The model predicting 'massive' increase in COVID-19 cases in Mobile badly missed the mark

A model that used social mobility data from May to predict a huge upswing of June COVID-19 cases in Mobile County didn't even come close.

Fox10tv.com posted this news item from respected investigative reporter Brendan Kirby on May 22, 2020.

Even as Alabama moves to lift more restrictions imposed to fight the novel coronavirus, a new model suggests the state’s hardest-hit county is poised for an explosion of new cases.

The forecast, produced by the Children’s Hospital of Philadelphia Policy Lab and the University of Pennsylvania, projects a six-fold increase in Mobile County’s daily average of new cases by the middle of June.

Unlike some other models, it is less concerned with the official polices in place and focuses more on what actually is happening on the ground. It includes factors like heat and humidity, which tend to slow the virus, and attempts to measure how well people are “social distancing.” As a proxy for that, the model uses cell phone GPS data from Unicast to determine how often people are going to “non-essential” businesses.

The model predicted a rate of new cases reaching 45/day on Memorial Day, then rising to more than 350/day by mid-June. The article is scary. Had you read it when it was published, you might have thought Mobile was in line for a disastrous few weeks.

It hasn't even been close.

I encourage you to follow the link to Kirby's report, and watch the two minute video. Pay particular attention to Rubin's comments.

Today is June 7 in the Year of our Lord 2020, a day on which there should have been a whopping 140 newly sickened patients. The actual number of new cases today: 56. That represents a significant increase from yesterday's observation of 25, but the jump has absolutely nothing to do with social distancing regime that was in place last month. Social distancing is the most influential predictor in the Children’s Hospital of Philadelphia, according to what Director David Rubin told Fox10. If anything, Mobile County has gotten worse in social distancing since  mid-May, not better. The intervening period includes high school graduations, the Memorial Day Weekend and days of public protesting. We may yet see an increase in positive COVID-19 cases as result of all of this highly social interaction, but we have no chance of seeing the dire trend predicted by this model. 

More than two months ago, this blog examined another model that predicted death and devastation for our fair state. That model also used specious data to produce a super scary forecast, and that one has also thankfully proved inaccurate. The Institute for Health Metrics and Evaluation's COVID-19 projections were garbage in April and are still garbage today. Yet Fox10 (and others) still portray them in pages on their websites. Why?

The chart below shows just how awful IHME has been in predicting the coronavirus pandemic in Alabama.

The bold, jagged brown line shows the daily confirmed infections statewide. The dashed line shows the model's predictions for infections. Since record keeping began and IHME has been producing forecasts, the model has horribly and consistently missed the actual case rate.

The chart I've included also shows a shaded area that represents the uncertainty interval for the model's results. The best way to understand that is to say, "there is a 95% chance that the actual result will fall with in the shaded area." The actual result almost never has.

The IHME model and the model produced by Children’s Hospital of Philadelphia both use their assessment of social distancing as a primary driver predicting infections, and both have produced garbage results.

Government officials and the media have been consistent in driving home the need for restrictive social distancing, and using these demonstrably worthless predictive models to convince people that if they don't practice proper social distancing, very bad things will happen.

Interesting, no?

Think of it this way: You saw the news reports in March and April. You saw the model predictions of a runaway virus killing people by the thousands in your community, and the officials' fervent pleadings and outright directives that you follow this new normal, or else. Now, you've learned that the models they used to force your compliance are garbage.

Grim state and local media coverage don't tell the whole Alabama COVID-19 story

As you read this, remember that the job of today's mass media is to sell cars,  drugs, legal services and laundry detergent. In order to do this job, they have to report on the state of current affairs, and they've found that fear, uncertainty and doubt sells more stuff than sunshine and optimism.

Nothing epitomizes the media mindset on the Coronavirus pandemic better than this op-ed piece from al.com's Kyle Whitmire. I hope you'll join me in walking straight past the bullshit to find the truth that there is every reason to be optimistic; that we aren't doing nearly as badly as our journalistic professionals would like you to believe.

The Alabama Department of Public Health has a website called the COVID-19 Data and Surveillance Dashboard, which includes case statistics and charts that help the public understand how the pandemic is affecting our state.

Real Clear Politics has another data aggregating website called the Coronavirus Tracker for the United States (they have a global aggregator page there, too).

These resources are updated in near real time, so when you load the pages the data you see are the freshest available and they are not edited.

Using the data provided by these resources, I've developed a set of interesting charts that strongly indicate that the worst of this disease's impacts are behind us (for this wave, anyway).

This shows the unweighted moving average of new cases reported to ADPH each day for the entire state. A moving average is used to smooth out the day-to-day "noise" in a time series and reveal trends that might not be discerned by looking at the noisy raw data.

Here's a plot of the same moving average for Mobile County. The shapes of the two plots are not radically different. Both show two peaks, occurring at about the same times. The state's second peak is higher, which is explained in part by higher levels of testing in more rural areas of the state. Neither of the two short term trends leading to the second peaks is sustainable because of the information revealed in the following chart.

This chart shows the percent positive vis-a-vis the number of positive test results. The percent positive line peaked at about the same time as the state's first peak in mid-April, but the total number of positive cases has trended generally higher. The only phenomenon that could cause such a divergence is the combination of increased testing and a decline in the proportion of the state population that has been infected with the virus. In other words, the state is looking harder for positive cases by increasing the number of tests, but is having a harder time finding positive test results. The total number of cases has gone up, but only because testing has gone up. The declining number of percent positive will probably continue, because stay-at-home, social distancing and quarantines have had their impact. All of the low hanging fruit has been harvested.

What about Alabama's status versus the rest of the country? According to the  RCP page in the link above, Alabama is:

• 26th in the number of confirmed cases
• 28th in the number of cases per 1 million population
• 25th in the number of deaths
• 25th in the number of deaths per 1 million population
• 31st in confirmed case fatality rate
• 22nd in tests performed 

So, an honest evaluation of how Alabama fares nationally would conclude that we're about average. About half of the country is doing better and about half is doing worse. In other words, it's not as bad as the tense, anxious news reports and op-eds would have you believe.