. . . OK, I’m a day late from my promised posting time. I ran into some data problems that I had to work through – details are below.

First, last week’s prediction. The stats were pointing to around 4,033 cases by 10Apr2020. The actual reported cases from the Johns Hopkins site was 3962 – a difference of only 71 cases. We’ll have to wait until next week though to see if this was a phenomenal bit of luck or whether the simple model I’m using is matching reality.

The problem I ran into with the data started on Wednesday. The day-to-day variation in the reported cases was wildly erratic. The growth factor is calculated directly from the daily cases and I think I’ve made my point about the sensitivity of predictions based on the growth factor.

I didn’t want to do anything too radical to the data otherwise it would appear that I was manipulating the numbers to make my own point. After a little research I decided to apply a three day running average on the new cases numbers. This is a common technique used to smooth financial and scientific data. If you have the following set of data:

DAY | 1 | 2 | 3 | 4 | 5 | 6 | 7 |

CASES | 80 | 40 | 110 | 100 | 100 | 190 | 60 |

Day 2 is (80+40+110)/3, day 3 is (40+110+100)/3, etc. This gives you the following smoothed data

DAY | 1 | 2 | 3 | 4 | 5 | 6 | 7 |

CASES | 80 | 76.6 | 83.3 | 103.3 | 130 | 116.6 | 60 |

Day 1 will always be 80 and day 7 will become a smoothed value if and when data for day 8 is collected.

Applying this technique to the Johns Hopkins data gives the following graph:

It’s hard to do a direct comparison with the new cases projection from last week because it wasn’t using smoothed data BUT from last week’s graph one would expect around 450-500 new cases to be reported today. The smoothed data predicts just over 300 which is much closer to the reported 287. While this looks promising I’ll be checking this as time progresses and will report the results next Friday.

Armed now with smoothed data what do the growth factors look like? Here’s a plot of the growth factors over time from the original and smoothed data:

and here is this week’s plot of accumulated cases and a projection for Friday the 17th of April.

A significant note is that over the last three projections the growth factors have been 1.16, 1.10, and 1.04 – a definite trend down. This is good news as it indicates that social distancing appears to be working and every incremental decrease in the growth factor spreads the number of cases out over time so that our hospitals don’t become saturated.

Until next Friday . . . be safe, be healthy!!