Research

COVID-19 Research

Title :

Simulating with confidence: Accurate estimation in the study of COVID-19

Area of research :

COVID-19 Research, Life Sciences & Biotechnology, Mathematical Sciences

Focus area :

Prediction model for COVID-19

Principal Investigator :

Dr Dootika Vats, Assistant Professor, Indian Institute of Technology (IIT) Kanpur

Timeline Start Year :

2020

Timeline End Year :

2020

Contact info :

Details

Executive Summary :

This study aims to develop an Open-source statistical software specifically for Markov chain Monte Carlo (MCMC) data that accurately quantifies the variability in the final estimates and informs scientists about the quality of their results and inference.

Outcome/Output:

The SimTools open-source software package has been developed for conducting reliable simulation in the statistical language R. The SimTools package produces plots that summarise the complex structure of the simulations and equip users with the tools to do reliable inference. The base plotting features in R provides no uncertainty assessment. By estimating the variability of Monte Carlo estimates, using batch means and spectral variance estimators, simultaneous confidence intervals have been made for any combination of means and quantiles for different marginal quantities. The width of these confidence intervals indicate that variability is high, and thus will allow users to obtain more samples in their simulation. This level of visual feedback is particularly useful for users not familiar with MCMC simulation, as is typically the case. Pertaining to COVID-19 modelling, many models run various simulations trying to project their forecasts. For such users, knowing an adequate length of their simulation is critical, and this visualisation tool is extremely informative. Additionally, for MCMC processes, visualising the correlation in the process is critical to assessing the rate of mixing of the Markov chains. This is done in the autocorrelation function plots, where the unique plots are able to present levels of autocorrelation from parallel Markov chains. The ease of usability for users is a big plus and the coding environment has been constructed to help future development of other diagnostic plots.

Organizations involved