Statistical Golden Rule

Guest blog by Bruce Ratner, PhD, Dec 5, 2013.

The statistician's utterance "statistical analysis and modeling involves art and science" implies a mixture of a skill acquired by experience (art) and a technique that reflects a precise application of fact or principle (science). The purpose of this cursory article is to bring forth a method to determine the quality of blend of art and science expended by the statistician on a given project.

I propose the statistical golden rule - an application of the well-known golden ratio (circa 490 - 430 BC) in the context of statistical practice. The proposed rule promises to service statistician as the golden ratio has guided artists and architects in the past and present.

"Many artists and architects have p roportioned their works to approximate the golden ratio ... believing this proportion to be aesthetically pleasing."

The end - product of a statistics project, say, an estimated logistic regression model - performed by a statistician who uses a "good" blend of art and science - is statistically complete in form (compact equation) and fitness (accurate equation).

The Statistical Golden Rule

The statistical golden rule (SGR) is the average of the two golden ratios expressions, in which the quantities a and b are, say, science units (e.g., measured in talent, time, mental strength, etc.) and art units (corresponding to the science units) employed during a statistical undertaking. The assignment of the units can be self-reported by the engaged statistician or can be allocated by an independent overseer of the statistician.

Note: the quantities a and b do not have to correspond to science and art, respectively; they can just as well be matched up with art and science, respectively. The interchangeability property of SGR renders it a symmetric relationship.

Read more.

Bruce Ratner an experienced statistician,
a founder of DM Stat-1 consulting,
the author of the best-selling book

"Statistical and Machine-Learning Data Mining: Techniques for Better Predictive Modeling
and Analysis of Big Data"

and a frequent speaker at industry events.

He can be reached at br@dmstat1.com.