Statistical Methods Links
Anna Lembke, the Hill Criteria, and social media; Robin Dunbar on Norse male violence; Jason Manning on what works; Emily Oster on alcohol and health
Of the dose-response relationship, the Bradford Hill paper states:
[I]f the association is one which can reveal a biological gradient, or dose-response curve, then we should look most carefully for such evidence. For instance, the fact that the death rate from cancer of the lung rises linearly with the number of cigarettes smoked daily, adds a very great deal to the simpler evidence that cigarette smokers have a higher death rate than non-smokers.
The literature for cigarettes, alcohol, and sugar all show a compelling biological gradient: The more of the substance ingested, the greater the harms. For social media, evidence for a biological gradient is robust. The more time teens spend using social media, the more likely they are to suffer adverse mental health consequences.
It was not feasible to use a controlled experiment to prove the causal connection between smoking and cancer. The Hill Criteria articulate ways to demonstrate a causal connection when a definitive controlled experiment cannot be run. The dose-response relationship is particularly compelling. Showing that people who smoke several packs a day are more prone to cancer than occasional smokers is strong evidence. Lembke, along with Jonathan Haidt and Zack Rausch, point to studies showing a dose-response relationship between social media and mental illness.
But I would argue that this differs from the smoking-cancer relationship. It could be that mental illness causes people to spend more time on social media, and that the dose-response relationship comes from this reverse causation. Reverse causation is much easier to dismiss in the case of smoking: it is not the case that getting cancer at age 60 causes you to smoke more at age 40.
Rob Henderson points to a study by Robin Dunbar and Anna Wallette on violence among medieval Norsemen.
despite having a 40% higher risk of being themselves killed than the average non-killer, killers had twice of many wives and offspring as matched non-murderers (men who were never recorded as murdering anyone) and nearly four times the inclusive fitness through their male siblings – providing they themselves survived to die in their own beds. More importantly, the brothers benefitted enormously: if the killer survived, their inclusive fitness was around three times that of the brothers of a non-killer.
In other words, violence does pay.
I looked at the paper, and I did not find it convincing. Suppose that A lives in a village with a lot of food and mates available, and B lives in a village where food and mates are scarce. A will have more descendants. But A may get into more conflicts and be more likely to kill and be killed. That does not show that violence pays. Maybe A could have had even more descendants if he had been peaceful.
Arthur C. Clarke said any sufficiently advanced technology is indistinguishable from magic. Not really. Magic is a concept we reserve for things outside of that which is so common and reliable that it’s part of the mundane world.
…Corrected formulation: Any sufficiently advanced technology is indistinguishable from nature.
…consider the way social scientists think of the experimental method and the concept of replication. We tend to have a model that involves sorting out the statistical differences between two study groups and looking at whether the results pass a significance test.
Contrast that with physical scientists building some experimental set-up and showing its reliability by repeatedly producing the same result and teaching others to do so as well. I think of something like the development of the laser.
In physics, your theory predicts what a laser will do. You try it, and it works. At that point, the theory is no longer speculative.
In economics you say that your statistical study proves some theory. But it remains speculative, because there isn’t the decisive test that is equivalent to a laser working or not working.
One is sometimes given the sense that if you combine many studies, that somehow fixes these issues, but it doesn’t. Adding more studies increases the statistical precision of your estimates; it doesn’t fix the bias. If all the individual studies are biased, their average will be biased too. This point cannot be overstated.
She is discussing studies of alcohol and health, which do not involve controlled experiments. Meta-analysis is not a cure for problems with observational studies.
substacks referenced above: @
@
@
The dose-response is not as clean as is believed. It can be very non-linear.
With the new administration appointment of RFK Jr., we will be deep into this problem with fluoride. Flouride is a required nutrient, and your body regulates the concentration, but only over a narrow input concentration range. If the input is too low, as I witnessed in 1962, both the rich and the poor had false teeth by 18 to 24 years old. That area also had a higher frequency of bone problems with their super high purity water (only containing some CaSO4). High concentrations above the optimal amount will cause discoloration of the teeth, and RFK jr claims IQ effects.
As almost all IQ research seems to have been restricted in the last half a century for political reasons, I have a feeling the science backing up the fluoride/IQ dose-response curve is a bit weak as it is probably based upon rural areas with high natural levels combined with many other factors that impact IQ such as inbreeding effects.
Now that we can measure every chemical everywhere, we face the problem of having N variable problems where N is a very large number. The relevant variables can also be interrelated; the resulting correlation can be misleading unless all are considered and measured. Think conic sections to see why you can't describe an N-dimensional problem in N-x dimensions. The fluoride concentration in nature is related to many other chemical components. Environmental activists and social science/humanities areas have long used the exclusion of relevant variables to obtain desired, but false results.
"Jason Manning writes,
Arthur C. Clarke said any sufficiently advanced technology is indistinguishable from magic. Not really. Magic is a concept we reserve for things outside of that which is so common and reliable that it’s part of the mundane world."
Jason Manning is very much missing the point of that statement. "Sufficiently advanced" is doing an important job he isn't quite getting, in that it is not common in our mundane world, and so in the context of the story and how we think of it it might as well be magic. 80's movies' computers are a good example, where, from the standpoint of today, they are basically magic beings that can come up with impossibly perfect calculations as major plot points.
This applies outside of stories to the real world when you see people with cargo-cult like approaches to technology, being so unable to grasp the basic rules and functions that they just assume it works by understanding what you want it to do. Think people who have car accidents because they set the cruise control while on the highway and took a nap.