2 The puzzle and challenge: Our ineffective giving
(I) Limited giving to those most in need: When faced with the “girl drowning in the pond” we are willing to sacrifice substantial wealth to save a life. However, most people don’t make large donations to the very poor (see, e.g., Meer and Priday (2020)), in spite of evidence suggesting that lives can be saved for less than 5,000 USD in expectation (and people’s apparent belief that it is even cheaper).3
(II) Inefficiency of charitable giving: There is a strong case that most donations go to charities that improve well-being far less per-dollar than others. The figure below depicts estimates of ‘cost per additional disease-adjusted-life-year (DALY)’ from the most promising health interventions.4
In contrast, charity-supported interventions such as providing guide dogs for the blind routinely cost over 5000 USD per dog per year, presumably yielding substantially less than a single DALY of benefit per year. Make-a-Wish ‘wishes’ for dying children cost over 11,000 USD on average; the same money may be able to save as many as three children’s lives (on average) if devoted to (e.g.) seasonal preventative anti-malarial drugs.6
In general, most charitable giving goes towards helping people in wealthier countries,* even though it is widely accepted that far more lives can be saved and improved when funds are devoted towards more effective interventions in poor countries (see fold).7
As Burum, Nowak, and Hoffman (2020) state: “We donate billions of dollars to charities each year, yet much of our giving is ineffective. Why are we motivated to give, but not motivated to give effectively?”
The above points (I) and (II) raises two related questions:
I. “Why don’t we give more to the most effective charities and to those most in need?”, and
II. “Why are we not more efficient with our giving choices?”
To address this, we must understand what drives giving choices, and how people react to the presentation of charity-effectiveness information.8
2.1 Motivation and descriptives
Individual donors, governments and firms demonstrate substantial generosity (e.g., UK charity represents 0.5-1% of GDP, US charity around 2% of GDP).9, 10
However, most donations go towards charities that are worthwhile but improve human well-being far less per dollar than basic medical interventions in poor countries, such as antimalarial bednets (see Givewell.org). Even within the same category, more can be achieved for less: e.g. while it costs around USD 50,000 to provide a blind person a guide dog, the equivalent amount may be able to be used to prevent many cases of visual impairment from trachoma for many people (Burton and Mabey 2009) through preventative measures and surgery, perhaps more than 50.11
Social science, biology and philosophy present a range of potential theoretical explanations of how values, preferences, and biases drive this ‘inefficient altruism’. However, evidence (e.g., for ‘availability bias’, or for ‘scope insensitivity’) comes largely from small-scale experiments in domains outside of charitable-giving. It is difficult to distinguish robust, credible findings from one-off results that are vulnerable to hype, p-hacking and publication bias (echoing the ‘replication crisis’ in experimental social science (Shrout and Rodgers 2018)). Given the limited, scattered findings, we do not have a definitive picture of which factors substantially impact ‘effective giving and support for policies that reduce extreme poverty’. (We give a review of papers surveying this evidence below.)
2.1.1 Descriptives of giving (US, international) and how ‘ineffective’ it is.
This subsection section needs further work, outlining the existing evidence and basic ‘world in data’ picture for the questions below.
Some relevant questions and issues
Who makes up ‘most of the giving’? What share of this comes from ‘regular’ (vs millionaire and billionaire) givers?
General giving statistics (academic work): Meer and Priday (2020), Rooney et al. (2019) - Meer’s work finds roughly constant (about 2%) shares of incomes are given, but (check?) this does not include the billionaires
Popular work/curated sites12
The ‘Study of High Net Worth Philanthropy’ reports that the wealthy give far higher shares … around 10%*13
“Those in the top 1 percent of the income distribution (any family making 394,000 USD or more in 2015) provide about a third of all charitable dollars given in the U.S” (Source philanthropy roundtable
Writeup answering key questions (needs checking) at balancingeverything.com
- ‘More than half of the US overseas assistance [43.9 billion, including charities, corporations, religious groups, etc] comes from private donations’
- ‘Almost one-third of all 2019 money donations went to churches’
- Approximately 31% of people [globally] donate to countries other than their own. (But note the top 3 receiving countries are USA, Israel and Canada; and the US rate of giving abroad seems substantially lower than other countries.)
Who DOES give effectively?
- EA Survey ‘Donation data’: full ‘bookdown writeup’, forum post
Fitz/Kagan: Understanding Effective Givers:
In this study we attempt to understand who is predisposed towards effective giving. After providing a description of the effective giving movement, we measure support for effective giving and measure a wide range of personality traits and demographics that may predict support for effective giving. We briefly define the EA movement as an important force “we” (economists, psychologists) need to discuss.
(Micklewright and Schnepf 2007)
What are the potential global welfare gains to changing ‘where we give’?
See discussion in Singer (2019), MacAskill (2016)14
2.2 (Lack of) previous synthesis on this
There have been some relevant prior reviews, such as Caviola et al, 2021).15
However, the current project uniquely** combines (or at least, aims to combine)
A focus on effectiveness,
considering ‘choices among charities’ as well as in isolation (see fold}
incorporating recent work and developments from the ‘EA movement’,
a rigorous, sceptical approach to evidence, and
advancing a research agenda while building tools that promote robust evidence.
2.2.1 Effectiveness-specific work
Caviola et al, 2021, “The Psychology of (In)Effective Altruism” (discussed below).
(jaegerPsychologicalBarriersEffective2021?) “Psychological barriers to effective altruism: An evolutionary perspective”
outlines ultimate [rather than proximate] explanations for ineffective altruism … evolutionary perspective …; three fundamental motives—parochialism, status, and conformity
Baron and Szymanska (2011) - Heuristics and Biases in Charity: Largely conceptual, minimal survey of specific empirical/experimental papers
Gertler, Charitable Fundraising and Smart Giving (not peer-reviewed but very useful)
(Caviola, Schubert, and Nemirow 2020) “The many obstacles to effective giving” … “on how both incorrect beliefs and preferences for ineffective charities contribute to ineffective giving.”17
Caviola, Schubert, and Nemirow (2020) is highly relevant and insightful (and has some connection to the structure we present below). From their abstract:
We review the motivational and epistemic causes of (in)effective giving. Many donors view charitable giving as a matter of personal preference, which favors decisions based on emotional appeal rather than effectiveness. In addition, while many donors are motivated to give effectively, they often have misconceptions and cognitive biases that reduce effective giving.
Still, Caviola et al does not drill deeply into the strength of the evidence and the relative importance of each factor. However, this may stem from a small amount of available evidence to survey. Ideas42 (ibid) wrote:
We did not find many field-based, experimental studies on the factors that encourage people to choose thoughtfully among charities or to plan ahead to give.
In the sections below, we (intend to) compare and incorporate our outline and discussion with Caviola et al’s work, as well as the other outlines. In the folds below, we briefly compare the outlines and structure of the syntheses mentioned above.18
Comparison of other outlines: unfold each
We are working on incorporating a connection to the concepts defined in these outlines into our own sections, particularly the ‘barriers’.
Caviola, Schubert, and Nemirow (2020)
Gertler, “Charitable Fundraising and Smart Giving”
Baron and Szymanska (2011) (chapter)
2.2.1.1 Other (potentially) relevant work and examples
Gifford (2011): “seven categories of psychological barriers…” that impede “climate change mitigation and adaptation”
Stefan Schubert’s ongoing work and book in progress
2.3 Definitions - “Efficiency” versus impact
It is important to define the concept of talking about. What, precisely, is this ‘effectiveness’ or ‘impact’ of a charity we are focusing on? It is not trivial to get this right and there are some delicate and hotly debated questions even within the EA movement. Nonetheless, I sketch the basic idea in the math below.
Aside: ’but what about uncertainty?20
\(G_j\): The total donations given to charity \(j\) during some interval; i.e., the charity’s income.
\(B_j(G_j)\): A function defining the beneficial outcome achieved by charity \(j\) with the total donations \(G_j\).
Here, we are referring to \(B_j(G_j)\) as (the improvement to) some ultimate outcome: Lives saved (i.e., deaths averted), quality adjusted life years (QALY) added, QALY weighted by age, Disease-adjusted (DALY), future happy lives generated, sentient suffering averted, etc.21
The important distinction here: \(B(G_j)\) does not refer to a simple intermediate ‘output’ such as ‘antimalarial nets provided’ nor ‘textbooks purchased’. We are referring to the social outcome of ultimate value; an outcome that could be valued in and of itself.
This naturally takes into account both the ‘technical efficiency’ in terms of how many units of output can be produced per dollar, and the rate at which each unit of this output boosts the ultimate outcomes of interest.
The ‘production function’ is (perhaps tautologically) the product of two terms:
(Total or marginal) impact per dollar = output per dollar \(\times\) impact per output
A donor may care about the ‘impact’ of her own donation; i.e., she may want to know the difference in outcomes that her donation achieves everything else equal. In other words, the difference in the ultimate outcome in a world with versus without her donation.
Small donor assumption: For a small donor (perhaps someone who donates less than USD 100,000), we may assume that this “rate of benefit” will be the same for both the first and the last dollar she donates.23
Thus we consider the marginal impact, as a simplification:
\(B_j^\prime (G_j)\) for the marginal donor.
I assert that \(B_j^\prime(G_j)\) is the quantity that GiveWell (and perhaps other EA charity raters) are attempting to measure.
We know (evidence cited in fold):
\(B_j^\prime(G_j)\) is much larger for the most impactful relative to the most popular charities.
Increased benefits could be achieved if donations were “reallocated” towards more impactful charities. An individual who gains value from her giving through it’s impact alone would naturally donate to only the one charity that has the greatest marginal impact, the charity \(j\) with the greatest \(B_j'(G_j)\) term. If every donor is doing this, then as an ‘equilibrium’ result, every charity receiving positive donations should have the same last-dollar marginal impact. In maths:
\[\begin{equation} B_k^\prime(G_k) = B_j^\prime(G_j)\forall j,k s.t. G_j>0, G_k>0 \end{equation}\]Caveats (unfold):
But we do not seem to be doing this. Again: billions are given to charity, and these charities clearly have vastly different marginal impacts, even among those that seem to target very similar outcomes.
2.4 Why (under which models) is this a puzzle?
Does the set of facts mentioned above constitute a “puzzle” for our Economics and Psychology models, or are there obvious existing explanations? In Economics terms, are donors mysteriously ‘’leaving money on the table’’ or are they simply optimizing given their their preferences and constraints? What could explain this?
We return to this in a later section, while classifying potential ‘barriers’ to effective giving.24
2.4.1 Aside: Economic models of giving
Traditionally, economic models are essentially ‘if then’ exercises meant to illustrate what factors might be driving outcomes, perhaps facilitating empirical testing. They are not, in general, claims about people’s preferences, how they behave, or what motivates them. The discussion of charity in the Economics literature has tended to focus on two extreme models, focusing on modeling assumptions that allow simple mathematical ‘if-then’ proofs. Most of these models involve an individual making choices between a private good and a homogenous charitable good so as to maximizing a “utility function” (value function), subject to their budget constraints
A adaptation of the ‘Public goods model’ (often attributed to Becker (1974)) in which individuals care about the total amount of a charitable good provided (no matter who contributes it), either because it benefits them directly or because they care about other people. (The utility of others might be subsumed into one’s own utility function). This is sometimes labeled (confusingly in my opinion) called “pure altruism”.
An extreme representation of a ‘Warm glow’ model (James Andreoni 1989) in which an individual’s utility increases (presumably at a decreasing rate) in the ’amount they sacrifice’ towards the charitable good.
It is also important to note that the ‘warm glow’ presented in the Economics literature does not map to only a single psychological mechanism or theory; it maps to many such mechanisms. There is some ambiguity as to which motivations should be subsumed in this ‘warm glow’. According to Andreoni (who coined the term):
The concept of warm-glow is only a convenient reduced-form representation for deeper and more complex considerations of givers (J. Andreoni 2006)
In fact, each of the above models is an extreme representation that was originally used to provide insight into how to think about issues such as “does government spending crowd-out private donations” and “how do individuals respond to tax-benefits for giving”? In its simplest form, each model yields predictions that are profoundly unrealistic and can easily be “rejected” by real world data.
Much is left ambiguous in each model, as well as in other fairly prominent models such as the “Impact” (Duncan 2004) or “Identification” (Atkinson 2009) models of giving, or reputation-based models (Harbaugh 1998)
For the present discussion, our focus, in considering these models, is:
(When and how) does the “actual impact of one’s gift” have an impact
into the ‘good feeling’ one gets from giving
or the thing the individual is maximizing
… in a way that drives donation choices?
Overview of research question/problem: Why don’t people give in an evidence-based way?↩︎
For an introduction to the field of effective altruism, see this article from effectivealtruism.org, as well as here for an introductory resource list.↩︎
This does not seem to be for a lack of overall generosity. As a prominent example, US Americans give roughly 2% of their income to charities (and ‘give far more’ if we include transfers to children and other family). However, giving mostly goes to local and domestic charities which are far less effective per dollar than the ones recommended GiveWell.org. G↩︎
Many of these interventions have room for more funding from private charity. Even the upper bound of these estimates is routinely below 100 (USD) per DALY.↩︎
Figure 7.1 is from original work by The World Bank. Views and opinions expressed in the present adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank.↩︎
See the bottom row of the google sheet here (accessed on 19 Nov 2021) for the cost-per-life-saved estimate.↩︎
See, e.g., evidence from a US nationally representative survey (self-reports) in Reinstein (2011), or lists of the most popular charities in wealthy countries, e.g., this YouGov poll for the UK.↩︎
Note that these two questions are not identical: the first asks about the amounts given to the most needy charities, and the second about the choice of charities conditional on giving.↩︎
(“CAF World Giving Index 2018 | Research into Global Giving Behaviour” n.d.; “U.S. Charitable Giving Tops $400 Billion for First Time” n.d.)↩︎
Jason Schukraft: Do the ‘masses of donors’ matter, or only the multimillionaire response? The average person … do small donations add up Also, knowing more about how average people to respond to analytical information (in an other regarding /social context) will inform how to influence good long term decision-making. … how to get USDA to care about animals and the government to care about the long term↩︎
The latter figure is an extrapolation from the more conservative estimates in the Givewell Blog. Note that while the effectiveness of the trachoma interventions seem to have been overstated in the past by Peter Singer and others. Still, they appear to be at least an order of magnitude more effective at prevventing harm from blindness as do guide dogs. See also Sullivan, 2013 on the cost of training a guide dog.↩︎
These numbers were gathered quickly for a timely answer. While the sources seem reputable, they should be examined more carefully.↩︎
We need to doublecheck this source. This is also consistent with a rough extrapolation from Meer and Priday (2020) appendix table 4↩︎
This is a placeholder: we should present and discuss the numbers and arguments more carefully here.↩︎
There has also been some unpublished or non-academic work: ‘Behavior and Charitable Giving’ (Ideas42, 2016), ‘Charitable Fundraising And Smart Giving’ (Gertler, 2015), and ‘The Psychology of Effective Altruism’ (Miller, 2016, slides only).↩︎
Greenhalgh and Montgomery (2020) performed ‘a systematic review of the barriers to and facilitators of the use of evidence by philanthropists when determining which charities (including health charities or programmes) to funds’ using PRISMA guidelines.↩︎
Mainly a series of MTurk experiments (and some with ‘effective altruists’ recruited through the EA Newsletter/Facebook) involving hypothetical donation choices, but also includes some literature review and conceptual breakdown↩︎
We (aim to) return to making some related comparisons, comparing our ‘breakdown of barriers’ in a later chapter to that of other papers.↩︎
Consider: The measures used are relevant to how we consider issues such as charity ‘quality ratings’ and ‘overhead aversion.’↩︎
Thanks to Jason Schukraft for bringing up this point. He noted “There’s a small pool of donors who deny that GiveWell has identified the most effective global poverty/health charities because (e.g.) GiveWell is too focused on”randomista” interventions and doesn’t give enough weight to “systematic” interventions.”↩︎
As noted, there are disagreements over how and whether we should trade off among these outcomes. Issues such as population ethics, and the importance of sentience and experience — come to the fore. We will ignore these for now.↩︎
This is obviously an oversimplification. To achieve the beneficial outcome the charity will require many intermediate inputs (or “outputs” as noted above), including ‘management’ and ‘careful targetting of programs’. Some charities may be able to acquire these inputs at better prices than others, and some may also use a more efficient mix of inputs.↩︎
See this talk/article from Owen Cotton-Barratt for more on some key economic effective altruist concepts, including heavy tailed distributions, diminishing marginal returns and comparative advantage↩︎
Economists love when we can say that something is officially a Puzzle. It is an achievement in itself to be the researcher who first discovered a Puzzle, even if we have no clue how to resolve it.↩︎
I return to this discussion in section @ref(subst-framework)↩︎