- Some samples have implicit bias. Watch out for them. (For example, surveys represent only the respondents. Stratified samples are hard to get.)
- "Average" can mean a lot of things. Arithmetic mean (weighted or not), median, or mode? When unspecified, you can assume that the writer put down the number that best advances their agenda.
- Data obtained from small samples can be used to make extravagant claims because they can exhibit erratic characteristics.
- When a statistic is about a range (e.g., IQ test scores), the differences in similar scores are not significant. More important is to compare confidence intervals.
- Graphs can be deceiving. (e.g., line graphs may not start at 0 on the y-axis, or be skewed to make the variance in values more significant.)
- Embelishing graphics to show quantities can be deceiving. (e.g., the area can square the significance--1 vs 1.5^2, or 3-d graphics can cube--1 vs 1.5^3)
- Unrelated claims can accompany honest-to-goodness statistics to insinuate unconfirmed relationships.
- Correlation does not mean causation.
- Some further examples: arbitrary misleading choices, back-of-the-napkin calculations taken as true statistics (Karl Marx), changing bases when comparing proportions.
- When looking at stats, ask these questions:
- Who is making the claim?
- What are the methods used?
- What is not in the stats?
- Are the subjects of the stats consistent? ("Average American" does X N.M hours a week!)
- Does it make sense?
It should be noted that the points that this book makes are still relevant. However, the book is old (originally published in 1954). Some figures (annual income) and references (True Story) are difficult to get, but it's not too bad.
The material is quite repetitive, and very short on technical details to really "lie" with statistics. It's more about how to deceive by making unrelated claims.
- For visual presentation, read Edward Tufte.
- For more technical issues surrounding the practice of statistics, see Statistics Done Wrong by Alex Reinhart.