by Jess Rees
On the 22nd of July, The ORSSA Johannesburg chapter hosted a webinar, to which all ORSSA members were invited.
We started with a lightning talk called "Sharpening your skills using online competitions" by MG Ferreira, from Ferra Solutions. MG explained how he has been involved in online data science competitions on the Zindi platform, and gave advice for how others can start competing. His advice for someone wanting to get involved is
you need to really commit and keep at your projects
you must use good coding and documentation practices, as you'll want to be
able to extend your project as the competition goes on
it's advantageous to compete in teams, as teams will likely have a greater variety of skills, and you'll get to learn new skills in the process of competing!
Zindi competitions often focus on applying Operations Research and Data Science for social benefit, and they are a great way to develop technical skills and enrich careers, so MG encourages everyone to participate!
The main talk was given by professor Paul Fatti, entitled "Some simple statistical ideas useful for understanding the COVID-19 pandemic". He explained how relatively simple statistical analyses can be used to provide valuable information about the pandemic, including the estimation of the number of infections and number of deaths, analysis of COVID-19 screening testing, and the testing of vaccines.
A variety of approaches for estimating the number of infections were highlighted, each of which has its own costs and benefits. In a discussion of estimating the number of deaths, professor Fatti explained how an analysis of the excess deaths in the country can be used to attribute deaths to the pandemic – either directly (through death from the disease) or indirectly (for example, as a result of the serious economic hardships faced due to lockdowns), and how such an analysis reveals the officially reported numbers to be severe under-estimates of the true number of deaths.
Professor Fatti showed how large-scale screening for COVID-19 in the general population does not make much sense, as the rate of falsely-positive test results can be can be unacceptably high when the true prevalence of infection is low, even when tests have very high sensitivity and specificity. Rather, tests should only be conducted when exposure to the virus is already suspected. He also explained how group testing can allow pathology laboratories to realise huge savings in time and materials without significantly impacting the tests’ effectiveness when the prevalence of the virus is relatively low, and how to estimate the optimal group size given the prevalence. Finally, professor Fatti showed how a Bayesian approach can be used to analyse the efficacy of vaccines in reducing infection rates.
A massive thank you to MG Ferreira and professor Paul Fatti for sharing their time, knowledge, and experience with the ORSSA community!