The single-factor trap in philanthropy

How bucketing and single-metric sorting lead to missing the best options

This is the third post in the RP Cross-Cause Prioritization series. In my last post, I argued that effective giving requires accounting for and appropriately calibrating on cost-effectiveness, evidence quality, and philosophical reasoning.

This post takes on a related problem: every funder who commits to one cause over another is making a cross-cause comparison, whether they call it that or not. Most make that judgment implicitly, through filters and category choices, rather than through explicit analysis. The goal should be to bring those assumptions to the surface, where they can be examined and tested.

When deciding how to allocate your giving, there are several steps you might follow: survey the philanthropic landscape, identify criteria that matter to you, and use said criteria to choose a cause area. Then do a careful analysis within this cause area.

This approach seems quite rigorous, but is it?

Imagine you’re apartment hunting. You’d probably spend 10 seconds setting browsing filters on neighborhood, price, and size before looking at a single listing. Then you’d spend hours scrolling through the results and touring apartments.

But somewhere in what you filtered out might be your dream apartment. It’s two-thirds the price for a 3-bedroom, and you set the view for 2. Maybe it sat on the edge of your preferred neighborhood. But because it’s outside of your filters, you never saw it.

Your filter wasn’t broken; it just answered the wrong question. You needed to ask what the best option is. Instead, you asked: Which of the options meets my criteria? The filter, not a careful analysis of the options, determined the outcome.

This is what simple heuristics can do to philanthropic allocation. Whether the heuristic we’re using is a single criterion or a cause-area label, the effect is the same: the most consequential choice in your allocation happens before the analysis begins.

The “single-factor” trap

A common way to narrow the field of charitable options is to sort by a single criterion. Sorting by a single factor makes budget allocation simpler, but it comes at a real cost: it forces you to treat very different interventions as if they’re basically the same, even when your own values would distinguish sharply between them.

Take risk as an example of a criterion. If you were to sort interventions purely based on their estimated probability of failure, you’d end up grouping a speculative policy intervention to prevent a pandemic alongside an effort to reduce suffering in farmed insects because they happen to share similar risk profiles. Whether the underlying evidence came from rigorous randomized controlled trials or a rough back-of-the-envelope calculation wouldn’t factor in at all. But single-factor sorting can make them indistinguishable, despite their very different impact potential.

In very rare cases, a single-factor approach may be appropriate. If you’re highly confident that a particular moral view is correct and that view holds one consideration that trumps all others, then a single-factor approach might prove to be beneficial. However, that requires very high confidence, which is usually unjustifiable, in what is likely a philosophically contested position. Overall, it is more rigorous to consider multiple factors and explicitly model them in line with your beliefs.

Sorting at the wrong level

Single-factor sorting isn’t the only way simple heuristics (our filters) can result in underoptimized philanthropic impact. The same problem arises when you apply the right criteria to the wrong unit: cause areas instead of interventions.

Consider neglectedness. Additional funding in an underfunded area should, in principle, carry high marginal value. But neglectedness gets misapplied when it’s assessed at the cause level rather than the intervention level. Knowing that global health receives billions of dollars per year tells you nothing about whether a specific, underexplored global health intervention is worth supporting. Against Malaria Foundation is one of the most heavily funded organizations in the effective giving space, and GiveWell still considers it a top charity worthy of additional resources. The relevant question was never whether the broader field is saturated. It’s whether your marginal dollar, directed to this specific intervention, produces more good than the alternatives.

Similarly, as with neglectedness, the criterion of tractability is misapplied when it’s assessed at the cause level rather than the intervention level. It’s common to hear that an entire field is simply too hard to be worth pursuing: policy change is hard, improving wild animal welfare is hard. But tractability isn’t a property of cause areas; it’s a property of specific interventions at specific moments. Some policy changes are achievable right now, in particular jurisdictions, under particular political conditions. Within wild animal welfare, targeted interventions and field-building are tractable today, even if ecosystem-level change remains out of reach. Ruling out a cause area because it’s generally intractable often means discarding something distinct and achievable, not because it can’t be done, but because it’s been grouped alongside something that can’t.

The structural problem with bucketing

Single-factor sorting is a special case of a broader failure of committing to categories before a thorough analysis has been done.

When you rule out entire cause areas based on a shallow initial investigation, you’re almost certainly overlooking specific interventions within those areas that would look very different under scrutiny. The less time you spend on the initial filter, the higher the risk that good options will fall through.

Consider what might happen when a new promising intervention doesn’t appear to fit into any of your existing categories. There are a few options to consider. You could ignore it because it doesn’t fit within your framework. You could awkwardly bundle it with existing categories, potentially displacing better-evaluated projects that it doesn’t actually resemble. Or, you could assess it as part of a new cause area, which introduces substantial work unless there’s an independent reason to believe that this cause area contains other promising options.

Each of these is unsatisfying. The first option is the worst because it wholly dismisses potentially valuable interventions, but all three options are symptoms of the same underlying problem.

The alternative is more modest than it might sound: evaluate interventions on their individual merits. A promising intervention stands or falls on what it actually does, not on the properties of the category it happens to be filed under.

But, cross-cause intervention comparison seems impossible?

You might argue that you can’t really compare preventing malaria deaths, reducing factory-farming suffering, and working on AI risk, for example, because they’re too different and it is better to stay within tractable, comparable categories.

But imagine two cause areas: one focused on providing momentary aesthetic experiences for the wealthiest people on earth, and the other on preventing a lifetime of suffering for millions of factory-farmed animals. If cross-cause comparison were truly impossible, you would not be able to say that anything in the second category was better than anything in the first. But almost no one actually believes that.

The moment you accept that some causes are better than others overall, you’ve already committed to cross-cause comparison. Once you accept that interventions within causes can be ranked on cost-effectiveness and evidence quality, it becomes hard to explain why that analysis can’t extend across the artificial line between categories.

Every time a funder commits to one area over another, they are making a cross-cause judgment, whether they call it that or not. Every dollar spent on X is a dollar not spent on Y. The choice to avoid cross-cause comparison doesn’t eliminate the comparison; it just makes it implicit. And implicit comparisons, made without scrutiny, are where the most systematic errors tend to hide, and where the most impact tends to get left on the table.

What this means in practice

None of this requires working across every cause area. A funder can rationally focus on a specific domain if that focus reflects the specifics of available interventions, the funder’s comparative advantages, or an explicit strategic bet. That’s different from a shallow analysis of many areas and selecting one based on simple heuristics.

The practical implication is to conduct cross-cause comparisons, even if they’re imperfect. Make your assumptions explicit so they can be tested. Evaluate interventions on their individual merits rather than on the properties of the category they happen to belong to. And when you discover a promising opportunity that doesn’t fit your existing framework, treat that as useful information about the framework, not a reason to ignore the opportunity.

Criteria and categories are useful tools. The problem isn’t in using them; it’s in mistaking them for analysis.

This post is part of a series on how cause prioritization can go wrong, and how to think about it better. See the previous posts here. Subscribe to receive the next part straight to your inbox.

Thanks to Sarina Wong, Urszula Zarosa, and Elisa Autric for feedback on this post.