Warming up Cooling Cooperators

The warm-glow of giving is one mechanism proposed to account for high levels of human cooperation. However, little is known about how warm-glow can be harnessed to sustain long-term cooperation to benet wider society (vaccinations, blood donation). We argue that the power of warm-glow to sustain long-term cooperation cools offs over time but can be warmed-up with a simple intervention message. We test, and conrm this prediction, in a eld-based experiment (n = 5,821) comparing warm-glow versus positive affect messages to predict long-term cooperation (blood donation) and conrm our ndings in a subsequent implementation study comparing donation attendance in a 3 year pre-implementation period, prior to the warm-glow message being sent to all 1st time donors in Australia (Ns = 90,317, 93,430 & 89,606) to a 2 year post-implementation period (Ns = 81,766 & 88,551). Exogenously manipulated warm-glow can encourage long-term cooperation that benets society. Wider societal implications are discussed.


Introduction
Human cooperation constitutes a behaviour that is costly to one individual (the actor) while bene cial to another (the recipient) [1][2][3]. The high levels of cooperation observed across human society are hard to explain from the perspective of natural selection [4][5][6][7]. Why would someone perform a behaviour that bene ts another person, often a stranger, at a personal cost? To address this problem a number of ultimate ('why') explanations (e.g., kin selection), supported by a number of proximal ('how') mechanisms, have been proposed to account for both one off and repeated acts of human cooperation [5][6][7][8][9][10][11][12][13][14][15][16][17][18]. This paper focuses on one of these proximal mechanisms: the warm-glow of giving [10,[19][20]. We argue that warm-glow acts as a reinforcement mechanism helping to sustain repeated cooperation, but that with larger time intervals between acts of cooperation the initial power of warm-glow to support cooperation cools off, leading to unintentional free-riding. We test the hypothesis that a simple intervention message can reignite feelings of warm-glow to sustain long-term repeat cooperation.
However, while human cooperation is high it can still be undermined by the counter-force of free-riding [40][41]. A number of mechanisms such as punishment of free-riders [4], opt-out defaults [42][43] and reputation management [12] have been proposed to ameliorate the free-rider problem. Warm-glow, or its absence, is another key mechanism to help explain why free-riding occurs and as such it should be possible to harness the power of warm-glow to reduce free-riding [44][45]. Speci cally, we consider warm glow to be a mechanism that reinforces future cooperation [46]. That is, warm-glow is known to activate the brain's reward centres [26], not to habituate with repeat acts of cooperation [32] and act as a future expected reward [25]. However, this potential reinforcing effect may diminish over time if the time interval between acts of cooperation is large. This may be attributable to people forgetting or not being able to recall their initial warm-glow [47]. This can result in unintentional free-riding, where the person wants to help but does not due to memory failures [48]. Thus, as the warm-glow associated with an initial act of cooperation starts to fade and cool, the person may be less inclined to repeat their initial act of cooperation. Therefore, re-igniting feelings of warm-glow should act to encourage repeat acts of cooperation.
One way to reignite feelings of warm-glow would be with a simple message prompt that encourages people to recall the feelings of warm-glow they experienced for a speci c act of cooperation. The literature on nostalgia and marketing is informative here. 'Nostalgic' memories boost current mood and esteem [49] and encourage charitable giving [50]. Nostalgia based campaigns can operate by anchoring the present in the feelings of the past [51]. Thus, while memories are reconstructive [47], having people recall the warm-glow they felt when initially cooperating, should reignite feelings of warm-glow linked to a speci c act of cooperation, acting to reinforce future expectation of the reward from feeling warm-glow and subsequently enhance cooperation [46,52].

Warm-glow, Repeat Blood Donation And Donor Identity
We test the above hypothesis using blood donation as our model of real world repeat long-term cooperation. Blood donation is an archetypal act of cooperation. The donor pays a personal cost to donate blood (time to donate, pain, loss of blood, the risk of nding they have an infection, fainting, and post-donation complications), which they donate to bene t a complete stranger (the recipient) in need. In most systems (other than family/replacement systems) the donor never meets or knows the recipient, and the recipient will never know who the donor is and is thus unable to repay the debt or express gratitude directly. Furthermore, blood donation is characterised by a large free-rider problem, with only 3-4% of the eligible population donating blood at any one time (97% free-ride) [53]. Thus, while everyone is able to receive blood, and blood products, if they wish to, the vast majority do not contribute. As such, it is a perfect real world analogue of the dictator and public goods games that constitute the lab based evidence for the role of warm-glow in cooperative behaviour [e.g., 19,[21][22]25]. Indeed, compared to non-donors, donors are more likely to express 'impure altruistic' motivations concerning a future blood donation [54]. Furthermore, blood donors are more likely to display warm-glow preferences in lab based economic games [55]. Experienced donors are more likely to express warm-glow preferences than rsttime or new donors [56][57] and expressed warm-glow predicts future donations [58].
A key nding from the work on warm-glow and blood donation is that warm-glow is a more salient predictor in those who identify with the role of being a blood donor (i.e., a cooperator), which normally emerges after the blood donor has made four donations [55][56][57][59][60]. Thus, it has been argued that de ning oneself as a blood donor -cooperative identity -catalyses the effects of warm-glow on future cooperation [56,61]. That is, some people get a bigger warm-glow kick than others and this acts as a stronger reinforcement mechanism, with those experiencing the greatest kick likely to repeat their donation. Thus, if a warm-glow message is going to work to encourage donors who have made their rst donation, their identity as a donor needs to be primed to kick-in earlier. Therefore, we predict that a warmglow intervention will be maximally effective if also accompanied by an extra message that primes a blood donor (cooperative) identity from their rst donation.
There is good evidence that general positive affect predicts future blood donation if experienced post blood donation, [62][63]. General positive affect is also associated with greater helping [64][65]. Thus, we test if it is the concept of warm-glow, speci cally, or positive affect, generally, that predicts future cooperation, and in particular, when these are catalysed by priming cooperative identity as a blood donor.. Some donors may feel initial stronger warm-glow than others and this may lead to a selection effect, whereby those who felt greater warm-glow may be more willing to donate again in the absence of any external warm-glow intervention [56,[60][61]. In the blood donation system this may be operationalised as those who make an immediate appointment after their rst donation to donate again, as opposed to those who do not. Thus, for those who do not make an immediate appointment warm-glow is potentially weaker and thus memories of the initial warm-glow of donation return to baseline quicker: these donors reach a cooling-off point off more quickly. Therefore, we predict that a warm-glow message should be more effective in reigniting feelings of warm-glow in those who do not make an initial appointment and warming them up su ciently to donate again.

The Current Studies
While there is extensive lab based experimental evidence revealing the presence of warm-glow preferences [e.g., 19,21,22,25], as well as eld based observations that warm-glow perceptions predict reported prosocial behaviours (e.g., environmental conservation: [27,33,35]), there are no experiments that exogenously manipulate individual warm-glow to predict real word costly future cooperation. In this paper we do just this. One study, however, has compared warm-glow messages to activate a new cooperative behaviour -donating refunds on used bottles to charity rather than keeping the money [66]. This study compared a warm-glow message ("Please donate your refund -think of the good feeling in helping others. Thank you for your donation."), to norm based messages (e.g., "Many of our customers from this store regularly donate their refunds. Thank you for your donation", or "Please donate your refund -We share an obligation to help those in need, Thank you for your donation"). These messages were manipulated at the level of organization displaying the message (supermarkets) and not used to directly in uence an individual's warm-glow. While the authors [66] report a signi cant effect on the amount donated in supermarkets displaying the warm-glow messages compared to normative ones, the authors report no signi cant effects on the individual level behaviour they collected. Thus, due to the lack of randomization in this study [66] the authors cannot draw a causal link between warm-glow and individual behaviour. Furthermore, they did not compare their warm-glow message to another affective message -so cannot isolate if the effect is attributable to warm-glow per se or a sense of general positive affect. Thus, we explore, for the rst time, the causal effect of a warm-glow intervention to reignite cooperation (blood donation) in individual donors (cooperators) who are cooling-off. We hypothesise that: (1) a warm-glow intervention combined with a prime for cooperative identity, will be more effective than a warm-glow message on its own, (2) a warm-glow intervention combined with cooperative identity prime, will be a stronger predictor than a general positive affect intervention (with and without a cooperative identity prime) and (3) that the most rapidly cooling cooperators (donors who have not made an immediate initial re-appointment to donate) will be the most likely to respond to the a warmglow intervention combined with a cooperative identity prime.
We achieve this by randomly manipulating warm-glow messages at the individual level in a large scale eld based experiment (n = 5,821 rst time blood donors) to predict subsequent blood donor behaviour (attending to make a subsequent donation). Within this eld-based experiment we further compare the relative e cacy of the warm-glow intervention, to a general positive affect-based intervention, to predict subsequent blood donor behaviour. We con rm the eld based experimental ndings in a large scale implementation study, where all new donors in Australia received the most effective warm-glow message.
In the implementation trial we compared the return attendance rate, for a 2nd donation, in all new

Results
Initially we present the results of the eld-based experiment followed by the implementation analysis.
Predicting 3 Month Donations: We conducted an intention to treat (ITT) analysis with assigned arm as the main predictor. A logistic regression (Table 1: columns 2-6) model with age, sex, blood group, arm, rebooked (Yes/No), and the interaction of arm by rebooked was used to predict attendance of all potential donors to donate 3 months later. As can be seen from Table 1, older new donors were more likely to return, as were those with an O-blood group (compared to those with O + blood group). Those exposed to the warm-glow plus identity message were 1.279 times more likely to return than the control group. None of the other messages had a signi cant effect. Those who rebooked were 3.242 times more likely to return than those who had not rebooked. Thus, initial rebooking is a powerful predictor of future cooperation. Importantly, rebooking status signi cantly interacted with condition and this was signi cant for the warm-glow plus identity message only.
To explore this interaction in detail we examined the effect of the warm-glow plus identity message versus the control arm (controlling for age, sex, blood group) in those who had rebooked and those who had not. In those who had rebooked there was no signi cant effect for the warm-glow plus identity message relative to the control arm (p = 0.077, OR = 0.772, 95% CI = 0.579, 1.029, n = 855, R 2 = 0.068: D Cohen = -0.143). There was a signi cant positive effect in those who had not rebooked (p = 0.025, OR = 1.273, 95% CI = 1.031, 1.573, n = 1.464, R 2 = 0.015: D Cohen = 0.133). Thus, those who did not initially rebook, which we argued were those less like to initially feel as much warm-glow and cool off (return to baseline) quicker, were more likely to be in uenced by the warm-glow plus identity message and attend to donate. These ndings support our hypothesis that warm-glow is maximally effective for cooperation when combined with a prime for the person's cooperative identity.

Implementation Study
We compared the return attendance rates, for the second donation, in those who initially rebooked an appointment vs those who did not initially rebook across three pre-implementation time windows and two post-implementation time windows. The number of 1 st time donors in each time window, the numbers who rebooked in centre and those who did not, and attendance return rates can be found in Table S1 (Supplementary le).
The results are shown in Figure 1. As can be seen there is a signi cant (Z = 8.52, p <0.00001: D Cohen = 0.04) increase in the overall return rate (Figure 1). Set against this we observe a steep shift pre-to postimplementation in the return rates for those who did not rebook but not for those who did rebook.
To formally test this we compared the return rates for those who rebooked and those who did not rebook for the per-implementation period just prior to the implementation (16 th Figure 2). Thus, those exposed to the warm-glow with identity message were signi cant more likely to return if they had not rebooked and this level of return was greater than for those who had rebooked.
On average there is a 6.73% increase in donations from those who did not rebook from the immediate pre-implementation period to the two post-implementation periods (

Discussion
Our predictions about the effect of a warm-glow intervention on future cooperation were all supported. Exposure to a warm-glow message, in conjunction with a cooperative identity prime, was the most effective in warming-up cooling cooperators and enhancing the probability of attendance to donate.
Thus, a warm-glow message based intervention is an effective low cost approach to enhance long-term cooperative behaviour. This effect was speci c to warm-glow and not just positive affect, as messages that focused on positive affect -even with an identity prime -were not effective in enhancing future cooperation. Furthermore, this effect was strongest for those cooperators who had cooled the most (new blood donors who had not rebooked).
Loss of warm-glow may be considered a mechanism that results in greater free-riding -probably unintentional free-riding -and these results show clearly that interventions to boost warm-glow may be an effective approach to sustain longer-term cooperation. In day-to-day social cooperative interactions, warm-glow may be continually reinforced through mechanisms such as reciprocity [6][7][11][12] and as such sustained [32]. However, for more intermittent, yet important acts of sustained cooperation (blood donation, vaccination, social distancing during the COVID-19 pandemic), this may not be the case. Thus, messages to boost warm-glow in these contexts are likely to be bene cial. Warm-glow also may be a more fruitful mechanism to support cooperation and reduce free-riding than other mechanisms such as altruistic punishment [4], as this suffers from the problem of 2nd order free-riding and retaliation. Thus, encouraging people to re ect on the positive feelings they have from helping may be a positive way to enhance cooperation without these problems and the studies reported here show that this has potential..
These studies show the real power of warm-glow as a proximal mechanism to sustain cooperation beyond the lab. While extensively supported by evidence from lab based studies [21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37] experimental evidence that warm-glow is effective with respect to real world cooperation is lacking, and these studies ll these gaps. Further, these studies show not only that there is something special about the concept of warm-glow, compared to general positive affect, but also that this has long-term bene ts is terms of sustaining cooperation.
We also highlight an important boundary condition for warm-glow to be effective: cooperative identity.
That is, the effect of warm-glow is stronger when people are primed with a cooperative identity. This may serve to enhance any reinforcing effect of cooperation [46] by strengthening the salience of the link between warm-glow and the behaviour. This hypothesis about the moderating role of cooperative identity with respect to warm-glow is something that can be tested both in lab studies as well as other eld based studies on cooperation (e.g., vaccination) and is a novel prediction arising from these eld based experiments. We also propose that warm-glow messages re-activate memories or at least help develop positive reconstructive memories of an enjoyable past cooperative event. While plausible this exact mechanism needs to be explored in detail in future studies.
These ndings also have important public health policy implications for blood transfusion. Repeat donors have a lower risk of experiencing vasovagal reactions (e.g., fainting) when donating and have a lower rate of discarded blood due to a lower incidence of transfusion-transmissible-infections (TTIs) [67][68][69][70]. Repeat donors, therefore, constitute a saving in terms of recruitment costs, improved donor safety and reduced waste from donated blood that cannot be used. Thus, improving conversion rates from rst to repeat donations is a key objective for transfusion services world-wide and warm-glow messages offers a simple, low cost and very effective means to help achieve this [6].
A number of other societal level behaviours are known to be motivated by other regarding preferences. For example, the altruism/prosocial vaccination hypotheses suggests that people are motivated to get vaccinated to protect others as well as themselves [71][72] and this is supported by a growing body of evidence [72-74 but see 75]. Similarly, behavioural restrictions introduced to combat COVID-19, such as social distancing and mask wearing, are likely motivated to some degree by concerns about protecting others as well as the self [76][77][78]. Thus, the idea of impure altruism seems relevant here also [10]. The work reported here suggests that emphasizing warm-glow can be added to self-other protective messages to emphasize, not only the protection of self and others but also, the personal warm-glow that the person may feel from helping others (e.g., "Get a vaccination to avoid getting and spreading the u and feel the warm-glow that comes from helping those around", or "Wear a mask and social distance to avoid getting and spreading COVID-19 and feel the warm-glow that comes from helping those around you"). Indeed, effective social distancing may be required to control the spread of viruses, such as SARS-Cov2and as such any ways to enhance the effectiveness of public health messages -such as a warmglow message -should be considered [79].

Methods
Timeline for the Field-Based Experiment and Implementation Study Figure 3 details the dates of the eld-based experiment, the implementation of the best warm-glow message and the dates for the pre-post implementation analysis. The methodology of the eld based experiment and implementation trial is detailed below.

Field-Based Experiment of Warm-Glow and Positive Affect Messages
Sampling and sample: A sample of 5,821 of new whole blood donors with A-, A+, O-and O + blood types across Australia who made their rst whole blood donation six weeks previously were recruited using the following criteria: (1) made their rst whole blood donation six weeks ago, and (2) had not donated previously. For each week of the eld-based experiment, all donors who met these eligibility criteria were selected and randomly allocated to conditions. Thus, the whole eligible population was sampled each week. This process continued until the pre-determined numbers per condition were achieved. Twenty donors were excluded: twelve due to email bounces and 8 due to being permanently deferred from donating blood after their initial donation (Fig. 4).
Procedure: New A-, A+, O-and O + donors were recruited to the trial when they made the rst donation between the 16th of April 2018 to the 8th July 2018, with the rst message issued from the 22nd of May 2018. These donors were randomly allocated (using simple randomization) to one of four active arms and a control arm. The control arm contained no message. All other communications from Lifeblood, including an SMS reminder at 12 weeks that the donor could donate whole blood again, were identical across the arms of the trial. Thus, the only difference was the warm-glow and positive affect messages with or without an identity prime. The four active arms crossed a warm-glow vs positive affect message with a prime ("… that's when you became a blood donor …") or no prime for cooperative identity (Fig. 5: Supplementary File S1 contains the full emails sent).
Measures: We collected the following data on the donors: (1) (4) if they had rebooked to make their next donation immediately after their 1st donation (scored 1) or had not initially rebooked (scored 0) Outcome: The main outcome was a veri ed attendance at the donor centre to make a donation (whole blood or plasma) three months after becoming eligible to donate. These data were collected using eProgesia. The data analysts who extracted the attendance data were blind to the experimental condition that the donor was assigned to. Attending to donate is a clear behavioural act of wishing to cooperate to help a stranger and the study is powered for attendance to donate as the outcome.

Power Analysis
Warm-glow has a small effect size with respect to predicting blood donor attendance [56,58]. In a simple regression model with 8 predictors (e.g., arms, age, sex, blood group, rebooked and interaction of rebooked with arms) to achieve 80% power, with an alpha of 0.05 and a small effect size requires 757 donors per arm. Thus, we aimed for 1,000 new donors per arm to allow for any exploratory analyses.
Pre-Registration: The trial was pre-registered on the Open Science Framework prior to 'prior to data collection commencing' (OSF reference: https://osf.io/5m69k).
Consenting: All donors, when they attend to make a donation, sign an individual general declaration consenting to assist blood donor research.
Ethics: The eld-based experiment was approved by the Australian Red Cross Lifeblood Ethics Committee on 7th May 2018, with the rst randomized message sent out on the 22nd of May for those who had made a donation on the 16th of April (Reference: Davison 04052018).

Implementation Study
Design. When recruitment for the eld-based experiment stopped, as the speci ed number of donors per arm had been achieved, the most effective message, based on marketing click through data (click to open) was selected to be rolled out nationally on the 9 th of July 2018. Once the eld-based experiment follow up data were collected and analysed it showed that the eld-based experiment results con rmed the click rate choice. This correspondence allowed us to conduct the implementation study.
The implementation samples consisted of all new donors across Australia who were whole blood donors with autologous and therapeutic donors excluded. We collated aggregate data on whether or not they had attended to make a 2 nd donation (WB or apheresis) within 3 months and whether or not they had initially rebooked in the centre or not. We compared the frequency of attending to make a second donation (whole blood or plasma) in a 3 year window prior to the message roll out to the frequency in the year after the message roll out, across two time windows. The pre-implementation period covered 3 time windows consisting of the 16 th of April to the 15 th of April for years (1)  Ethics: As these data for the implementation analysis are aggregated at the population level, and all donors when they attend to make a donation sign a general declaration consenting to assist blood donor research, no speci c additional ethical approval was needed for these analyses.
Statistical Analyses: Field Based Experiment and Implementation Data All data were analysed using standard statistical packages (IBM SPSS v26, ZumStat, Psychometrica). All tests are two-tailed and effects sizes for all analyses are reported as Cohen's D. Cohen's D was derived for comparison across multiple group means using the procedures described in [80] and from Z scores using procedure described in [81], with both implemented in Psychometrica [82]. Odds Ratios are converted to D using procedures described in [83]. Comparison of percentage across groups, including interactions, used procedure detailed in [84][85].

Declarations
Data Accessibility: The authors declare that all data supporting the ndings of this study are available within the paper and its supplementary information les.
Funding: Australian governments fund the Australian Red Cross Lifeblood to provide blood, blood products and services to the Australian community Con ict of Interests: None of the authors has con icts of interest to declare.
Authorship: All authors have read the content of the paper are happy with this and the order of authorship. Comparison of Pre-and Post-Implementation Periods. Error Bars = 95% CIs.  Active intervention arms

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.