- Statistics that aren't sufficiently complex can be misleading.
- Polls differ from one another because everyone has biases.
- Anecdotes can be easier for people to relate to, but usually aren't more informative than data-driven information.
Mark Twain wrote, "There are three kinds of lies: lies, damned lies, and statistics."
This phrase is often taken to mean that statistics are another form of lying. Not so, says Aaron Fisher, a statistician with his doctorate from the Johns Hopkins Bloomberg School of Public Health Biostatistics department. He writes that when read in context, it is clear Twain isn’t arguing that statistics are just another form of lying. Rather Twain is warning that statistics that aren’t sufficiently complex can be misleading.
Twain’s warning not to accept all statistical analysis as truthful applies equally to all sorts of data analysis. The most glaring example of data reaching the wrong conclusions is the polling results from the last two presidential elections. The projections weren’t lies, but they were misleading because polling is exceedingly difficult to conduct. Polls have to ensure that the sample is representative, a difficult accomplishment in sensitive areas such as politics and sex. Polling also relies upon respondents giving truthful answers. The questions asked must be sufficiently clear and concise so that respondents know what they are answering.
Everyone has biases, so those creating the studies must be as careful as possible to create studies that are as neutral as possible. Complete objectivity is impossible. This is why studies must be replicable and why polls differ from one another.
Science depends upon statistical analysis. That is how we know whether a vaccine is both effective and safe. Even here, when conclusions about efficacy and safety are presented, there is a margin of error. In other words, even the best results come with a caveat: that there is a chance the results are faulty. How certain the researchers are that the drug does what it is supposed to do is expressed statistically.
Statistics are useful for making generalizations. However, there are always exceptions to generalizations. For example, a friend of mine had an accident and was thrown from the car. If she had worn a seatbelt, it is likely that she would have been crushed to death. From this, she has concluded it is safer not to wear a seatbelt than to wear one. In reality, she had beaten the odds, and it would be foolish to take her particular experience and encourage others to do the same.
Anecdotes are stories of particular events, like my friend’s above. It is tempting to accept an anecdote as proof rather than statistics because it's easier to relate to individuals than it is to numbers. Anecdotes draw pictures; statistics present abstractions. While particular stories elicit empathy, they aren’t especially useful in providing guidance.
Scientists, economists, and statisticians shouldn’t have the last word about how to conduct our lives. In the end, data are simply numbers. Numbers by themselves have no meaning. They are important to the extent that they further our goals. A meteorologist may predict a 100 percent chance of rain, but you may choose not to use an umbrella because you enjoy walking in the rain. Still, you wouldn’t want pilots to ignore weather charts and forecasts.
There are times when we want to lead with our hearts; there are trade-offs to be made between competing goals. There is no statistic that teaches us how to lead meaningful or ethical lives. Statistics have an important role to play in modern life and that role needs to be in support of creating a flourishing life for all.