The Cross-Cause Fund allocates money to high-quality philanthropic funds supporting effective work in their respective cause areas.
Methodology
Learn how we select funds for re-granting, estimate cost-effectiveness across cause areas, and model our allocation recommendations.
How it works
Fund selection
Funds and areas under consideration
- Democracy protection (non-partisan): We are conducting research to determine whether we can responsibly include this area while remaining compliant with our legal obligations.
- Climate change: We are assessing whether existing climate-focused funds meet our quality and data requirements for inclusion.
- Effective giving (meta): We may investigate funds focused on movement-building, cause-prioritization research, and similar activities that aim to increase the overall effectiveness of philanthropic giving.
Some funds that we intend to vet for inclusion in the future: Ambitious Impact funding circles (Mental Health, Meta Charity, Seed Network, Global Health, Strategic Animals), Giving Green, The Life You Can Save, AI Safety Tactical Opportunities Fund, Astralis Foundation, Survival and Flourishing Fund, Animal Charity Evaluators' Recommended Charity Fund (ACE), Senterra Funders, Macroscopic Ventures, EA Infrastructure Fund, Coefficient Giving Funds (e.g., Global Growth Fund, Abundance & Growth Fund).
How did we select these funds?
Broad cause coverage
Fund quality
Data availability
Decision relevance
Legal and organizational restrictions
Cross-cause prioritization model
This section describes the model used by RP Cross-Cause Fund and RP Donor Compass. The model delivers allocation recommendations to users based on a range of inputs, including the fund-by-fund cost-effectiveness estimates and several other variables described below.
Rethink Priorities’ Default Inputs
You can find a table summarizing the default worldview inputs Rethink Priorities currently uses for the Cross-Cause Fund here.
Our worldview-independent fund-by-fund cost-effectiveness inputs for the Cross-Cause Model (behind both the CCF and the Donor Compass) can be found here.
A more detailed description of our model’s methodology is available here.
Model inputs
Fund-by-fund cost-effectiveness estimates
As described in the section below.
Worldview credences
The model proposes multiple worldviews (coherent bundles of normative commitments and empirical beliefs regarding the parameters listed below). Each worldview is assigned a credence reflecting its weight in the final aggregation.
Multipliers for effects of transformative AI
It’s possible that AI will significantly affect the cost-effectiveness of the causes we model in ways we cannot predict. For example, it may increase funds’ cost-effectiveness by speeding up the pace of research and scientific discovery. Or, AI-related developments might undermine a fund’s theory of change or reduce its counterfactual impact. We allow users to incorporate these uncertainties by assigning a multiplier to each fund representing their best guess for how its expected cost-effectiveness is impacted by developments in AI (outside of extinction risks cutting short the impact of the fund, which is modelled separately). A multiplier between 0 and 1 means that the expected cost-effectiveness will be lower after factoring in AI-related changes, whereas a multiplier above 1 means the user anticipates a fund’s cost-effectiveness increasing.
Wash-out and cluelessness discounts
Wash-out is the possibility that a fund’s effects fade over time, or that the future is smaller than assumed, so that RP’s cost-effectiveness best-guess estimates overstate its longer-term effects. Cluelessness is the possibility that our best guesses about whether those effects are good or bad overall are arbitrary, in which case a worldview sets them aside. Each worldview sets a wash-out factor and a cluelessness factor, from 0 to 1, for each of six time periods: 0-4, 5-9, 10-19, 20-99, 100-499, and 500+ years. A factor of 1 counts a period in full, and 0 leaves it out.
Skeptical profiles
A worldview’s skeptical profile says how much it distrusts (and hence adjusts downward) the most extreme simulations of a fund’s estimated effects. For example, one upside-skeptical profile gives simulated cost-effectiveness estimates above the 97.5th percentile less and less weight, pulling the most extreme values down. This mostly affects funds whose estimates include a few extreme values at the top of the distribution.
Probabilities of moral status
Each worldview gives a probability that each kind of non-human animal has moral status, meaning that its welfare matters morally at all. Each animal’s conditional moral weights (below) say how much its welfare counts if it does. Keeping the two apart lets risk-averse worldviews treat the chance that an animal does not matter (e.g., because non-sentient) as a risk. Helping a chicken that has a 50% chance of (not) feeling pain is a gamble, while a certain weight of 0.5 (meaning a chicken can/does suffer only half as much as a human) is not. (See Moral weights conditional on moral status, below.)
Risk profiles
Risk-averse worldviews penalize funds with, e.g., a chance of having no impact, or a chance of having negative effects. Each fund's impact is modeled as a cost-effectiveness distribution. Each worldview is assigned a risk profile, which determines how the distribution is collapsed into a score. A risk-neutral profile will collapse a distribution to its expected value; risk-averse profiles will give different scores based on the level of risk.
Moral weights conditional on moral status
To compare funds with different types of effects, worldviews assign weights. Each human year lived with disability (YLD) is assigned a weight of 1. Other effects are given different weights.
The other human effects are:
- human life-years saved
- human life-years enabled (lived by people born later)
- income increases, measured in doubling the income of one person for a year
Animal effects are measured in years of suffering averted (calibrated to be animal-YLD equivalents). Each animal gets a weight, reflecting the value of averting a year of suffering for an individual in that species.
Worldview aggregation method
As explained above, the model assigns credences to a suite of worldviews. These worldviews must make joint decisions on how to allocate funding. A process for joint decision-making is known as an aggregation method. The model uses six aggregation methods. Each one allocates the whole budget, based on the worldviews and their credences. We are unsure which method is best, so each method gets its own credence, separate from the worldviews’ credences, which indicates how confident we are in that method. The final allocation is the weighted average of the six methods’ allocations, weighted by these credences.
Steps we follow to output allocation recommendations
Step 1: Estimate each fund’s cost-effectiveness
Produce worldview-independent cost-effectiveness estimates for each potential recipient fund. This looks like a cost-effectiveness distribution for each fund, based on thousands to millions of model simulations, in units native to the fund’s work (e.g. human life-years enabled, or years of chicken suffering averted).
Step 2: Define worldviews
The model user specifies one or more worldviews, each defined by their best guesses on the factors listed above. These worldviews make explicit the user's normative and empirical assumptions underlying the allocation.
Step 3: Score each fund under each worldview
Each worldview applies its views on moral weights, cluelessness, how to deal with risk, etc., to each fund’s cost-effectiveness distribution (Step 1), to produce a single score for each fund under that worldview (see Model inputs above).
Step 4: Aggregate fund scores across worldviews to build a portfolio
Based on the scores that each worldview gives to each fund, the worldviews jointly “decide” on an allocation, via some compromise we call an aggregation method (e.g., Nash bargaining or Maximize Expected Choiceworthiness). Because no single method is universally accepted as correct, the method actually used is the weighted average of different aggregation methods.
The model allocates the available funding in $2M increments. Each increment is allocated to the “best” funds according to the aggregated worldviews.
Which funds are “best” may change as the portfolio is built. Each time a fund receives money, its marginal cost-effectiveness declines according to that fund’s diminishing-returns factor. A fund whose factor reaches zero receives no more funding.
The output is a recommended allocation across funds expressed as a percentage of the total budget (e.g., our Cross-Cause Fund allocation assumes $200M).
Model limitations
Uncertain inputs
Several parameters the model relies on are not well established, and reasonable changes to them can meaningfully shift the final allocation. We've consulted experts in our network to arrive at best guesses, but readers should treat these as informed estimates rather than settled facts. We don't think the effective altruism community currently has good answers to many of these questions, and we plan to involve more experts in future iterations. Several examples are worth flagging:
- Attributing extinction risk across causes. How to split total extinction risk among AI, biological, and nuclear risks is genuinely uncertain, and small changes in these shares can shift significant portions of the allocation. Similarly, how much these extinction risks can be reduced by specific interventions is an uncertain assumption in our modeling work.
- Far future and extinction-risk impact. It is very difficult to confidently assess the long-run impact of a project. We mitigate this by assigning probability distributions to all inputs, by considering different scenarios with different levels of conservatism (for example, capping the time horizon at roughly 100 years in some worldviews), and applying different risk-weighting functions (some are risk-neutral, others penalize outcomes that are very unlikely to occur). In the future, we may expand upon the methods we're using to treat future uncertainty.
- How much credence to give to each moral theory. How much weight to assign to different moral theories also affects the result, and there is no clear-cut answer to how to set these credences.
- Cost-effectiveness estimates at the fund level. We estimate each fund's cost-effectiveness by combining estimates for the interventions we expect it to support, using a mix of historical and forward-looking data. The underlying data varies in quality: not all funds have detailed cost-effectiveness estimates available, and those that do vary in how transparent or self-critical their assumptions are. We have tried to level the playing field through adjustments so that conservative, self-critical funds are not penalized for their conservatism, but this requires subjective judgment. The model also does not capture our subjective confidence in particular grantmakers, and funds may reasonably change direction over time, so what each fund actually supports going forward may differ from the mix we have assumed. We hope to keep improving this as capacity allows.
In addition to involving more experts, we are working on a more advanced sensitivity analysis to assess how changes in the distributions of our inputs affect our ultimate recommendations.
Limited set of funds and cause areas included
We've included a small number of funds, and several cause areas are not represented at all. Most notably, climate and democracy protection are not currently covered, and neither are "meta" causes aimed at improving giving decisions or expanding the number of people working on priority causes. We hope to include democracy and potentially climate work later this year. Rethink Priorities is a 501(c)(3) organization and is legally prohibited from supporting or opposing political candidates or parties, so giving to political candidates is also outside the model. Beyond these specific exclusions, we did not include many strong funds due to capacity constraints. Absence from this list should not be read as a judgment that a fund is less cost-effective than those we've covered.
The ability to capture interaction dynamics
Interaction effects between grantmakers
In certain cases, what one funder does will impact what another funder does. Sometimes they may pull a new funder into the space or crowd out a different actor (this is sometimes known as "funging"). If what we care about is overall impact, then we should consider these effects. However, these effects are difficult to model due to a lack of data on what someone would do in a counterfactual scenario, and they can differ substantially within and between areas.
For our purposes, we have selected funds that are in touch with other actors in their respective spaces, which should limit the impact of this concern. However, actors not in those spaces, particularly outside effective giving circles, may be causally affected by their actions in ways that are difficult to account for.
This may be a smaller issue in smaller fields, where it's less likely that other funders will step in. And a bigger issue in fields with more money is that a number of actors can make any given grant and may be interested in doing so.
Interaction effects between grant recipients
In some areas, grant recipients largely operate independently of one another, but in others, particularly in the policy space, a single bad intervention or action could harm the prospects or outcomes of related interventions. In our model, this type of concern may be most salient for AI grantmakers (where multiple actors are trying to influence policy in the same countries and locations) and less relevant for GHD (where interventions operate under different theories of change in very different countries). For animal interventions, certain actions may be more tightly correlated, like corporate campaigns where success may depend on keeping the public on your side, and a single rogue actor could harm the chance of success for all.
Using judgment calls, reasonable people could disagree with
A number of the methodological decisions behind the model do not have a clear best answer, and choosing a different one would change the recommendation. For example, one of the risk profiles we use is skeptical of benefits that are very unlikely to occur. We currently implement this by fully counting all outcomes except the top 2.5% and discounting those in the top 2.5%. The exact threshold is a judgment call, not a settled methodological standard, and a different threshold would have produced different results. Similarly, all interventions are currently modeled as having diminishing overall effectiveness as funding scales, with no distinction between an intervention whose probability of success falls and one that becomes more likely to backfire or have no effect. Different moral theories and risk attitudes would treat those situations differently. We are running sensitivity analyses to better understand how much these choices matter, and we hope to refine our treatment in future iterations.
No settled answer for how to combine recommendations across moral views
Aggregating across moral views is a young and unsettled area, and the choice of method does influence results. We have extensively reviewed the available literature on this topic and commissioned a report from Harry Lloyd, a philosopher specializing in moral uncertainty. Our model implements several aggregation methods and combines them using the weighted factor model described in this document.
Representing non-utilitarian moral theories in simplified form
Our model uses simplified versions of non-utilitarian theories, and does not yet include virtue ethics or the capabilities approach. We hope to add both in future iterations. One simplification is worth flagging.
To produce a recommendation, the model has to combine impacts across many people or animals (for example, weighing small harms to many against a large harm to one) and assign each option a single number so options can be compared. This naturally fits the reasoning of consequentialist theories. But non-consequentialist theories, such as deontology, do not fit this structure. The non-consequentialist theories represented in our model are better thought of as proxies for the recommendations these theories would make, rather than as reflecting the structure of their reasoning. We plan to keep iterating on our treatment of these theories in future versions.
More broadly, in its current form, our methodology rests on the assumption that the goal of giving is doing as much good as possible, where 'good' is defined by the donor's own worldview weights across welfare, risk reduction, and moral views. It is aimed at readers who broadly accept this quantitative lens.
More on our methodology
Why explicit modeling is so important
Learn about why explicit quantitative models, despite their limitations, remain more reliable than intuition or heuristics for high-stakes resource allocation decisions, and why the alternative of informal reasoning often hides the same problems while adding new ones.
Read The Spreadsheet in Your Head Is Misallocating Your Giving.
Dive deeper into this topic with In Defense of Modeling.
Aggregation methods
Learn about how to combine views across multiple moral theories when making donation decisions, why no single aggregation method should receive more than 50% of your credence, and why a weighted average across approaches may be the most defensible strategy.