22 September 2021 : ChartJunk of the day, from the NYTimes

I saw the headline in Memeorandum, but skipped the click because I anticipated fearmongering. Never in my wildest dreams could I have anticipated that bar chart on the right. It came my way via Twitter.

What am I calling chart junk? Charts that mislead readers. However, I’ll also critique graphics that are hard to read/interpret. The chart on the right fails both tests.

Here’s what the FBI reports on its website; these numbers are not in the NYT story.

• In 2019, the US had an estimated 16,425 murders, which was a 0.3% increase from the 2018 estimate (Table 1). This was equivalent to 5.0 murders per 100,000. This rate was slightly lower (0.2%) than that in 2018. However, the murder rate was up 1.1% when compared with the 2015 rate and 5.1% compared with the 2010 rate.

• In 2020 … the page does not exist.

According to the NYT:

The F.B.I. data shows almost 5,000 more murders last year than in 2019, for a total of around 21,500 (still below the particularly violent era of the early 1990s).

How bad is this? Let’s count the ways.

1. Normalize data.
   When reporting time-series data (whether population or dollars), it’s important to normalize the data rather than use absolute numbers. Population changes over time; both risk and “rates” are a function of a base number.
   
   The “murder rate” line chart on the left does this. It shows at a glance that the rate per 100,000 for 2020 is equivalent to that of the early 1970s and the late 1990s, and that it has jumped. We are supposed to deduce that straight “up” line is from 2019 to 2020. I think. The headline mentions 2021.
   
   The chart on the right does not normalize data: it is a bar-chart of percentage changes over time. Read that again. It’s chart junk.
   

2. Use the appropriate visualization.
   A bar chart best illustrates comparisons between categories of data — not time-series data, categories of data. In any given year, a bar chart would be an appropriate choice for showing types of crime, for example.
   

3. Differentiate between data and projections.
   In small print on the left-hand chart, we see that the 2020 number is an estimate. That part of the line should be dashed and marked in a different color and/or with an asterisk.
   
   If you read the story, you’ll see the claim that FBI projections for 2020 are available but that they may change next week. There are currently (22 September, 3 pm Pacific) no 2020 projections on the link used in the story. Regardless, projections should be marked as such on the chart.
   

4. Tell the whole story.
   Buried beneath the fearmongering: “Even with the rise in murders and a roughly 5 percent increase in violent crime, the new data shows that overall major crimes fell about 4 to 5 percent in 2020.” Is there a chart showing that major crime fell? You know the answer to that: nope.
   

5. Contextualize risk.
   Humans are lousy at estimating risk.
   
   That bar chart suggests that your risk of being murdered in the US increased by 30 percent last year. Which. Feels. Horrific.
   
   There were 5.0 murders per 100,000 people in 2019.
   
   There were an estimated 6.5 murders per 100,000 people in 2019, per NYT data.
   
   Each of those contextualizations is missing from the story.
   
   What about comparative risk in 2019?
   - Odds of dying in a car accident: 1 in 8,393
   - Odds of an accidental death: 1 in 1,897
   - Odds of being murdered: 1 in 20,000
   - Odds of dying due to complications in medical and surgical care: 1 in 61,595

Source: Murder rose by almost 30% in 2020. It’s rising at a slower rate in 2021. Jeff Asher, NYTimes (22 September 2021). Note: the author is not a NYT staffer: Jeff Asher is a crime analyst based in New Orleans and co-founder of AH Datalytics.

🤓 Recommended reading

• Avoid chartjunk in presentations: Three lessons from Tufte. Ambika Post, Bright Carbon blog (12 May 2016).

• Calling Bullshit: The Art of Skepticism in a Data-Driven World. Carl T. Bergstrom and Jevin D. West (2020).

• Chart junk. Edward Tufte, The Visual Display of Quantitative Information (1983, 2001), pages106-121.

• Choosing a chart type. Berkeley Library (23 August 2021).

• Financial Times Visual Vocabulary. Financial Times (05 October 2018).

• How to choose a type of chart. U.S. Department of Education (n.d.).
  

🔐 This is an example of posts that will be available to paid subscribers.

👓 See COVID-19 resource collection at WiredPen / See all newsletters
👓 See primary news sources

👏🏻 Thank you for your time! Please spread the word about the newsletter.