Maybe 20 years ago people would've said "Someone who would advise their public to inject bleach into their veins is unelectable". Now it's a bunch of supporters playing word games, he didn't tell them to do it, he merely implied something like detergent would be effective if applied inside the body!

Thanks. I've had a look at the IEP and re-read the exchange in that light. It now makes a bit more sense, I think. You're saying that certain emotional responses resulting from the model might betray some higher goal even though they're the correct output from the model - so infelicitous, not "wrong"? That there could be a situation where we confuse the accurate function of the model for an accurate output? Like presuming that if a car is running really well it must be taking us where we want to go? — Isaac

That's about where I was going, aye.

I think "infelicity" is a good touchstone to describe what the "prediction errors" of active inference, in terms of emotion, might be. They're still "optimal predictions" (in some sense) given their constraints and priors, but that does not mean the priors reflect the relational dynamics of the body and its environment, and their potential developments given the interventions I propose.

In the back of my mind I was contrasting classification error in a machine learning model (does a model trained to recognize bridges in pictures recognize a bridge in this picture? When there is no bridge? When there is a bridge?) to one where the "success criteria" are more complicated; do I succeed in this goal, more or less? Is it the right goal? Am I operating in a fruitful cognitive frame for this task? Does "getting angry" help here? What about being remorseful?

If so do you not think that the infelicitous output would simply constitue a prediction error of some higher model? Tom Fitzgerald has done some work with Karl Friston on active inference and habit formation which covers some of that ground. I might PM you with it though, I suspect the online equivalent of a series of blank looks if we start discussing it here!

I guess that would depend on the higher model considered. I tried to suggest what I mean above (but would like to see the paper if you have it on hand).

Barrett refers in her book to what she calls ‘affective realism’, which I understand to be a misapplication similar to transcendental illusion: — Possibility

That's very cool. I think I might have to pick up that book. Thank you.



Will have a read of that tomorrow, and will try to make a post about that "infelicity" thing regarding it. If you wanted to read about it, IEP talks about Austin's use of the term; it's a more general failure/unsuccess category than "right" or "wrong", and he applies it to speech acts. My motivation for using it was to stress that goal/task relevance acts as a constraint in active inference, the "failures" we have with it are also rooted in comparisons to what we're trying to do.

A manager whose only mode of negotiating is angry discipline is being "infelicitous" in the sense I was using it; contrary to their goal in reality, signalling a predictive failure they will not recognise or their heuristics do not deem relevant.

No, it doesn't. It doesn't refer to anything. — Banno

What would your take on a formal semantics approach to 1's referent be? Like, taking it to be by definition the successor of 0, or the equivalence class under bijections of { { } }.

I've been trying to think about this and I think you're entirely right to think about this in terms of modality. I hope this is not a case of me just trying to curve-fit, but given how messy the production of emotions can be, would it not be the case of something like a necessary production of contingent valuation errors? Like, given the exigencies of bodily developmental history, the openness of context, the instability of (emotional) meaning (in Derridian terms one might speak here of a necessary play of différance involved in emotion), the overlapping patterns of cultural meaning, etc, etc - that in some sense, we're almost guaranteed to have errors crop up often. — StreetlightX

Maybe one way of condensing Barrett's points about errors is that they are more like infelicities of speech acts; in the regard that promoted actions (including attaining a specific emotional state) can be in reality unlikely to aid in achieving the goal , or that the evidential basis that furnishes the body's inferential transformation into an emotional state is flawed; accessing the wrong information (like an error of context) or drawing the wrong heuristic, aleatory and associational conclusions based on mechanisms of association that do not track the associations of the phenomenon in question. All these types of misfits leverage exploratory information about one's body and its relationship to its environment (representation of their joint causal structure in terms of salient task oriented features). These are errors the process of active inference can make in its course.

What I'm thinking of as a transcendental illusion would perhaps be a flaw in the process itself; whether there is some mechanical failure in the process of active inference that makes it attempt to exceed its bounds. Perhaps the one thing I can recall that resembles this in "The Emotion Paradox" paper is the enduring error that people have distinct states like "anger", "sadness", "joy" which behave like natural kinds. Even then, we can think otherwise from these. So perhaps a transcendental illusion in this context would be an enduring or widespread heuristic bias that the machine of active inference is likely to pick up (on a population level), and even then they could not easily be distinguished from cultural effects. But perhaps, if we constrained the discussion to tendencies of active inference that will always be infelicitous, and arise from nothing more than how the process works in itself (as in, how it works as a mechanism of prediction/association independent of its content), maybe these infelicities are close to the idea.

A "true" transcendental illusion, a reapplication of Kant's doctrine, would in my mind be an inescapable tendency of this process to confuse its representations of causal structures (given a goal) with the causal structures themselves. And that's partly what the models outputted from the active inference machine are for anyway, and have it as an inbuilt feature (the "raw data", like a body's true heart rate, ambient temperature etc, is summarised in a way that makes the active inference machine depend only upon a representative summary of the raw data when promoting actions). This isn't so much as a reapplication of the doctrine of transcendental illusion, but a restatement of it in another context (maybe). We actively infer in a manner where the true propensities of the world's development given our interventions and goals are (ideally) closely approximated by our predictions given those same interventions (and we weight discrepancies/imprecisions/mismatches highly in the process to maintain allostasis).

... Barrett's point about cognitive systems maintaining allostasis is, I think, more than just a neuroscientific one. It emphasises the role of interoception as forward-acting (it generates some following set of reactions, rather than just backward-acting, suppressing existing responses) something else afterwards . So a model, is a kind "this is like/is due to/arises from that" (to borrow fdrake's expression), but it cannot really be separated from "...and this is what I do about it" because the doing is part of the loop updating the priors. — Isaac

To draw a really special emphasis on this point (which was lacking in how I presented it); the predictions of our self models are also in part proposed interventions; what can I do which is appropriate for my goal and the rest of my current model? It isn't like the whole model is just some epistemic device we use to learn about the world, the model is also sending out things like motor signals; lean closer to hear better, stuff like that. Our behaviour itself, our emotions themselves, are a sample from the model, and "exactly what we do" is a collapsed down form, a representative summary, of the model's state given (its own representation of) a current goal (and our expectations of environmental/bodily behaviours).

Edit: I'm not comfortable enough with the Construction of Emotion paper yet to talk about its mechanics in detail though, so I'm largely improvising in this based on my understanding of the work (in active perception) she's trying to integrate it with.

Philosophy rarely contains interpretations of dimension reduction techniques applied to psychometrics ; it isn't quite the same game here.



I think your "this is all the same as before" detectors are malfunctioning in this instance. There are predictive differences between the natural kinds view and the one Barrett's proposing, and they are referenced in what we're discussing. It might be the same "explanatory category making" game, but here the explanations do inform what predictions are made about how emotion functions (see the hidden thing * for an example). Experimental results about emotion make more (or less) sense depending on the view's content.

Of course, to both of you, being largely a philosophy discussion, we'll maybe be playing the same game in interpreting the view and relating it to "philosophical positions", but researchers using Barrett's paradigm will make sense of and predict different results than those of (at least some of) the views she's criticising.

Continuing from here.

The titular "Emotion Paradox" from the paper is that:

People are compelled by their own experiences to believe that emotions exist as natural-kind entities, yet a century of research has not produced a strong evidentiary basis for this belief. To date, there is no clear, unambiguous criterion for indicating the presence of anger or sadness or fear.

So there are two thrusts of it: (1) people are compelled by their experiences to believe that emotions exist as natural kind entities (categories of experience, analogising "anger" to "the human leg") but (2) there's no evidence that emotions actually work like that.

The paper addresses this by trying to explain how those beliefs in emotions as discrete/partitioned entity types come about by describing a mechanism of emotion; the categorisation arises as a prediction and contextualisation of one's bodily state in a task which is cognitively, discursively and culturally mediated while weighing all those things in the light of past and current experience.

I imagine it that we have a (pliable, modifiable) emotional vocabulary of concepts that our self models evaluate in terms of (@Isaac, dunno if this is actually a good analogy, you'd definitely know more about it); this is like/is due to/arises from that, most broadly this is associated with that. The "this" and "that" are rarely articulated, due to being fuzzy uncategorised interminglings of all the signals we have in the body and from the environment. Something which someone articulates about their emotional state, then by necessity, must already have been weighed and measured by the process of valuation that mediates these signals and found to be a "best fit" summary of the state (given prior experience).

Edit: regarding the "fuzziness" of the categories; it reads like the process of categorisation is something that's evaluating all the time, categorisation itself isn't an "on/off" thing, or a single step mapping from "fuzzy core affect" to "distinguished emotion", it's that core affect is always more or less categorised; maybe parametrising it or thinking of it in terms of an intensity is helpful; there's a sliding scale from "completely uncategorised" to "completely distinct" that our affect(s) are constantly charted on by the process of valuation; but it's always in this or that category (what box things are put in matters), and categories are learnable.

Categorizing is fundamental cognitive activity. A category is a class of things that are treated as equivalent. To categorize something is to determine what it is, why it is, and what to do with it. A concept can be thought of as a collection of mental representations for
a category that people draw on during the process of categorization. Once conceptual knowledge is brought to bear to categorize something as one kind of thing and not another, the thing becomes meaningful. It then becomes possible to make reasonable inferences about
that thing, predict about how best to act on it, and communicate our experience of the thing to others.

There are inbuilt tendencies in these associations (central pattern generators), I imagine they give rise to whatever cultural universals we have regarding emotion.

The structure is incredibly leaky and pliable. Affects and unacknowledged emotions circulate beneath the level of explicit emotion (cognitively-recognized emotion), and, as per 'misfelt feelings', there can be all sorts of crossed wires and potentially 'misidentified' and misconstrued feelings. — StreetlightX

Yes! One consequence of emotions (the kind of thing we have when we say "I am sad" or whatever) being predictive, task relative, valuations is that they can be wrong, flawed, not fit for purpose. They can be misattributions, inaccurate or mis-focussed summaries of the current bodily state relative to its context(ualised task), we can find the wrong things meaningful (salient), they can suggest ineffective actions - I imagine there are other nuances of the errors we can make in feeling; special emphasis, feeling itself; but I can't think of more now.

Which comes back, of course, to your question about transcendental illusions and their applicability to emotion. — StreetlightX

I'm beginning to think that transcendental illusions are separate in character from the predictive errors spoken about in this approach; insofar as transcendental illusions are necessary failures of reason generated by its misapplication, I don't think they'd apply to the contingent error prone-ness of valuations. I'm not saying that there aren't transcendental illusions for emotion, but I can't see a neat way of linking the paper to the question I wrote to you (summarised: "Are there analogues of transcendental illusions in emotion?").

I wanna share what I think to be Feldman Barret's positive account from the paper "The Emotion Paradox", rather than dealing with the details of the evidence she presents for it.

Emotions are interoceptions. They are the body's perception of itself (through some modalities, like "somatovisceral information"). That means what we leverage to explain perception applies to emotion. Perception is task relative; what you see depends upon what you're doing; and mediated by a bunch of things. Your past experiences, what you've learned, how you conceptualise stuff. What you've learned influences what you see, what you're doing influences what you see, how you talk about what you see influences what you see. Those effects of "what you've learned", "how you conceptualise stuff" and "what you're doing" get put into the process of valuation.

Valuation is one part of how our bodies do emotions. The other part is core affect. Core affect, inn Barret's words is:

Core affect has been characterized as the constant stream of transient alterations in an organism’s neurophysiological state that represent its immediate relation to the flow of changing events

Core affect (i.e., the neurophysiological state) is available to consciousness and is experienced as feeling good or bad (valence) and to a lesser extent as activated or deactivated (arousal).

Core affect and valuation run in parallel all the time. Insofar as emotion is concerned, we are core affect machines and valuation machines. Valuation draws on resources outside core affect, and it modulates (or mediates) the emerging experience of emotion. The extra resources are anything that we could bring to bear to contextualise information; all these extra resources can be labelled as concepts. They're not only the "clear distinct ideas" of Russel, they're more general representations.

How they work together is that core affect updates its environmental and bodily information quickly, valuation updates itself a bit more slowly. If the environment and the body are food, core affect is chewing, valuation is digestion, emotion is the whole thing insofar as it's bodily.

To be able to label a state of core affect "anger", we have to have the cognitive resources in place not just to categorize it, but to devote enough of our attentional and representational resources to it that that a categorization ("I feel angry") emerges as an individuated description of our emotional state. As Barrett puts it, emotions are "perceptual symbols", like "red".

I argue that the process of categorization is fundamental to both color and person perception
and draw a parallel between these categorization processes and the way in which people use their knowledge of emotion categories to shape an experience of
emotion.

The usual folk psychology ways we talk about emotion are all post categorisation; we have distinct but overlapping (descriptions of) states like happiness, anger, shame, horniness. What we learn about these categorisations, the folk psychology stuff, influences how we feel. This intervention of resources outside core affect upon our emerging experience of emotion through learning simultaneously makes it discursively/culturally mediated and something that can be honed and practiced, a skill. Husserl called phenomenology "relearning how to see", Barrett may suggest that we can relearn how to feel.

The intervention learning has upon the emerging experience of emotion is not limitless; the processes that constitute core affect are linked to how the body is extremely likely to typify its own sensations/state; the body is predisposed to some emotions and responses. There are some biological primitives (central pattern generators) that valuation acts improvisationally upon, draws boundaries between, and leverages contextual information to interpret and prescribe what to do upon their (and core affect's) basis.

The way I've written it above may suggest that core affect and valuation are still reactive; they are means for digestion of environmental stimuli, and this always goes stimulus->response or behaviour->feeling. This isn't true; valuation is predictive/prescriptive, and this is fleshed out in some of her later work. In the paper "The Theory of Constructed Emotions", Barrett leverages Bayesian brain ideas about concepts and perception to refine how emotions are predictive/prescriptive, not simply reactive. Emotion is in some regard anticipated behaviour that is expected to be effective and salient in summarising our body's situation (including current task and goals).

I hypothesize that in assembling populations of predictions, each one having some probability of being the best fit to the current circumstances (i.e., Bayesian priors), the brain is constructing concepts (Barrett, 2017) or what Barsalou refers to as ‘ad hoc’ concepts (Barsalou, 1983, 2003; Barsalou et al., 2003). In the language of the brain, a concept is a group of distributed ‘patterns’ of activity across some population of neurons.

Valuation becomes a patterning of core affect(s); patterning is always going on in the emerging experience of emotion, so it modulates emotion; emotions are a "what's now? what's next" rather than just a "this just happened", and "this just happened" is part of informing "what's now? what next?".

As Barrett puts it:

A brain can be thought of as running an internal model that controls central pattern generators in the service of allostasis (for more on pattern generators, see Burrows, 1996; Sterling and Laughlin, 2015; Swanson, 2000). An internal model runs on past experiences, implemented as concepts. A concept is a collection of embodied, whole brain representations that predict what is about to happen in the sensory environment, what the best action is to deal with impending events, and their consequences for allostasis (the latter is made available to consciousness as affect). Unpredicted information (i.e. prediction error) is encoded and consolidated whenever it is predicted to result in a physiological change in state of perceiver (i.e. whenever it impacts allostasis). Once prediction error is minimized, a prediction becomes a perception or an experience. In doing so, the prediction explains the cause of sensory events and directs action; i.e. it categorizes the sensory event. In this way, the brain uses past experience to construct a categorization [a situated conceptualization; (Barsalou, 1999; Barsalou et al., 2003; Barrett, 2006b; Barrett et al., 2015)] that best fits the situation to guide action. The brain continually constructs concepts and creates categories to identify what the sensory inputs are, infers a causal explanation for what caused them, and drives action plans for what to do about them. When the internal model creates an emotion concept, the eventual categorization results in an instance of emotion.

I think I'm on the border of not knowing what we're talking about. :razz: I just wanted a reason to drop in my comment about collective action and how amazing it is. Internet and whatnot. — frank

:up:

Some people become infected or colonized by this coronavirus and have no symptoms. Some become ill enough to die. I think there is a hidden factor involved. — frank

I believe in that sense, comorbidity presence and severity explain a lot of the variation between those cases (they are the common factor). In general, the closer something gets to being a mysterious hidden factor (patternless unstructured variation), the closer it gets to being noise (unstructured individual level variation). Signals tend to announce themselves.

When meaningful, categorical propositions or their negations are supposed to be true with respect to appropriate criteria, the latter phrase the lubricant that keeps the whole enterprise from grinding to a halt. I wish to focus on that: with respect to appropriate criteria. Truth without criteria is a chimera - how indeed would it be known to be the truth without criteria? — tim wood

I guess linking to both; we rarely articulate the criteria by which our arguments are seen to flow. Or how our assertions/attempts to reframe things are intended to work. Or in what context our evidence works as evidence. These failures in articulation crop up more in contexts (like on the forum) where a common background of understanding is not present, and when the discussants are operating under different criteria of relevance.

In contexts like this: "The Earth orbits the sun" is followable by "How are you certain of that?", which draws the image of a square peg smashing angrily into a disagreeable round hole.

The "unstated" point in such things may not be null, it's often what matters most; a pivot around which the disagreement turns. It has that status in part because it remains unarticulated and unevidenced.

A: "Assertion!"
B: "On what basis do you believe this is true?"
A: (supporting argument or citation)

Is how it should go with a fixed question, or analyzing a given argument. I'd've answered "yes" to 3 if I thought assertions weren't usually more like:

A: "Assertion that reframes something unstated as relevant"
B: "Upon what basis do you believe that is relevant?
A: (supporting argument or citation for assertion, rather than the relevance of the unstated)

Is how it usually plays out. Context requires chewing. It lives in what's left out when using an enthymeme. We're almost always using enthymemes. If we could only agree on what assertions were relevant to what, things would go more like the first case and not the second.

Answered no. It's difficult to perturb a wrong question with right evidence.

I don't think so personally.

I believe we agree that's not the case; that there is some effect of culling the terminally ill (whether known or unknown) but it's not a large effect. — boethius

But it does kill people now who are likely to die soon that aren't likely to die now otherwise, right?

The original claim I have contention with, is that respiratory deaths may go so low after Covid as to balance Covid deaths. — boethius

If COVID kills the majority of people that would've otherwise died from these issues (pulmonary, heart, cancer patients). and there isn't a corresponding increase of intake in those risk groups over the relevant time period, the number of people that could die from those issues after this year (in the relevant short time period) is going to be lower. It isn't so much that this will cancel out the deaths attributable to COVID, it's that it'll constrain the number of people that could die from other causes in that group (irrelevant of what they are); since the dead people won't be in that group any more, and certainly can't die from other causes if they're already dead from COVID. (@Isaac 's use of the word "cohort" in his responses probably is used to emphasise this time moving property when fixing a group to study over time, cohorts have a fixed initial size that depletes through death/over time)

we can't simply assume Covid is killing off some unknown sub-group who are very likely to die; — boethius

I mean, COVID is more likely to kill people who are more likely to die anyway. This complicates whatever attribution of death to COVID you do.

Even if the infection risk is constant across the factors that increase the likelihood of death. If COVID infection initiates the causal chain that leads to their death, even if it's the knock on effects (like it interacting with comorbidities) that ultimately kill people, that's going to make the number of deaths attributable to COVID much higher than it would otherwise be - IE, if COVID's the thing that's making people in the underlying risk groups die, and they wouldn't've died now without COVID, then it's COVID's fault they died now.

It can still be that COVID infection is what killed them, even if the majority of people end up dying from heart failure or kidney failure, or other complications (rooted in comorbidity or not).

It's an over simplistic assumption to postulate Covid deaths is selecting for "least healthy" within a risk group. It seems intuitively correct, but is not correct. — boethius

Whether it's a random sample with no causal mechanism or not depends on the model. Some models are better.

The question is: who are those people who are most likely to die of COVID? That's people who are elderly and have comorbidities.

If the question is: how many deaths are attributable to COVID? That's a bit different. If the question is restricted to: how many deaths of those in a given comorbidity + age risk group are attributable to COVID? It revolves around the counterfactual: If you gave someone with those characteristics COVID, how much more likely are they to die than if they did not have COVID? This is much harder to answer, requires an explicit model of how COVID interacts with the comorbidities, and can't be immediately read off the risk of death of those people who have those characteristics (comorbidity + age) who have confirmed cases and died in hospital (that group selects for comorbidity severity already!)

Edit: if you have a link to a study which is doing this kind of calculation, or something similar, already (trying to quantify EXTRA deaths from covid within risk categories), I'd like to see it!

We cannot assume those that die from Covid are "least healthy" within their risk group. — boethius

It isn't necessary that those within a risk group that are "least healthy" will die, it's just more likely. If you found any factor or variable which contributed to risk, and it wasn't aliased with the risk group already, those in the sub group of that risk group that have the extra risk are more likely to die than those in the base risk group. If you condition on death, you're already pre-selecting for demography and other characteristics which make death more likely; as in, within the group of those who have died from COVID (or any other thing), the characteristics that make death more likely will be more prevalent than they are in the general population (on average). It's the same mechanism that makes the more severe cases of COVID be more likely to be tested for COVID, but in a different form.

To be super clear about this; if your sample is those who have died while having COVID, that sample is more likely to contain a higher number of people on average who fall into the groups that amplify risk. Like... those who have died from COVID are more likely to have comorbidities and be older.

If fairly extensive random testing were carried out, would that not give a reasonably accurate picture of just how much Covid-19 infection is in the community? — Janus

I think it would.

From what I understand, there's no other way to do it, as there's no way to "know" who really will die absent Covid. We can only put people into risk groups and then calculate the probability of death from Covid of people in their respective risk group. — boethius

Aye. When someone is usually tested based on severity the data's going to have things which will come from colliders, they'll tend to inflate sample correlation between whatever's collided; but it's extremely unlikely that the correlation observed between severity/mortality and comorbidity is explained entirely by the collider bias induced by the testing.

there is simply a large number of additional deaths from Covid and the risk profiles going forward remain constant — boethius

From what I know, I'd agree with that; the disease doesn't seem to have changed nature, and if it already had changed we'd probably be talking about it, though the risk profiles can increase through exogenous stuff like healthcare system failures; if the infection risk goes way up within hospitals due to resource shortages (not sure how likely this is), that increased risk can partially be attributed to COVID presence, but also on the broader stuff that lead to the shortage.

The people with comorbidities are more likely to die if they're in the "we have covid" group than in the "we do not have covid" group. Covid's an extra influence, and its knock on effects can be (partially) attributed to it; quantifying the effect of it on health outcomes should also include its knock on effects on other outcomes. It's like uh.... Hitting someone in the head with a clawhammer, I'm responsible for the brain damage, even though it's an effect of the impact rather than my intention to hit someone on the head with a clawhammer.

I think "comorbidities" has given a lot of people on the internet the impression that most people who die from Covid are essentially on deaths doorstep and the virus was a, perhaps even in their minds a merciful, coup de grace. However, these underlying conditions that increase the probability of death are very large, fairly banal groups; people with heart disease, diabetics and cancer survivors that have relatively long life expectancies. Large groups where there's no reason to believe those that would die of some respiratory disease anyways, not to mention just any cause of death, are significantly more likely to get Covid fist. — boethius

That wasn't my intention.

I was just talking about whether partitioning the data based upon risk group removes the collider bias you mentioned, and giving some justifying statements that comorbidity complicates the attribution of the deaths to coronavirus, seeing as it also contributes to deaths from influenza.

This applies especially if we're analysing only people who have died, will die, or would die, that group's going to have corona virus presence collided with other virus presence because it's already known that health outcome severity is influenced by comorbidity presence and severity. Anything regarding population infection rates at large isn't going to remove it (at least, I don't see an easy way to control, the randomisation introduced by infection risk won't touch the death risk given infection and its influences; we're already conditioning on infection for one of the considered groups!)

The relevant scenario to consider is whether a person with a given comorbidity is more likely to die (or other health outcome, or increased prevalence of larger values of negative health outcome) from coronavirus than from influenza, rather than considering the population at large based on risk groups that do not track comorbidity while still being influenced by it (through the dependence introduced by subsampling based on death)

For instance, if there is a risk group with 10% chance of respiratory death this year, and getting the virus increases those chances by double, then there is a of collision between Covid deaths and other respiratory deaths. But if those diseases are distributed randomly the collision is not much. E.i. if this group is 100 people, then the odds that the 10 people expected to die from Covid would happen to be the 10 people expected to die from influenza, is extremely low. — boethius

The latter odds of the two coinciding likely depend on comorbidity to begin with though. I think if you stratified based on comorbidity the reasoning holds though.

P(person dies of flu | confirmed comorbidity presence, risk group) is much higher than P(person dies of flu| risk group).
P(person dies of coronavirus | confirmed comorbidity, risk group) is much higher than P(person dies of coronavirus | risk group)

So long as the confirmed comorbidities are comparatively rare on the population level or within the risk group anyway (so if your idea of "risk group" includes accounting for comorbidity explicitly, that would be fine too).

It's like flipping two coins which can result in heads or tails (heads = death, tails = not death) with loads of blu-tack on tails (blu tack on not death), they'll coincide on heads (death) a lot even if the flipping mechanisms (joint infection probability of coronavirus with another disease, given comorbidity and risk group) are unrelated (P(i have flu, i have coronavirus| comorbidity, risk group). = P(i have flu| comorbidity, risk group)*P(i have coronavirus|comorbidity, risk group))

Edit: also notice the transition between discussing death probability and infection probability, P(infection | comorbidity, risk group) might behave much differently than P(death | comorbidity, risk group), so the causal colliders in one need not transfer to the other, or death might have colliders that infection does not.

Ah good. I'm glad.

I saw this in the related news the other day, "First ever iceberg wearing a bad wig found on large landmass situated between Atlantic, Pacific and Arctic oceans".

I hope so. I don't think I've ever seen @frank use the word cool when it wasn't sarcastically insulting someone though.

An overlooked super spreader. Cool. — frank

Ok. I asked a doctor. They said I'm not increasing the transmission risk by washing the sink how I wash it. Absent a good justification, I'm gonna take their word over yours.

Edit: Ok, that's 2 doctors who agree that it's good to wash the sink how I wash it.

Compliments aren't supposed to be just objective appraisals. They're not like... surveys, I mean "You look beautiful" is not "I have done a survey of 100 people regarding your looks, adjusted for prompting bias somehow, and you have an attractiveness score in the 95th percentile based on previous studies". It's much closer to "I declare you are beautiful", a performance of appreciation.

An overlooked super spreader. Cool. — frank

Why?

Something can't come from nothing.
There is something.
There was always something.

but I think you missed a few — Isaac

I did! Thank you.

I was just saying that people who have extra angst could direct that energy more locally. — frank

I disinfect the cleaning station at my supermarket when I go. Ironically it doesn't seem to be cleaned often.

@frank

I just wanted to provide some info about the things that make death statistics fluctuate a lot, based on (what I assume is) the same Lancet paper.

(1) Testing. If you want see how many people have died with coronavirus, they need to have tested positive for coronavirus. If you test certain groups of people, but not others, this makes the death rate calculated from the tested group depend upon the group's demography.
(2) Group demography effects in testing; tested people are more likely to be severe cases; whatever factors contribute to the severity of the case will be more present in the tested population, this is a positive bias to the death rate.
(3) Severe infection demography. If the virus is more likely to inflict a severe case dependent upon a demographic factor, people in those demographics are more likely to be tested due to the things stated above.

Correcting for these things requires weighting the observed statistics by the demographic factors then calculating for the population at large based upon the correction. It is especially hard to correct for people outside of the tested population; which includes the mild and asymptomatic cases disproportionately, and those demographic categories which are not as likely to be tested due to having those milder cases; so observed death rates in hospitals and in the spreadsheets derived from there are likely to be over-estimates of the overall population death risk, the latter means the case where demography and case severity are no longer making the tested population non-representative of the infected population at large.

However, it should always be stressed that the effects of the disease when considered in some circumstance need not reflect the population death rate. EG: If 1 in 50 infected individuals die, thinking about it that way it wouldn't be surprising not to know anyone who died. But part of that 1 in 50 is the people aged over 60, in which 1 in 8 are expected to die. You maybe don't know anyone who's died personally, but it's way more likely for there to be someone you care about who's lost their grandparents or parents to the disease and is grieving.

To be honest I could've ranted on at you @ssu for a very long time, but I wanted to add one more thing that's important to notice; the delays didn't just exacerbate the deaths, they will exacerbate the long term economic damage too.

I acknowledge that there is the crowd that put basically the economy before anything, but I don't the chief epidemiologist Tegnell in Sweden had (and has) that in mind. Or Wittkowski above. Even my little country, which now has emergency laws and has quarantined the whole Capital region from the rest of the country doesn't have a curfew in place. To argue that people should stay inside their homes and not venture out is dismissed as humbug by doctors here. You can choose something between a) doing nothing and b) having a curfew. — ssu

You're right, there were multiple ways of impeding the spread of the virus. The less socially intrusive ways were less effective. The European countries that I'm aware of (except Italy and Hungary) have taken to politely requesting people to adhere to social distancing and increased hygiene, mandating businesses to close to remove central hubs etc.

I am not disagreeing that there were multiple ways of responding to the virus; corresponding to a trade off between social intrusion/impediment and decreasing growth rate.

It wasn't that. It wasn't about implementing quarantine measures, but any kind of response to the pandemic. Basically it was about denying there to be any serious pandemic at all. That's a huge difference. — ssu

This is precisely what the use of "herd immunity" by politicians was for. It was not used as a statement of the uncontroversial fact that eventually populations will immunise. The fact was used rhetorically as a stalling tactic. Eventually all countries effected which used the rhetoric have responded somehow, because they needed to.

My only point is that there really is the medical/health policy discourse on the subject, something that you seem to deny. — ssu

Nah. I've helped a couple of doctors I know understand their epidemiology bulletins due to the virus and have been studying the global case records personally occasionally, also keeping somewhat up to date with stats papers on it. I'm sure that fellow statisticians will be working on the data set for years to come. The discourse within epidemiology (as far as I am aware) is not about the plain fact that decreasing social connectivity decreases growth rate. It's about using the data to quantify the hows and whys and to track and predict spread.

I'm not of the opinion that curfews and using the military to keep people in their homes is particularly desirable, even though it would obviously make the transmission rate go down. (Edit2: even if middle road measures need not decrease the total number of infected long term, it will decrease the mortality by not overloading the bandwidth of healthcare systems by even more)

The use of "herd immunity" by politicians was a stalling tactic against every response. Any stalling is well understood to lead to inflated death tolls; it's killed people who would not have died otherwise, and that number is increasing with time due to how the growth works. It took catastrophic economic circumstances to be likely, like global economic collapse, to get these people's heads in gear and actually take the situation seriously.

I'll put this numerically; politicians advocated a strategy (non-response) that their intelligence networks and consultants must have informed them would yield up to 2% of their populations dying. Because they did not want to risk a big recession that would come from curtailing the loss of life. If at the end of the pandemic, you piled all the bodies that would have come from the advocated non-response strategy, that number would have been bigger than political north troop deaths in Afghanistan (using just UK figures, that would have already been passed a while ago). That was seen, at the time, as an acceptable risk to prevent a recession.

To make matters worse; right news media in the US and UK has reported to support this interpretation; portraying it as patriotic to die to save the country from a recession; there are internal memos from the UK Tory party being nonchalant about killing off all the old people.

Until the global economy started shuddering, all of the above were acceptable risks.

Edit:{

The pandemic would have been forecasted to overload the healthcare system by so much it would increase the death rate beyond that, especially for risk groups (if schmucks like me could see it coming, I'm sure people who know more than a university course dealing with some epidemiology models and related private study would have done so and passed it on) and cause untold more losses if responses were not taken to mitigate the risks. This was seen as an acceptable risk.

When it actually started happening, they finally responded "Oh no! All the things we were obviously told beforehand in intelligence briefings confidently by consultants are actually coming true, let's start doing something about it!"

So consider; what made the politicians more afraid of a predicted recession than more predicted deaths of their own citizens than your average war?

}

So what's the "purely ideological" reason for Swedish social democrats to choose the more lax measures? — ssu

I can understand attributing a view to me I don't actually hold in these circumstances.

It goes beyond this even - a pandemic like this is immediately political not only because of politically and ideologically motivated responses - responses ought to be politically and ideologically motivated - but because the virus's effects are immediately socially socially distributed along class and even racial lines. Aside from the fact that - in the US at least - CV has killed disproportionately more black people than others (because less likely to have access to good healthcare, because more likely to work in so-called 'essential jobs', because less able to have the privilege of self-isolating) the virus kills the poorest of the population at incredibly high rates: — StreetlightX

:up:

es, a big part of it is the present media environment which instantly reports everything. We are also very intolerant to deaths from pandemics. We don't accept that many people die of infectious diseases, when we could avoid them. — ssu

I'm sure you already know this, but regardless of media reporting, healthcare systems would have been way more likely to fail, and even more people would have died, if their intake wasn't controlled through quarantine measures.

The "herd immunity" policy isn't totally crazy and we cannot now just brush aside the path that Sweden has opted with it's chief epidemiologist Anders Tegnell as utterly wrong. — ssu

Eventually people will recover or die. The reason "herd immunity" was wrong wasn't because eventually the majority of the population (albeit an ageing one) will adapt and what's the point, it's because people advocating herd immunity explicitly did not want the economic risks of quarantine measures, despite the massive death toll and healthcare system failure that recklessness would have caused.

The problem is that it now days everything becomes political and too many people see a political / ideological agenda in everything. — ssu

The reasons people resisted quarantine measures were purely ideological, it isn't just the discourse, it's, unsuprisingly, policy being politically/ideologically motivated rather than just looking to the epidemiologists and scientists for cues on how best to manage the pandemic. The delays and resistance from our politicians to implementing quarantine measures were ideologically motivated, later they conformed because they realised they must.

So it is absolutely bonkers to claim that the issue isn't a political one, when the management of a pandemic is an economic, scientific and political project.

You would not be saying "it's all so difficult now that politics is in the mix" if your reference points were Indian police beating the shit out of Muslims breaking quarantine for worship, or the use of a state of emergency for Orban to seize power indefinitely. How it's managed and responded to is political from the get go, unless somehow the world lives inside an epidemiology journal or WHO bulletin.

I'd need access to a keyboard to talk about it more. It would be genetics in combination with other factors, not genetics alone. — frank

Aye, it would have to be.

Ah. Yeah, I read about that. Attributing the observed risk inflation between countries to something which is constant over those populations is strange. I was hoping for a link that shows, say, Norwegians are way, way more likely to have good heart tissue genetics vs UK people. To my understanding, it's a good tool for explaining the weird what appears to be post recovery heart failure, but explaining such a large inflation of observed risk (on the population level) by something that each population has in approximately the same degree does not make sense.