See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/332533000

Some Biases in Likert Scaling Usage and its Correction

Article  in  International Journal of Sciences: Basic and Applied Research (IJSBAR) · April 2019

CITATIONS READS

88 67,428

1 author:

Jonald L. Pimentel
University of Southern Mindanao
20 PUBLICATIONS   306 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

collaboration project View project

personal project View project

All content following this page was uploaded by Jonald L. Pimentel on 22 April 2019.

The user has requested enhancement of the downloaded file.

 International Journal of Sciences:
Basic and Applied Research
(IJSBAR)
ISSN 2307-4531
(Print & Online)

http://gssrr.org/index.php?journal=JournalOfBasicAndApplied
---------------------------------------------------------------------------------------------------------------------------

Some Biases in Likert Scaling Usage and its Correction

Jonald L. Pimentel*

University of Southern Mindanao, Kabacan, Cotabato 9407 Philippines

Email: jonaldpimentel24@gmail.com

Abstract

In this paper, common biases or errors in the construction of intervals under the Likert scaling methodology in
both odd and even scales are shown. Examples of researches that uses this scales especially in survey that
requires perception response from the respondents will be presented, discussed and analyzed for possible biases.
Suggestions for the corrections of these biases are presented in order to minimize the bias leading for a better
labeling and interpretations of the results.

Keywords: Bias; difference; even; Likert scale; odd; scoring.

1. Introduction

Likert scaling is a psychometric concept commonly used in survey research that uses guide questionnaires that
need responses in scales from the subject or respondents of interest. It is a widely used scale in many field of
discipline most particularly in the social science research like in education and psychology. It is a scaling
method in which a statement can be responded positively or negatively. This scaling was invented by a
psychologist named Rensis Likert in 1932, whose purpose is to look for an effective and efficient means of
describing attitudes of humans and its influences that affected them. Likert developed that scale that carried his
name until this present time. This scale is also considered a rating scale. As an example, if a respondent wanted
to respond a Likert questionnaire item, this respondent specifies a level of agreement or an order/ranking to a
particular statement which can be describe in a description which in turn has an equivalent numerical values.
Likert scale has many uses in psychology like on the study of behavior, marketing surveys as to the preferences
of a product, education, medicine, nursing, finance, engineering and human study and the likes [1, 4, 8, 15].

------------------------------------------------------------------------
* Corresponding author.

183
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

Further, it is used either as a total score that is when the scale is sum up for all items and in this case the result
produced an index, this is used most commonly in psychology researches or a single ordinal score that is
common among survey research like in marketing and business where they need to have an exact rank of a
certain product. If a statement being ask on which a person need to respond of any criteria whether subjective or
objective criteria, that statement is called a Likert item. Generally, Likert scaling is a measure on the level of say
agreement or a disagreement on that particular statement.

Most social science researchers preferred to use response categories that are on odd scale (example three, five,
seven or nine) because they are interested the scenario in the middle response. Further, psychometricians
preferred to used seven and nine response categories [3] for reasons that they can stretch the interval on which
they can describe the respondent’s preference or view. Most common practiced on the other hand used five
response categories while others used seven response categories, however according to reference [5] in his
recent empirical study, the use of five or seven response categories produced slightly mean scores higher
relatively to a possible maximum score and their difference is significant. Others researchers preferred to use the
response categories that are even (example 4,6, 8 or 10), which are termed force-number Likert scale in some
literature for some reasons that they want to escape the in between scenario of response scale like the three
(neutral/undecided) in the five Likert scale. Most common in this case is the scale that have four response
categories. Other data characteristics on the use of Liker scales shows a very little difference among the scale
formats in terms of variation about the mean, skewness or kurtosis.

Among researchers, some controversies and confusion sometimes arises due to some several causes on the use
of Likert scaling. Hence, an elaboration is needed. First is the treatment of the scale whether it is an ordinal scale
or a measurement scale (ratio or interval) [13, 16]. This sometimes need an explanation and perhaps requires an
assumption so the method applied is valid and consistent. In this study we include this an assumption so to avoid
debates and confusion. Second question is the existence of central tendency bias due to the subject’s avoidance
of extreme response choice that is most respondents tend to go to the middle to escape form the outlying
responses. Third is the presence of acquiescence bias that is an agreement with the given statements as
presented, well this to refrain controversy. Fourth question is known as the social desirability bias that is putting
a subjects or their group or establishments in a more favorable way hence the resulting conclusion is
questionable due to this kind of error. The thought of designing a balance keying (an equal number of positive
and negative statements) is always a challenge in the choice of selecting a scale that can minimize those biases
and problems that may appears. Reference [12] also mentioned on the criticism among reviewers on the
application of statistical methods which is not appropriate when dealings with Likert scales. Some of the
criticism are well founded. For instance, violations to the assumptions of the use of some of parametric methods
for example the analysis of variance, regression analysis and correlations which requires the normality
assumptions on the distributions of the data or per see the appropriateness of using parametric statistical
methods for Likert scaling. However, reference [12] argued that many studies since the 1930s consistently show
that parametric studies can handle with respect to violations of these assumptions. Thus, those claims of
inappropriateness are unfounded according to him hence, parametric methods can still be utilized without the
concern of answers being invalidated. In this study, we do not attempt to apply any parametric statistical method
but to do some trial and error method to further improve the method introduced by reference [13] that is to

184
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

enhanced specifically the use of Likert scaling for both even and odd number categories as well as its final
equivalent description and interpretations. In this study, the researcher will illustrate in practice, for example the
format of a typical five- point level Likert item on agreement coded with a numerical notation and a description
legend being labeled. For example, we agree assigning 1 to strongly disagree, 2 to disagree, 3 to neither agree
nor disagree (undecided or neutral), 4 to agree and 5 to strongly agree.

1.1 Limitation of the Study

This study probably is applicable only to those study that uses instruments that needs responses in scales either
in an ordinal or maybe a measurement (interval or ratio) scales. The study assumes that the data in an ordinal
scale can be considered to be in continuous scale that is either in interval or in ratio scale so that labeling a
weighted mean (which is considered continuous) to its corresponding verbal description after it was computed is
valid.

2. Methodology

2.1 Scoring and analysis

If items in a survey questionnaire instrument is responded using a Likert scale, each item maybe in two ways
first it is analyzed separately given the response scale of the respondents or responses is totaled to create a score
for a group of items. This is usually a common practice in some psychology and educational researches (the
score is considered an index). One issue being raised is whether Likert items is an ordinal data or an interval
data. This is always a subject of disagreement among researcher and users of this scale. In most practice
however, they are considered ordinal or interval level data depending on the assumptions being considered. In
this study we consider this as an assumption. Whether individual Likert items be considered as ordinal level
data, many researchers regard such items only as an ordinal data because, in a special case when using only five
levels, one cannot assume that respondents perceive all pairs of adjacent levels as equidistant. On the other
hand, often the wording of response levels clearly considered both directions (symmetry) of its response levels
assuming there is a middle category at. In that situation such item would fall between ordinal and an interval
level measurement since treating the item as just merely ordinal, we would lose information. In the case of a
scale where we consider an equal spacing of its response levels, we can immediately treat this case as an interval
level data which we can argue is more strongly. Now, if we consider Likert responses an ordinal data, This
responses can be displayed in a graph particularly bar charts or can describe in a median or mode, ,used as a
center but not the average, or the variability using the range across quartiles but not the standard deviation since
the average and standard deviation are inappropriate measures for ordinal data as mentioned by reference [6].
Other alternative for the analysis of the data is using a non-parametric tests on which it depends on the
researcher’s objective are the chi-square test (a categorical statistical method usually or count data), Mann–
Whitney test, Wilcoxon signed-rank test, or Kruskal–Wallis test (comparison of mean ranks). If guaranteed by
the Central Limit Theorem that ordinary averages of the Likert scale data behaves or can be assumed as an
approximately normally distributed in which parametric statistical analysis can be performed. In psychological
researches, Answers to several Likert questions are summed up (totaled) with the assumptions that all questions

185
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

use the same Likert scale and the scale is a defendable approximation to an interval scale, for which case it may
be treated as an interval data that measures an unobserved (or latent) variable. Thus, parametric statistical tests
such as the analysis of variance may be applied. Further, one consideration the method can be applied is to
have more than five Likert item questions that are totaled.

2.2 Level of measurement

The five response categories are often believed to represent an interval level of measurement but this can only
be in the case if the intervals between the scale points correspond to empirical observations in a metric sense.
We bear in mind that the consideration of ordinal scales as an interval scales has a lot of controversial issues as
mentioned previously, however in matter of agreement of its principle, it will be used as a basis for obtaining
estimates in an interval level consideration on a continuum by applying item response models in which the data
can be obtained in the form of an ordinal scale that fit the model particularly known as the polytomous Rasch
model [20,21, 22],. The researcher recommends that the data has to be thoroughly checked to fulfill the strict
formal axioms of the model before it is being analyzed.

3. Results and Discussions

Following reference [13], When items are Likert scaled and is assumed to have an interval measurement, often
in practice, the information to all the respondents are summarized in the form of a weighted mean that is the
scale is used as a weight multiplied to the frequency divided by the total frequency as to obtain the weighted
average or mean. After this procedure the resulting weighted mean is being interpreted using an interval where
in return had a corresponding verbal description. Most of these Likert scales are used in giving the level of
agreement, level of frequency, level of importance or the likelihood of appearance in most application. To give
an illustration in its use. Reference [17] in their paper, used four point Likert scales in the assessment of
students’ self-level of motivation in mathematics. Seven survey question items were constructed for their
assessment. From these items, one item was selected for discussion, the item “How much have you liked
mathematics this year” which was answered using the four point Likert Scale with corresponding verbal
description: 1. Not much, 2. A little, 3. some, 4. A lot. Now, some researchers in these kind of surveys, in
practice created an interval of means in order to give interpretations for the weighted mean. Results of most
studies would treat the problem as practiced by some researchers in which we will illustrate below

Table 1: Four point Likert Scale

Likert Scale Interval Verbal Description

1 1.00-1.50 Not much
2 1.51-2.50 A little
3 2.51-3.50 Some
4 3.51-4.00 A lot

186
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

Table 1 shows a table of a four point Likert scale with the constructed interval and its corresponding equivalent
verbal description. This kind of interval is bias in the sense that the difference in the upper interval and lower
interval of the first and last intervals is much lower as compared to the two middle intervals (see table 2 below).
This observation will result items with the computed weighted mean considered more in the middle. This creates
an unbalanced difference as we observed.

Table 2: Four point Likert Scale

Likert Scale Interval Difference Verbal Description

1 1.00-1.50 0.50 Not much
2 1.51-2.50 0.99 A little
3 2.51-3.50 0.99 Some
4 3.51-4.00 0.49 A lot

To correct this problem, we attempt to reduce or eliminate the bias using trial and error, by making the
difference in each interval having a uniform difference. A new interval was constructed as shown in Table 3
having a majority uniformed difference in each interval (except for one interval). This new interval may
improve the description of the weighted mean.

Table 3: Four point Likert scale

Likert Scale Interval Difference Description

1 1.00-1.75 0.75 Not much
2 1.76-2.51 0.75 A little
3 2.52-3.27 0.75 Some
4 3.28-4.00 0.72 A lot

The correction of the bias in the interval makes the difference “uniform” except for one and perhaps might
change the descriptive interpretation of the item. Going back to the study of reference [17], As shown in the
results of the overall weighted mean of the 5 student survey items by which yielded a values of 3.47 for IWB
group and 3.22 for the control group. A description of this weighted mean leads to the interpretation for the
verbal description of “some” for IWB group when using the table 2 but will have different description “a lot”
when using the table 3 in which there is the presence of the correction. For the control group it has the same
“some” interpretation when table 2 and table 3 are both used.

For the next tables which will be presented below, we will show the improvements for some even and odd scale
categories specifically for the three, four (as shown above), five six, and seven response categories. Intervals
with two decimal places are recommended below given their differences and possible description (Note

187
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

description can be change for appropriate description depending on the objective of the study)

Table 4: Three point Likert scale

Likert Scale Interval Difference Description

1 1.00-1.66 0.66 Bad
2 1.67-2.33 0.66 Undecided
3 2.34-3.00 0.66 Good

Table 4 shows that for a three point Likert scale, we can create an interval in which their differences are uniform
which is what we wanted to achieved.

Table 5: Five point Likert scale

Likert Scale Interval Difference Description

1 1.00-1.79 0.79 Never
2 1.80-2.59 0.79 Rare
3 2.60-3.39 0.79 sometimes
4 3.40-4.19 0.79 Often
5 4.20-5.00 0.80 Always

Next is the table 5 above for a five point Likert scale in which an intervals were created with majority of the
differences are similar except for one which is wider and has slighter difference of 0.1 among the rest.

For table 6 below, in a six point Likert scale an intervals were created with majority differences are similar
except for one interval which is wider and has a difference of 0.03 among the rest.

Table 6: Six point Likert Scale

Likert Scale Interval Difference Description

1 1.00-1.82 0.82 Very bad
2 1.83-2.65 0.82 Rather bad
3 2.66-3.48 0.82 Bad
4 3.49-4.31 0.82 Slightly good
5 4.32-5.14 0.82 Good
6 5.15-6.00 0.85 Rather good

188
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

Lastly, table 7, a seven point Likert scale with created intervals produced a majority of similar difference except
the upper interval which is shorter but a very small difference of 0.01 among the rest. These per observation is a
good table.

Table 7: Seven point Likert scale

Likert Scale Interval Difference Description

1 1.00-1.85 0.85 Very bad
2 1.86-2.71 0.85 Rather bad
3 2.72-3.57 0.85 Bad
4 3.58-4.43 0.85 Neither good nor bad
5 4.44-5.29 0.85 Good
6 5.30-6.15 0.85 Rather good
7 6.16-7.00 0.84 Very good

For all the tables created above which shows the intervals, differences and verbal descriptions, These tables can
be used as a guide or a tool for describing the verbal description of a computed weighted mean for presentation
which can perhaps be considered valid and has less error (bias) preventing which is usually common in practice
to some intervals created by some researchers whose intervals usually are concentrated at the middle of the scale
leading to more items whose weighted assigned mostly in the middle verbal description.

4. Conclusions

Using this suggested interval perhaps will appropriately improve proper labeling for the descriptive
interpretation of the computed weighted mean or average for any researches that considered Likert rating scales
in each items for the whole responded questionnaires according to the number of response categories of the
items.

5. Recommendations

The researcher recommends that more study should be done to refine the interval in each table by any statistical
methods so to obtain uniform difference in each point Likert Scale table similar to column difference in table 4.
It is also worth to discover the intervals for 8,9 and 10 point Likert scale.

References

[1]. Anonymous. The Effect of Right or Left Placement of the Positive Response on Likert-Type Scales
Used by Medical Students for Rating Instruction. Education for Health, vo1.1, pp.122, 1998

[2]. E.R..Babbie. The Basics of Social Research. Belmont, CA: Thomson Wadsworth. p. 174 , 2005

189
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

ISBN 0534630367.

[3]. J. Dawes. "Do Data Characteristics Change According to the number of scale points used? An
experiment using 5-point, 7-point and 10-point scales". International Journal of Market Research, vol.
50 (1), pp.61–77, 2008

[4]. S. Grant, T. Aitchison, E. Henderson, J. Christie, S. Zare, J. McMurray and H. Dargie..‘A Comparison
of the Reproducibility and the Sensitivity to Change of Visual Analogue Scales, Borg Scales, and
Likert Scales in Normal Subjects during Submaximal Exercise’ Chest, vol.116, pp 1208-1218, 1999

[5]. J.Jacoby, and M.S. Matell. ‘Three-Point Likert Scales are Good Enough’ Journal of Marketing
Research, vol. 8, pp. 495-501, 1971

[6]. S. Jamieson. Likert scales: how to abuse them Blackwell Publishing Ltd MEDICAL EDUCATION
2004; vol.38, pp.1212–1218, 2004

[7]. G.P. Latham. Work Motivation: History, Theory, Research, And Practice. Thousand Oaks, Calif.: Sage
Publications. p. 15, 2006 ISBN 0761920188.

[8]. K.Lindhorst,, L. Corby, S. Roberts, S. Zeiler. ‘Rural Consumers’ Attitudes towards Nutrition Labeling’
Canadian Journal of Dietetic Practice and Research, vol.68, pp.146-150, 2007

[9]. R. Likert, "A Technique for the Measurement of Attitudes". Archives of Psychology vo.140: pp. 1–55,
1932

[10]. L. S. Myers, G. Anthony, G. Glenn. Applied Multivariate Research: Design and Interpretation. Sage
Publications. p. 20, 2005, ISBN 1412904129.

[11]. N. Mogey, (March 25, 1999). "So You Want to Use a Likert Scale?". Learning Technology
Dissemination Initiative. Heriot-Watt University. Retrieved April 30, 2009.

[12]. G. Norman.Likert scales, level of measurement and the “laws” of statistics. Advances in Health
Sciences Education. vol.15(5), pp. 625-632, 2010,

[13]. J.L..Pimentel. A Note on the Usage of Likert Scaling for Research Data Analysis. USM R&D Journal,
vol. 18, (2), 2010. ISSN 0302-7937.

[14]. W.E. Saris, I. Gallhofer, W.Van der Veld, W. A scientific Method for Questionnaire Design: SQP.
University of Amsterdam, 2003

[15]. M.Seal, M. Patient Advocacy and Advance Care Planning in the Acute Hospital Setting. Australian
Journal of Advanced Nursing, vol.24, pp.29-37, 2007

190
 International Journal of Sciences: Basic and Applied Research (IJSBAR) (2019) Volume 45, No 1, pp 183-191

[16]. G.M.Sullivan, A.R. Artino. Analyzing and Interpreting Data from Likert-Type Scales, Journal of
Graduate Medical Education vol. 5(4), pp.541-542, 2013

[17]. B.Torff, R.,Tirotta. Interactive whiteboards produce small gains in elementary students’ self-reported
motivation in Mathematics. Computers & Education vol. 54, pp. 379-383, 2010 Elsevier Ltd.

[18]. W.M. Trochim, (October 20, 2006). "Likert Scaling". Research Methods Knowledge Base, 2nd
Edition. Retrieved April 30, 2009

[19]. J.S. Uebersax, J. S. (2006). "Likert Scales: Dispelling the Confusion". Retrieved August 17, 2009.

[20]. E. Muraki. Fitting a Polytomous Item Response Model to Likert-Type Data. Applied Psychological
Measurement.vol.14.no.1. pp.59-71,1990

[21] F. Samejima. Estimation of latent ability using a response pattern of graded scores. Psychometrika
Monograph Supplement, no. 17. 1969

[22]. G.N. Masters. A Rasch model for partial credit scoring. Psychometrika, vol. 47, pp.149-174, (1982)

[23]. D.R. Hodge, D. Gillespie, Phrase Completions: An alternative to Likert Scales, Social Work Research,
vol.27 pp. 45-55. .2003

[24]. J.C. Westland. The information content of financial survey response data. Financial Innovation .vol.1.
no.1. 2015

191

View publication stats