Una curiosidad interesante

«All models are wrong but some are useful»

Box Robustness in the strategy of scientific model building 1979

Obra invitada: David Robert Grimes On the Viability of Conspiratorial Beliefs 2016 (link)

«Something currently lacking that might be useful is a method for ascertaining the likelihood that a conspiracy is viable, and the factors that influence this. The benefits of this would be two-fold; firstly, it would allow one to gauge whether a particular narrative was likely and what scale it would have to operate at. Secondly, and perhaps more usefully, it would help counteract potentially damaging anti-science beliefs by giving an estimate of viability for a conspiracy over time. The parameters for this model are taken from literature accounts of exposed conspiracies and scandals, and used to analyse several commonly held conspiracy theories, and examine the theoretical bounds for the magnitude and time-frame of any posited conspiracy theory.»

No soy estadista como para evaluar 1.1, pero la idea me parece de lo más curiosa. La determinación de los parámetros en 1.2 parece en orden («This essentially yields a “best-case” scenario for the conspirators»), aunque su muestra es sólo de 3 conspiraciones. Citándole:

«Given the lack of clarity in getting precise numbers and time-frames, there is inherent uncertainty in this work on the estimated parameters and better estimates would allow better quantification of p. There is also an open question of whether using exposed conspiracies to estimate parameters might itself introduce bias and produce overly high estimates of p—this may be the case, but given the highly conservative estimates employed for other parameters, it is more likely that p for most conspiracies will be much higher than our estimate, as even relatively small conspiracies (such as Watergate, for example) have historically been rapidly exposed.».

«The estimates also make the assumption that all agents in the estimate are considered to have knowledge of the conspiracy at hand; if this wasn’t the case, then only those with adequate knowledge of the deception would count towards the number No. This might potentially be the case for some political or social conspiracies, yet for a hypothetical scientific conspiracy it is probably fair to assume that all agents working with the data would have to be aware of any deception. Were this not the case, fraudulent claims or suspect data would be extrinsically exposed by other scientists upon examination of the data in much the same way that instances of scientific fraud are typically exposed by other members of the scientific community.»

(no voy a entrar en su discusión del modelo, pero su comentario sobre Eq 6 y su propuesta de mejora vía «agent based models» me parecen también relevantes)

Aun con todo, son interesantes de ver las tablas 1, 3 y 4.

«The analysis here predicts that even with parameter estimates favourable to conspiratorial leanings that the conspiracies analysed tend rapidly towards collapse. Even if there was a concerted effort, the sheer number of people required for the sheer scale of hypothetical scientific deceptions would inextricably undermine these nascent conspiracies. For a conspiracy of even only a few thousand actors, intrinsic failure would arise within decades. For hundreds of thousands, such failure would be assured within less than half a decade. It’s also important to note that this analysis deals solely with intrinsic failure, or the odds of a conspiracy being exposed intentionally or accidentally by actors involved—extrinsic analysis by non-participants would also increase the odds of detection, rendering such Byzantine cover-ups far more likely to fail. Moreover, the number of actors in this analysis as outlined in Table 2 represent an incredibly conservative estimate. A more comprehensive quantification would undoubtedly drive failure rate up for all considered conspiracy narratives.»