Page 446 - Contributed Paper Session (CPS) - Volume 2

CPS1922 Imran K. et al.

Errors of the fourth kind - A click away…key lies
in statistical collaboration
1
2
Imran Khan ; Eric A. Vance
1 Sher-e-Kashmir University of Agricultural Sciences and Technology of Kashmir, Srinagar,
Jammu and Kashmir, India
2 University of Colorado Boulder, USA

Abstract
Statistical software is nowadays widely available to researchers not well-
trained in statistical analysis and therefore is often used recklessly and blindly.
These researchers may commit errors in their research, simply with a click of
the mouse, which we call “Errors of the fourth kind” or “Type IV errors.” Such
errors give the wrong answer to the right research question and have been
observed by the authors while examining research problems in other
disciplines. Errors of the fourth kind can spring from poor statistical know-how
among researchers while using statistical software on their research data and
lead to erroneous results and incorrect conclusions. Statistical collaborators
seek to help researchers to answer their research questions correctly by
applying the right statistical tool to the right problem in the right manner. This
paper highlights six case studies from the literature and the authors’
experience that show how Type IV errors were committed and how they could
have been avoided through improved statistical collaboration.

Keywords
errors; software; collaboration

1. Introduction
Observations are innocent and, we should avoid torturing them. To play
with observations, by using a friendly software in a friendly way can prove
dangerous, without proper knowledge of statistics. Many of the scientists who
use designed experiments in their research have had little or no exposure to
more than elementary statistical methods. The exposure they have had has
been concerned largely with the process of applying statistical techniques in
the analysis of data. Little attention has been paid to the consideration of what
the data can tell the scientist about whether the experiment has met its
objectives or about what the results really mean within the context of the
problem. As a consequence, too often the data are subject to procedures with
which the scientist is familiar, whether they are appropriate are not (Petersen
1985). This is particularly true of pair wise multiple comparison procedures,
which are frequently used in situations where they are entirely inappropriate
(Petersen 1977).

435 | I S I W S C 2 0 1 9