Statistical Probabilities

Take a simple game of chance, the toss of a coin, and a nominal bet.  Heads you win, tails you lose and pay up.  Your opponent flips the coin...sorry, tails.

You go again...tails.

Again...tails.

Again...tails.

Perhaps once more?  Tails. Again.

At what point do you call out your opponent and declare him a cheat?

After the first throw?  That seems far too premature.  The third?  The twenty-third?

In each case the probability of the coin toss is clearly unchanged but the repeated result becomes cumulative and increasingly more and more unlikely.  The statistical analyis of probability is a huge area and not the focus here.  The focus is the statement, the declaration that the coin is fixed.

On March 3rd Dr. Deborah Persaud from John's Hopkins University described the first case of a child being 'cured' of HIV at the Conference on Retrovirsuses and Opportunistic Infections.  Eventually it trickled through to the major news organisations and, around 9pm began to circulate widely on Twitter.

As things go, the journalistic nature of science announcements is generally good but is harmed by the wish to break it down to simple soundbites.  Here was the first patient cured of HIV.  Success for medical science.  Hope for millions of people.

In reality this is not what was claimed by Dr. Persaud's team or by the news networks but was hailed with "Wow!"'s, retweets and shared with the addition of triumphant comments.  The sad breakdown of the scientific process for a media environment primed for snappy assertions.  While the reality was explained within the articles the word cure was present in almost every headline, most in emphasis marks but far too many without.

The difficulty is the vast difference between medicine and their associated trials, and how the rest of the scientific community work.  Here Dr. Persaud was presenting the equivalent of the OPERA neutrino result; an experimental finding that doesn't have a full explanation by the team but is presented in a way for other scientists to read, discuss and investigate further.  Both have suffered from the prevalence for scientific research to be public (quite rightly) but also for it to be translated into news-speak.

"Eating red meat is good for you! Scientists say"

"Shampoo makes you fat! A new study today shows"

The reality is much more boring and mundane but today's society expects immediate results, absolute statements and quick conclusions.  Science needs to move methodically but this is too often is presented as going too slowly.

In this HIV story the main factor is that is based around one little girl who had HIV which can not be detected now.  One.  Medical trials are carried out on hundreds, often thousands of people before the confidence levels are high enough to be sure that the coin is truly weighted.  Even then it's not necessarily 'good' science.  What percentage of people have actually been tested out of the global population?  At what point do you take this tiny fraction to be good enough to warrant licencing?

In a few weeks time CERN will be holding a conference to announce the results of their yearly work and data analysis, and the feeling on site was that the Director General, Professor Rolf-Dieter Heuer, will go from discussing Higg's-like particles being detected to discussing Higgs particles being detected.  To many people who I speak to this surprises and confuses them.  To those who only scratch the veneer of science, that which is presented by the media, the Higgs particle was found last year.  In some way this is true; a particle behaving in an identical way to the predicted nature of the Higgs has been identified in the appropriate energy range however sighting it is not confirmation.  The five sigma level of the results showed that there was 99.977% certainty that the results were accurate but this is all down to data collection.

In the medical trials the samples are in people, in this case one person.  In particle physics the samples are in the number of collisions, it's given a technical unit called the inverse femtobarn and isn't recommended for public use in CERN output; they suggest converting it to a raw number of collisions.  In 2012 alone the LHC's detectors recorded over 23 fb⁻¹.  To follow CERN's advice and convert that into an actual number that's 2300 trillion collisions in one year.

What irritated me so boils down to two points:

1. The media is very effectively promoting science with one hand but undermining it with the other by announcing experimental findings prematurely which then leads to speculation rather than holding off and waiting for confirmation by scientists and not presenters of the Today programme.

2. The word 'cure' in a headline or opening paragraph, even if the article later describes the results correctly is bad journalism, bad science and inherently misleading.

More people talking about science is great in my opinion but having people going into work or school talking about the cure that has been found for HIV is exactly what science education and the science curriculum is trying to combat in this country.  Where the media and journalism could help they are instead part of the problem.