![\"\"](\"https:\/\/i0.wp.com\/radioislam.org.za\/a\/wp-content\/uploads\/2020\/09\/STATSCOV.jpg?resize=600%2C400&ssl=1\")
<\/p>\n<\/p>\n
By Jacques Raubenheimer<\/span><\/strong><\/p>\n Senior Research Fellow, Biostatistics, University of Sydney<\/p>\n 15:09:2020<\/strong><\/p>\n If we don\u2019t analyse statistics for a living, it\u2019s easy to be taken in by misinformation about COVID-19 statistics on social media, especially if we don\u2019t have the right context.<\/p>\n For instance, we may cherry pick statistics supporting our viewpoint and ignore statistics showing we are wrong. We also still need to correctly interpret these statistics.<\/p>\n It\u2019s easy for us to share this misinformation. Many of these statistics are also interrelated, so misunderstandings can quickly multiply.<\/p>\n Here\u2019s how we can avoid five common errors, and impress friends and family by getting the statistics right.<\/p>\n Social media posts<\/a>\u00a0comparing COVID-19 to other causes of death, such as\u00a0the flu<\/a>, imply COVID-19\u00a0isn\u2019t really that deadly<\/a>.<\/p>\n But these posts miss COVID-19\u2019s infectiousness. For that, we need to look at the infection fatality rate (IFR) \u2014 the number of COVID-19 deaths divided by all those infected (a number we can only estimate at this stage, see also point 3 below).<\/p>\n While the\u00a0jury is still out<\/a>, COVID-19 has a\u00a0higher IFR<\/a>\u00a0than the flu.\u00a0Posts implying<\/a>\u00a0a low IFR for COVID-19 most certainly underestimate it. They also miss two other points.<\/p>\n First, if we compare the\u00a0typical flu IFR<\/a>\u00a0of\u00a00.1%<\/a>\u00a0with the\u00a0most optimistic COVID-19 estimate<\/a>\u00a0of 0.25%, then COVID-19 remains more than twice as deadly as the flu.<\/p>\n Second, and more importantly, we need to look at the basic reproduction number (R\u2080) for each virus. This is the number of extra people one infected person is estimated to infect.<\/p>\n Flu\u2019s R\u2080\u00a0is about 1.3<\/a>. Although COVID-19 estimates vary, its R\u2080 sits around a\u00a0median of 2.8<\/a>. Because of the way infections grow exponentially (see below), the jump from 1.3 to 2.8 means COVID-19 is vastly more infectious than flu.<\/p>\n When you combine all these statistics, you can see the motivation behind our public health measures to \u201climit the spread\u201d. It\u2019s not only that COVID-19 is so deadly, it\u2019s deadly\u00a0and<\/em>\u00a0highly infectious.<\/p>\n Read more:\u00a0How deadly is the coronavirus? The true fatality rate is tricky to find, but researchers are getting closer<\/a><\/strong><\/em><\/p>\n A simple graph might plot the number of new COVID cases over time. But as new cases might be reported erratically, statisticians are more interested in the rate of growth of total cases over time. The steeper the upwards slope on the graph, the more we should be worried.<\/p>\n Read more:\u00a0Coronavirus is growing exponentially \u2013 here\u2019s what that really means<\/a><\/strong><\/em><\/p>\n For COVID-19, statisticians look to track\u00a0exponential growth<\/a>\u00a0in cases. Put simply, unrestrained COVID cases can lead to a continuously growing number of more cases. This gives us a graph that tracks slowly at the start, but then sharply curves upwards with time. This is the curve we want to flatten, as shown below.<\/p>\n However, social media posts routinely compare COVID-19 figures with those of other causes of death that show:<\/p>\n Even when researchers talk of exponential growth, they can still mislead.<\/p>\n An Israeli professor\u2019s\u00a0widely-shared<\/a>\u00a0analysis claimed COVID-19\u2019s exponential growth \u201cfades after eight weeks\u201d. Well, he was clearly wrong. But why?<\/p>\n His model assumed COVID-19 cases grow exponentially over a number of days, instead of over a succession of transmissions, each of which may take several days. This led him to plot only the erratic growth of the outbreak\u2019s early phase.<\/p>\n Better visualisations truncate those erratic first cases, for instance by starting from the 100th case. Or they use estimates of the number of days it takes for the number of cases\u00a0to double<\/a>\u00a0(about six to seven days).<\/p>\n Read more:\u00a0The bar necessities: 5 ways to understand coronavirus graphs<\/a><\/strong><\/em><\/p>\n Then there\u2019s the confusion about COVID-19 infections versus cases. In epidemiological terms, a \u201ccase\u201d is a person who is diagnosed with COVID-19, mostly by a positive test result.<\/p>\n But there are many more infections than cases. Some infections don\u2019t show symptoms, some symptoms are so minor people think it\u2019s just a cold, testing is not always available to everyone who needs it, and testing\u00a0does not pick up all infections<\/a>.<\/p>\n Infections \u201ccause\u201d cases, testing discovers cases. US President Donald Trump was close to the truth\u00a0when he said<\/a>\u00a0the number of cases in the US was high because of the high rate of testing. But he\u00a0and others<\/a>\u00a0still got it totally wrong.<\/p>\n More testing does not\u00a0result<\/em>\u00a0in more cases, it allows for a\u00a0more accurate estimate<\/em>\u00a0of the true number of cases.<\/p>\n The best strategy, epidemiologically, is not to test less, but to test as widely as possible, minimising the discrepancy between cases and overall infections.<\/p>\n Estimates vary<\/a>, but the time between infection and death could be as much as\u00a0a month<\/a>. And the variation in\u00a0time to recovery<\/a>\u00a0is even greater. Some people get really ill and take a long time to recover, some show no symptoms.<\/p>\n So deaths recorded on a given date reflect deaths from cases recorded several weeks prior, when the case count may have been\u00a0less than half<\/a>\u00a0the number of current cases.<\/p>\n The rapid case-doubling time and protracted recovery time also create a large discrepancy between counts of\u00a0active and recovered cases<\/a>. We\u2019ll only know the true numbers in retrospect.<\/p>\n Some social media users\u00a0get angry<\/a>\u00a0when the\u00a0statistics are adjusted<\/a>,\u00a0fuelling<\/a>\u00a0conspiracy theories<\/a>.<\/p>\n But few realise how\u00a0mammoth, chaotic<\/a>\u00a0and\u00a0complex<\/a>\u00a0the task is of tracking statistics on a disease like this.<\/p>\n Countries and even states may count cases and deaths differently. It also takes time to gather the data, meaning retrospective adjustments are made.<\/p>\n We\u2019ll only know the true figures for this pandemic in retrospect. Equally so, early models were not necessarily wrong because the modellers were deceitful, but because they had insufficient data to work from.<\/p>\n Welcome to the world of data management, data cleaning and data modelling, which many armchair statisticians don\u2019t always appreciate. Until now.<\/p>\n<\/div>\n1. It\u2019s the infection rate that\u2019s scary, not the death rate<\/h2>\n
\n\n<\/div>\n<\/div>\n<\/div>\n
\n
\n2. Exponential growth and misleading graphs<\/h2>\n
\n
\n![](\"https:\/\/i0.wp.com\/cdn.theconversation.com\/static_files\/files\/890\/Flatten_the_curve1.gif?w=1080&ssl=1\")
\n
\n\n<\/div>\n<\/div>\n<\/div>\n
\n
\n3. Not all infections are cases<\/h2>\n
\n\n<\/div>\n<\/div>\n<\/div>\n4. We can\u2019t compare deaths with cases from the same date<\/h2>\n
\n\n<\/div>\n<\/div>\n<\/div>\n5. Yes, the data are messy, incomplete and may change<\/h2>\n
\n\n<\/div>\n<\/div>\n<\/div>\n