The most popular stats book in the 2nd half of 20th century was "How to Lie with Statistics".
The #COVID19 portion of Trump's @axios interview should be titled "How NOT to Lie with Statistics.
Here are 5 concrete takeaways for #DataScience. No cheap Trump-bashing.
(1/N)
The #COVID19 portion of Trump's @axios interview should be titled "How NOT to Lie with Statistics.
Here are 5 concrete takeaways for #DataScience. No cheap Trump-bashing.
(1/N)
1. Nice figures aren't sufficient to make a good argument.
Trump erred in bringing plots he couldn’t explain well. Data can't speak for itself. Visualizations can’t speak for the data. Visualizations only help humans speak for the data.
2/N
Trump erred in bringing plots he couldn’t explain well. Data can't speak for itself. Visualizations can’t speak for the data. Visualizations only help humans speak for the data.
2/N
2. Presentation matters.
The interview has cringe-tinged moments of an agitated POTUS shuffling through pieces of paper as camera struggles to get angle and Swan looks on, bemused.
Plastic sleeves would've fixed that. As a born showman, he should've anticipated this.
3/N
The interview has cringe-tinged moments of an agitated POTUS shuffling through pieces of paper as camera struggles to get angle and Swan looks on, bemused.
Plastic sleeves would've fixed that. As a born showman, he should've anticipated this.
3/N
3. Anticipate statistical counterarguments
Trump's focus on stats that look good for him is understandable. Getting caught off-guard by @jonathanvswan's focus on stats that make him look bad is not.
Especially when those stats hit hard. "I'm talking about death!" said Swan.
4/N
Trump's focus on stats that look good for him is understandable. Getting caught off-guard by @jonathanvswan's focus on stats that make him look bad is not.
Especially when those stats hit hard. "I'm talking about death!" said Swan.
4/N
My wife compared this to having a company all-hands meeting after a round of surprise layoffs. The CEO starts talking about how great things are, not mentioning layoffs. Someone asks, and the CEO is clearly unprepared to answer. That would be weird.
5/N
5/N
4. Causal inferences have to make sense
Trump focused on the death rate among test-positive cases in
, saying it's better than
and
.
The implicit causal inference is his admin's policies are better at saving lives of the sick than those countries.
Implausible.
6/N
Trump focused on the death rate among test-positive cases in



The implicit causal inference is his admin's policies are better at saving lives of the sick than those countries.
Implausible.
6/N
5. Inferences should be consistent
POTUS pumps the lower mortality rate among test- positives.
He says it's because of more widespread testing.
But he also calls for less testing (b/c of absolute death stats?)
The stories you tell about the data should not conflict.
7/N
POTUS pumps the lower mortality rate among test- positives.
He says it's because of more widespread testing.
But he also calls for less testing (b/c of absolute death stats?)
The stories you tell about the data should not conflict.
7/N
Write-up: https://altdeep.substack.com/p/trump-shows-how-not-to-lie-with-statistics
Curious about #DataScience #Statistics and #epitwitter thoughts.
Let me know if you think I missed something, or forward this thread to a friend if you think they'd have points/counterpoints.
Curious about #DataScience #Statistics and #epitwitter thoughts.
Let me know if you think I missed something, or forward this thread to a friend if you think they'd have points/counterpoints.
Links to video at various time-stamps in the post.