[I will suggest/request that anyone who wants to link to the video please link to this post instead. I would like to encourage comments and discussion here, and will probably not monitor the comments on the youtube page itself. Also, there is more background that might be useful.]
The heart of the presentation is a social dynamics model of how THR (e.g., switching from smoking to e-cigarettes) occurs in a community thanks to the education and communication of social norms that come from social interaction. It starts out with a general overview of THR since many in the audience were not familiar with that. If you are not interested in the overview, you might want to skip to about 9:30 and just see the presentation of the new model. (On the other hand, I have been told that it is one of the better existing presentations about the core concepts and justifications for THR. Not as good as what I presented at the Beirut IHRA conference, unfortunately, but I do not have a recording of that. So you might want to view that part even if you already are familiar.)
The presentation speaks for itself so I will not try to summarize it here. But to provide a bit more background on the modeling (and if this is confusing, just watch the video -- it is less technical than what follows, but still explains what you need to know): There is an interest in predicting THR behavior, in part for obvious reasons, and in part because of a make-work exercise that the US FDA is imposing on anyone trying to promote THR. As with any modeling of population dynamics, there are various methods available.
The simplest is to just project a trend by extending past numbers. This is largely useless for anything that involves conscious choices by people, and utterly useless when there are emerging technologies involved. Despite this, these are the models that are used when people make simplistic predictions about how many smokers there will be 20 years in the future, which others then report as fact. Such projections about tobacco/nicotine use are perhaps slightly better than trying to project a trend about how many people will be using 11-inch tablet computers 20 years from now, but not much better.
Next simplest is Markov modeling, which basically divides people into different bins (smoker, e-cigarette user, non-user, etc.) and assumes that knowing how many people are in each bin is all you need to know to know about them to determine what happens in the next period (i.e., the next day or year). This allows for much more robust modeling of some interacting influences, but under the hood, it is still based on projections of population level trends (e.g., what portion of current smokers will adopt THR as a function of how many have already done so). Allowing for subpopulation-based trends is an improvement over just projecting graphs into the future, as it were, but at its core it is still just a version of that, with all its limitations.
Agent-based models are based on the recognition that the behavior of a population, when considering a decision-based process like THR, is really the aggregation of a lot of individual decisions. Thus, such models are based on individual actors rather than just population percentages, and the population statistics are emergent properties of the actions of individuals. The individual decisions are based on economic motives (i.e., considerations of costs and benefits) which are affected by various global factors as well as social interactions. Individuals can be realistically modeled as having different preferences and other characteristics rather than being all the same. The agent-based models also allows for social interactions at an individual level -- i.e., people can affect their neighbors and those they encounter, and the results of this may not be the same as treating everyone as if they just have the "average" experience ever period.
The model that we have created is about the simplest model possible that still captures the social dynamics, individual variability, and economic decision making that affects a population's adoption of THR. It allows for THR adoption to be a social contagion, with someone's chance of adopting it being a function of how much of it they encounter, as well as global forces. People learn (and their level of learning persists through time) and decide (based on individual motives). This contrasts with a simple projection or subpopulation-based model, where the future is basically determined by the choice of a single function -- e.g., "P% of the population smokes and that is trending down at a rate of R, so next year the number of smokers will be...." or "if X people have adopted THR in period t, then D% of the rest will adopt it in period t+1, for a total of X+D". As shown in the video, this produces population outcomes that are not just the obvious immediate result of the choice of those functions.
Update: I discuss some of the implications of this model in the context of anti-THR claims at the antiTHRlies blog.
Update (13 Nov 12): The "live" version of this (the presentation I actually gave in London) has been posted by the conference. As is usually the case, it is a bit rougher than the studio version, but for those who are are completists (are there any Phillips completists? I doubt it -- I am not even one :-), there it is. I think there are also some bootlegs, but I don't have them. Unlike most live versions, this one is a bit shorter (the studio version includes a bit more information).
Alan, We could probably model that as a dramatically higher level of utility needed to be willing to switch. It would be fairly similar to have very strong ongoing anti-THR dis-education campaign. And yes, such a change can certainly keep the population from ever tipping into large-scale adoption, or at least push it so far into the future that it is outside the model (i.e., generational change -- which is not allowed for in this model -- starts to occur).
ReplyDeleteI might do one run of this as a simple demo for a more complete presentation of the model sometime. But I hesitate to make much of it -- I do not believe I have a good way to account for black market activity in a model like this. Perhaps someone does and I can find their work.
A clear, precise and accurate description of THR, and why it works when other methods don't. An interesting model for how accelerated spread of reduced-risk behaviours can occur - crowdsourcing of a kind, I guess.
ReplyDeleteHow was it received by colleagues at City Health?
Thanks. It was very well received. I got a lot of comments/questions (kind of crowded out attention to the other presenters in the session -- a bit unfair to them). They ranged from intelligently neutral to highly positive. People were genuinely interested in the implications, were already making new suggestions for the model, and seemed positive about THR. All in all, it was quite outstanding. Many people remarked on how impressive and convincing the message was.
DeleteI got a few negative comments from the "public health" (i.e., science-impaired prohibitionists) types that were in attendance during breaks later. They clearly did not want to speak up and have their criticism publicly dismantled (though why they then wanted to confront me and have it privately dismantled is beyond me -- I think it is a personality disorder).
It was an impressively good conference. In terms of appreciation for good science and practical messages like this, it was probably the best health policy conference I have ever been to.
the model is quite fascinating, thanks for posting the vid. I wanted to go to the conference itself, but couldn't make it (and also you never replied to my email about it... tut tut.)
ReplyDeletesome questions came to mind and more detail on the modelling decisions would be interesting, two things in particular:
1) it is very reminiscent of a (real life, not abstracted) social network 'contagion' of an incentive scheme for smoking cessation (grocery vouchers)in a Scottish city I saw quite recently. uptake to the service spread through word of mouth etc and could be visually mapped onto the city layout, spreading throughout areas in a similar fashion to your model. a difference is that the uptake in my example appeared constrained at a certain point to 'islands' of use - i.e. it spread throughout one network 'node' quite well, but didn't always make the jump to nearby ones to continue the spread. obviously the THR intervention is delivered in a different manner, but there seems to be an underlying assumption in your model that each node has equal resistance to the THR contagion. (or to continue the infectious disease analogy, each subpopulation is equally susceptible to the 'disease'). I wonder whether this is a reasonable assumption to make?
2) regarding the effect of 'anti THR' activity suppressing spread. what was the theoretical or evidential backing you used to to determine the magnitude of effect of this parameter on the model outcome? obviously, in the presentation the effect is very large, which makes the point, but as the model builder you could have set it to be any value that gives the desired outcome of a very large delay in spread of THR. how did you justify the parameters that you chose?
0) Oops, sorry for not responding to the email.
Delete1a) Any "grab this offer now!" type situation has the social networks contagion element, but it free nicotine or a particularly good Groupon. However, it does not have the gradual learning element of the THR model. Someone either likes the offer and acts immediately, or not. A more subtle difference is that someone does not need to covert (take the offer) to become a source of information for others, whereas in our model, only someone who converts becomes a source of education.
1b) Not everyone is equally interested in switching. There is a random distribution of types of agents in terms of their attachment to smoking. Right now the utility function there is very simplistic -- definitely one place we plan to improve the model. But it is not uniform.
2) All of the education parameters are in arbitrary units that feed into the utility function (or the simple substitute for a real utility function we have now). There is obviously no natural unit to use for someone's level of education (though there is a ceiling on it, representing "cannot learn anything more positive about THR than she already knows"), and so by extension the impacts on an agent's education (from any source) are in arbitrary units. The relative size of different sources of education has some meaning, though. In the case shown, the flow of dis-education from the ANTZ (per individual) is small compared to the positive impact of encountering a THR user. That is part of what is so striking -- even something that small can dramatically increase the pre-tipping time.
yes, I think the fact the anti-THR message had such an apparent extreme effect raised questions for me.
ReplyDelete(some difficulties I have in wrapping my head around this arise from the fact I'm pretty ignorant regarding agent modelling, so might be making some mistakes.)
so to check if I get this right... the probability of any agent in your model moving to the 'adopted THR' state is a function of the exposure the agent has to proximal THR users (which contribute 'positive education' that pushes the agent further in the direction of THR by increasing the perceived benefit of switching to them in some kind of decision heuristic)... which is at the same time suppressed by a global parameter related to 'negative education' - that influences all agents simultaneously - which pushes them further away from THR (plus some randomness + time).
if that is correct, then I'm still struggling to see how the large magnitude of difference observed (as in the audience gasps 'isn't that line on the graph so very far away from that baseline one') in delay of tipping point isn't mostly just a result of the model specification: the weighting placed to negative eduction, in its arbitrary units, compared to the positive. but how does this relate to how people in the real world actually perceive conflicting messages of this kind from different sources? you say that in the examples in the video the ratio of negative to positive for any agent was small, but *how* small?
I'm questioning this because I'd take the view that for many health-type decisions, people seem to put a very very large amount of weight to their experiences with their social network and much less to conflicting views from the 'authorities'. the enduring popularity of some alternative medicine treatments is a testament to this - authorities say 'don't use this to treat your ailments, it doesn't work', but that seems to carry very little weight in personal decision making when compared to having a friend who proselytises about homoeopathic remedies for their migraine. so it's a question over whether you have been conservative enough about the power of negative messages from powers-that-be compared to more salient messages from social networks.
Rory, I realize we were talking past each other a bit because I was thinking of the run of the model I presented as a movie and you were probably thinking about the graphic that showed the impact of four different interventions. I believe the latter had rather larger education/propaganda flowing to the population. (I say "I believe" because my colleague ran that bit and we did not have a chance to review all of each other's work. Keep in mind this is a work-in-progress -- though it is still probably tighter and better reviewed than most of what the tobacco control industry calls "peer reviewed publications" -- which also means that your comments are appreciated and helpful.)
DeleteYou have that first bit basically right. The dis-education campaign adds (negative numbers) to the education level in the same way that positive encounters add, rather than directly suppressing the positive education. The randomness mainly takes the form of who a particular agent encounters in each period, there there is also a random element affecting how much someone "takes his learning to heart" (i.e., whether it works to tip him each period). Time passing does not have any effects (other than being one more chance for all of the above to act, of course).
Of course we can select any magnitudes to test that we want to. That is the nice thing about modeling. We do not presume to be able to figure out which of the parameterizations is "right", and I would call "liar" on anyone who claimed otherwise. There have only be maybe 10 examples of this behavior in history, and we do not have good data for most of them. And even if we did, they would all be so different from the present circumstance that we would not be able to get precise estimates from any of them. But the general trends are still interesting. How small can a TCI propaganda campaign be and still keep people smoking for substantially longer? It can produce dis-education that is three orders of magnitude smaller than the positive effect of someone seeing all his friend vaping and still have a substantial effect at the social level. That is interesting. But also interesting is the fact that it takes a really huge dis-education effort to stop progress entirely.
Re social networks and alternative medicine, I would argue that campaigns by those in power to dissuade people from these choices (setting aside any questions of their accuracy or motives) have been remarkably effective. Were it not for those campaigns, a much larger part of the population would be persuaded to try various remedies. But it is difficult to extend the analogy because in my model, THR works (as in real like) whereas most of these remedies do not. So most people who trial them are going to quickly give up on them. It may well be that social connections do cause a lot of people to trial various remedies -- the strong effect of social networks -- but then they give up. So there is not the population shift that accelerates the social effects.
yes, I think we were talking at cross-purposes a bit.
ReplyDeleteagree that we can't or shouldn't set definitively what the correct parameters are in this instance. perhaps something of interest instead would be to work the other way around from what my question was implying. that is, instead of applying observed real world knowledge about how people value different forms of educational input to the model, we start with the model output and go out to find out to what value people *do* place on anti harm reduction educational messaging, in the current context. if an effect is there(even if it is quite small) the model tells us that this is something that is important.
I suppose you could argue that you don't really need research on that, because there's a strong enough logical argument that negative messages would have some effect (and we only need some effect to be something that is worth caring about), but it might nevertheless be of some interest. particularly in a context where we don't know very much about the experiences of the average (i.e. not a self-selecting member of the vaping community) e-cig user, and, I think, even less about what the average non e-cig user thinks about e-cigs and the messages they are receiving that relate to them.
oh, and as a completely unrelated last point, I noted in the presentation that you used some phrase like 'harm reduction is the only method proven to reduce adult smoking prevalence below a fifth of the population'. given Canada is now down to about 17% and Australia around 16%, both continuing to decrease, I wondered whether it might be better to rephrase that. Sweden is lower still of course (I think around 13% last I checked), but it seems to me that to raise it in that way offers the opportunity to derail into technical arguments over whether other countries that haven't adopted harm reduction approaches have or have not gone below 20%.
The model gets us from the impact of dis-education on agents to the aggregate results. Figuring out how a particular real-world bit of dis-education maps into impact is pretty much impossible. Really the only way to do it would be to get the model "right" and then infer the effects based on the results of many natural experiments. Needless to say, this is never going to happen -- in economics we are lucky to get natural experiments like this for simple big events like minimum wage changes and such. Even changes in excise taxes -- which are relatively clean and simple -- are nearly impossible to sort out.
DeleteThe pseudo-scientists in the tobacco control industry are fond of doing junk lab studies and making claims about how people react in the real world (e.g., the absurd claim that plain packs "have been shown" to have particular effects). But that is all so worthless that, as you point out, existing basic knowledge is far more informative.
As for those numbers, I am still quite comfortable with 20%. The Canada number is based on one cherry-picked survey that has not been replicated and that has some obvious problems (rate of "ever smoking" has decreased impossibly fast; no apparent consideration of First Nations/aboriginal/Innuit poplns). Every other estimate is higher. Besides, the difference is about the rate of THR use, so that is kind of my point anyway. As for Australia, I do not know as much, but I would guess the same issues apply, only more so because the government is so thoroughly captured by obviously dishonest TCI people. How can we possible trust any statistics from the same people behind the plain packs insanity? All of these surveys do nothing to account for the presumably growing bias to give the socially acceptable (and perhaps penalty-minimizing) answer as smoking becomes more vilified and there are punishments for admitting smoking status (especially as more people people turn to the black market). Indeed, the slow downward trend over recent years seems like it is *less* than would be explained by this bias alone.
As for the "continuing to decrease", I say AHA! That, not the exact number, is the point of the statement. Those of you in the TCI -- even the extremely rare ones like you who actually step outside the echo chamber and care about science -- state as if it were fact that decreases will keep happening indefinitely. My point is that this is religion, not fact, and the claim that there will be any substantial decrease below 20% (other than via cohort replacement -- i.e., changing the population over time) is pure speculation. It is the opposite of what all the evidence shows -- except when THR has worked.