Running head: THE TREATMENT OF LIKERT DATA

On the Treatment of Likert Data

Tyler Rinker
University at Buffalo
Department of Learning and Instruction
CEP 510: Psychometric Theory in Education

THE TREATMENT OF LIKERT DATA

On the Treatment of Likert Data

Most researchers in the social sciences will come across Likert data derived from Likert
scales at some point in their career. For myself this encounter occurred, while I am yet a doctoral
student, when I was asked by a professor to analyze reading attitudes data derived from a Likert
scale. As an eager student, I readily accepted the challenge, not really understanding the rabbit
hole I was about to jump down. Likert scales have been available for psychometric purposes for
80 years, and few measurement tools are as sorely misunderstood and hotly contested than Rensis
Likerts attitudinal metric known as Likert scales (Edmondson, 2005; Zand and Borsboom, 2009).
This paper will guide the reader through (a) an introduction to Likert scales, including terms associated with Likert scales, (b) the historical evolution of the Likert scale (c) known biases of
Likert scales, (d) the debate over the treatment of ordinal data as interval and (e) a discussion in
how to approach Likert data. Failure to understand and reflect on issues surrounding Likert data
can lead a researcher to make faulty inferences (Zand and Borsboom, 2009). It is important to become aware of Likert scales as a historical method with clear design, intent and underlying theory
that has been thoroughly examined and debated by some of psychometrics greatest minds.
An Introduction to Likert Scales
Likert scales are the creation of Rensis Likert and were first introduced to the field in
1932 in an article entitled, A Technique for the Measurement of Attitudes in Archive of Psychology (Edmondson, 2005; Likert, 1932). The Likert scale was designed to simplify the complexity of the Thurstone scaling technique (Edmondson, 2005, p. 127). Likert constructed his
scale as a means of capturing an infinite number of attitudes in an ordinal scale format (Likert,
1932). Likerts scale presumes the existence of an underlying (or latent or natural) continuous
variable whose value characterizes the respondents attitudes and opinions...[e]ach Likert-type
item provides a discrete approximation of the continuous latent variable. (Clason and Dormody,
1994, pp. 31-32). Likert was quite specific in his use and understanding of the scale that bears
his name yet the Likert scale is often misunderstood and misused (Jamieson, 2004; Likert, 1932).
Before we continue our discussion of Likert scales, it is important to define what is meant by the

THE TREATMENT OF LIKERT DATA

term, examining the associated terminology.

Carifio and Perla (2008) state that a Likert scale is the summation of a combination of
multiple Likert items. A Likert item is a single item in the form Likert proposes but in itself does
not have the properties of the Likert scale and should not be used for analysis, as this is contrary
to Likerts intentions (Carifio and R. Perla, 2007; Carifio and . Perla, 2008). It is common for
researchers to confuse the term Likert scale with a Likert item leading researchers to incorrect
analysis on individual Likert items (Carifio and R. Perla, 2007).
Uebersax (2006) offers a succinct list of descriptive characteristics original to Likerts
(1932) article:
1. The scale contains several items.
2. Response levels are arranged horizontally.
3. Response levels are anchored with consecutive integers.
4. Response levels are also anchored with verbal labels which connote more-orless evenly-spaced gradations.
5. Verbal labels are bivalent and symmetrical about a neutral middle
and
6. In Likerts usage, the scale always measures attitude in terms of level of agreement/disagreement to a target statement (para. 11)
Uebersaxs (2006) list provides a clear framework for defining the constitution of Likert scales
and the items that comprise them. Characteristic five, which requires that the responses are to be
symmetrical with a neutral center, indicates an odd number of choices with the non-neutral responses on either side being equal in magnitude yet opposite in polarity (Likert, 1932; Uebersax,
2006). Some researchers, in an effort to avoid central tendency bias, or the tendency to choose
the neutral response for items the examinee lacks confidence with, stray from the odd number of
response, eliminating the neutral choice (James, Demaree, and Wolf, 1984).
The terms Likert and Likert-type are often used interchangeably and may result in a use
contrary to Likerts explicit framework for the scales design (Likert, 1932). Likerts (1932)
original summed scale was was derived from a five point (or number of discrete points of the
responses) response score of multiple items, though he did not specify this quantity, utilizing the

THE TREATMENT OF LIKERT DATA

following five point responses: strongly approve, approve, undecided, disapprove, and strongly
disapprove (see Figure 1).

How do you feel about treating Likert data as interval?

Strongly
Disapprove

Disaprove

Undecided

Approve

Strongly
Approve

Figure 1: Example Likert Item

It is important to note again that the responses of an individual item do not constitute the scale,
rather the summed scores from the responses. The summation of the items is the Likert scale
where as the term Likert item is generally in reference to the format of individual items following
in the arrangement of Likert (Clason and Dormody, 1994, p. 31). Items, and thus scales, that
deviate mildly from Likerts defined characteristics are termed Likert-type items, though a radical
departure should not be called Likert-type (Likert, 1932; Uebersax, 2006). Often such scales
have more general names (i.e. visual analogue scale) that should be utilized (Flynn, van Schaik,
and van Wersch, 2004; Uebersax, 2006). Though the distinction between Likert and Likert-type
must be respected and reflected in a researchers analysis, in order to avoid repetitive use of the
clause Likert and Likert-type the remainder of this paper will refer to the term Likert to also
subsume the term Likert-type.
A Brief History
Likert (1932) first proposed his scaling technique as a response to the labor intensive
method of Thurstone scaling that required multiple judges to assess values for questions (Edmondson, 2005; Edwards and Kenney, 1946; Likert, 1932). The Thurstone scaling method
was the first widely used attempt to capture latent variables on a continuous scale, however,
the method suffered several drawbacks including the use of multiple judges which was costly
and could potentially lead to judging bias(Edmondson, 2005; Edwards and Kenney, 1946; Likert,

THE TREATMENT OF LIKERT DATA

Roslow, and Murphy, 1993). Likerts approach was to use a summation of carefully crafted, symmetric scored responses (Likert, 1932). Likert (1932) indicated that reliability scores for his new
method were at least as good as Thurstones and in some cases superior. Edwards (2005) cited
that a possible rationale for the increased reliability over Thurstones method may be attributed to
the increased steps in the scale (p. 77). Likert responded to criticism of the reliability analysis
by publishing a similar study in 1934 with similar results, though Edmondson (2005) finds design
faults with this analysis as well (Edmondson, 2005, p. 128). Currently, Likert and Likert-type
scales are used in many fields related to the social sciences and the scales have evolved to display
differing number of points and responses, but the essence of Likerts original scale remains the
same (see Vagias (2006) for a sample selection of these formats).
Biases Associated with Likert Scales
Researchers that produce or consume studies derived from Likert scales should be aware
of potential biases and weaknesses attributed to the scale when constructing or examining a scale
and items. James, Demaree,& Wolf (1984) describe one form of bias, central tendency bias, as
the bias of examinees to choose the neutral response in an odd point scale, termed forced choice,
as a way of avoiding items that they are not comfortable or confident in answering. Some attempts to overcome this bias have included the use of an even point scale, however, researchers
need to be aware that this could alter the distribution of the data in ways that are more likely
to lead to departures from the assumption of normally distributed error terms for linear models
(Likert, 1932). A second known potential bias, acquiescence bias, is a phenomena in which the
examinee tends to give positive responses to the survey questions is sometimes approached by
reversing the polarity of the item (Lichtenstein and Bryan, 1965). Semon (2000) notes that acquiescence bias displayed differently depending on the cultural group of the respondent. This
differences among cultures in responding to an item is referred to as cultural bias (Flaskerud,
2012; Semon, 2000). Chung & and Monroe (2003) describe another form of bias common to Likert scales, social desirability bias, in which [p]eople have a need to appear more altruistic and
society-oriented than they actually are, and social desirability (SD) is the tendency of individuals

THE TREATMENT OF LIKERT DATA

to deny socially undesirable actions and behaviors and to admit to socially desirable ones (p.
291). Researchers must be conscious of different biases that may affect the inferences that can be
made from a studys findings.
The Problem of Likert Data
When I began my initial review of the literature around Likert scales it became abundantly clear that the scales use was mired in hefty debate among psychometricians since its
inception in 1932. The battle in the literature is over the treatment of an ordinal scale as interval.
This argument affects the choices a researcher makes in analyzing the data and interpreting the
results (de Winter and Dodou, 2010). The conservatives consider it a sin to use parametric
tests to analyze what they consider to be strictly ordinal data (Knapp, 1990). The liberals contend that the robustness of the F ratio allows for Likert data to be treated as interval (Carifio and
R. Perla, 2007; Knapp, 1990). The stance you take will likely affect how you conduct and interpret research and more importantly have to potential to affect type I and II error rates (Anderson,
1961). This sentiment is captured by Jamieson (2004):
The legitimacy of assuming an interval scale for Likert-type categories is an important issue, because the appropriate descriptive and inferential statistics differ for ordinal and interval variables and if the wrong statistical technique is used, the researcher
increases the chance of coming to the wrong conclusion about the significance (or
otherwise) of his research. (p. 127)
I do not make a determination as to precisely how a researcher should approach Likert data but
instead present the two major viewpoints and the resources to make an informed decision regarding a particular data set and research hypothesis.
S. S. Stevens: The Beginning of a Firestorm
Much of the literature traces the beginning of the ordinal-interval debate back to Stevens
(1946) piece, On the Theory of Scales of Measurement, released 14 years after Likerts article
(1932). Stevens (1946) wrote the article in an attempt to provide some guidance and unity in

THE TREATMENT OF LIKERT DATA

measurement, particularly that of human perception. Stevens (1946) first defines measurement,
measurement in the broadest sense, is defined as the assignment of numerals to objects according to rules and then proposes classifying data into one of four scales of measurement: nominal,
ordinal, interval and ratio (pp. 677-678). He defines the nominal scale as data belonging to
different categories with no clear order or zero point (pp. 678-679). It was here that using football numbers as an arbitrary assignment to nominal data was first discussed. The ordinal scale
shares all the same properties as the nominal scale, but the categories could be ordered according
to some rank. This is particulary relevant to Likert data as many, particulary the conservative psychometricians, would typically classify this scale as being ordinal (Edmondson, 2005; Jamieson,
2004). Here Stevens (1946) warns:
In the strictest propriety ordinary statistics involving means and standard deviations
ought not to be used with these scales, for these statistics imply a knowledge of
something more than the rank-order of data...On the other hand, for this illegal
statisticizing there can be invoked a kind of pragmatic sanctification: In numerous
instances it leads to fruitful results. (679)
Stevens (1946) third scale type is interval, which contains order, as the ordinal data, but the
spacing between the ranked categories is equidistant. This scale classification is the first type
mentioned that he considers quantitative, though, like the ordinal and nominal scales, lacking a true zero point (p. 679). It is between the ordinal and interval levels that Stevens (1946)
acknowledgs some ambiguity of such terms as intensive and extensive. Both ordinal
and interval scales have at times been called intensive, and both interval and ratio scales have
sometimes been labeled extensive (p. 678). The final scale of measurement, ratio, are possible
only when there exists operations for determining all four relations: equality, rank-order, quality
of intervals and equality of ratios (Stevens, 1946, p. 679).
Though Stevens (1946) article is an attempt to unify the field, it had the opposite effect.
Seven years after Stevens piece, Lord (1953) wrote a parable in an attempt to produce a logical
counter argument to Stevens proposed scales and their application; this is considered the second

THE TREATMENT OF LIKERT DATA

blow in the ensuing 80 year debate.

Frederic M. Lord: The Debate
Lord (1953), not wholly satisfied with Stevens (1946) scale classification, used a football
numbers story as a logical contradiction, based on Stevens own mention of football numbers as
nominal data, to dispute Stevens claim that parametric statistics can not be used with data of the
nominal scale. It should be noted that Lords story does not mention Likert data, nor is it about
treating ordinal data as interval, but is very relevant (and often sited) in that it goes further than
using parametric tests for ordinal data, suggesting that such statistics can be applied to nominal
data (Lord, 1953).
Lord (1953) essentially argues that a test designed for interval data can, in some instances,
be applied to nominal data. He develops a story of a professor gone mad because of a love of
calculating means and standard deviations of students test scores, driven to insanity by his
own quantitative hypocrisy (p. 750). In retirement the professor sells football numbers, a supposed arbitrary numeric assignment to nominal data, and is faced with accusations of selling a
disproportionate amount of low numbers to the freshman class (p. 751). Without a method of
testing the charge, the professor enlists the help of the campuss statistician, who promptly employees parametric measures to the nominal data (p. 751). The professor protests, But you cant
multiply football numbers, the professor wailed. Why, they arent even ordinal numbers,
like test scores., to which the statistician retorts, The numbers dont know that (p. 751).
The statistician defies the professor to disprove his applying parametrics to nominal data and the
professor promptly sets out to do so through random sampling of his football numbers. After
repeated samplings the professor is convinced that you can apply means and standard deviations
to nominal data.
Following Lords (1953) football numbers story a debate has raged over the appropriateness of treating nominal or even ordinal data with tests designed for interval and ratio data.
Behan & Behan (1954) quickly retort, But, let us note that when we are all finished, we know
something about the number signs, not something about the football players (p. 262). Bennett

THE TREATMENT OF LIKERT DATA

(1954) humorously replied to Lord:

So it is with Lords parable. The freshman-sophomore argument settled by the statistician was one of cardinal highness or lowness in a set of numbers used in an entirely
different context to identify football players. Our Professor X had best re-retire; his
helpful statistical friend had best return to his TV set. I at least shall continue to lock
my door when computing the means and standard deviations of test scores. (p. 263)
Lord (1953) addressed Bennetts (1954) concerns with a more relaxed stance:
It would be unfortunate if what has been written here were to lead anyone to ignore
the very serious pitfalls actually present. Let me hasten to agree with Dr. Bennett
that incorrect or meaningless conclusions can easily be reached...The conclusion
to be drawn is that the utmost care must be exercised in interpreting the results of
arithmetic operations upon nominal and ordinal numbers; nevertheless, in certain
cases such results are capable of being rigorously and usefully interpreted, at least for
the purpose of testing a null hypothesis. (p. 265)
This indicates his true intention, to warn against arbitrarily applying a statistical test without
considering the data and measurement, to caution against apply rules rather than reason. Unfortunately, this article is cited much less frequently and the message is lost in the debate between the
conservatives and liberals (Zand and Borsboom, 2009). Though Likert scales werent explicitly
mentioned in either Stevens (1946) or Lord (1953) the Likert battle of ordinal-interval uses both
Lord and Stevens as spring boards for arguments of their position.
Current Views
The debate between ordinal and interval treatment of Likert data is still raging (Zand
and Borsboom, 2009). Often the debate is in the logical realm, in theory, rather than practice
(Lord, 1954). Lord (1954) gives permission to stop the debate by testing the charge, It seems
very fortunate that any fundamental disagreement here between critic and statistician need not
long remain a matter of opinion since the question is so readily submitted to wholly objective,

THE TREATMENT OF LIKERT DATA

10

practical verification (p. 265). Knapp (1990) provides a great deal of insight into testing Lords
challenge, summarizing many of the arguments held by both sides. It becomes clear that the F
test is quite robust, that one can usually tease normality and homogeneity- of variance quite
a bit without doing serious injustice to t or F, particularly with equal sample (Knapp, 1990, p.
122). Knapp (1990) also discusses the possibility of the quantity of break points for the scale
effecting the distribution of data, with more points tending to continuize (p. 123). Perhaps the
most striking claim Knapp (1990) makes is regarding power, one of the underlying rationale for
preferring parametric tests:
But both camps are mistaken regarding... the, alleged power superiority of parametric tests over non parametric tests. The wilcoxon tests for independent samples and
for paired samples are never much less powerful than t, and when the population
distribution is not normal (for ordinal or interval measurement) they can be much
more powerful (Blair & Higgins, 1980; 1985)...If you claim that-you have an interval scale, you are more likely to prefer parametric techniques, but should you have
qualms about normality and/or homogeneity of variance and elect some nonparametric counterpart, dont be apprehensive about losing power; it maybe even-higher. (pp.
122-123)
Denny Borsboom, a respected leader in the field of psychometrics, co-authored a piece
that takes a radical and sensible approach to the Likert scale debate. Zand & Borsboom (2009)
attack Lords (1954) argument in a different way than previous critiques. Rather than confront
the theory and logic they discredit the logical contradiction of Lords football number selling
professor. Lords (1954) argument rests on the fact that nominal data with numeric representation
can be treated as parametric. Zand & Borsboom (2009) show that the numbers in Lords story are
not serving a nominal role but are a representation of the bias of the vending machines theyre being distributed from (p. 72). At this point Lords contradiction is debunked and no longer serves
as an argument for treating Likert data as interval. Zand & Borsboom (2009) are not taking the
position that Likert data should be consistently treated as non parametric; instead, they contend,

THE TREATMENT OF LIKERT DATA

11

as Lord did, that Stevens rules should not be applied mindlessly (p. 74). Zand & Borsboom
(2009) further the point, The numbers dont have to know where they came from; researchers
have to know where they came from, since they assigned them in the first place. (p. 74). The
argument is transformed from one of ordinal versus interval to that of sound measurement
methods, reflective research practices and consideration of the inferences made from statistical
tests.
The Treatment of Likert data
Zand & Borsboom (2009) discredited Lords logical contradiction and called for attentive research practices, therefore, it is necessary to understand the direction a researcher must
approach after making decisions regarding measurement, scale, analysis and inferences. It is important to realize, as stated by Carifio & Perla(2007), that the F test is actually quite robust to use
of Likert data, even skewed data:
The non-parametric statistical analyses only myth about Likert scales is particularly disturbing because many (if not all) item fixated experts seem to be completely unaware of Gene Glass famous Monte Carlo study of ANOVA in which
Glass showed that the F-test was incredibly robust to violations of the interval data
assumption (as well as moderate skewing) and could be used to do statistical tests at
the scale and subscale (4 to 8 items but preferably closer to 8) level of the data that
was collected using a 5 to 7 point Likert response format with no resulting bias. (p.
110)
Carifio & Perla(2007, 2008) also make it clear that this robustness only holds true when Likert
data is analyzed as a scale, that is a summed composite score, not individual items. Anderson
(1961) and Knapp (1990) warn that interactions may be affected more harshly than the main
effects and need to be considered and analyzed carefully in using Likert data.
The consideration of power is the major concern for researchers in choosing parametric
vs. nonparametric (Knapp, 1990). Knapp (1990) and Anderson (1961, 2004) both indicate that

THE TREATMENT OF LIKERT DATA

12

under equinormality, power for both the parametric and nonparametric are close, and may actually be greater for the nonparametric under violations of the normality assumption. Zand &
Borsboom (2009) also discuss the tendency to use parametric tests because of their availability
and ease of use. As statistical computer programs, such as R (2012), become reflective of psychometric theory they have come to incorporate procedures for handling non parametric data
with ease. It is absurd to allow ease of use to affect scientific inquiry that may lead to policy
changes. Osterlind (2010) makes mention of more sophisticated graded response IRT models
suited for Likert data that should become another tool in the researchers tool box (p. 298). It is
incumbent upon the researcher to become familiar with appropriate techniques or to hire a professional statistician so that measurement levels...guide the choice of statistical test (Zand and
Borsboom, 2009, p. 69).
Conclusion
It was the intention of this paper, not to give a ready to follow road map for analyzing
Likert data, but to provide the insight and knowledge necessary for researchers to properly approach measurement, analysis and interpretation with a more cautious and informed perspective.
When I first jumped down the rabbit hole of Likert data I was amazed at the misconceptions and
ignorance I had. This paper is by no means comprehensive, but does provide the major considerations the reader should be aware of and can serve as a guide in exploration of this deeply seeded
and inherently important topic that has been a source of debate and contention for generations
of physchometricians. Zand & Borsboom (2009)offer great insight for the probing researcher
attempting to grasp measurement with Likert scales:
Research findings and conclusions depend on arbitrary, and usually implicit, scaling decisions on part of the researcher. This hinders scientific progress because it
obscures a factor, namely the choice of scaling, that is influential in determining
conclusions based on empirical research. It is important, therefore, to have a clear
understanding of how level of measurement can affect our conclusions. (p. 69)

THE TREATMENT OF LIKERT DATA

13

It is our duty to scrutinize our own data and the research of others as we attempt to build our
collective understanding of various issues in social sciences. The Likert scale is a tool that may
be useful but must be used with sensible understanding of the scale, its intended use, potential
weaknesses, analysis approaches, interpretations of the results and of inferences gathered.

THE TREATMENT OF LIKERT DATA

14

References
Anderson, N. H. (1961). Scales and statistics: Parametric and nonparametric. Psychological
Bulletin, 58(4), 305316. doi:10.1037/h0042576
Behan, F. L., & Behan, R. A. (1954). Football numbers (continued). American Psychologist, 9(6),
262 263. doi:10.1037/h0053500
Bennett, E. M. (1954). On the statistical mistreatment of index numbers. American Psychologist,
9(6), 264. doi:10.1037/h0059284
Bertram, D. (n.d.). Likert scales ...are the meaning of life.
Carifio, J., & Perla, . (2008). Resolving the 50-year debate around using and misusing Likert
scales. Medical Education, 42(12), 1150fffd1152. doi:10.1111/j.1365-2923.2008.03172.x
Carifio, J., & Perla, R. (2007). Ten common misunderstandings, misconceptions, persistent
myths and urban legends about Likert scales and Likert response formats and their antidotes. Journal of Social Sciences, 3(3), 106116. doi:10.3844/jssp.2007.106.116
Chung, J., & Monroe, G. S. (2003). Exploring social desirability bias. Journal of Business Ethics,
44(4), pp. 291302. Retrieved from http://www.jstor.org/stable/25075038
Clason, D. L., & Dormody, T. J. (1994). Analyzing data measured by individual likert-type items.
Journal of Agricultural Education, 35(4), 3135. doi:10.5032/jae.1994.04031
de Winter, J. C. F., & Dodou, D. (2010). Five-point Likert items: t test versus Mann-WhitneyWilcoxon. Practical Assessment, Research & Evaluation, 15(11), 112. Retrieved from
http://pareonline.net/pdf/v15n11.pdf
Edmondson, D. R. (2005). Likert scales: A history. CHARM, 12, 127133. Retrieved from https:
//www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0CCkQFjAA&
url=http%3A%2F%2Ffaculty.quinnipiac.edu%2Fcharm%2FCHARM%2520proceedings%
2FCHARM%2520article%2520archive%2520pdf%2520format%2FVolume%252012%
25202005%2F127%2520edmondson.pdf&ei=8AqKT4WaD4W40QGN1K3mCQ&usg=
AFQjCNHITtvhd9XFkFBUWVWseN7yuOmZPA&sig2=ERTm7gL9B8yUit44t84kAA

THE TREATMENT OF LIKERT DATA

15

Edwards, A. L., & Kenney, K. C. (1946). A comparison of the Thurstone and Likert techniques
of attitude scale construction. Journal of Applied Psychology, 30(1), 72 83. doi:10.1037/
h0062418
Flaskerud, J. H. (2012). Cultural bias and Likert-type scales revisited. Issues in Mental Health
Nursing, 33(2), 130132. doi:10.3109/01612840.2011.600510
Flynn, D., van Schaik, P., & van Wersch, A. (2004). A comparison of multi-item likert and visual
analogue scales for the assessment of transactionally defined coping function. European
Journal of Psychological Assessment, 20(1), 49 58. doi:10.1027/1015-5759.20.1.49
James, L. R., Demaree, R., & Wolf, G. (1984). Estimating within-group interrater reliability
with and without response bias. Journal of Applied Psychology, 69(1), 8598. Retrieved
from http : / / people . sabanciuniv . edu / gokaygursoy / ISTATISTIK OLD / BOLUM
CALISMALARI/ILL/2007/OnlineSaglananDokumanlar/12192.pdf
Jamieson, S. (2004). Likert scales: How to (ab)use them. Medical Education, 38, 12121218.
doi:10.1111/j.1365-2929.2004.02012.x
Knapp, T. (1990). Treating ordinal scales as interval scales: An attempt to resolve the controversy. Nursing Research, 39(2), 121123. Retrieved from http://www.mat.ufrgs.br/viali/
estatistica/mat2282/material/textos/treating ordinal scales[1].pdf
Lichtenstein, E., & Bryan, J. H. (1965). Acquiescence and the mmpi: an item reversal approach.
Journal of Abnormal Psychology, 70(4), 290 293. doi:10.1037/h0022412
Likert, R. (1932). A technique for the measurement of attitudes. Archives of Psychology, 22(140),
155.
Likert, R., Roslow, S., & Murphy, G. (1993). A simple and reliable method of scoring the Thurstone attitude scales. Personnel Psychology, 46(3), 689 690. doi:10.1111/j.1744-6570.
1993.tb00893.x
Lord, F. M. (1953). On the statistical treatment of football numbers. American Psychologist,
8(12), 750751. doi:10.1037/h0063675

THE TREATMENT OF LIKERT DATA

16

Lord, F. M. (1954). Further comment on football numbers. American Psychologist, 9(6), 264
265. doi:10.1037/h0059284
Osterlind, S. J. (2010). Modern measurement: Theory, principles, and applications of mental
appraisal (2nd ed.). Boston: Pearson Education.
R Development Core Team. (2012). R: A language and environment for statistical computing.
ISBN 3-900051-07-0. R Foundation for Statistical Computing. Vienna, Austria. Retrieved
from http://www.R-project.org/
Semon, T. T. (2000). No easy answers to acquiescence bias. Marketing News, 34(3), 7. Retrieved
from http://search.ebscohost.com.gate.lib.buffalo.edu/login.aspx?direct=true&db=bth&
AN=3568208&site=ehost-live&scope=site
Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103(2684), 677680.
doi:10.1126/science.103.2684.677
Uebersax, J. S. (2006). Likert scales: Dispelling the confusion. Retrieved from http : / / john uebersax.com/stat/likert.htm
Vagias, W. M. (2006). Likert-type scale response anchors. Clemson International Institute for
Tourism & Research Development, Department of Parks, Recreation and Tourism Management. Clemson University. Retrieved from http://www.clemson.edu/centers- institutes/
tourism/documents/sample-scales.pdf
Zand, A. S., & Borsboom, D. (2009). A reanalysis of Lords statistical treatment of football
numbers. Journal of Mathematical Psychology, 53(2), 6975. doi:10.1016/j.jmp.2009.01.
002