Monday, October 07, 2013

Fighting our Biases, Empathy Edition

Here is an example of how reporting on social science research can mislead rather than inform. The author tells us about new studies that show rich people are less empathic (i.e., they care less about others) than poor people. While this may be true on average (and the author gives several reasons why this might be so), the article likely inflames biases rather than illuminates reality.

Why? Because the author neglects to tell us: a) how big the difference in empathy between the two groups is, and b) the underlying distribution of empathy within the rich and poor groups.

The almost inevitable result is that the article will spur one group to be more self-righteous and confident and the other group to be more defensive and angry.  

The underlying analyses generally involve complex regressions, but let's simplify a bit so we can imagine two competing realities behind the results of the study cited.

Figure 1 reveals what most readers will take away from a newspaper story like this.  It shows two distribution curves, one of rich people, and one of poor people.

Figure 1: Huge difference; no overlap
In this scenario, not only do rich people have much less empathy (C) on average than do poor people (D), but ALL rich people have less empathy than ALL poor people. The two distribution curves don't even intersect.

Contrast this with Figure 2:

Figure 2: Small difference; large overlap
In this scenario, rich people do have less empathy on average than poor people.  But the difference (B-A) is much smaller.  And there is huge overlap between the two groups.  There are plenty of rich people who are empathic, and plenty of poor people who are not.

Figure 2 more accurately represents the results of most social science research.  When scientists get lucky, they find a characteristic (in this case, wealth) that has a statistically significant impact on another characteristic (in this case empathy).  But statistical significance does not mean that the difference is large. And even more rarely does it mean that the underlying groups have little or no overlap.

[PS:  Apologies for the poor graphing skills.]


Share on Facebook

2 comments:

chewychunks said...

A good example of a community blog that runs counter to this paper's conclusion:

(rich 1 percenters writing essays of why they stand with the 99 percent)

http://chewychunks.wordpress.com/2011/10/22/1-percent-stories-meta-analyzed/

steve said...

Nicely put. I thought the charts help explain it well.