Lacent’s Vaccination Model Says What Its Told To Say

I beg your indulgence, dear reader. Please stay with me for these introductory paragraphs. I want to teach you how to read these papers so that you are well equipped to debate those who only see headlines about them in the “news”.

All models only say what they are told to say. This is a truism, and should not be in the least controversial. It doesn’t matter if any one person, or all people, did not anticipate any model’s output. The model they built (figuratively or literally) and they commanded to run in a specific way can only say what it is programmed to say, and can say nothing but what it is told to say.

This is neither good nor bad, a crucial point. Models’ inflexible servitude just is. It is never a moral or quality judgment about any model to point this out. Even stronger, models only become good or bad depending on the uses to which they are put.

From this we deduce any model’s output is never proof of the model’s premises. You cannot for example create a model that is premised on the proposition “Vaccines work at saving lives,” run that model, examine its output, then say “Lo! The model has proved vaccines saved lives.” No. It does not matter whether vaccines do in fact save lives, or that vaccines kill (many reading this post will forget this point). You have proven nothing about vaccines from your model. You assumed vaccines work. Of course your model will, and must, say they worked.

Observations about vaccines’ effectiveness are another matter. They can of course be evidence one way or the other. If observations say “This vaccine has shown itself in the wild (i.e. outside the hands of the company that sells it) to be x% efficacious in preventing death,” then that information can be used in various ways. But it is still a premise if it appears in any new model about vaccines (or anything). If the new model with that premise announces, “We discovered vaccines to have saved x% of lives” our response ought to be a scientific “No kidding.”

Then if a modeler says, “I don’t have that x%, because data is missing, but I’ll make up a plausible guess, which will have some plus or minus attached to it, of course, but I’ll ignore that for now,” and he uses that guessed x% as a premise to his new model, which he is allowed to do, if he announces, “We discovered vaccines to have saved x% of lives”, our response ought to be “Dude.”

With all this in mind, let’s examine the peer-reviewed paper “Contribution of vaccination to improved survival and health: modelling 50 years of the Expanded Programme on Immunization” by Andrew J Shattock and bunch of others in The Lancet.

The authors conclude based on their model this: “The study found that, in 2024, both children and adults are more likely to survive to their next birthday than if no vaccinations had occurred since 1974.” I’ll remind you of this at the end.

From their Methods:

In this modelling study, to quantify the impact of EPI, we estimated the number of deaths averted, the life-years gained, and the years of full health gained (ie, disability-adjusted life-years averted) by vaccination (henceforth vaccine impact) against 14 pathogens (ie, diphtheria, Haemophilus influenzae type B, hepatitis B, Japanese encephalitis, measles, meningitis A, pertussis, invasive pneumococcal disease, poliomyelitis, rotavirus, rubella, tetanus, tuberculosis, and yellow fever) in 194 WHO member states between June 1, 1974, and May 31, 2024, through coverage achieved by routine and supplementary immunisation activities. We developed a standardised analytical framework to estimate vaccine impact per fully vaccinated person over time, synthesising the results of 22 models and applying regression-based imputation methods to ensure geographical and temporal completeness.

No covid, and no flu you see. Recall Haemophilus influenzae type B, or Hib, is a bacteria (not virus) that can cause meningitis in kids. I’ll pass by the poor English of substituting the perfectly good vaccination with the abominable vaccine impact.

You see, I trust, the Lancet model is a model of models, and you remember that “imputation” means they didn’t have observations or models in some cases and so made guesses what these might have been, which should have plus-or-minuses attached. Further, each of the input models has, or should have, a plus-or-minus attached to its output, which ought to have been carried forward into the model of models. These uncertainties were not carried forward, the importance of which I’ll show you in a moment.

From their Procedures:

We synthesised age-specific vaccine coverage estimates from four data sources…Where country coverage data between 1974 and 1979 were unavailable, for low-income and middle-income countries we linearly extrapolated from known coverage in 1980 to an anchored 0% coverage in 1974…

Saying “synthesised” and “extrapolated” sounds a lot more scientific than “guessed”, I suppose. And there is nothing wrong with guessing, mind you. It’s that when you guess you ought to account for the uncertainty in that guess, which they did not.

The rest of the paper (mostly tucked away in the Supplementary material, the modern publishing trick of hiding the details) is giving their model, which is simple. They assume functional forms for vax efficacy. For instance, Diptheria is a simple exponential decay. The others are similar.

Coverage models are similar. These are fed into the “the equivalent number of disease-specific deaths in the absence of vaccination” model, a function with one number divided into another, the numerator of which is a number plugged in from a vaccine database.

Indeed, all the parameters of these models are straight inputs, which their either got from other papers (no uncertainty attached) or guessed.

That’s basically it, except the whole is dirtied in their imputation models (regressions) by adding in things like Gini index, gross domestic product (yes; with lags; don’t worry if you don’t get that), public health spending and the like. The usual “economics” confusion of correlation and causation.

Now they do show some uncertainty in their model outputs. They claim, using causal language, that between 1974-2024, 353,000 Diptheria deaths were averted globally because of vaccination, rounded to nearest thousand. They give parametric uncertainty to this, i.e. “95% credible intervals generated from 100 samples of Monte Carlo Markov Chain posteriors from impact function fits.” For Diptheria this is “[325,000 – 375,000]”.

For Poliomyelitis they say 1,570,000 deaths averted globally with uncertainty bounds of 0; that is, “[1,570,000 – 1,570,000]”. The number is exact. And so on similarly for all the pathogens they model.

None of this can be taken that seriously. This kind of precision is not possible in these circumstances, given all the reasons above. These credible interval numbers reflect details about model innards, and are not to be taken as predictions about the world. There is no way to know in Reality that the polio vax was 100% effective everywhere and everywhen (and injury free). The model assumed this. The model then said this. There was no other possibility.

Understand carefully that it may be that Diptheria vaccination did indeed save 353,000 lives, and so on for each vax and bug. Our criticism is not about whether this is true. It is about whether the authors’ model is justified to say so in the certainty they are stated. Mixing these statements up accounts for much confusion and error. This is the point where many become unreasonably emotional. For many, questioning their reasons for belief is seen as a unwarranted attack. Recent events prove this out.

There is nothing untoward with guessing how many lives were saved by a vax. But that guess to be worthy should have a proper accounting of all the uncertainties involved. And that justification for that guess cannot be the model itself, for the model only said what it was told to say. This model was told to say “Vaccinations saved X lives, +/- W, for Disease Y”, and that is what it reported. Then, even in model terms, those Ws are far too small.

The authors would have been better served had they shown, by the same regions as they report, the number of reported diagnoses of each disease, and the number of reported deaths of each disease, along with a some indication of the chance of error in each count. Believe it or not, dear reader, doctors make mistakes in diagnoses, and records of causes of death are not in the least simple. There is always a plus or minus attached.

Of course, for those regions that did not have such reports, or reports of vaccinations, it would then become instantly clear that uncertainty is much greater than that reported in the model. For instance, in Africa, the model reports 210,000 lives saved with model uncertainty of “[203,000 – 216,000]”. This is one of the areas which has the worst reporting. Can those narrow bounds really be justified over that entire continent over that wide span of years? No, sir, they cannot.

Really, not much has happened in this paper. The authors began with the presumption “These vaccinations prevent death at the stated efficacy”, and that’s what they ended saying. They concluded, you recall, “The study found that, in 2024, both children and adults are more likely to survive to their next birthday than if no vaccinations had occurred since 1974,” which is (and other similar “findings” are) of course precisely what they told their models to say. The paper is thus more like the kind of report bureaucracies demand and publish to justify their existence. It provides very little new information; it repackages what was known in a shiny mathematical wrapping.

It falls in the class of scientism of the first kind, and justifies my (long-running) contention that there is too much money in science and too many scientists. Nothing would have been lost had this paper never been known, and something would have been gained; namely, a greater appreciation of uncertainty.

Subscribe or donate to support this site and its wholly independent host using credit card click here. Or use PayPal. Or use the paid subscription at Substack. Cash App: \$WilliamMBriggs. For Zelle, use my email: matt@wmbriggs.com, and please include yours so I know who to thank. BUY ME A COFFEE.