Subject: Re: EPR Approach to Intro Stat: 
             Relationships Between Variables

   Date: July 18, 1996 2248 EDT

   From: "Donald Macnaughton" <>


On July 16, 1996 Brian Griffin wrote

> The emphasis [in the EPR approach] is clearly placed on pre-
> diction and control, yet I think the most important goal of
> research is totally ignored in your discussion--explanation. 

I agree that explanation (sometimes referred to as "understand-
ing" or sometimes linked to the concept of "understanding") is an 
important goal for many *researchers*.

However, I don't think that we should concentrate on the goals of 
*researchers*.  Instead, I think we should concentrate on the 
goals of empirical research *for society*.

We should concentrate on the goals for society because if re-
searchers are working to satisfy societal goals (which here in-
clude commercial goals), the researchers are more likely to get 
funded, and their work is more likely to make a useful social 

Thus Brian's posting leads to the following important question:

   Which of the following goals of empirical research are 
   more important *for society*:  
     - prediction (including control) or 
     - explanation (including understanding)?

Consider a thought experiment:

a. Suppose you are offered a technique that yields substantially 
   more accurate predictions (or controls) in some area of life 
   than are now available.  But suppose the technique provides 
   *absolutely* no explanation or understanding of how the pre-
   dictions work or how the phenomena under study work.  If you 
   can use the predictions to your benefit, you will probably be 
   very interested in using them even though you don't understand 
   how they work.  (Of course, most people would prefer that the 
   prediction ability be *accompanied* by explanations and under-
   standing because the explanations and understanding would make 
   us more comfortable that the predictions are reliable.  But if 
   we can't have the explanations and understanding, so be it--we 
   will still be pleased to take advantage of the predictions.)  
   Thus accurate predictions (or controls) have high societal 
   value, regardless of whether they are accompanied by explana-
   tions or understanding.

b. Suppose you somehow know that certain explanations (or under-
   standing), although correct about the past, contain *abso-
   lutely* no information relevant to the prediction or control 
   of variables in the future.  (Thus the explanations would ap-
   pear to be of no practical use.)  Most practical people (and 
   most funding agencies) will be uninterested in these explana-
   tions and will be unwilling to support efforts to obtain fur-
   ther such explanations.  That is, explanations (or understand-
   ing) with no hope of future prediction or control have low so-
   cietal value.  

Since accurate predictions have high societal value *regardless* 
of whether they are accompanied by explanations, but correct ex-
planations have low societal value *unless* they are accompanied 
by accurate predictions (or hope of predictions), therefore pre-
diction (control) is more important to society than explanation 

Therefore, it is reasonable to emphasize prediction and control 
of the values of variables in the introductory statistics course.  

Approaching from a different point of view, note that for most 
scientists the acid test of whether someone has fully "explained" 
something is whether that person can accurately *predict* the 
values of relevant variables.  This suggests that the essence of 
an explanation lies in its ability to make accurate predictions 
(or enable accurate control).

Finally, it's important to note that explanations and understand-
ing are sometimes very helpful as middle steps in discovering how 
to predict or control the values of variables.  So I'm not saying 
that explanation and understanding are unimportant.  However, in 
view of the arguments above, I maintain that, from the point of 
view of society, the goals of explanation and understanding are 
subordinate to the goals of prediction and control.  

Material about the entity-property-relationship approach to the 
introductory statistics course is available at

Donald B. Macnaughton      MatStat Research Consulting Inc.         Toronto, Canada