Likert Scale: Misuse of Mid-Point Anchor
Abstract
Validity of responses to surveys is affected by how respondents
interpret the items and the anchors used in the rating scale. We show
that under some circumstances, the mid-point anchor of the traditional
5-point Likert scale could be interpreted as a non-substantive response
reflecting a lack of knowledge answer. An effective way to encourage
respondents to read the mid-point anchor as a point representing their
attitude in the middle of the continuum–is to provide a “Don’t
Know” option at the end of the rating scale with a gap to indicate that
it is a non-substantive response option.
Key Words: Likert Scale, Survey, Scale Construction
Likert Scale: Misuse of Mid-Point Anchor
Surveys have long been recognized as an efficient method to gather
opinion or factual information and to assess attitudes for large numbers
of respondents (Tourangeau, Rips, & Rasinski, 2000). Decades ago,
different scaling methods, such as Likert, Thurstone, Guttman, and
Semantic Differential, were developed to measure attitude. According to
Carifio and Perla (2008):“Likert methodology is one of the most
commonly used methodologies in all fields of research, but particularly
so in allied health, medicine and medical education” (p.1151).
In Likert’s (1932) original formulation, the rating scale consisted of
five points with “agree” and “disagree” labels on which the
respondent indicated direction and intensity of agreement with
declarative statements. In 2004, Jamieson echoed the first of the
“seven deadly sins of statistical analysis” presented by Kuzon et al.
(1996) and argued that use of a Likert scale yields data that are
ordinal and hence parametric statistics should not be used to analyze
these data. Since then, strong reactions from researchers have refuted
her “ordinalist” view on Likert scale data analysis. (Pell, 2005;
Carifio & Perla, 2007, 2008; Norman, 2010). [Her response to critics
was published in 2005 in the same journal.]
The counter-argument is, even though the Likert scale format yields
ordinal item responses, the derived scale scores are
interval, and, even if ordinal, as long as there are 4 to 8 items per
scale or subscale and 5 to 7 anchors points are used (Carifio & Perla,
2007), parametric statistics can still be applied because parametric
tests are robust. The debate appears to have been settled in favor of
the “intervalist” position. Norman (2010) declared: “Parametric
statistics can be used with Likert data, with small sample sizes, with
unequal variances, and with non-normal distributions, with no fear of
‘coming to the wrong conclusion.’ These findings are consistent with
empirical literature dating back nearly 80 years. The controversy can
cease (but likely won’t).” No response from the “ordinalists” has
been heard since. In this paper, we put forth another often ignored
problem in the use of the Likert scale.
Despite the controversies about the Likert scale, two issues all
researchers have agreed on are (1) anchors must be chosen so they are
roughly equidistant on the continuum underlying attitude being measured,
and (2) the selection of an anchor must be a substantive response, a
unit of information reflecting a point on the continuum.
Theoretically, every abstract construct exists on an underlying
continuum. Discrete anchor points along the continuum are often provided
for respondents to choose from. Choice of anchors in an attempt to
create an equal interval rating scale has been researched for over three
decades (e.g., Lam & Klockers, 1982). In addition to subjective
judgment of the intensity of various anchors, through a rank ordering
procedure, “agreement,” “evaluation,” and “frequency” anchors have
been scaled. The derived numerical indices show the positions of the
anchors on the underlying continuum to assist researchers in “choosing
equally spaced response categories for summated rating scales”
(Spector, 1976, p.374).
Likert used the five anchors: “strongly agree,” “agree,” “neither
agree nor disagree,” “disagree,” and “strongly disagree.” Since
then, the middle anchor has been variously labeled as “Neutral”
(Armstrong, 1987), “Undecided,” (Armstrong, 1987; Lock, 2001),
“Uncertain” or “not sure,” (Klopfer & Madden, 1980) and “?”
(DuBois & Burns, 1975; Klopfer & Madden,1987).11In a
dissertation read by the first author, “don’t know” was used as the
mid-point anchor. In all of these cases, the distance between the
mid-point anchor and the two adjacent anchors, “agree” and
“disagree,” seems the same. However, it is not clear if all these
middle anchors are interpreted by respondents as an attitude with
intensity at the middle of the agreement-disagreement bipolar continuum
and not used as “don’t know” (DK). Under some circumstances,
the mid-point response to an item could be a non-substantive response
reflecting a lack of knowledge answer. When such an item response is
scored and included in the computation of the scale or sub-scale score,
validity of the aggregated score suffers. We believe that this error has
occurred but is typically undetected in survey research. The focus of
this paper was to review the potential misuse of the middle anchor of
the Likert scale and offer a strategy to minimize this potential misuse.