proper corona admin, vol 93 - vaccines (leftovers - cribsheet)

Emily Oster included a few comments about vaccines in Cribsheet that I thought might be of interest given the current state of world affairs. First, she notes that vaccination rates tend to go down in areas with more education, indirectly addressing a perception about the types of people who oppose them. This may be due to the origin of the anti-vaccination movement, which she traces to a deliberately misleading paper published by a since-discredited doctor - the purpose of the work was to build a case for a lawsuit against manufacturers, but to me it seems the effect was to rubber-stamp a fringe idea. The government may have unwittingly lent credibility to these (and similar) theories when they protected companies from vaccine-related lawsuits - some felt this was an admission that errors were commonplace.

A good way to frame the opposing point of view is to note that although side effects from approved vaccines are exceedingly rare, the possibility of vaccine harm generally far outweighs the known debilitations of the disease. In the case of COVID, the death toll should speak for itself, but there are also concerns about the long-term effects on respiratory function for its survivors that should contextualize any downside effects of a vaccine. It's also important to note that the most significant problems - catastrophic side effects, if you will - are often observed in the first few weeks of a vaccine being administered; we're past this mark for most of the COVID vaccines currently under trial (1).

But the best argument for a vaccine is one Oster presents in her book - there are people restricted by disease or immune deficiencies who cannot get vaccinated. When I hear it like this, I recognize that my own apprehension about a fast-tracked vaccine isn't particularly relevant in the context of how I see a certain obligation healthy people have to their less fortunate peers. It's vital for each of us to think individually - there is no other way to think, really - but with rare exceptions, it's critical to remember that our thinking should always take a backseat to communal action.

Footnotes

1) I pulled this note from How to Vaccinate the World, a BBC podcast series hosted by longtime TOA favorite Tim Harford. I'd recommend it to anyone who is currently wasting time scouring news sites for vaccine information.

reading review - cribsheet

Let's have a closer look today at Cribsheet, a book I mentioned in passing back in May in a context tangentially related to the main idea of this work, which is how to use data to inform the parenting of young children. It's not to be confused with a "how-to" guide, however, for despite the occasional straightforward remark like "data says swaddling improves sleep", the difficulty of collecting reliable data for all parenting scenarios prevents the existence of such a book - how do you track cause and effect for incidents like "baby cried"? The most important thing, as Oster hints throughout, is to understand the quality of the data before incorporating it into a decision, which is a point that reemerges constantly within my book notes.

Cribsheet by Emily Oster (February 2020)

The biggest obstacle to having high-quality data in this (or any) field that studies human behavior is that the cause and effect relationship is often confounded by underlying differences that define people's choices. For example, studying the effect of a stay-at-home parent on a child's development is complicated by the ways such a household is fundamentally different from one where both parents work full-time. It's a similar challenge to understand the effect of TV, as homes with a TV differ from those without. The problem with studying various parenting techniques under controlled conditions is that parents cannot be forced into applying any methods with their kids, which means researchers have no choice but to allow subjects to determine their own degree of involvement and commitment to the study; the fact of choice always has an immeasurable bias in the result. The same applies in any study where a control group is recruited, or participants must self-report their findings - again, the decisions made by the participants compromise the standards of the randomized control study.

This does not rule out the possibility of gleaning productive insights from data analysis. One specific recommendation I loved was to rely on distribution information about infant weight loss to better contextualize what might otherwise be alarming changes in those first few days. It speaks to the power of disaggregated data to help people relate across their differences - in this case, parents redefine their standard for "alarming weight loss" so that they can care for their infants. Disaggregation is a regular theme in Cribsheet, appearing in sections about breastfeeding (locations with poor water supplies should always encourage breastfeeding, since alternatives would involve direct use of that water) or reading (children who are read to don't see additional gains in other areas like math when compared to those who were not read to yet received a similar amount of attention, suggesting something specifically important about reading).

The ideas I'm likely to remember best are those I'd categorize as informed hunches, which seem to come not from the data but rather the wisdom of the aggregated experiences across a countless number of parents - that consistency is a critical factor in sleep training, for example, or that interactive reading (asking open-ended questions as you go) is a productive way to read to children. I also liked the note that repeated exposure is a helpful way to get children used to new foods as well as the observation that discussion rarely improves a child's behavior. Of course, as I look over these one last time, I'm struck by an odd realization - these all apply to adults, in varying degrees; perhaps we can all benefit in some ways from Cribsheet, regardless of our parenting status. 

TOA Rating: Three cribs out of four.

tales of two cities – helmet data analytics

I recently read Emily Oster's Cribsheet, her data-driven examination of the endless recommendations about parenting young children. Her primary tactic is determining the validity of study methodologies in order to understand the strength of various statistical conclusions. Her consistent refrain is to be careful with any data that does not separate the type of person from the parenting decision because no researcher can ever conclude from such studies whether the type of person mattered more than the decision in terms of influencing the observed outcome.

The book got me thinking about the ways I’m exposed to phony analytics. The bike helmet is a great example. Is the type of person who uses a helmet different from the type of person who does not use a helmet? I'd say yes, and would like to know the extent to which the safety gains are due to the nature of the person rather than the helmet. The question, unfortunately, has many layers. The helmet itself isn’t a going concern until there is some kind of head impact, and in those moments it's clear that you'd rather have the helmet. But do helmeted riders crash more often? Some researchers have studied whether wearing a helmet makes a person more likely to get hit by a car, generally exploring one of two hypotheses: (a) the helmet invites riskier driving or (b) the helmet encourages riskier cycling. It's tough to answer these questions and I'm grateful for those like Emily Oster who do their best to educate us with the truth.

One angle I haven't heard much about is how helmet ownership changes riding decisions. I'll offer my story as anecdotal evidence – after getting a helmet, I've felt less safe without it, and this had led to fewer rides. (Those in lockdown may be able to relate to a different version - having to wear a face covering, I feel less safe, and I go out less often.) I ride less often because when I forget to bring my helmet with me, I don't ride a bike, whereas in the past since I never 'forgot' the helmet I didn't own, in these situations I would have pedaled on as was my original intention. If we assume that bike accidents happen with equal probability regardless of helmet use, one sure way to lower the number of accidents is to ride less, and people who feel less safe will ride less.

Again, I simply pose the question of whether helmet use increases safety by emphasizing danger, which leads to fewer bike rides – I’ll leave any official verdict to the experts. It's not going to change my decision to ride with a helmet. But from my point of view, whatever the perceived benefit, it surely is being exaggerated by ignoring the marginal decisions made by people like me. The failure to account for such an effect is the exact sort of thing that leads people to (correctly) question the accuracy of the metric, and although the difference may be insignificant, whenever doubt is cast on the truth, the skeptics have an opportunity to misrepresent the error, discredit the overall idea, and make it more difficult for others to make good decisions for their health and safety.