Spreadsheet snafu, ‘Long Covid’ quantified, and the birth of probability

Episode Summary

Executive Summary: This episode examines several ways numbers can mislead during the pandemic: England’s test-and-trace data were delayed and distorted by an Excel spreadsheet error; claims about “half a million” people with long COVID are based on early app data and uncertain definitions; public guesses about COVID deaths are badly skewed by averaging methods; and “exponential growth” is often misused in everyday language. The recurring theme is that careful interpretation matters as much as the raw figures.

Main Topics: England’s missing COVID test results (Priority: 5/5): Public Health England lost 15,841 positive test results because an old Excel file format could not handle all incoming rows, delaying contact tracing and temporarily understating case growth. Long COVID estimates and uncertainty (Priority: 5/5): The episode scrutinizes media claims that half a million Brits have long COVID, showing the figure depends on assumptions about total infections and a 28-day definition, while longer-term app data suggest persistent symptoms in a smaller but still significant share. Public perceptions of COVID deaths (Priority: 4/5): A poll claiming people thought 7% of the UK had died of COVID is unpacked as a mean distorted by outliers; the median guess is far lower and closer to a general belief that the death rate is small but underestimated. What counts as exponential growth (Priority: 3/5): The show addresses the misuse of 'exponential' as a synonym for 'fast-growing,' explaining that true exponential growth means growth whose rate compounds over successive generations, like rabbit reproduction or epidemic spread. Using multiple data sources to understand the pandemic (Priority: 4/5): Experts explain that case counts alone are unreliable when testing/reporting is incomplete, so surveys, symptom apps, hospital admissions, and ONS data give a fuller picture of spread. Limits of app-based health research (Priority: 4/5): The COVID Symptom Study app produced rapid, useful insights, but its self-reported nature, selection effects, and lack of clinical verification create uncertainty about symptom attribution and prevalence.

Key Arguments: Using Excel as a database for national-scale health data was a design flaw because it is not built for large datasets or robust audit trails. The missing 15,841 cases mattered operationally because delays in contact tracing reduce the chance of interrupting transmission. The apparent dip in reported cases was an artifact of missing data; after correction, the upward trend in late September continued. Case numbers alone were not a reliable indicator of spread because testing access and reporting were incomplete; multiple independent indicators pointed to rapid September growth. The “half a million” long COVID claim depends on assuming 5 million total infections and defining long COVID as symptoms lasting 28 days; both are uncertain. More conservative app data suggest about 1.5% to 2% of test-confirmed cases still had symptoms after three months, implying a substantial burden but not a precise headline number. Survey averages can be misleading when outliers are extreme; median estimates better represent what a typical respondent believes. True exponential growth is compounding over time, not merely any increase that looks steep to the eye.

Data Points: Missing positive test results: 15,841 - England’s test-and-trace data lost over eight days due to a spreadsheet limit Spreadsheet row limit: 65,000 rows - Old Excel file format used by Public Health England could not handle more than this number of rows Reported COVID cases in England: 386 hospital admissions - Used to compare with March lockdown levels March hospital admissions benchmark: 1,100 admissions - Admissions on 23 March in England Lockdown comparison: Less than two doublings - Current admissions were described as less than two doublings from March lockdown levels Long COVID claim: 500,000 people - Headline figure cited in newspapers, derived from app data and infection estimates Symptoms after 28 days: 10% - COVID Symptom Study app estimate used in the half-million calculation Estimated total UK infections: 3 million to 5 million - Range used to infer how many may have long COVID Symptoms after three months: 1.5% to 2% - App-based estimate among people with confirmed COVID tests App participants: Over 4 million - Number of users contributing symptom reports to the COVID Symptom Study app Smaller analytic sample: About 3,500 people - Subset used for the long-COVID three-month estimate Reported death estimate in poll: 7% of UK population - Average guess published from KextCT survey and criticized as misleading Actual UK deaths registered with COVID: About 58,000 - Deaths on death certificates at the time discussed Median public death guess: Around 1% - Bobby Duffy’s explanation of a more representative estimate than the mean Infection fatality rate: About 1% - Used as a benchmark for comparing long COVID prevalence Doubling time in March: Every 4 days - Approximate early-pandemic doubling rate mentioned for UK cases Doubling time in early September: About 1 week - Rate cited during rapid September growth Doubling time at time of broadcast: About 3 weeks - Growth had slowed, though still remained exponential

Pivotal Quotes: "It was a design flaw of using Excel when we know Excel isn't really meant to kind of store large amounts of data or be used as a database." — Christina Pargel: Explaining why the test-and-trace spreadsheet approach failed "The average guess is around 1%." — Bobby Duffy: Describing how the median/public typical estimate differs from the misleading mean of 7% "So it was using the wrong software and using it badly to run this." — Christina Pargel: Critiquing the broader test-and-trace setup and its data-handling failure

Implications: The episode shows how pandemic decisions can be distorted by bad data handling, weak definitions, and poor statistical reporting. Listeners should treat headline figures cautiously and look for methodology, uncertainty, and multiple sources of evidence.