2.2 Sampling Bias - two examples

There are quite a few glaring examples of sampling bias in history. One of them has to do with a rather famous photo:

2.2.1 Dewey Defeats Truman?

Figure 2.1: Dewey Defeats Truman? (1948)

After defeating Thomas Dewey with a comfortable margin of 303 electoral college votes to Dewey’s 189, President Harry Truman holds up a Chicago Daily Tribune stating the exact opposite. While the Truman Library would like to think that this iconic photo is an example of tenacity, perseverance, self-confidence, and success - it’s actually a result of sampling bias.

The reporting error stems from the fact that the newspaper conducted a poll using phone numbers obtained from a list of vehicle registrations. Most people didn’t have phones in 1948, and the people that were being polled had both phones and automobiles. This skewed the sample distribution to wealthy, white, males - which was obviously not sharing similar views with the overall voting population.

2.2.2 98.6?

Everybody knows the following fact about their body…

The average human body temperature is 98.6 degrees F (37 degrees C).

Is it really? To put this issue in the context of our terminology, the average human body temperature is a population parameter. The population here is every human that has ever lived and ever will live (i.e., an unobtainable sample). This average is actually a sample average obtained by a German physician in 1851 - a time believed by many current physicians to be one where many suffered from chronic infectious diseases resulting in inflammation and mild fevers. Current studies are suggesting the average human body temperature is more like one degree lower than previously thought.

Now to be clear, there is a bit of a semantic argument about this last example. Some empiricists do not call this necessarily a sampling bias issue in 1851, because if a large portion of the population did regularly suffer from mild fevers then the sample was an accurate subset of the population at the time. Of course, if one is saying that the 1851 estimate of 98.6 degrees F is a representation of the current population - then that can be regarded as sampling bias.