Transitory Income Windfalls and Charitable Giving: Evidence from Norwegian Register Data, 1993–2021

Abstract

This paper studies the impact of unearned, transitory income shocks on charitable giving using Norwegian administrative data. We exploit the random timing and size of lottery wins and our long time period (1993–2021) to estimate both short- and longer-term impacts. We find no meaningful effect of small windfalls. Yet, windfalls exceeding $10,000 induce a long-lasting increase in the likelihood to donate, the absolute level of donations and the share of annual income donated (conditional on donating). We show that this is consistent with individuals thinking of large transitory income shocks as a long-term addition to their annual income.

Macro-, public and labour economists share a keen interest in how economic actors respond to income shocks in terms of, for instance, their consumption patterns and labour supply (Johnson et al., 2006; Jappelli and Pistaferri, 2010; Christelis et al., 2019; Powell, 2020; Fagereng et al., 2021; Commault, 2022). This interest derives at least in part from the implications that such responses have for various ‘economic questions, including the effect of fiscal policies, the relation between income and consumption inequalities, and the dynamics of business cycles’ (Commault, 2022, p.96). While standard economic theory suggests that economic actors should not react very strongly to transitory income shocks, empirical evidence highlights that such responses can be substantial in the short term. For instance, Johnson et al. (2006), Fagereng et al. (2021) and Commault (2022) all found a marginal propensity to consume around 0.25–0.3 over the months following a transitory income shock.

Similarly, behavioural economists are interested in how experimental subjects respond to endowment windfalls, which reflect (transitory) shocks to individuals’ income in a laboratory setting. A key finding here is that experimental subjects treat income/endowment windfalls differently from earned income. For instance, a larger share of windfall income is likely to be shared in dictator games relative to earned income (Cherry et al., 2002; Oxoby and Spraggon, 2008; Carlsson et al., 2013). One potential explanation of this behavioural pattern is that ‘an individual's sense of entitlement over income is related to his contribution in generating it’ (Tonin and Vlassopoulos, 2017, p.112).

This article extends our understanding of the effect(s) of unearned, transitory income shocks by using population-wide Norwegian register data to study their impact on one specific type of consumption, namely, charitable giving. In extant research, two main approaches have been implemented to study how income shocks influence charitable giving. The first approach exploits aggregate-level shocks. List and Peysakhovich (2011), for instance, showed that stock market booms are associated with higher levels of charitable giving, while downturns have smaller negative effects. Meer et al. (2017) studied the impact of the Great Recession and showed that it caused a sharp decline in charitable giving. However, aggregate income shocks such as the Great Recession or stock market booms and busts represent a highly compounded treatment, such that it is hard to isolate the pure income effect.¹ A second approach relies on lab experiments implementing a charitable version of the dictator game—which requires participants to split (un)earned income between themselves and a charitable cause (Carlsson et al., 2013; Cartwright and Thompson, 2023; Eckel et al., 2023)—or simply offering participants the opportunity to donate part of their earnings (Drouvelis et al., 2019; Drouvelis and Marx, 2021). Recent reviews of this literature by Umer et al. (2022) and Cartwright and Thompson (2023) illustrate that 30–50% of windfall income is commonly allocated to charities. However, levels of generosity are commonly much lower in the real world, suggesting that results from the lab may have limited external validity (List, 2007; Carlsson et al., 2013).

Using Norwegian register data over the period 1993–2021, our study offers new empirical evidence on the impact of unearned, transitory income shocks on individuals’ charitable donations. Our identification strategy relies on the random nature of lottery winnings. These create a random and transitory income shock for the winner (Imbens et al., 2001; Lindahl, 2005; Cesarini et al., 2017; Fagereng et al., 2021; Ghomi et al., 2024), which we exploit using an event-study analysis (Schmidheiny and Siegloch, 2023). The lottery data encompass large wins (exceeding 100,000 NOK) for the entire 1993–2021 period and small wins (exceeding 10,000 NOK) up to 2006 (10 NOK ≈ 1 USD ≈ 1 EUR in the period under analysis). We link lottery wins to detailed data on charitable giving and other individual-level characteristics (including age, gender, education, immigrant status, income and assets). Data on individuals’ charitable donations are available from 2004 onwards, and cover all tax-deductible donations exceeding 500 NOK (i.e., the legally defined lower limit for the tax deductibility of personal donations to charitable groups; more details below). The resulting dataset offers several unique advantages.

Firstly, lotteries and sports betting in Norway are exclusively provided by two state-owned companies, and each year about 50% of the Norwegian population participates at least once in the numerous games they organise. As all prizes above a minimal threshold are automatically reported to the tax authorities (even though they are tax exempt), our administrative data offer full coverage of all such wins. The random nature of wins (conditional on playing) and the very high number of observations facilitate precise estimation of causal effects. Secondly, in contrast to lab experiments, winning a monetary prize in real life is not automatically followed by a prompt to donate. This avoids any bias towards over-estimating generosity in our analysis (List, 2007; Carlsson et al., 2013). Thirdly, lottery prizes in our dataset range from very modest sums to prizes that vastly exceed the winners’ annual income. This diversity enables us to analyse the potential heterogeneous effects of different windfall sizes on charitable giving. Finally, our long time period enables estimation of the short- as well as longer-term consequences of transitory income shocks.

Our empirical analysis starts by showing that individuals experience a significant increase in both net and gross assets following a lottery win. While the immediate increase in assets accounts for about 55–60% of the windfall, the long-term impact of large lottery wins stabilises at around 20% of the windfall. This highlights that most of the transitory income shock is spent within four years. The remainder of our analysis then focuses on how this windfall spending affects charitable donations. Our main findings illustrate that small prizes (above 10,000 NOK; roughly $1,000) display negligible and often statistically insignificant impacts. When we focus on wins exceeding 100,000 NOK, we find that winners are up to 7% (or 1 percentage point) more likely to give. In the first years after their win, charitable donations conditional on donating rise by 364 NOK (from a pre-win average donation level of 3,284 NOK among donors), which subsequently dwindles to an insignificant long-term increase of 121 NOK. Donations as a share of annual earned income initially jump by 0.09 percentage points (from a pre-win baseline average of 0.83% among donors), and settle at a persistent increase of 0.05 percentage points. Finally, when we translate the transitory income shock into an equivalent lifetime annuity (Cesarini et al., 2017; Lindquist et al., 2020), we find that a 10,000 NOK increase in the win annuity increases donation levels by just under 10 NOK (conditional on donating). While this marginal propensity to donate out of windfall income is small, we estimate that it is still about five times larger than the marginal propensity to donate out of equivalent earned income. Overall, these findings show that a transitory income boost leads to a small—but remarkably persistent—increase in charitable giving.

These findings differ from two previous studies exploiting lotteries to assess the impact of income or wealth shocks on charitable donations. Kotsadam and Somville (2024) recently found that winning a housing lottery in Ethiopia had no significant impact on winners’ charitable behaviour. Their research design differs from ours, however, in important respects. First, the illiquidity of housing assets may limit the ability of this wealth shock to increase donations. Second, Kotsadam and Somville (2024) derived their donations data from a survey, which might be prone to measurement error and differential reporting bias (Ring and Thoresen, 2024). Finally, they measured donations using a modified dictator game where the charity is pre-selected by the authors (cf. Drouvelis et al., 2019; Drouvelis and Marx, 2021), which creates both an artificial donation situation and a highly restrictive choice set. Closer to our setting, Ydnesdal (2023) relied on the same Norwegian data source, and reported very weak short-term effects of general income shocks on charitable donations. Yet, he investigated a much shorter time period (i.e., 2012–7), aggregated the data to the household level and only relied on lottery wins as an instrument for changes in income levels (rather than directly estimating the local average treatment effect of a lottery windfall).

Our findings also complement a vast and largely US-based literature studying how individuals’ income relates to their charitable contributions (Meer and Priday, 2021). The evidence shows that more affluent individuals tend to donate more to charitable causes (Andreoni and Payne, 2013), but there is less agreement that they also donate a larger share of their income (Neumayr and Pennerstorfer, 2021; Adena et al., 2024). Yet, empirical studies based on such field data are susceptible to several confounding factors. For instance, a ‘hard-coded’ preference to care for others could affect both charitable contributions and labour market earnings. Natural talents might induce individuals towards pursuing higher education, which could lead to higher earnings (Card, 1999) as well as shaping preferences towards charitable giving (Bettinger and Slonim, 2006). Reverse causality could also be an issue if pro-social behaviour is rewarded in the labour market (Hackl et al., 2007; Sauer, 2015; Baert and Vujić, 2018). Our exogenous income shock offers a stronger identification of the income-donation relationship.

1. Institutional Setting and Data

1.1. Charitable Giving

The voluntary sector in Norway accounts for about 2% of GDP, and Norwegian households provide approximately 45% of voluntary organisations’ revenues (mainly via membership fees and donations). Personal donations to charitable groups—including non-profits, voluntary organisations and religious institutions—became tax deductible in Norway in the year 2000. Norwegian tax legislation thereby imposes restrictions on the donation level and the recipient charity. With respect to the latter, only donations to approved entities are tax-deductible (see Online Appendix B for the legal criteria in tax law §6–50). Non-profits and voluntary groups can apply to national tax officials to be part of this tax-deductibility program, and become registered in a public directory upon approval. As the tax-deduction scheme is set up to support organisations with a broad impact (and explicitly excludes political parties), only non-profits and voluntary groups with a (inter)national scope can qualify for tax-deductible status. Currently, about 500 organisations are included in the registry, and the top beneficiaries include humanitarian groups like ‘SOS Children's Villages’, ‘Plan International’, ‘Doctors without Borders’ and ‘Save the Children’.

With respect to the level of tax-deductible donations, Norwegian law stipulates both an upper and lower limit.² Annual donations need to exceed 500 NOK (about $50), but remain below a set cap, to qualify for the tax deduction. This upper cap has changed over the years, and was set to 6,000 NOK in 2004, 12,000 NOK in 2005–13, 16,800 NOK in 2014, 20,000 NOK in 2015, 25,000 NOK in 2016, 30,000 NOK in 2017, 40,000 NOK in 2018 and 50,000 NOK in 2019–21. Because of the tax deduction, the tax system creates an implicit subsidy for charitable giving. The relevant income tax rates were 28% from 2004 to 2014, 27% in 2016, 24% in 2017, 23% in 2018 and 22% in the 2019–21 period. This means that donors’ tax bills are reduced accordingly, resulting in net outlays of 72–8% of the gift for income earners (see also Ring and Thoresen, 2024).³

For every donation received, organisations on the list of approved entities must submit a digital report to the tax authorities with detailed information about the donated amount and the donor's social security number. This not only ensures that most deductible gifts are made electronically instead of in cash (as is common in door-to-door collections), but also that the relevant tax deductions can be automatically processed as part of individuals’ tax returns. Our dataset covers these donations included in individuals’ tax returns starting in 2004, and for the entire Norwegian population over age eighteen. While the donations data we obtained from Statistics Norway are censored at both the lower and upper caps for the 2004–8 period, we have access to uncensored data on individual-level donations starting in 2009. Observe also that although the tax deduction cap is not very high, only 11.0% of donors hit the cap in 2004, which drops to 5.5–6.8% of donors in the 2005–8 period following a doubling of the upper cap in 2005. As the cap is gradually raised further over time, the share of donors hitting the cap reaches 1.0% of donors in 2021.

Online Appendix Figure A.1 shows the evolution of the total level of donations observed in our sample. This increases from roughly 1 billion NOK in 2004 to almost 4.5 billion NOK in 2021. Still, the donation data available to us do not cover all charitable giving since people may also contribute to organisations outside the tax-deductibility scheme or contribute in cash (which may not be registered via the deductibility scheme). Survey data reported in Sivesind (2015) indicate that Norwegian households self-report contributions to charitable purposes for a total of 4.5 billion NOK in 2013. Although survey data could be prone to over-reporting as a consequence of social desirability and self-selection biases, Online Appendix Figure A.1 suggests that taking this number at face value would imply a coverage of just over 50% of total annual contributions in our sample.

Figure 1 displays descriptive statistics on charitable giving over the period 2004–21 as a function of individuals’ annual income (displayed in deciles, calculated separately for each year). We display the share of the Norwegian population (aged eighteen or older) with at least one year with annual donations exceeding the tax-deductible limit of 500 NOK (left-hand plot), the average contribution exceeding the tax-deductible limit of 500 NOK among those donating (middle plot) as well as the average share of income donated by donors (right-hand plot). Note that individuals in the lowest income decile often report very low income levels. As this can lead reported donations to be several hundred times larger than total annual income, we remove the bottom income decile from the sample when creating the right-hand plot. We take the same approach in all analyses reported below where the dependent variable is the share of income donated.

Fig. 1.

Income and Charitable Giving (2004–21).

Note: The plots display the relationship between charitable donations and individuals’ income. Income is displayed in deciles on the x axis (calculated separately for each year in the dataset). In the left-hand plot, we show the share of the Norwegian population (aged eighteen or older) with at least one donation exceeding the minimal tax-deductible limit of 500 NOK. The middle plot displays the average tax-deductible contribution among donors. The right-hand plot displays the average share of income donated by donors (removing individuals in the lowest income decile from the sample).

On the whole, Figure 1 indicates a near-monotonic increase in the likelihood (from 5% among the poorest to more than 20% among the richest) and level (from 3,000 NOK among the poorest to 4,500 NOK among the richest) of charitable donations. In sharp contrast, the share of donations in total earned income is monotonically decreasing from more than 2% among the poorest donors to roughly 0.5% among the richest donors. This downward trajectory is consistent with research showing that the poor often ‘contribute substantial shares of their income to charitable causes’ (Adena et al., 2024, p.633). As discussed in the introduction, these substantial differences along the income distribution need not reflect the causal impact of income on donations due to several potential confounders. Hence, we rely on exogenous income shocks from lottery wins in our analysis.

1.2. State-Owned Lottery Monopoly

Norwegian law grants an exclusive right to two state-owned companies—Norsk Tipping and Norsk Rikstoto—to offer lottery and betting services. The latter company specialises in horse racing, while the former company offers a wide range of games including lotteries, sports betting, scratch cards and so on. Norwegian authorities actively prevent foreign companies from offering similar services within the Norwegian territory. This ensures that most—if not all—lottery and betting activities in Norway take place via its two state-owned companies. In 2021, about 2.1 million individuals participated in one of the games offered by Norsk Tipping and Norsk Rikstoto, which is about 50% of the Norwegian population aged eighteen or older. It is important to note, however, that Norsk Tipping has a very dominant position with revenues of roughly 47 billion NOK in 2022, against merely 3 billion NOK for Norsk Rikstoto. Furthermore, sports betting accounts for only about 10% of Norsk Tipping's revenues (Lotteri- og stiftelsestilsynet, 2024). The vast majority of revenues—and of win payouts since, on average, 81% of revenues are paid out to players—thus derives from various games of chance. As such, we henceforth refer to ‘lotteries’ and ‘lottery wins’ for ease of reference.⁴

All prizes won through the games offered by Norsk Tipping and Norsk Rikstoto are—with very few exceptions—paid immediately and in full rather than spread over time. They are also exempt from taxation, and thus are excluded from annual earned income statistics. However, larger wins are automatically reported to the tax authorities. The threshold for this automatic reporting was 10,000 NOK (about $1,000) up to 2006, and 100,000 NOK (about $10,000) from 2007 onwards. This implies that we have overlapping data on charitable donations and lottery wins exceeding 10,000 NOK for the three-year period 2004–6, and for the period 2004–21 for wins exceeding 100,000 NOK. While individuals can choose to include wins of any size on their tax returns, this is not legally required below the applicable legal threshold. To avoid any self-selection concerns for individuals reporting small wins, we restrict our analysis to wins exceeding 10,000 NOK for the period up to 2006 and to wins exceeding 100,000 NOK for the entire available time period.⁵ We analyse these two samples separately throughout our analysis.

Figure 2 displays descriptive statistics for lottery prizes in the periods with overlapping data on charitable donations and lottery wins, i.e., 2004–6 for wins exceeding 10,000 NOK (left-hand panel) and 2004–21 for wins exceeding 100,000 NOK (right-hand panel). In both cases, the histogram displays the number of lottery wins (on the primary y axis) in each win size interval displayed along the x axis. The black solid line shows the ratio of lottery wins relative to the median annual earned income of the winners in each win size interval (on the secondary y axis). It is clear from both histograms in Figure 2 that small(er) wins are by far the most common, even though our data also include a substantial number of large wins. Importantly, the black solid line in Figure 2 highlights that wins exceeding 100,000 NOK generally account for a substantial share of annual income, and that wins in the 300,000–400,000 NOK interval approximate the median reported annual income.⁶ On the extreme right of the distribution, we observe a significant number of income windfalls far surpassing typical annual incomes in Norway. This in part reflects the very long upper tail of the win distribution with some individuals receiving windfalls (far) exceeding 100 million NOK. We return to these extreme observations below.

Fig. 2.

Distribution of Lottery Wins.

Note: In each panel, the histogram displays the frequency of lottery wins (along the primary y axis) for specific win size intervals presented along the x axis. The black solid line displays the median ratio of the lottery wins to the annual income of winners (along the secondary y axis). The horizontal dashed line reflects where the win size equals the median income level.

Table 1 provides more comprehensive summary statistics depending on individuals’ status as a lottery winner. The top panel shows data for the 2004–6 period, while the bottom panel covers the period 2004–21. The observation numbers in Table 1 first of all illustrate that most people do not experience lottery-related income windfalls, and that the majority of such windfalls are relatively small (see Figure 2). Note, however, that those not playing are included in columns (1)–(2) as non-winners, since non-winners and non-players are impossible to separate in our data (see below). The top rows of panels A and B in Table 1 highlight that the share of individuals donating is 2–3 percentage points higher among those recording a large income windfall (i.e., exceeding 100,000 NOK) relative to those not recording any income windfall. Both panels also indicate that the average donation is marginally higher and the share of annual earned income (excluding the windfall) that is donated tends to be lower among individuals experiencing an income windfall. The top panel of Table 1 furthermore suggests that any increase in absolute donation levels depends on witnessing a sufficiently large income windfall. Indeed, charitable contributions and donations as a share of annual income among individuals with 10,000–100,000 NOK windfalls are lower than among those without windfall gains, even though they are somewhat more likely to donate.⁷

Table 1.

Summary Statistics.

	No lottery prize		Lottery prize between 10,000–100,000 NOK		Lottery prize exceeding 100,000 NOK
	Mean	SD	Mean	SD	Mean	SD
	(1)	(2)	(3)	(4)	(5)	(6)
Panel A: 2004–6 time period
Donation (= 1)	0.11	0.32	0.12	0.32	0.12	0.33
Donation (NOK)	332.63	1,289.86	302.89	1,120.40	348.36	1,262.28
Donation (% of income)	0.13	0.56	0.10	0.43	0.11	0.46
Income (deciles)	5.49	2.87	6.19	2.61	6.43	2.70
Assets (net; deciles)	5.50	2.87	5.79	3.06	6.05	3.24
Higher education (= 1)	0.26	0.44	0.20	0.40	0.22	0.42
Age (years)	47.35	18.41	52.88	14.33	50.08	14.23
Female (= 1)	0.51	0.50	0.40	0.49	0.37	0.48
Immigrant (= 1)	0.12	0.33	0.06	0.24	0.07	0.25
N	10,537,588		124,596		15,673
Panel B: 2004–21 time period
Donation (= 1)	0.16	0.37	–	–	0.19	0.39
Donation (NOK)	653.43	2,793.41	–	–	677.92	2,408.64
Donation (% of income)	0.18	0.74	–	–	0.16	0.62
Income (deciles)	5.49	2.87	–	–	6.29	2.64
Assets (net; deciles)	5.49	2.87	–	–	6.11	3.07
Higher education (= 1)	0.31	0.46	–	–	0.24	0.43
Age	47.71	18.53	–	–	55.38	14.60
Female (= 1)	0.50	0.50	–	–	0.40	0.49
Immigrant (= 1)	0.19	0.39	–	–	0.07	0.26
N	70,395,843		–		584,815

Note: The upper panel displays summary statistics for the years 2004–6, where we have overlapping data on charitable donations and lottery wins exceeding 10,000 NOK. The lower panel displays comparable statistics for the years 2004–21, where we have overlapping data on charitable donations and lottery wins exceeding 100,000 NOK. Columns (1)–(2) cover all individuals not winning a lottery prize in the respective periods (including, therefore, also those that do not play), while the remaining columns cover individuals winning exactly one prize of a given size (multiple winners are excluded for consistency with the remainder of our analysis; see also below).

The remaining rows in Table 1 show that lottery winners tend to be older and male, have lower education levels and are less likely to be immigrants. Nonetheless, their annual incomes (excluding the lottery win) and net assets (including the lottery win) are located in a higher decile. These patterns persist when we identify all winners in the period 1993–2021, including about 4,700 individuals who self-report wins below the legally applicable reporting threshold. This approach maximises the likelihood of identifying winners in any given year even though a significant share of people who never win will be de facto players. Comparing individuals who never won during the 1993–2021 period with those who won one or more times, we continue to observe some differences between the two groups. The potential underlying selection issues in terms of who plays should be kept in mind when interpreting our results below, since they may affect the generalisability of our findings to the entire population.

2. Empirical Specification: Event-Study Analysis

To assess the impact of exogenous and transitory income shocks from lottery wins on charitable giving, we implement an event-study methodology (Schmidheiny and Siegloch, 2023). Let t (ranging from |$\underline{t} $| to |$\bar{t}$|⁠) reflect the ‘observation window’ where we observe our outcome variable of interest, and assume that our prime interest lies in evaluating what happens to this outcome in the ‘event window’ from |$\underline{l}$| periods before an event until |$\bar{l}$| periods after the event. The event of winning the lottery thus arises for each winner at l = 0, which effectively ‘stacks’ the data for each winner on a common timescale around this win event (with l < 0 reflecting pre-win periods and l > 0 capturing post-win periods). Using subscript i for individuals and t for years, our empirical specification can then be formulated as

$$\begin{eqnarray} {Y}_{i,t} = {\varphi }_{\rm{*}}Q_{i,t}^{\rm{*}} + \mathop \sum_{l = \underline{l} }^{\bar{l}} {\varphi }_lQ_{i,t}^l + {\varphi }_{{\rm{**}}}Q_{i,t}^{{\rm{**}}} + {\theta }_i + {\vartheta }_t + {\epsilon }_{i,t}, \end{eqnarray}$$

(1)

where |${Y}_{i,t}$| represents three charitable donation outcomes described below. In (1), the treatment variable |$Q_{i,t}^l$| equals 1 when we are l years before or after individual i wins a prize exceeding the threshold for automatic reporting to the tax authorities, and 0 otherwise. As mentioned, this limit equals 10,000 NOK up to 2006, and 100,000 NOK starting in 2007. Since donation data are available from 2004 onwards, we can at most assess two pre-win years for lottery wins below 100,000 NOK (i.e., donations in 2004 linked to a win in 2006). Hence, we set |$\underline{l} = - 1$| and |$\bar{l} = + 2$| when estimating the impact of income shocks exceeding 10,000 NOK, and limit the observation window to the period 2000–6. When evaluating income shocks exceeding 100,000 NOK, we set |$\underline{l} = - 3$| and |$\bar{l} = + 3$|⁠, and extend the observation window to the period 1993–2021 (as lottery data are available since 1993).⁸

The endpoints outside our effect window are ‘binned’ to capture any persistent effects of wins occurring outside the specified window (Schmidheiny and Siegloch, 2023). That is, we set |$Q_{i,t}^* = \mathop \sum_{l = - \infty }^{\underline{l} - 1} Q_{i,t}^l$| and |$Q_{i,t}^{**} = \mathop \sum_{l = \bar{l} + 1}^\infty Q_{i,t}^l$|⁠. The estimated end-point effect |${\varphi }_*$| thus captures the effect of lottery wins on charitable contributions in all years prior to |$l = \underline{l}$|⁠, whereas |${\varphi }_{**}$| captures the effect of all years after  |$l = \bar{l}$|⁠. Furthermore, as is common practice, we define the reference group as |$Q_{i,t}^{ - 1}$|⁠. Hence, |${\varphi }_{ - 1} = 0$| and all other effect estimates are normalised with respect to the year before the income windfall (Schmidheiny and Siegloch, 2023; Gardner et al., 2024). Given the ‘stacked’ nature of our dataset and the exclusion of non-winners from the estimation sample (see below), the comparison group l = −1 is the set of lottery winners in the year before they win their lottery prize. Our empirical approach thus allows us to investigate the local average treatment on charitable giving before and after individuals experience an income shock due to a lottery win at some point in time.

We study three different outcome variables (⁠|${Y}_{i,t}$|⁠) for individual i in year t: an indicator variable equal to 1 when individual i donated at least the tax-deductible limit of 500 NOK (0 otherwise), the absolute amount donated by individual i and the share of income donated by individual i. Hence, our analysis addresses both the extensive margin (i.e., the likelihood of giving) and the intensive margin (i.e., the level of contributions). To the extent that transitory income shocks have a positive impact on the likelihood and level of charitable donations, we expect |${\varphi }_l > 0$| for time periods where |$l \ge 0$|⁠. We can also verify the absence of any pre-trends by evaluating whether |${\varphi }_l = 0$| when |$l < 0$| (Schmidheiny and Siegloch, 2023; Gardner et al., 2024). Crucially, including non-donors when estimating the impact of lottery wins on the donation level/share may conflate the intensive and extensive margin estimates. In line with, among others, Huck and Rasul (2011), Meer (2011) and Adena et al. (2017), we therefore restrict the sample to donors when estimating the impact of lottery wins on the level/share of donations, which provides the local average treatment effect conditional on donating.⁹

Since (1) includes a full set of individual fixed effects, our identification relies on within-individual changes over time. This accounts for any individual-level characteristics that do not change over time, such as gender, innate levels of altruism or any individual-specific preference to participate in lotteries. Inclusion of these fixed effects is also important because it allows us to plausibly assume that the timing of a (big) win is random conditional on playing, thereby ruling out potential anticipation effects and pre-trends.¹⁰ Such ‘random timing of events’ is important for the identification of our relationship of interest (De Chaisemartin and D'Haultfoeuille, 2023; Schmidheiny and Siegloch, 2023; Borusyak et al., 2024). We also include a full set of year fixed effects to control for any time-specific events affecting all individuals equally.

Finally, our dataset does not allow identifying those who participate in lotteries (at least from time to time) from never players. As these two groups may differ along (un)observable characteristics, we include in our main analysis only those individuals who record exactly one lottery-related income windfall within our observation window(s). Excluding individuals who do not record any lottery win restricts the sample to those engaging at least sometimes in lotteries, which brings us closer to a ‘most-similar’ research design (Schmidheiny and Siegloch, 2023; Geys and Sørensen, 2024). This is consistent with our argument above about the randomness of a lottery win conditional on playing. By excluding individuals recording multiple lottery wins, we avoid difficulties in determining the relevant pre- and post-event periods (Schmidheiny and Siegloch, 2023; Gardner et al., 2024).¹¹ We return to these sample restrictions in our robustness checks.

3. Results and Discussion

3.1. The Effect of Lottery-Based Income Windfalls on Assets

Before we present our main results on the income-donation relationship, we first use the model in (1) to assess how lottery-based income windfalls affect individuals’ assets.¹² This can be viewed as an initial validation exercise since one would expect a lottery win to induce a substantial boost in individuals’ assets, at least when it is not immediately spent on items with no or low investment/asset value (such as cars or clothing rather than housing, stocks or bonds). The effect on gross assets may thereby exceed the effect on net assets (defined as gross assets minus debt) whenever individuals use part of the windfall to settle existing debts. The results from estimating (1) using net and gross assets as the dependent variable are displayed in Figure 3 (full details are given in Online Appendix Table A.1). The plots in the top and bottom rows respectively show event-study estimates of lottery wins exceeding 10,000 NOK using the 2000–6 period (average win equals 143,972 NOK) and exceeding 100,000 NOK using the 1993–2021 period (average win equals 830,503 NOK). The left-hand and right-hand plots display results for net assets and gross assets, respectively, and we include 95% and 99% confidence intervals.

Fig. 3.

Effect of Lottery Windfalls on Assets.

Note: The figure shows event-study estimates (with 95% and 99% confidence intervals) of lottery wins exceeding 10,000 NOK (top row plots) or exceeding 100,000 NOK (bottom row plots) on net assets (left plots) and gross assets (right plots). The event window covers two (three) years before and after the win with binned estimates for time periods outside that event window. The year before the income windfall (l = −1) is the omitted reference category, such that its coefficient is set to 0 in the figure. The average win exceeding 10,000 NOK is 143,972 NOK, while the average win exceeding 100,000 NOK is 830,503 NOK. The confidence intervals are calculated using robust SEs clustered at the individual level. Full details are given in Online Appendix Table A.1.

The results in Figure 3 first of all confirm the absence of substantively meaningful pre-trends. They also uncover a very substantial boost in individuals’ net as well as gross assets starting in the year of the lottery win. The estimated effect size in the win year lies at roughly 55–6% of the average windfall when the lottery win exceeds 10,000 NOK (in the top row plots) and 58–60% when the win exceeds 100,000 NOK (in the bottom row plots). In subsequent years, the effect of a small lottery win falls to zero, while the long-run effect is estimated at 20–1% for large wins (l > +3 in the bottom row plots of Figure 3). This would suggest that roughly two-fifths of a lottery win is ‘consumed’ immediately, and a substantial majority is ‘consumed’ within four years (see also Fagereng et al., 2021).¹³ This observation is consistent with recent findings that unearned income is often spent more liberally compared to earned income (Carlsson et al., 2013; Jackson, 2022), and may reflect that such income lands on a different ‘mental account’ than earned income (Arkes et al., 1994; Thaler, 1999). In the remainder of our analysis, we evaluate whether and to what extent charitable donations are part of this consumption pattern.

It is important to note that Norway has a wealth tax, which is applied to individuals’ net assets above some minimum threshold (which has been increasing over time). Recent work by Ring and Thoresen (2024) suggests that exogenous shocks in wealth tax exposure (such as those deriving from legislative changes) cause a reduction in charitable donations. Wealth taxation may thus affect our analysis since 14,097 out of 140,143 lottery winners (i.e., 10.1%) within the 1993–2021 period are pushed over the minimum wealth threshold in the win year (not necessarily just due to their lottery win). We find no evidence, however, that lottery winners altered their donation behaviour in the win year to remain below the wealth tax threshold. It thus appears unlikely that strategic donations due to wealth taxation bias our inferences, but we return to this issue in our robustness checks.

3.2. The Effect of Lottery-Based Income Windfalls on Charitable Giving

Figure 4 summarises our main event-study findings on how temporary income windfalls impact upon charitable donations. The top and bottom row plots respectively show the effect of lottery wins exceeding 10,000 NOK (using the 2000–6 period) and exceeding 100,000 NOK (using the 1993–2021 period).¹⁴ In each case, we provide three sets of results. The left-hand plots show the effect of income windfalls on the probability of donating. The middle and right-hand plots respectively look at the effect on the level of charitable donations and the share of income donated (conditional on donating). Full details are given in Online Appendix Table A.2.

Fig. 4.

Effect of Lottery Windfall on Donations.

Note: The figure displays event history estimates (with 95% and 99% confidence intervals) where the dependent variable is an indicator variable equal to 1 when an individual donated at least the minimal tax-deductible limit of 500 NOK (left-hand plots), the absolute amount donated by an individual conditional on donating (middle plots) and the share of income donated by an individual conditional on donating (right-hand plots). The top row plots focus on lottery wins exceeding 10,000 NOK (focusing on donations in the 2004–6 period), while the bottom row plots display estimates for lottery wins exceeding 100,000 NOK (1993–2021 period). All models cover two (three) years before and after the win with binned estimates for time periods outside that event window. The year before the income windfall (l = −1) is the omitted reference category, such that its coefficient is set to 0 in the figure. The confidence intervals are calculated using robust SEs clustered at the individual level. Full details are given in Online Appendix Table A.2.

Several findings arise from Figure 4. First, most of the point estimates in the period prior to a lottery win (l < 0) are statistically insignificant at conventional levels. Overall, therefore, we find little evidence of substantively meaningful pre-trends. Second, the top row plots of Figure 4 indicate that wins exceeding 10,000 NOK have at best very weak and short-lived effects on the likelihood of a charitable donation. In the first years following such a lottery win (⁠|$l \in \{ {0,1} \}$|⁠), the likelihood of a charitable donation increases by 0.3–0.4 percentage points (from a pre-win baseline of 11 percentage points). While this initial impact is statistically significant at the 95% confidence level, it disappears in the long run. Similarly, conditional on donating, the absolute level of donations increases by 229 NOK (or 9.6% compared to a pre-win baseline of 2,393 NOK), while the share of income donated increases by 0.06% (from a pre-win baseline of 0.81%).

Third, the bottom row plots of Figure 4 show that wins exceeding 100,000 NOK have a considerable and persistent impact on individuals’ charitable donations. The likelihood of a charitable donation increases by up to 1 percentage point after witnessing a large lottery windfall, and stabilises at 0.7 percentage points even in the long run (i.e., l > +3). Similarly, in the year of the lottery windfall, individuals on average contribute 364 NOK more to charity (roughly 11% relative to the pre-windfall baseline of 3,284 NOK). While the next three years still record higher average donation levels (218–73 NOK), this subsequently settles at a statistically insignificant long-term increase of 121 NOK. In contrast, donations as a share of annual (earned) income initially jump by approximately 0.09%, and settle at a persistent increase of roughly 0.05% (relative to a pre-windfall baseline of 0.83%).¹⁵

Even though the lottery-based income windfall by its very nature is highly transitory, our findings illustrate that large income windfalls can have immediate and highly persistent effects on donating behaviour. While the type of income smoothing expected to arise under the permanent income hypothesis can explain a small, long-term increase in the income share donated, the rise-and-fall pattern of the observed adjustment to the new steady-state would be harder to square with this theory's predictions (Auten et al., 2002; List, 2011; Fagereng et al., 2021). The observed pattern therefore appears more consistent with psychological processes that lead to the prompt spending of unearned income (Carlsson et al., 2013; Jackson, 2022), and a habit-forming effect of charitable behaviour (Rosen and Sims, 2011; Meer, 2013). Recent experimental work indeed suggests that donating in one period increases the likelihood of donating again in—usually one—subsequent fundraising period (Adena and Huck, 2019; List et al., 2021; Heger and Slonim, 2022). Our findings go beyond such short-run donation persistence, and provide evidence in favour of longer-term giving habits. Putting both psychological processes together, a material spike in charitable donations would arise when the income windfall occurs (spent from a ‘mental account’ containing unearned income; Arkes et al., 1994; Thaler, 1999), and this increase persists above pre-windfall levels even in the long term due to a donation ‘habit’ formed during the initial windfall giving (Adena and Huck, 2019; Heger and Slonim, 2022). Naturally, the (psychological) mechanisms underlying this tentative interpretation would require substantiation in future research.

As mentioned in footnote 9, one might worry that our two-step approach conflates the treatment effect on the level (and share) of donations with a selection effect. To address this potential concern, we replicate our analysis while conditioning on individuals’ pre-win gift-giving behaviour. Specifically, we estimate the effect of wins exceeding 100,000 NOK on charitable giving using data from the 2012–21 period, and focus on four subsets of individuals based on their contribution pattern in the 2004–11 period, i.e., individuals who donate every year within this eight-year period (‘always givers’), individuals who donate 5–7 times (‘frequent givers’), individuals who donate 1–4 times (‘sometimes givers’) and individuals who never donate (‘never givers’). Online Appendix Figure A.2 and Online Appendix Table A.3 show that our main results are driven predominantly by both always donors and never donors (Online Appendix Table A.4 shows summary statistics on the likelihood and level of donations before and after lottery wins across all four donor groups). As such, large income windfalls boost contributions among those who were already inclined to give, which reinforces the idea that the estimated effects on donation levels in Figure 4 are driven by a treatment effect rather than a selection effect. Furthermore, these auxiliary results confirm the long-term habit-forming effect of donations since always donors as well as never donors also display statistically significant impacts of their lottery win after three years have elapsed (Rosen and Sims, 2011; Meer, 2013; Adena and Huck, 2019; List et al., 2021; Heger and Slonim, 2022). Individuals that were initially infrequent donors, however, display little if any persistence in their donation levels after a lottery win (which seems reasonable in light of their inconsistent pre-win donation behaviour).

3.3. Robustness Checks

This section summarises key findings from a number of robustness checks (see the Online Appendix for full details).

First, the distribution of lottery wins has a very long upper tail with some individuals winning extremely large sums. We therefore exclude the 95th, 99th and 99.5th percentiles of all lottery wins above the 10,000 NOK threshold, and show that this leaves our main findings unaffected (Online Appendix Figure A.4). Large lottery wins may also push winners above the minimum wealth threshold, which could affect our inferences when wealth taxation plays a role for donation decisions (Ring and Thoresen, 2024). Hence, we replicate our main analysis including only the subset of lottery winners pushed above the minimum wealth threshold in the win year. Online Appendix Figure A.5 illustrates that this does not affect our main inferences. It should be remembered, however, that the subset of individuals pushed above the wealth tax limit is by construction wealthier than the total Norwegian population. As such, they are also likely to be less liquidity constrained, and may thereby be more responsive to a lottery windfall.

Second, our donations data are censored at both the lower and upper caps for the 2004–8 period. Even though only few individuals hit the upper donations cap (see above), we verify that our results are unaffected when including only observations from 2009 onwards (for which we have uncensored data on individual-level donations; see Online Appendix Figure A.6). Third, we extend the number of leads and lags in our model to explore the long-term effects of lottery wins. Online Appendix Figure A.7 shows that the effect of a large lottery win on the probability to donate and the absolute level of donations (conditional on donating) falls short of statistical significance after five years have elapsed. The estimated effect sizes also rapidly decline at that point, suggesting that the impact of the win is truly disappearing. The increase in donations as a share of annual (earned) income remains significant at the 90% confidence level or better until six years have elapsed, and the effect size remains remarkably persistent between 0.055–0.062 in the period four to six years after the lottery win. Overall, these longer-term impacts are consistent with the habit-forming effect of charitable behaviour (Rosen and Sims, 2011; Meer, 2013).

Finally, it is by definition impossible to observe how lottery winners would have developed in the counterfactual situation where they had not won the lottery. Yet, this counterfactual is critical for the validity of the inferences drawn from our analysis. We therefore turn to imputation-based two-stage difference-in-differences models (Butts and Gardner, 2022; Gardner et al., 2024) to provide us with an alternative approach to predict this counterfactual. In a first stage, we regress our three donation outcomes for individual i in year t against a full set of individual and year fixed effects using only pre-event observations. The residual values from this estimation across both pre-win and post-win periods impute counterfactual donation outcomes as if the lottery win had not occurred (Butts and Gardner, 2022; Gardner et al., 2024; Geys and Sørensen, 2024). Using these residuals as the dependent variable in our main estimation equation (1) then again allows estimating dynamic treatment effects, conditional on being left with as good as random variation after the correction for individual- and year-specific donation profiles in the first stage. Online Appendix Figure A.8 shows that this alternative approach to the counterfactual leaves our main inferences unaffected.

4. The Marginal Propensity to Donate from Lottery Win Annuities

Thus far, we have shown that lottery wins exceeding 100,000 NOK lead to long-lasting increases in individuals’ assets (Figure 3) and charitable donation behaviour (Figure 4). Both of these observations may be closely related, since a persistent increase in assets may lead winners to adjust their future ‘income’ expectations. Previous research indeed suggests that many individuals prefer to distribute windfall gains over extended periods, thus treating positive transitory income shocks as a long-term addition to their annual income (Cesarini et al., 2017). With that in mind, a natural extension to our analysis lies in estimating the marginal propensity to donate out of annualised windfall income. Following Lindquist et al. (2020), we convert lottery wins exceeding 100,000 NOK into equivalent annuity payouts using winners’ expected remaining life years (conditional on gender, age and win year; obtained from Statistics Norway) and a 3% annual interest rate. We then award winners these annuities for the win year as well as all subsequent years until death (setting the annuity to zero in pre-win years).

Figure 5 displays the distribution of these annuities for specific annuity size intervals presented along the x axis. This shows that most annuities are naturally relatively small, although a substantial number of them exceeds even the median annual earned income level over the relevant time period (i.e., about 350,000 NOK). Hence, these individuals would effectively more than double their (perceived) annual income.

Fig. 5.

Distribution of Lottery Wins as Annuities (2004–21).

Note: This histogram displays the frequency of lottery win annuities (along the y axis) for annuity size intervals presented along the x axis. We include all wins exceeding 100,000 NOK for the period 2004–21 (for which we also have data on individuals’ charitable donations). The median annual earned income level over the period 2004–21 is about 350,000 NOK.

To assess the marginal propensity to donate from lottery win annuities, we run a set of linear regression models using the same three outcome variables as before. The main independent variable is the lottery win annuity, and we also include a full set of year fixed effects and cluster SEs at the individual level. We focus on the subset of individuals documenting one win exceeding 100,000 NOK in the 1993–2021 observation window, even though the estimation itself only includes the period 2004–21 (since we have charitable donation data only starting in 2004). As before, our identification relies on the randomness of a large lottery win, which switches the win annuity from zero to a strictly positive number in the win year (and all subsequent years until death). This implies that the marginal effect of the jump in the annuity from the pre- to the post-win years is indicative of causal effects. The results are summarised in panel A of Table 2.¹⁶

Table 2.

Marginal Propensity to Donate from Lottery Win Annuities and Earned Income.

	Donation (dummy)	Donation (level)	Donation (share)
	(1)	(2)	(3)
Panel A: effect of lottery win annuities
Annuity (in 10,000 NOK)	0.0004***	9.922***	0.0017***
	(0.0001)	(3.281)	(0.0005)
N	947,263	109,857	107,609
R2	0.104	0.012	0.002
Panel B: effect of annual earned income
Income (in 10,000 NOK)	0.0001***	1.796***	−0.0009***
	(0.00003)	(0.432)	(0.0003)
N	944,537	109,601	107,609
R2	0.105	0.013	0.002

Note: The table shows estimates from linear regression models covering the period 2004–21, where the dependent variable is an indicator variable equal to 1 when an individual donated at least the minimal tax-deductible limit of 500 NOK (column (1)), the absolute amount donated by an individual (column (2)) and the share of income donated by an individual (column (3)). In panel A, the main independent variable equals the annuity from a lottery win exceeding 100,000 NOK (in 10,000 NOK). In panel B, the main independent variable equals annual earned income (in 10,000 NOK). The estimation sample includes only individuals winning exactly one lottery prize in the 1993–2021 period. Robust SEs (in parentheses) are clustered at the individual level. Significance: *** p < .01.

The top panel of Table 2 illustrates that the estimated impact of win-generated annuities on the likelihood of charitable donations (column (1)) remains small. A 10,000 NOK increase in the annuity yields an increase in the contribution probability of 0.04 percentage points. The effect on the absolute level of donations—conditional on donating—is just below 10 NOK, while the effect on donations as a share of annual income is 0.002%. Given our assumption that the annuities derived from lottery prizes represent permanent increases in income, we can compare these findings to the correlation between income levels and charitable contributions in panel B of Table 2 (using the same estimation sample, time period and regression model).¹⁷ Clearly, the annuity effects substantially exceed the observed correlation when it comes to both the probability of donation and the absolute level of donations. Furthermore, while the lottery annuity increases the share of annual earned income donated, higher levels of earned income are associated with a decrease in the share of annual earned income donated (as already reflected in the right-hand panel of Figure 1). These observations are consistent with the common observation that unearned income is spent more readily than earned income (Carlsson et al., 2013; Jackson, 2022).

Although all models in Table 2 are estimated without individual fixed effects, extremely similar results are obtained when including such fixed effects. This makes sense whenever (un)observable individual-level characteristics—such as individuals’ innate generosity—are relatively unimportant to explain charitable behaviour. Hence, taken together, these auxiliary findings strongly suggest that high-income individuals contribute more to charities primarily because of their higher (perceived) income, and not due to inherently greater pro-social tendencies. If individuals with lower earnings had incomes comparable to those of higher earners (e.g., following an income shock due to a lottery win), their charitable contributions would tend to become more similar to those of their higher-earning counterparts.

5. Conclusion

This study investigated the impact of unearned, transitory income shocks on economic decision-making. This issue has occupied generations of macro-, public and labour economists (Johnson et al., 2006; Jappelli and Pistaferri, 2010; Christelis et al., 2019; Powell, 2020; Fagereng et al., 2021; Commault, 2022). More recently, it also lies at the heart of behavioural economists’ endeavours to grasp how endowment windfalls affect behavioural responses in the lab (Cherry et al., 2002; Oxoby and Spraggon, 2008; Carlsson et al., 2013; Tonin and Vlassopoulos, 2017; Drouvelis et al., 2019).

Our analysis of large-scale Norwegian register data from the period 1993–2021 uncovers substantial and remarkably persistent effects from large transitory income windfalls on charitable donation behaviour. The persistence of the observed impact suggests that windfall gains could be a catalyst for long-lasting charitable gift giving, rather than creating just a temporary boost in outlays (Rosen and Sims, 2011; Meer, 2013; Adena and Huck, 2019; List et al., 2021; Heger and Slonim, 2022). While the rise-and-fall temporal pattern in our findings is inconsistent with predictions based on income smoothing over one's lifetime (Auten et al., 2002; List, 2011; Fagereng et al., 2021), the exact underlying (psychological?) mechanisms should be explored in more detail in future research.

Our findings offer an important modification to the experimental literature showing that individuals are extremely generous with windfall income (Carlsson et al., 2013; Drouvelis et al., 2019; Drouvelis and Marx, 2021; Umer et al., 2022; Cartwright and Thompson, 2023). This is clearly much less the case in our real-world setting. We believe that this difference may be explained by the fact that the timing of a windfall in a real-world setting is institutionally and temporally separate from the decision to donate. Unlike in the lab, people in the real world are not prompted to donate immediately following a lottery win. If this is indeed a key driving force, it could rationalise a framework where lottery winners are offered the possibility to donate part of their win upon payout by the organising institution. Similarly, charities may wish to ‘target’ lottery winners immediately following their win, as commonly observed in the US and UK settings (where the identity of winners often becomes public knowledge).

Our work also bears relevance to the relationship between individuals’ income and their philanthropic contributions (Meer and Priday, 2021; Neumayr and Pennerstorfer, 2021; Adena et al., 2024). Exploiting the random timing and size of lottery windfalls (Imbens et al., 2001; Lindahl, 2005; Cesarini et al., 2017; Fagereng et al., 2021), we find that a substantial portion of the higher donation levels of high-income individuals (in absolute terms) is due to their higher income levels rather than a higher pro-social inclination of the wealthy. Hence, our study offers little support for the idea that the wealthy are naturally more munificent, and that their larger donations reflect their charitable nature instead of their wealth. It would thus be hard to rationalise reduced pro-poor redistribution by arguing that the wealthy are more generous.

Naturally, our analysis opens several avenues of further research. For instance, lottery winnings are tax exempt in our Norwegian setting, while charitable donations are tax deductible. This set-up invites further analysis into the effects of tax prices on charitable giving (Feldstein, 1975; Fack and Landais, 2010; Duquette, 2016; Almunia et al., 2020). Additionally, extending our research to other contexts and types of income shock can help to generalise our findings, and thereby serve as a guide towards policy decisions aimed at encouraging charitable giving.

Additional Supporting Information may be found in the online version of this article:

Online Appendix

Replication Package

Notes

The data and codes for this paper are available on the Journal repository. They were checked for their ability to reproduce the results presented in the paper. The authors were granted an exemption to publish parts of their data because access to these data is restricted. However, the authors provided the Journal with temporary access to the data, which enabled the Journal to run their codes. The codes for the parts subject to exemption are also available on the Journal repository. The restricted access data and these codes were also checked for their ability to reproduce the results presented in the paper. The replication package for this paper is available at the following address: https://doi.org/10.5281/zenodo.14047863.

The authors gratefully acknowledge financial support from the Norwegian Research Council (Grant No. 314079; PI: Jon Fiva), and insightful feedback on a previous version from the editor (Steffen Huck), three anonymous referees, Maja Adena, Christine T. Bangum, Pierre-Guillaume Méon, Zuzana Murdoch, Salmai Qari and seminar participants at Universidad Carlos III, BI Bergen and the 2024 EPCS meeting.

Footnotes

1

Recent work by Ring and Thoresen (2024) disentangles the income and substitution effects from a wealth tax reform in Norway. They showed that reductions in disposable income due to higher wealth tax payments cause a decrease in charitable donations.

2

https://www.skatteetaten.no/en/business-and-organisation/reporting-and-industries/industries-special-regulations/voluntary-and-non-profit-organisations/tax-for-voluntary-and-non-profit-organisations/giftsdonations.

3

Previous research generally shows a positive impact of such tax incentives on the intensive and extensive donation margins (Feldstein, 1975; Fack and Landais, 2010; Duquette, 2016; Almunia et al., 2020). Similar responsiveness to the ‘price’ of donations has been observed in studies where individual donations are matched with an additional donation (Adena and Huck, 2017; 2022; Adena et al., 2024).

4

One might worry that sports betting and some forms of gambling may be affected by a player's skills, which could undermine the randomness of a win. Although our data do not differentiate wins from games of chance (e.g., lotteries and scratch cards) and sports betting, the vast majority of income windfalls de facto derive from (predominantly online) games of chance (Lotteri- og stiftelsestilsynet, 2024).

5

Our data reveal that individuals who self-select into reporting wins below the applicable legal threshold are more likely to be female and young, and also have higher education and income levels.

6

The median income level in the period 2004–21 is about 350,000 NOK (excluding individuals with zero income).

7

The absolute level of donations and the share of income donated appear substantially smaller in Table 1 relative to Figure 1. The reason is that Figure 1 reports these data conditional on donating, while Table 1 includes all observations in our sample. Hence, to recover the numbers displayed in Figure 1, the absolute level of donations and the share of income donated in Table 1 must be scaled using the share of donors. Note also that, unlike in some other European countries including Spain, there is no secondary market for winning tickets in Norway.

8

Extending the effect window reduces the number of observations for time periods further away from the event year, which leads to lower statistical power and larger SEs for their point estimates. Even so, we experiment with longer leads and lags, and show that this corroborates our main inferences.

9

Note that this approach may lead the treatment effect (i.e., an increase in donation levels due to a lottery win) to be conflated with a selection effect (e.g., individuals who start to donate after a lottery win may be ‘small’ donors). We are grateful to an anonymous referee for pointing this out, and return to this potential concern below.

10

One might worry that individuals’ propensities to play are higher (lower) in good (bad) years. Although this would not affect the randomness of a win conditional on playing, it would undermine the random nature of a win (since winning is impossible without playing). Fortunately, data on the number of unique players by year obtained from Norsk Tipping highlight a weakly increasing trend over time at the aggregate level (along with the size of the Norwegian population), with no consistent breaks in economic crisis years.

11

Multiple winners are a relatively rare occurrence, especially for wins above 100,000 NOK. Only 2,867 of the 38,449 individuals winning more than 100,000 NOK document two or more such wins in the 1993–2021 period (i.e., 7.46%). This includes 1,728 individuals winning exactly two large prizes in our twenty-eight-year period (60% of repeat winners), and reinforces our earlier presumption that any role of skill is likely to be minimal in our setting.

12

The Norwegian tax authorities calculate individuals’ assets based on official data including property-ownership registries, individuals’ financial assets reported by banks and so on. Hence, the underlying data are rarely self-reported, such that under-reporting of assets is at best a minor problem in our Norwegian setting.

13

We place ‘consumption’ between inverted commas since the taxable asset value of an owner-occupied main domicile is set to 25% of its market value in the Norwegian tax system (with an increase to 50% for the share of its value exceeding 10 million NOK). Hence, investing part of a lottery win into a new main domicile would create a substantial difference between the recorded value of the lottery win and the observed increase in net as well as gross assets. Note also that winners could give some of the windfall to their spouse, children or other family members. This would naturally limit the impact of a lottery windfall on winners’ net and gross assets, and may create spillover effects of a lottery win onto close family members. We consider such potential ‘ripple effects’ an important avenue for further research.

14

Separating the analysis by win size may be beneficial when there are heterogeneous treatment effects from the size of a lottery win (Sun and Abraham, 2021; Callaway et al., 2024).

15

Remember that our measure of income does not include the lottery win. Using a measure of total income that includes the lottery win leads to a negative point estimate for the share of total income donated in the year of the lottery win (Online Appendix Figure A.3). This suggests that individuals donate a considerably lower share of the income windfall than they had been donating from their earned income.

16

Note that the residual terms are not normally distributed. This does not affect the estimated coefficients in the regression models, but may affect the SEs. We consider this a minor concern given the high number of observations in our models.

17

In this case, we cannot exploit random income shocks, and our findings thus reflect the relationship between any change in income and charitable donation behaviour.