Prediction: Active COVID-19 Cases in Fiji

The data on this page is a prediction based on plotting actual data and using it to extrapolate (forecast/predict) future case numbers for Fiji. I will use the term ‘predict’ for simplicity.

Based on the types of data objects (tables, charts), this page is best viewed on a desktop computer, or if on a mobile device, in landscape (rotated) view.

Updated-Modelled Forecasts

Due to the recent fluctuations in daily new the reported case numbers and the general slowing down of the cumulative case number trend, I have revised my forecasts and present these below in Table 1.1. These are tentative forecasts, so I will wait and see how they compare for now before commenting further. I have not changed anything in the cumulative case numbers yet.

I must add though that the daily case number fluctuations between ~700 one day and double that the next is really distorting predictions. The stats are based on positive cases identified and affected by several factors such as sample delivery, capacity to test the samples etc. It means that the data previously used to construct the predictions are perhaps not as valid going forward. I have therefore amended the source data on which the predictions are based off. Tentatively, I have discarded the June stats and kept the July 1 data onwards for the revised forecast. While I have not shown the graph just yet, I can share the predictions and the fact that the fit of the amended data is linear. Considering how the data trends start to vary unpredictably (pardon the pun), it is prudent to predict for shorter periods, which is why Table 1.1 predictions only extend to August 15, 2021.

The amended predictions are starting from June 29 data, and in Tables 1 and 2, below.

UPDATE: Please note that I have now amended predictions of Cumulative Case Numbers (Table 2) as the error margin had started to exceed the 20% accuracy mark using the previous modelling (Table 4).

Because we are still interested in how the old predictions compare, I have retained them in the tables below.

Previously-Modelled Data

For the interested: The prediction is based on a polynomial fit of the active cases. Depending on when the source data (for preparing the predictions began, there are different values for Day 1 in each table). I predicted the future stats at different times for different metrics. I am watching the unfolding case numbers each day on www.ourworldindata.org (which explains why my stats seem a day late), and using their data to predict future case numbers. It is how science works, and while it may yet seem like an imperfect process, it is based on recorded and reported data, and all my reasonings behind the predicted numbers are shared with you. It gives you an insight into my thought process and rationale, and allows you to give me feedback on anything that may be useful that I may have missed. I would appreciate your feedback on these stats and interpretations. Email me here.

I must emphasize that that the predicted stats are unfolding data and as a result, subject to several external factors, all of which are moving parts. These include:

• The effectiveness of the medical system in Fiji which has recently been boosted with experts from Australia and New Zealand (there could be more coming in which would hopefully help address the spiraling cases);
• The resistance to vaccination on the ground which slows the response against the virus;
• The nature (severity) of cases of infection at any time (which affects the rate of recovery); and
• we are dealing with persons, and each person is different from another in terms of their physiology etc.

The forecast will be regularly updated with the new cases data as it comes.

Here is the predictions table (Table 3). Note that this is a recent prediction and it remains to be seen how accurate these predictions are.

The graphed data for cumulative case numbers was started in mid-July, so much more predictions (Table 4) have been available.

Both data sets will be updated regularly as well.

The rest of the information is fairly self-explanatory.

Please note that these are large tables and this page is best viewed on a computer or, at best, a mobile device in rotated (landscape) view.

The predictions would rarely be 100% accurate (or with 0% error). The four factors listed above are some of the causes of this. When the error starts to change in a particular direction (negative or positive), that is called a bias. A bias can be in either direction (increase or decrease). That is usually we revisit the prediction (forecast) and tweak it.

Tweaking a forecasting model can be done in several ways. The first is to inspect whether the trends have changed in the latest data, causing the predictions to be distorted. When that is the case, we re-check the values within the data set that were used to compile the model and produce the predictions. It may require some data to be discarded (usually the oldest), the new data to be included, or both.

The accuracy of the predictions depends on a large extent to how well the data is ‘fit’. As at 22 July 2021, the fit is reasonably good. We can continue to use the models to make predictions as they are.