4 ms·

Clearly I was not claiming that random assignment is un ethical generally, but specifically in the case of assigning the degree of physical affection/cuddling t

Clearly I was not claiming that random assignment is un ethical generally, but specifically in the case of assigning the degree of physical affection/cuddling to a baby is probably unethical. As someone who has been involved in multiple longitudinal studies, although not in epigenetics, I can say that they are more expensive. In terms of what I said about prior evidence, I was coming from a cognitive psych background where we do not typically deal with record-based data. We deal with participants directly. It is expensive and time consuming. Longitudinal studies usually follow cross-sectional evidence.

gwern 9y ago
> but specifically in the case of assigning the degree of physical affection/cuddling to a baby is probably unethical. Alright, since you refuse to think, I'll give it to you for free: do an intent-to-treat design with incentives for additional cuddling for either the parents or hired helpers like nannies, or passive controls; as no infant receives less physical affection than they would otherwise, no one is hurt by the randomization and it is not unethical, but you still get a causal effect. (Not all of them will use it, and the ones who do, the difference will be smaller than if you could remove cuddling, but that's merely a matter of power.) Such designs are often used in economics, and are applicable to many topics where randomization is considered infeasible for 'ethical' reasons: if you have a limited supply of X, it can be allocated by lottery; if you want to measure the impact of health insurance, surely this is unethical? How could you ever study it? But no, you have limited Medicare funds, so you simply allocate by lottery. Charter schools, lead poisoning, vaccines, basic incomes, daycare, enrichment programs, cuddle subsidies etc. The winners get the additional treatment while the losers serve as baseline controls, finessing the ethical problems. As it is usually the case that there's not enough money to deliver the claimed-beneficial intervention to all possible beneficiaries, it's quite a general strategy. > As someone who has been involved in multiple longitudinal studies, although not in epigenetics, I can say that they are more expensive. I didn't say that they weren't, just that they were only somewhat more expensive, and they can be done cheaply in a number of ways. There are considerable economies of scale in these things - look at the UK Biobank.
- SubiculumCode 9y ago
  First, you may not be aware of yourself sufficiently, but you are being a real asshole. Second, you are correct that I didn't consider that design. This may be because it is a little stupid, or at least impractical without the right target population. For example, if I wanted to study the effects of malnutrition it would make sense to do random assignment to individuals to food and food+extrafood conditions, but only if my sample contained a number of malnourished kids. If I took my sample from middle class America, and I assigned them to food and food+plusmorefood conditions (via force feed I guess?), I'd be studying obesity more likely than malnutrition. Levels matter. I'd guess that the primary interest in the present research is the effect of not getting physical affection as an infant. The effect between some and more affection, is probably qualitatively different than none and some affection.
  - gwern 9y ago
    > First, you may not be aware of yourself sufficiently, but you are being a real asshole. Uh huh. > Second, you are correct that I didn't consider that design. This may be because it is a little stupid, or at least impractical without the right target population. It's neither stupid nor impractical, and I gave several examples, many of which could or have been done in the US. Even healthcare can be randomized in the US this way - consider the Oregon Medicaid or the RAND healthcare experiment. (Both of them, BTW, deliver the profoundly surprising result that correlation!=causation and additional healthcare does little for health.) As it is a priori unlikely that all US children are above some hypothetical threshold for cuddling - in which case OP would be even more dead on arrival that it looks - the design would be perfectly adequate to investigate the causal effects of cuddling. One can screen for low cuddling levels if need be to increase power. > but only if my sample contained a number of malnourished kids. If I took my sample from middle class America, and I assigned them to food and food+plusmorefood conditions (via force feed I guess?), I'd be studying obesity more likely than malnutrition. This is a bad example as middle class American kids are not 100% micronutrient replete. Iron deficiency/anemia is common, and even iodine deficiency is actually quite common according to the urinary surveys. (If you look, you'll find at least two iodine RCTs claiming noticeable cognitive benefits in New Zealand college students, a country typically considered better off than the US on public health.) Vitamin D is another question mark - the official RDAs are very low but they're heavily criticized and if you accept some of the higher recommendations, then middle class American kids are quite likely to be deficient (all that time indoors). > The effect between some and more affection, is probably qualitatively different than none and some affection. That would be very unfortunate for OP, then, as I doubt any of the kids in their study had literally zero affection, as they were not recruiting from Cold War Romanian orphanages.

gwern 9y ago

> but specifically in the case of assigning the degree of physical affection/cuddling to a baby is probably unethical. Alright, since you refuse to think, I'll give it to you for free: do an intent-to-treat design with incentives for additional cuddling for either the parents or hired helpers like nannies, or passive controls; as no infant receives less physical affection than they would otherwise, no one is hurt by the randomization and it is not unethical, but you still get a causal effect. (Not all of them will use it, and the ones who do, the difference will be smaller than if you could remove cuddling, but that's merely a matter of power.) Such designs are often used in economics, and are applicable to many topics where randomization is considered infeasible for 'ethical' reasons: if you have a limited supply of X, it can be allocated by lottery; if you want to measure the impact of health insurance, surely this is unethical? How could you ever study it? But no, you have limited Medicare funds, so you simply allocate by lottery. Charter schools, lead poisoning, vaccines, basic incomes, daycare, enrichment programs, cuddle subsidies etc. The winners get the additional treatment while the losers serve as baseline controls, finessing the ethical problems. As it is usually the case that there's not enough money to deliver the claimed-beneficial intervention to all possible beneficiaries, it's quite a general strategy. > As someone who has been involved in multiple longitudinal studies, although not in epigenetics, I can say that they are more expensive. I didn't say that they weren't, just that they were only somewhat more expensive, and they can be done cheaply in a number of ways. There are considerable economies of scale in these things - look at the UK Biobank.

SubiculumCode 9y ago
First, you may not be aware of yourself sufficiently, but you are being a real asshole. Second, you are correct that I didn't consider that design. This may be because it is a little stupid, or at least impractical without the right target population. For example, if I wanted to study the effects of malnutrition it would make sense to do random assignment to individuals to food and food+extrafood conditions, but only if my sample contained a number of malnourished kids. If I took my sample from middle class America, and I assigned them to food and food+plusmorefood conditions (via force feed I guess?), I'd be studying obesity more likely than malnutrition. Levels matter. I'd guess that the primary interest in the present research is the effect of not getting physical affection as an infant. The effect between some and more affection, is probably qualitatively different than none and some affection.
- gwern 9y ago
  > First, you may not be aware of yourself sufficiently, but you are being a real asshole. Uh huh. > Second, you are correct that I didn't consider that design. This may be because it is a little stupid, or at least impractical without the right target population. It's neither stupid nor impractical, and I gave several examples, many of which could or have been done in the US. Even healthcare can be randomized in the US this way - consider the Oregon Medicaid or the RAND healthcare experiment. (Both of them, BTW, deliver the profoundly surprising result that correlation!=causation and additional healthcare does little for health.) As it is a priori unlikely that all US children are above some hypothetical threshold for cuddling - in which case OP would be even more dead on arrival that it looks - the design would be perfectly adequate to investigate the causal effects of cuddling. One can screen for low cuddling levels if need be to increase power. > but only if my sample contained a number of malnourished kids. If I took my sample from middle class America, and I assigned them to food and food+plusmorefood conditions (via force feed I guess?), I'd be studying obesity more likely than malnutrition. This is a bad example as middle class American kids are not 100% micronutrient replete. Iron deficiency/anemia is common, and even iodine deficiency is actually quite common according to the urinary surveys. (If you look, you'll find at least two iodine RCTs claiming noticeable cognitive benefits in New Zealand college students, a country typically considered better off than the US on public health.) Vitamin D is another question mark - the official RDAs are very low but they're heavily criticized and if you accept some of the higher recommendations, then middle class American kids are quite likely to be deficient (all that time indoors). > The effect between some and more affection, is probably qualitatively different than none and some affection. That would be very unfortunate for OP, then, as I doubt any of the kids in their study had literally zero affection, as they were not recruiting from Cold War Romanian orphanages.