Hi all,
I’ve come up with an intervention that I may pursue. I want to see if you find the cost effectiveness model sensible, or what you would change.
I appreciate that CSA is not an EA cause area, but I’m nevertheless applying an EA lens to improve on this problem.
The intervention is a virtual, scalable training program for people who work with children (teachers, social workers, police officers, health staff). The aim is to make adults better at helping children to disclose abuse, and therefore to reduce the duration of CSA after the first incident.
I appreciate candour/a well-intentioned roasting, so don’t hold back!
Assumptions about the problem:
1. 1 in 20 children in the UK experience CSA.
- That stat is from the NSPCC.
- Prevalence research varies, with the highest I’ve seen citing 1 in 5.
2. CSA costs UK society £89k per child victim.
- That number comes from a 2021 Home Office report. It estimated that the total cost of CSA was ‘at least’ £10.1 billion in 2019… and that’s only England and Wales, not Scotland and Northern Ireland. The report estimated a ‘per unit’ cost of just over £89k per victim.
- I’ve gone with £89k, despite the not-full coverage of the whole UK and despite inflation since, to err conservatively.
3. The average duration, from first incident of abuse to cessation, is three years.
- I couldn't find research which clearly states the average duration, so this is subjective and based on my anecdotal experience from talking to survivors. I consider it a conservative estimation.
4. The resulting harms of CSA are evenly distributed throughout those three years. So, each year of ongoing CSA costs £29,667.
- This is a blunt assumption without empirical backing. There’s very little research to go on regarding which ‘phase’ of abuse is the most damaging/expensive to society. Common sense tells me that the first incident, or few incidents, are the most damaging to the victim; but that might not correlate with societal cost. Whether or not a legal process is initiated, for example, probably has a big impact on cost but isn’t correlated with abuse duration. For simplicity, I’m dividing the £89k cost per victim by 3.
Assumptions about the intervention, which are all very subjective:
1. Every individual who goes through the training interacts with 30 children in the 12 months post-training.
2. 1 in 20 of those children will be experiencing CSA.
3. The base likelihood (without training) of those child victims disclosing to the professional in question (their teacher, their doctor, or whoever) during that year is 10%.
4. Coming into contact with a professional who has gone through the training increases that likelihood to 20%.
5. Therefore, our training results in 0.15 additional disclosures per professional, per year.
6. Where those additional disclosures occur, they all result in the abuse being stopped.
7. Each case of abuse that is stopped would otherwise have continued for another year, of the three years average duration assumed.
8. Each of those disclosures averts £29,667 in societal costs.
From here, we can calculate an expectation of £4,450 in averted costs, per trained professional, per year.
This would be about a 44x ‘ROI’ if the training costs £100 per person.
How this compares with other EA mental health interventions
If we convert the £4,450 in avoided harm per professional-year into DALYs using NICE’s £20–30k per QALY/DALY benchmark, this equates to 0.15–0.22 DALYs averted. At £100 per training, that works out to £450–670 per DALY.
For comparison (I relied on ChatGPT to pull the benchmark figures, which match HLI/GiveWell analyses. Please correct me if I’ve misquoted):
- StrongMinds, providing group interpersonal therapy for depression in sub-Saharan Africa, is estimated at $77–115 per DALY (£60–90).
- Other community-based mental health programmes (e.g. BasicNeeds) often fall in the $100–500 per DALY (£80–400) range.
- ChatGPT also offered more optimistic scenarios for calculating DALY’s, but this was the most conservative. Under promise and over deliver, as they say…
I thought of ending this post with more info on the intervention itself and why I think it would double the chances of a professional being able to help victims to disclose CSA, but that would deviate from the overall model which I’d like to focus on today. DM me if you're curious about that.
Thanks for sharing this!
My first reaction is that 30/20*10% = 0.15 years of avoided CSA per £100 training seems implausibly effective.
Hi Larks, thanks for your comment. Yes, that's perhaps the hardest assumption to justify. I need to find some way of bolstering it, but unfortunately, tying training outcomes to actual disclosures is not done, anywhere, by any current provider. Probably because it is so hard to pin down, but also, runs the risk of revealing that current efforts don't actually make a clear difference. The EA in me wants to be able to show how our offering led to that solid outcome, though. Needs more thought!
I think your assumptions around the costs and their even distributions over three years would need to be refined a bit more to work out more precisely the areas in which the proposed intervention would or would not reduce costs. To my reading of the research you cite here, the costs include things such as prevention programs and and police, court, and prison costs in dealing with the perpetrator etc., as opposed to just costs experienced by the victim (which I think you are right are also not just linearly experienced over the years). I therefore think it would be useful parcellate out more specifically which aspects of the costs you think would be reduced.
This is a good point. I may be misinterpreting the cited research, but if the proposed training increases disclosures of CSA, that will (a) presumably reduce 'costs as a consequence' (which includes a monetised value of the physical and emotional harm suffered) but (b) plausibly increase 'costs in response' (due to increased police, prison, and safeguarding costs)
That's exactly the kind of nuance I was hoping to get. Sharp by name, sharp by nature! Yes, if there are increased disclosures then it follows that costs as a response will go up. However, that in turn could have a deterrent effect.
If it would turn out that the response costs outweigh the savings, then from a purely costs perspective, the most cost efficient thing could be to... do nothing. Don't try to prevent CSA once the harm is already there.
Response costs should probably be treated as either neutral or as part of the justice system’s baseline functioning, rather than as part of the marginal calculation. Which, of course, would change the model...
Hi Jamie, thank you - I initially read Matt's comment below as being part of this comment. So, thank you also for your astute point about the parcellation (?) of costs. Definitely needs more thought.
Yes I think Matt's point is a specific example of what you might find if you split out (parcellate) the different forms of cost and assess how each is affected by the intervention. I was thinking, for example, that court or police costs are not divided by three by reducing the duration of the abuse by 3x, because they aren't dealing with one third of the perpetrator.
Got it, thank you. Yes, that's a good next step as I can estimate at which point each category of costs comes into effect.
According to this, there are only ~50,000 reported cases of CSA in the UK, while there are 800,000 cases (1/20 of 5-19 year olds). This gives us a current reporting rate closer to 6%.
As opposed to my assumed 10%, I take it. I'll update the model accordingly, thanks. Were there any other objections?
This was the only one I thought of. I will be interested in hearing about the intervention if you're willing to share. Good luck regardless!
Thanks for writing this Siobhan, and sorry this comment is very late. I currently see a few key issues (this comment), and a couple of broader concerns (future comment).
Combining these, the new cost effectiveness is (1/5 * (19.8/29.7) * 2 * £4450 = ) £1190 averted per professional per year, which is £1680-£2520 per DALY.
I think it's possible that I've misunderstood some/all of these, so would appreciate sanity checks from others.
Here are some less important/certain factors that I think you could also take into account with your model:
Much for me to reflect on there. Thanks again!
You haven't misunderstood and I very much appreciate the clarity and candour of your feedback!