he study protocol was approved by Lifeblood's Human Research Ethics Committee on 15th November 2016 (Reference: 2016#24).Study 6: The experiment was approved by the University of Nottingham, School of Psychology Ethics committee (ref: S1309).
Study 3: The study was approved by the Australian Red Cross Lifeblood Human Research Ethics Committee on 31st July 2020 nature research | reporting summary