1.  Introduction

2.  Institutional Background and Data

3.  Empirical Analysis

4.  Discussion and Conclusion


Nicola Lacetera

University of Toronto

National Bureau of Economic Research

Mario Macis

Johns Hopkins University

Institute for the Study of Labor


Every year, millions of people devote time, effort, and money to voluntary, pro-social activities such as donating cash for a cause or organization, helping the elderly and disabled, cleaning beaches, or donating blood. These activities involve single citizens as well as small and large, private and public organizations, and represent a sizeable share of a society’s life.1 Yet, for many of these  activities, supply is still below societal needs.

A notable example is given by blood donations. Blood transfusions are required in such situations as blood loss due to trauma or during surgeries, the treatment of premature babies,and for several chronic  diseases. There is no available substitute for human blood, and, in recent years, the demand has increased dramatically due to, among other causes, an aging population and new surgical procedures such as organ transplants (Di Rado, 2004; www.bloodbook.com). However, blood supply frequently does not keep pace with demand.2 Neither eligibility criteria nor a lack of information fully explain these shortages. Although about 50% of adults are eligible to donate blood in developed countries, and donations can be made between 4 and 7 times per year according to dfferent legislations, only 3.8 donations are made, on average, per 100 individuals (World Health Organization, 2009). Also, information campaigns and communications about shortages are often conducted by such organizations as the Red Cross, government agencies, schools, hospitals, etc. An alternative cause to be explored concerns people’s behaviors and incentives. Individuals might simply not find it worthwhile to dedicate time to donate blood if the private benefits of donating blood fall short of the opportunity costs. This implies that policies offering explicit incentives to donate might play a role in encouraging participation in activities that are, in most countries, basedon voluntary and unpaid contributions.

In this paper, we study the effects of Law 584, a legislative provision passed in 1967 that gives Italian blood donors the right to a paid day off work on the same day that they donate blood or blood components. The law applies to all donors who are employed at any private or public organization, and salary and contributions are reimbursed by the state. We evaluate whether this legislative provision induces blood donors to make more donations, and quantify this effect.

One would not necessarily expect a paid day off to increase blood donations. Although standard economic theory would deliver this prediction, research in psychology and, more recently, behavioral economics argues that extrinsic incentives might crowd out the quantity and quality of the supplied altruistic activity. For instance, the introduction of economic rewards might create doubts about the true reason behind pro-social behavior, thus potentially crowding out intrinsic motives. Economic incentives, therefore, could backfire instead of reinforcing altruistic motivations (Bénabou and Tirole, 2006; Deci, 1975).3 Given the presence of competing predictions about the impact of economic incentives on pro-social behavior in general and on blood donation in particular, empirical examinations are in order.

Figure 1 reports the number of members of the Italian Association of Blood Donors (AVIS), both in absolute terms and per 1,000 persons in Italy in the years immediately preceding and immediately following the passage of the law (individuals have to be affiliated with a blood donor organization, like AVIS, in order to donate blood. Further details in Section 2 below). The aggregate evidence indicates that there was no reduction in the number of donors (i.e., no net crowding out), and hints to an increase above the previous trend.4 This evidence, however, is obviously not sufficient to draw causal implications of the addition of this extrinsic incentive. In particular, information on donors’labor market status is needed.

Even though AVIS’s national headquarters does not collect data on the labor market status of its donors, local units of the association sometimes do. Our analysis is based on a unique, longitudinal dataset comprising the individual histories of blood donations of the hole population of donors in an Italian mid-sized town ("The Town" hereinafter) located in the north center region  of the country. In addition to demographic information and the number and dates of donations made by about 2,600 unique donors in the periods 1985-89 and 2002-06, the database includes information on the donors’labor market statuses and occupations. Because our data do not include the years both before and after the introduction of the policy, we cannot evaluate the effects of the policy on the extensive margin of donations (i.e., on inducing more people to become blood donors). However, the data do  allow us to study whether blood donors are responsive to the paid-day-off incentive. In particular, our empirical strategy exploits  cross-sectional variation in donors’ labor market statuses, and, crucially, variation in job-switching by donors over time.

We exploit the fact that the incentive benefits only donors who are employed by using variation in donors’occupations and changes in donors’' labor market status to identify the effect of this incentive on blood donations. The longitudinal nature of our data enables us to perform regressions of each donor’s donation frequency on his or her labor market status with individual fixed effects, which absorb any heterogeneity in time-invariant personal attributes that might be correlated with donation behavior (including "intrinsic" altruism). Thus we use within-donor employment shifts to determine whether changing labor market status is associated with different donation frequencies.

Our findings indicate that when donors are eligible to benefit from the day-off incentive (i.e., when they are in paid employment) they make, on average, one extra blood donation per year, a substantial effect that represents a 40% increase. Notably, our findings are unchanged if we focus on the transitions between being an employee and being out of the labor force (i.e., being a student, homemaker, retiree, or unemployed), thus excluding self-employment.

We are able, therefore, to rule out a potential alternative explanation for our findings, i.e. that people donate more when their opportunity cost of time is lower (as it would happen for an employee as compared to a self-employed individual), because being out of the labor force arguably leads to even more available time to donate. We also investigate whether ceasing to be an employee is associated with a lower donation frequency. Although our point estimates are negative, thus indicating a reduction in donations when donors cease to benefit from the day-off incentive, they are (in most cases) small in magnitude and not statistically significant. Even though more data would be required to make conclusive claims, this finding is consistent with some form of "persistence" in behavior whereby those with higher donation frequencies tend to maintain a high frequency even after they have lower incentives to do so. Evidence from patterns of repeat donations by new donors is consistent with this persistence being due to learning and selection, although we cannot rule out that exposure to the incentive can also lead to habit formation.

These findings are robust to a variety of specifications and sample restrictions that are aimed at addressing issues such as measurement error, serial correlation, in labor market status, and the concern that changes in labor market status might be capturing other life changes that might be correlated with donation frequencies, such as changes in health condition or family structure (proxied by age). These tests also allay the concern that the higher donation rates of employees are not due to an incentive effect but, rather, to an income or wealth effect.

Further indication that the day-off incentive affects donor behavior comes from our analysis of patterns in the choice of donation day, and from actual take-up rates of the day-off incentive. We document that a substantial fraction of donors who are employees choose to donate on a day that extends their weekend (notably Friday, which leads to 2:5 consecutive days off because donations can be made only in the morning), whereas no such preference is found for donors of a di¤erent labor market status. The "Friday effect" that we detect for employees suggests that a substantial share of this group of donors exploits the full potential benefit from the day-off provision. As a further corroboration of this view, we find that the take-up rates of the incentive, as represented by the percentage of donors who request a document attesting to their donation to be presented to their employer, average 70%, with spikes on Fridays and Mondays. Thus, most donors who are employees do take advantage of the incentive as opposed to, say, donating and returning to work. Interestingly, about 30% of employees who donate on a Saturday request the day-off document; therefore, a substantial share of employees donating on Saturday are, indeed, donating on their closest workday to the statutory free day (i.e., Sunday).

A number of studies have investigated the e¤ects of material rewards on pro-social activities. Early laboratory experiments by Deci and his collaborators found that adding explicit rewards for the performance of activities that are motivated by intrinsic reasons leads to a reduction in the performance of those activities (Deci, 1975). Similar findings have been obtained by, among others, Frey and Oberholzer-Gee (1997) and Gneezy and Rustichini (2000). Gneezy and Rustichini, however, find that "large enough" incentives do stimulate pro-social behavior. With specific reference to blood donation, in an artefactual field experiment Mellstrom and Johannesson (2008) offered Swedish college students small monetary rewards to undertake a health test in order to determine their eligibility to donate blood, finding no effects for males and negative effects for females on the willingness of taking the
health test (the study did not observe actual donation behavior). Goette and Stutzer (2008), in a field experiment in Switzerland, find that lottery tickets used to promote donations increase turnout at blood drives. Lacetera, Macis, and Slonim (2012) present observational and experimental evidence from American Red Cross blood drives that offering small material rewards increases donations.5 In addition to providing novel findings from a new source of data, our study complements the existing literature on the impact of explicit incentives on the performance of altruistic activities in at least two other ways. First, the fact that in most industrialized countries there are tight restrictions to rewarding blood donations makes it difficult to analyze the issue empirically in the field. To the best of our knowledge, this study is the first to analyze the actual behavior of an entire population of blood donors in response to a naturally occurring incentive defined by the law. Second, in our context the individuals are free to not enjoy the economic benefits (e.g., by choosing the day of the week on which to donate, or by returning to work after donating). The "natural" occurrence of the incentive and the ability of donors to not benefit from it reduce concerns about social desirability bias and limited sorting in the experimental literature (Harrison and List, 2004; Lazear, Malmendier, and Weber, 2012).

The paper proceeds as follows. Section 2 describes the institutional context of this study and the data. Section 3 presents the empirical analysis and findings. Section 4 offers a summary of the findings and considerations on their implications for organizations and policymakers.

Institutional Background and Data

The data used in this study originate from hand-collected information on the entire blood donation histories of all donors in an Italian town located in the north-central part of the country.6 Before describing the data in detail, we provide institutional details on the blood donation system in Italy and in The Town, and we describe the day-o¤ incentive introduced by Law 584.

2.1 Blood donation in Italy and in The Town
Blood donation in Italy is organized through blood banks run by volunteer donor associations. The associations have a central headquarters as well as town-level units. To donate blood, an individual is required to become a member of one of these associations. The three major associations, which are  present in di¤erent parts of the country and therefore do not "compete" with one another, are Associazione Volontari Italiani del Sangue (AVIS), the largest association with about 1:1 million members in 2007; Federazione Italiana delle Associazioni Donatori di Sangue (FIDAS), with about 400; 000 members (Caligaris, 2007); and Fratres, with 150; 000 members (in 2000).7 The affiliation is to a local unit of the national associations, and blood donors predominantly donate in the town where the unit with which they are affiliated is located. In The Town, blood donation is managed by AVIS, and donations of either whole blood or blood components take place in The Town’s public hospital, Monday through Saturday from 8 to 11 a.m. Donors do  not make appointments, and donate on a "first come, first served" basis.

Only individuals between 18 and 65 years of age are allowed to donate blood.  Italian law limits the frequency of donations of blood and blood components. Male donors must wait at least 90 days between donations of whole blood, and females 180 days (since 1991).8;9 The time required for a platelet or plasma donation is about one hour, compared to an average of twenty minutes for a whole blood donation. Including the time to reach the donation site, the waiting time before the donation and the resting time at the hospital after the donation (which is longer for donations of whole blood), on average, a blood donor should expect a commitment of about two hours.

2.2 An explicit incentive to blood donors: A one-day paid leave of absence for employees
According to the National Law 584 of 1967, all donors who are employed at a private or public organization have the right to a paid day off work on the same day that they donate blood or blood components. Employers are refunded by the state for the related salary expenses they incur (including social security and other contributions). From the employee’s standpoint, this provision is equivalent to sick days in addition to those that he or she has by contract, with an important difference. Unlike an illness absence, when the employee is required to stay home (a medical inspector can be sent to  check for the sickness claim), this requirement does not hold in case of an absence for blood donation.10 Employees in other countries are typically  allowed time during work hours to donate blood, usually without deductions in salary or accrued leave. However, these provisions just give donors the  material time to make their donation, and, typically, the donor has to return to work after donating. The benefit from Law 584 goes well beyond giving some rest to donors after their donation; thus it represents an explicit, substantive reward to blood donors.

The typical work week of an Italian employee is Monday to Friday. Most businesses, as well as most public workplaces, do not operate on Saturdays. Some exceptions are represented by hospital doctors, teachers (in Italian public schools, attended by nearly all students, only Sundays are off), and some employees in public workplaces such as City Hall clerks in offices open to the public. A further notable exception is given by stores, most of which are closed, by law, on Monday mornings and are open on Saturdays.11

2.3 The data
Searching the archives of both AVIS and The Town’s hospital, we identified all of the Association’s members (and, therefore, all of The Town’s blood donors) from 1983 to 2006. For each individual donor, we obtained the entire donation history over this 24-year period. Information on donors includes sex, age, blood type, and the date when each individual became a donor. Following AVIS’s practices, and to mitigate the risk of including in the sample individuals who have moved out of The Town or ceased to donate due to health reasons, we consider a donor to be "active" in each year if that donor has made at least one donation in the previous two years or in the current year, and exclude her from the sample otherwise.12 Crucially for the goal of this study, we have information on the donors’employment statuses. AVIS asked its members to report their occupation. In some cases, donors report it with great precision, indicating their exact job and even the name of the employer. In most cases, however, donors report very broad categories, such as "employee" or "self-employed." Therefore, we cannot distinguish jobs in the finest way, but we can reliably define three categories: employees, self-employed, and donors out of the labor force. The self-employed category also includes business owners, and individuals out of the labor force are students, homemakers, retirees and those who report to be unemployed. This classification (together with a number of robustness tests described below) will allow us to identify the relationships of interest. AVIS does not update its members’employment statuses on a regular (e.g., yearly) basis. Within the period covered by our data, updates were made in 1985 and 2002. We therefore limit our analysis to two subperiods, 1985-1989 and 2002-2006, attributing the same employment status in the initial year of each period to the following four years. This introduces the possibility that, for some donors and in some years, occupation is recorded with error. We show below, however, that our results are essentially unchanged when we restrict the analysis only to the years 1985 and 2002, when occupation is accurately measured.

Table 1 presents descriptive statistics of the donors. The left panel shows data on all active donors, while the right panel focuses on donors who were active in both periods, and for whom we have complete labor market information. The number of donors has increased over time, going from 845 in 1985 to 2,332 in 2006, and so has the fraction of female donors, from 24% in the mid 1980s to 30% in more recent years. The pool of donors has aged slightly over time, moving from an average age of about 37 to about 40 years. The average number of donations per year was 1.9 during 1985-89 and 1.8 during 2002-06. In 1985-89, 68% of donors (62% in 2002-06) were employees, 9% (12% in 2002-06) were self-employed, and 23% (25% in 2002-06) were out of the labor force. Of the 845 donors who were active in 1985-89, 338 (40%) were still active in 2002-06. Reaching the age limit is what mostly explains why donors cease to donate and hence drop out of our database between 1985-89 and 2002-06. 62% of the donors who were below the age of 40 in 1989 were still active in 2002 (67% for males and 46% for females).13

Of the 159 donors who were active in both periods 1985-89 and 2002-06, 59 changed labor market status between 1985-89 and 2002-06 (73% of the transitions were between being an employee and being out of the labor force). Compared to the overall sample, the donors who were active in both periods had been members of AVIS for a longer period (7.3 years vs. 5.1 years in 1985-89), and tended to make more donations per year (2.5 vs. 1.9 in 1985-89). The distribution of these panel donors across labor market statuses is not too dissimilar from the overall distribution. Finally, only 9% of the donors in the panel are females, compared to 24% overall in 1985-89, but this discrepancy is mostly due to patterns in missing occupation information, as shown in Appendix Table 1.14

Empirical Analysis

We begin our empirical investigation by providing descriptive evidence on the behavior of donors of di¤erent job market statuses. We will consider the donation frequency and the choice of the donation days of these categories of donors, as well as the take-up rates of the day-off benefit. We then use regression analyses to assess whether Law 584, by increasing the economic benefits of donating for employees, leads to more blood donations.

3.1 Descriptive evidence

3.1.1 Donation frequency by employment category
Figure 2A shows the distribution of the number of donations per year by labor market status: employees, self-employed, and out-of-labor-force donors. The differences in the distribution of donation frequencies across these groups are substantial. The percentage of active donors not making any donation in a given year is 15% on average, 13% among employees, almost 18%
among the self-employed, and 20% among the non-employed. Similarly, the share of donors making one donation is 23% among employees, 27% for the self-employed, and about 30% for the non-employed. Conversely, while about 44% of employees make 3 or more donations a year, this fraction is just 37% among the self-employed and 35% for those out of the labor force. The findings are very similar when we limit the analysis to male donors, as reported in Figure 2B, to account for the fact that female donors have tighter restrictions on the number of donations they can make per year. Kolmogorov-Smirnov tests confirm that the distribution of donation frequencies for employees is signi…cantly di¤erent from that for the self-employed and that for the donors out of the labor force (p-value < 0:001 for both the overall and the male distributions). In contrast, no statistically significant di¤erences are found (p-value = 0:133 overall, and 0:260 for males) between the donation frequency distributions of theself-employed and those out of the labor force.

Even though the higher donation frequency of employees suggests a positive e¤ect of the day-off policy, there are other possible explanations for this finding. First, employees might di¤er from non-employees along observable characteristics (e.g., gender, age) correlated with donation frequency; second, employees might possess a higher degree of (unobservable) intrinsic altruism than  non-employees; and third, employees might donate more frequently because their opportunity cost of time is lower compared to other categories of donors. Our regression analysis below will address these concerns. First, the analysis will include controls for observable donor characteristics. Moreover, to account for heterogeneity in altruistic attitudes, we will include individualized effects in the regressions, thereby identifying our e¤ects of interest from within-individual variation in employment status. Finally, we will address the third concern by focusing, in one of our specifications, on transitions between paid employment and out-of-the-labor-force status. In fact, although it could be argued that employees donate more often than do the self-employed because their opportunity cost of time is lower, this objection does not apply to the transition between being an employee and being out of the labor force, because in this latter status, if anything, people would have a lower opportunity cost of time. We will show that our findings also hold when we compare paid employment and out-of-the-labor-force status.

3.1.2 Occupation and choice of donation day
Further descriptive evidence consistent with a positive response to the incentive of Law 584 is given by the analysis of the choice of donation day. Donating on different days of the week entails di¤erent costs and benefits for di¤erent categories of workers. Donors who are employees would need to ask for time off, or they could donate on a day when they do not work. Law 584, however, gives them the option of taking a day o¤ when they make a blood donation. Because donations can be made only in the morning in The Town’s hospital, and because most employees do not work on Saturday, donating on a Friday (if leisure is taken as a good) maximizes the number of consecutive days o¤ (2.5) that a donor-employee can enjoy. A number of studies show that workers, in fact, favor work schedules with more consecutive days off (Facer and Wadsworth, 2008; Moores, 1990; Pierce and Dunham, 1992). Therefore, one would expect donor-employees to show a preference for donating on Friday if they respond to the  incentive in the sense of making the most out of it. The behavior of this group can be contrasted, again, with that of the other categories of donors, who do not enjoy the day-off benefit. First, donors who are out of the labor force (e.g., students, homemakers, retirees) should not display strong preferences over different days of the week, because costs and benefits are very similar. Second, for donors who own and operate a business (i.e., those we labeled "self-employed"), the cost of donating varies across days of the week; they would incur a cost if they donated on a working day, but not on the day their business is closed. In particular, this group also should not have any particular preference for donating on Friday.

Figure 3 reports the distribution of donations across the days of the week by donors’labor market status, pooling the data from the two periods 1985-89 and 2002-06.15 The figure shows that a higher-than-average number of donations by employees, around 20%, take place on Fridays. The "excess fraction" of Friday donations (fraction on Friday minus 16:70%) is +3:3% for employees (t-ratio = 8:8), against -0:93% (not statistically different from zero) for donors out of the labor force, and a statistically significant -4:8% for the self-employed (t-ratio = -6:3). Donations by the self-employed are clustered on Mondays and Saturdays.Specifically, store owners’donations are clustered on Mondays, when stores are closed, and the other employers’ donations are concentrated on Saturdays, when the other businesses are closed. Finally, the donations by donors who are out of the labor force are essentially uniformly distributed across days of the week. This evidence is, again, consistent with all categories of donors making cost-benefit considerations when deciding when to donate.16

Together with Friday, note that Saturday also emerges as a favorite day by many employees. This is not surprising, for at least two reasons. First, it is possible that, for some employees, taking a full day off work is too costly (i.e., too disruptive given their role andresponsibilities in the workplace). Second, some donors might prefer not to take advantage of the day-off benefit out of a concern that their blood donation might be misinterpreted as a self-interested act rather than as a purely altruistic act, and therefore they might decide to donate on a non-workday to keep their signal "clean" (Bénabou and Tirole, 2006).

As an additional investigation of this "long-weekend effect," we divided the sample of donor-employees based on a-priori differences in the economic advantage of donating on a given day, as given by the type of employer, on which we have information for a subset of employees and only for the period 1985-89: small firms and large firms. We do not have direct information on such  measures of size as number of employees or revenues. However, given the economic structure of The Town, the vast majority of private, local employers (i.e., those firms that are not part of a large company with a national presence) can be reasonably classified as small. Large firms include banks and other companies with a presence on a national level that also operate in The Town. Arguably, workers in small firms will be more constrained in their ability to take a day off. Anecdotal evidence indicates that this is indeed the case; workers in small firms who take a day o¤ are typically required to make up for the time lost by exerting extra e¤ort upon their return. Workers in large firms, instead, might be less "indispensable," therefore bearing a lower cost from leaving work during a business day. Furthermore, although employers cannot prevent a worker from taking the day off for the purpose of donating blood, the fear of being fired might induce employees to limit their use of the incentive. This concern is likely to be more serious in small firms because the Italian law grants far stronger protection in cases of "unfair" layoffs of workers in large firms than to those in smaller firms (Garibaldi, Pacelli, and Borgarello, 2004).17 We indeed find the preference for Friday to be much more  pronounced among employees in large firms, where it reaches 30%. There are no systematic di¤erences between employees in large and small firms in the number of donations per year. Thus, it appears that differences in work practices and regulations between small and large firms translate into different norms, thereby a¤ecting the choice of the donation day but not the frequency of donations.

3.1.3 Take-up rates of the day-off benefit
Evidence from actual take-up rates confirms that donor-employees do take advantage of the day-o¤ provision. Upon donors’request, The Town’s hospital provides the donors with official documentation that they can present to their employer to prove that they donated blood so they can actually enjoy the paid day off (the employer then presents the same document to the state when asking for reimbursement). The hospital, unfortunately, does not keep complete records of these documents. We obtained only partial records (about half a year in 2006), and report the information in Figure 4. On average, Monday through Friday about 70% of donor-employees requested official documentation to prove that they had donated. This indicates that they actually took the day off rather than donating and then returning to work. Spikes in the take-up rate (about 80%) are found on Mondays and Fridays.

Figure 4 also shows that the take-up rate is about 30% on Saturday; therefore, a non-negligible fraction of donor-employees who choose Saturday also enjoy the day off and extend their weekend. As reported above, for some categories of workers Saturday is not a free day; therefore, in this sense, it is "equivalent" to a Friday for the other salaried workers. Between the Friday donors and the Saturday donors who request the doctor’s certificate, it would appear that about a quarter of donations by employees result in an extended weekend. Thus, although it is plausible that for some donors, donating on Saturday may be a way to reduce the noise in their signal of being altruistic, as discussed in Section 3.1.2 above, the high take-up rates throughout the whole week and the non-negligible rate on Saturdays are also consistent with many donors (and also the public) not perceiving any conflict or contradiction between the altruistic act of donating blood and receiving a (visible) reward for it.18

The high take up rates are also in contrast with the possibility that the "excess" donations on Fridays by employees (or the Saturday donations, for the people who work on that day) might be driven by the desire to minimize disruptions for the employer. If donors wanted to minimize disruptions to their employers, and if such disruptions were smaller on Fridays (although we have no specific evidence to believe that the last days of the week are "slower" work days in general), they could donate and then go back to work; however, the high take-up rates are inconsistent with this being the case. Finally, if causing less disruption was the prevalent explanation for the choice of the donation day rather than the willingness to fully enjoy the day-off, we should not see an increase in the donation frequency, which is, instead, what emerges from the descriptive evidence above, and is confirmed by the regression analysis that follows.

3.2 Regression analyses

3.2.1 Main results
As a first step toward establishing a causal relationship between the day-off incentive and the frequency of donations, in columns (1), (2), and (3) of Table 2 we report the results from the estimates of the following linear regression model:

Xit + "it; (1)

where the number of donations in year t by individual i is regressed on a dummy variable, EMPLOY EEit, equal to 1 if a donor i is an employee at a given point in time t, and 0 otherwise; the regressions control for period effects (1985-89 and 2002-06), four age-group dummies (18-29, 30-39, 40-49, and 50+), and individual fixed effects ( i). The individual fixed-effects, as mentioned above, control for any unobservable, time-invariant donor characteristics (including "intrinsic" altruism) that may be correlated with the frequency of donations. The inclusion of period effects ensures that the results are not driven by common trends, and theage-group dummies control for possible effects due to aging.19 In all of our regressions, the standard errors are clustered at the donor level, to account for potential heteroschedasticity and auto-correlation of the error term within individuals. The sample includes all donors with labor market information in column (1) whereas it is limited to the panel of donors active in both periods in column (2). In column (3) and beyond, we further restrict the analysis to the subsample of donors for whom we have complete information about their transitions across employment categories in the two periods. At the top of each column, we report the mean of the dependent variable. The coefficient estimate on the EMPLOY EE indicator, b , is the difference-in-differences estimator, identified out of the donors who switch labor market status. In these specifications, this coefficient estimate is positive and statistically significant.  It indicates that when donors are eligible to benefit from Law 584 (i.e., when they are in the "employee" status) they make, on average, about 0:7 extra donations per year. The effect is sizeable; it amounts to 27%-36% of the average number of yearly donations by the donors in the sample.

As mentioned above, if the day-o¤ incentive does a¤ect donors’behavior, one might expect donors who become eligible for the day-off benefit (by becoming a paid employee) to increase their donations, and donors who lose the benefit (by ceasing to be employees) to reduce them. Model (1) assumes these two effects to be identical in absolute size. However, one might not found those e¤ects to be symmetric because, as found in other contexts (e.g., Charness and Gneezy, 2009), some form of "persistence" in behavior might exist, whereby those who were induced to donate more frequently by some incentive maintain a high frequency even after they lose the incentive. To explore this, we estimate the following model:

DONATIONSit = i + 0D2002��2006 + 1NEV ER_EMPLOY EEit  D2002��2006 (2)
+ 2EMPLOY EE_TO_OTHERit  D2002��2006
+ 3OTHER_TO_EMPLOY EEit  D2002��2006 +
Xit + "it:

Here we divide donors into four categories: (a) donors who were employees in both 1985-89 and 2002-06 (the omitted category); (b) donors who were non-employees in both 1985-89 and 2002-06; (c) donors who were non-employees in 1985-89 and became employees in 2002-06; and (d) donors who were employees in 1985-89 and became non-employees in 2002-06. We define dummy variables for these groups and interact them with the dummy variable for period 2002-06, D2002-2006. We perform the analysis on male and female donors jointly (column 4), as well as separately for males (columns 5 and 6).20 The coeffcients of interest are those on the interactions of the 2002-06 dummy and the labor-market-transition indicators (because our specifications include person fixed effects, the main effects of those dummies are not identified).

The coefficient estimates on D2002-2006 are small and not statistically different from zero. Thus there was no significant change in the yearly frequency of donations for the baseline group (i.e., the donors who were employees in both periods). The coefficient estimates on NEV ER_EMPLOY EEit D2002-2006 are also small and not significant. This implies that donors who were not employees in both periods also did not experience a change in the number of yearly donations between the two periods. In both columns (4) and (5), however, the coefficient on OTHER_TO_EMPLOY EEit  D2002-2006 is estimated to be positive and statistically significant, indicating that donors who became employees increased their donations compared to donors who did not change labor market status, net of period, age-group, and individual fixed effects. The magnitude of the estimates indicates an increase by around 1 extra donation per year, or 39%-43% increase over the baseline. As for the coefficient on EMPLOY EE_TO_OTHERitD2002-2006, the estimates have a negative sign (in the case of males, this coefficient indicates that donors who ceased to be employees reduced their donations by 0:54 donations a year), although they are not statistically significant.

In column (6), we report estimates from a further restricted sample, for which we only consider transitions between being an employee and being out of the labor force (as reported above, these transitions are 73% of all transitions). We do so because a potential alternative explanation for the higher donation frequency during the "employee" status is that people might donate more when their opportunity cost of time is lower. This would be the case for transitions to and from self-employment, as part of which donors, plausibly, have a higher opportunity cost of time. However, this does not apply to the transition between being an employee and being out of the labor force (e.g., student or retired), because in this latter status, if anything, people would have a lower opportunity cost of time. The estimates in column (6) are not consistent with an opportunity cost explanation; the coefficient on OTHER_TO_EMPLOY EEit  D2002-2006 is positive, statistically significant, and larger than in columns (4) and (5), indicating that becoming an employee is associated with nearly 1:5 extra donations per year (a 57% increase over the baseline).

The negative sign on the coefficient estimate on EMPLOY EE_TO_OTHERitD2002-2006 suggests, at least directionally, a negative e¤ect of leaving employee status, and therefore losing the possibility to enjoy the day o¤. However, the small absolute size of the estimate, and its statistical insignificance, indicate some form of persistence in behavior, whereby those with a higher donation frequency (induced by the day-off incentive) tend to maintain it even after they have lower incentives to do so. On the one hand, the initial higher incentive to donate might accelerate the learning process about the costs and benefits to donate blood (e.g., about the amount of time and physical pain it entails), and thus also lead to selection. On the other hand, individuals may have full information about these costs and benefits, and the initial incentives to donate might facilitate the formation of a habit. Both mechanisms could be at work. The structure of our data makes it difficult to determine which, if any, dominates. In fact, the individuals in our panel (i.e., present in both periods) had already made several donations by the year 2002, and therefore any learning process or habit development would have occurred earlier.21 In an attempt to assess the presence of learning (and selection) in blood donation, we consider the subsample of 180 individuals who have started to donate between 2002 and 2004. We do not have information on the number of past donations for the sample of donors appearing only in 1985-89, and for the most recent years, we consider donors who started no later than 2004 in order to be able (at least potentially) to observe a sufficient number of donations by the end of 2006. For these individuals, in Appendix Figure 1 we plot the share who make the nth donation, conditional on having made n - 1 donations. The figure shows that the share of returning donors is decreasing in the number of previous donations, especially when the number of donations is low. After the 8th donation, the share of returning donors increases sharply, to reach 90 percent at the 10th donation. Thus, it would appear that any learning and updating of beliefs is exhausted after 10 donations. This is consistent with learning occurring in the first few donations when people adjust their beliefs, and some select out from donating. If there were full information ex ante, and more donations led to habit, we would have observed a  non-decreasing pattern of return donation rates. It could still be, however, that among those who selected into remaining blood donors, additional donations helped creating a habit. A longer panel would be needed to make conclusive claims. Interestingly, the share of returning donors, especially for the first few donations, is higher for the employees than for the non-employees. While merely descriptive, this evidence is consistent with an incentive e¤ect generated by the dayoff provision that, holding learning (and possibly habit) constant, provides an extra reason to donate again.

3.2.2 Robustness tests
We perform a series of tests to assess the robustness of the findings reported above. First, we estimate model (2) after collapsing the data into two periods: 1985-89 and 2002-06. In columns (1) and (2) of Table 3, each donor has only two observations (one per period), and the dependent variable is the average number of yearly donations made by each donor in each period. We do so because our key explanatory variables do not actually change from one year to the next but only from one period to the next, and, as suggested by Bertrand, Du‡o, and Mullainathan (2004), collapsing the data at the appropriate level helps in obtaining proper estimates and standard errors (clustering the standard errors, which we do in all our other regressions, is another strategy). Our results are confirmed both in magnitude and statistical significance. In column (1), where we include all donors in the panel, the coefficient estimate on OTHER_TO_EMPLOY EEit D2002-2006 is about 1, and in column (2), where we limit the sample to males, it is roughly 1:6. Thus the estimates are virtually identical to those reported in columns (5) and (6) of Table 2.

Our second robustness check addresses potential errors in attributing the same job to donors for each whole period, while having exact information only on the years 1985 and 2002. Thus, in columns (2) and (3) of Table 3 (again, on males and females first, and then only on males), we use information for the years 1985 and 2002 only, and the results still hold. If anything, the estimated effect of becoming an employee is bigger, which might indicate that measurement error in the occupation variable was biasing our estimates toward zero. Finally, we further test the key identifying assumption that the average outcomes for the "treated" (i.e., those who change employment status) and "control" (i.e., those who stay in the same employment status) groups would have followed parallel paths over time in the absence of the policy. Our interpretation of the effect of the day-off law on the frequency of donation is valid provided that no other factor that is associated with changing occupation is systematically associated with donation frequency. In particular, although in our regressions we are including individual fixed effects and age-group dummies, it is possible that the variables associated with "becoming an employee" and "ceasing to be an employee" are capturing other changes that take place when individuals reach certain ages (e.g., changes in family structure or health). This is a possibility because donors who were out of the labor force and then became employed were, in 1985-89, of relatively young age (28 years old on average). Conversely, those who were employees and then left the labor force were relatively old in the first period (40 years old on average). Our identification test here is based on the fact that the unobserved life changes that one might be concerned about are correlated with age; moreover, they should occur not only for the employment-status switchers, but also for those who remain in the same employment status. Thus, we added to the regressions a set of dummy variables for six age categories to which the donors present in both periods could belong in the period 1985-89 (18-24, 25-29, 30-34, 35-39, 40-45, and 45-55) interacted with the 2002-06 indicator D2002-2006. The coefficients on these interactions represent the change in yearly donation frequency experienced in the period 2002-06 by individuals who were of a certain age group in the period 1985-89. This specification controls for the e¤ect of aging on donations, allowing for donation trends to di¤er for different initial ages. Note that this strategy exploits variation in initial age among individuals in all employment transition categories (switchers and non-switchers). As can be seen in columns (5) and (6) of Table 3, the coefficient estimates on these interaction terms, in the full sample as well as in the sample limited to males, are small and never statistically significant, and the presence of these additional regressors does not affect the estimates on the (labor-market-transition)*(2002-06) dummies. Columns (7) and (8) report results from an even more demanding specification, i.e. one where we include a full set of age dummies (for the period 1985-89) interacted with the 2002-06 dummy, as well as a full set of contemporaneous age dummies. Once again, our coefficients of interest are largely una¤ected.

These exercises make us confident that the coefficients on the (labor-market-transition)*(2002-06) variables are capturing the e¤ect of being eligible for the day-off benefit rather than that of other life changes.22 The analyses in columns (5)-(8) of Table 3 also allay the concern about another possible confounding factor: that the relationship between higher donation rates and occupational status might be due to income differences among groups. For example, if people with higher income have a higher propensity to donate blood, and employees have on average higher income than the self-employed and those out of the labor force, then this might drive the donation patterns that we observe, independent of the presence of the day-off provision. It is difficult to attribute precise income figures to these three broad categories of workers because detailed information about the precise job is available only for a small subset of donors and, in addition, the jobs included within each category span a high variety of earning levels. However, for the typical Italian employee, salary is largely based on seniority, and wealth can be taken as accumulating over time, on average. Thus the finding of a limited e¤ect of aging is, again, evidence in favor of a genuine day-of incentive effect. Further, the evidence of a preference for donating on Fridays is consistent with donor-employees trying to take full material advantage of the privilege. If the main determinant of the higher propensity to donate was only income, we would see a more uniform donation distribution throughout the week.

Discussion and Conclusion

We have analyzed the e¤ects of a legislative provision that grants a one-day paid leave of absence to Italian blood donors. Our results indicate that the policy induces donors who are employees to make, on average, one extra blood donation per year, which represents an increase of around 40%. Because we identifed this effect by exploiting within-donor variation in employment status, our findings cannot be attributed to unobservable heterogeneity across donors of different labor market statuses. The effect was found to be robust to a variety of specifcations and sample restrictions, which corroborates our causal interpretation.

We did not find the effect to be fully symmetric; ceteris paribus, donors who become employees make about one extra donation per year whereas the reduction in donation frequency by donors who cease to be employees is small in magnitude and not statistically significant. Although more data and further research would be needed to draw former conclusions, this result is consistent with habit formation in behavior, whereby those with a higher donation frequency (induced by the incentive) tend to maintain a high frequency even after they have lower incentives to do so (somewhat similar to what Charness and Gneezy (2009) find for gym attendance). Further indication that the day-off incentive a¤ects donor behavior comes from our analysis of patterns in the choice of donation day, combined with actual take-up rates. We documented that a substantial fraction of donors who are employees choose to donate on a day that extends their weekend (notably Friday) whereas no such preference was found for donors of a di¤erent labor market status, and that is consistent with donor-employees maximizing the extrinsic, economic benefit of donating.

This paper contributes to the larger debate on the role of extrinsic incentives in stimulating pro-social behavior. Evidence from donor surveys (Lacetera and Macis, 2010b) and from research where the subjects were aware of being part of a study on the e¤ect of incentives on blood donations (Mellstrom and Johannesson, 2008) shows some indication of negative responses to rewards. However, recent field studies of actual populations of blood donors (and actual donation behavior) such as Goette and Stutzer (2008) and Lacetera, Macis, and Slonim (2012) find that donors are actually positively a¤ected by material rewards, and that these effects increase with the economic value of the reward. This is consistent with Gneezy and Rustichini (2000), who found that "large enough" rewards do enhance pro-social behavior.

In this paper, we have analyzed a large, naturally occurring economic incentive that is a step-removed from cash and, as such, a good candidate to have a substantial, positive effect on donations.23 The incentive that we have analyzed and its behavioral effects have social welfare implications in that they impact the voluntary supply of blood. The evidence indicates that the paid-day-off provision does stimulate more donations. Therefore, removing this policy (a measure, in fact, recently advocated by the Italian Employers’Association as well as by the central Government; see Il Messaggero, 2009) would likely result in a reduced number of donations by existing donors. Welfare comparisons, however, have to balance the gains from the policy with its costs to taxpayers. We found that the policy leads to about one extra donation per donor per year. The state, however, must finance all donations made by donor-employees (i.e., about three donations per donor per year). Evaluated at an average labor cost of 139 euros per day (inclusive of social security and other contributions [ISTAT, 2007]), that one extra blood donation has a cost of around 400 euros related to the day-o¤ incentive. To that, one would then add the production costs for the additional collected unit such as labor and equipment costs, and the cost incurred to separate the di¤erent blood components. Figures published by the Italian Health Ministry set the production costs of one additional unit of whole blood at around 250 euros. Thus, the incentive put in place by Law 584 would appear to be cost-e¤ective provided that the full social value of one unit of whole blood is at least about 650 euros.

Other countries have recently introduced similar provisions in related contexts. In the United States, for instance, a number of states allow certain categories of employees (e.g., public employees) to take a paid leave of absence for the purpose of being bone marrow or organ donors. Although empirical analyses are necessary to determine the effects of those provisions in di¤erent contexts, our results do suggest that rewarding altruism with paid time off can be an effective way of stimulating it. While effective in stimulating more donations, and at a unit cost likely below the social benefit,24 the day-off incentive might not be the most cost-effective way of raising donations, however. In fact, Lacetera, Macis and Slonim (forthcoming), in their analysis of American Red Cross blood drives, find that small gifts (e.g., T-shirts, mugs, coupons) lead to an increase in donations, and they calculate the cost of an extra donation to be, on average, about $250. In this study we have focused on a population of existing blood donors. Further research is needed to establish whether incentives have an e¤ect in attracting new donors (or inducing existing donors to cease to donate), and possibly facilitate learning about this activity and develop a habit to perform it. Also, the specifc incentive studied in this paper is designed so that donors can decide not to take advantage of it (e.g., by returning to work after donating, or by donating on a non-working day) and therefore allows for di¤erent tastes and attitudes by di¤erent donors to be satisfied. It would be interesting to examine whether the positive effects of the day-off incentive that we documented are due to such flexibility, or whether also other, less flexible incentive structures would deliver similar outcomes.