Friday 18 January 2013

Death Bias

Siouxie Wiles asks an interesting question: do Kiwis' expectations of what they'll die from, and expectations of current death rates, match up with aggregate death rates?

If people overestimate the risks of some kinds of death, this could lead policy to be biased towards spending too much trying to avoid that kind of death and too little on other risks.

Mike Dickison produced this infographic on actual causes of death in 2009.

They then contrast this with the survey findings:

The full set of slides is here.

I love this initiative. But there's a pretty big problem. If I expected to die of a heart attack, emphysema, or cancer in my 90s, I'd probably say "Old Age" rather than any of those specific disorders on a survey. I might even fail to think about "people in their 90s dying of cancer" when answering how many Kiwis I think die of cancer every year. Somebody in his 50s, that's dying from cancer; somebody in her 90s... I might just not think of it that way. Maybe I should think of it that way, but it's also pretty plausible that somebody dying of cancer in their 90s would have died of something else a few weeks later but for the cancer.

So I'm not sure we can say people are underestimating their risk of dying from diseases associated with old age, like cancer, circulatory disease, or respiratory disease, when "old age" is a survey response. Similarly, I can imagine answering "suicide" to that questionnaire even though I have absolutely no intention of killing myself except in old age if my expected future utility stream is sufficiently low.* Those expecting law changes around euthanasia and thinking that a reasonable end may well answer "suicide".

I'm not sure what the best way of fixing this might be. You could go back to the MoH data and restrict the actual death sample to deaths among those under a plausible "old age" cutoff line - say 70 - and see whether survey expectations among those providing an answer other than "Old Age" matched those expectations. You could re-run the survey, noting explicitly and up-front that the New Zealand death statistics do not consider "Death by Old Age" to be a category, and that respondents thinking about people dying in old age should think of the proximate cause of that death.

This is important. In the slides, when asked whether suicide, melanoma, or road accidents were associated with the most deaths, melanoma got the most votes despite being responsible for the fewest deaths. By survey response, respondents thought that melanoma was the worst, followed by suicide followed by road accidents. In reality, the rank order is suicide, road accidents, then melanoma. This could very plausibly lead to underinvestment in initiatives that prevent suicide relative to investments targeting melanoma. Note also that the "old age" confound is likely to attenuate the degree of measured public bias here if some people would count melanoma deaths as just being old age if the death were experienced by an older person - real numbers could be worse.

I look forward to seeing where Siouxie's team goes with this. It's worth their following up.

* Does "Voluntarily going in for brain plastination or cryonics in old age in hopes of later uploading if the singularity hasn't happened yet" count as suicide, or does failing to do so count as suicide? I'd lean towards the latter.


  1. Great to see our efforts are generating discussion already :). Just to be clear though, the question we posed on how people thought they would die was an open one, we didn't give people any options. Paul Gardner then went through everyone's answers and sorted them to give us the graph.

  2. I'd expected that; you wouldn't have included "old age" as a toggle option. But the open-ended answers makes it hard to match up perceptions and reality, except on the questions where you have folks rank-order death causes where "old age" is unlikely to have attenuate responses to any of the items.

  3. very frightening, I feel anxious already, i think i die of anxiety shortly, can I go on your list