Академический Документы
Профессиональный Документы
Культура Документы
PRESENTED TO:
DR PAUL ODUNDO
December, 2010.
0|P ag e
Table of Contents
Page of content ...................................................................................................................................... 1
1.0 Introduction ..................................................................................................................................... 2
1.1Definition of Measurement Scale……………………………………………………………………………………………………. 2
1|P ag e
1.0 INTRODUCTION
This paper outlines types of measurements, levels of measurement, and measurement errors.
Measurement is a relatively complex and demanding task especially when it involves qualitative
or abstract phenomena. Measurements involve both physical objects as well as abstract concepts.
Measurement can be defined as the process of assigning numbers to objects or observations, the
level of measurement being a function of the rules under which the numbers are assigned.
events, developing asset of mapping rules and applying the mapping rule to the events observed
There exist two main types of measurements and may be defined thus:
Qualitative Measurement: This involves the use of numbers and can further be classified into
vector and scalar measurement respectively depicting magnitude with direction and magnitude
Qualitative Measurement: Is non-numeric and takes the form of descriptive text. Examples
2|P ag e
2.3 DEFINITION OF A MEASUREMENT SCALE
A scale may be defined as the possible range of measured values and constitutes the basis for
tools used in performing actual measurements. In measuring, we therefore devise some form of
scale in the range and then transform or map the properties of objects from the domain onto the
scale, for example we may measure the Degree of significance of an environmental as being high
There are different levels of measurements. These levels differ as to how close they approach the
variable in research because the levels of measurement determine the type of statistical analysis
that can be conducted and therefore, the type of conclusions that can be drawn from the research.
Levels of measurement typically refer to the theory of scale types also referred to as
Measurement scales. A measurement scale can therefore be said to be a mapping rule devised
measurement scale requires the considerations of several factors that influence the reliability,
I. Research objectives
3|P ag e
III. Data properties
The appropriate choice of scale therefore depends on the underlying assumptions about how
numerical symbols correspond to the real-word situations. The following assumptions are usually
made:
iv. The number series has unique origin indicated by zero (absolute zero point).
i. Nominal scales
4|P ag e
3.1.1 Nominal Scale
Nominal scales are the lowest level of measurement. It is simply a system of assigning symbols
to events in order to label them. The numbers assigned to an object is only a symbol. For
instance, we can use number “1” and “2” to represent Male and Female respectively. As a rule,
we should categorize exhaustively ( for example with no case that includes all cases of that type,
) and mutually exclusive (that no case can be classified as belonging to more than one category).
The numbers are just convenient labels for the particular class of events and as such have no
quantitative value.
It is essentially labeling it can only establish whether two observations are alike or different for
example , sorting a desk of cards into piles red cards and black cards . The status of marital status
may be measured by two categories , married and unmarried but these must each be defined so
that all possible observations will fit into one category but not more than one : Legally Married ,
, common law marriage , religious marriage , , civil marriage , living together , never married ,
In all nominal measurements, all observations in one category are alike property and they differ
from the objects in the other (or categories) on that property (example, code, sex) There is no
ordering of the categories ( No category is better or worse or more or less than another .
Nominal scale is the least powerful level of measurement. It does not indicate order or distance
relationship and has no arithmetic origin. It simply describes differences between things by
assigning them to categories. The scale wastes all the information that it may have about varying
degrees of the variable. The main statistics used for Nominal scale are the mode, measures of
qualitative variation and appropriate measures of association. Chi-square test is the most
5|P ag e
common test of statistical significance. For measures of correlation the contingency coefficient
This is a level of measurement that shows the relative importance of variables in order of
Magnitude, Size and Preferences. Ordinal scale emphasizes order, which is expressed in degree
of quality. The typical relations are, “Higher”. “Greater”, “More desired” and so on. In most
cases ordinal scales indicate Rank Order for example in military to distinguish categories of
soldiers .This level of measurement uses symbols to classify observations into categories so that
they are not mutually exclusive and exhaustive ; in addition the categories have some explicit
An interval scale possesses all the characteristics of an ordinal scale with one additional feature;
the distances between the points on this scale are equal. For example the distance between a
This level of measurement is used where particular data and information collected has
quantifiable magnitude such as population size, weight and distance, which are measured against
temperature, time and test score among others. In the interval level, a common and constant unit
of measurement has been established between the categories. For example, the commonly used
measures of temperature are interval level scale. Number may be assigned to the observation
6|P ag e
because relationship between the categories is assured to be the same as the relationship between
numbers in the number system. For example, 74+1= 75. The interval between categories are
equal, but they organize from some arbitrary origin, that is, there is no meaningful zero point on
interval scale
This is the highest level of measurement that entails expressing the number of persons, and other
attributes such as proportions of the total population. It is a scale that possesses an actual, or zero
point. Variables such as weight, time, length and area have natural zero points and are measured
at the ratio level. In many cases internal ratio scale in public policy and administration, budgets
and the number of programmes participants are measured on ratio scales are treated all in terms
of statistical tests that one applied. Variables measured at high level can always be converted to a
lower but not vice versa. For example observations of actual age (ratio scale) can be converted to
categories of older and younger (ordinal scale), but age measured as simply older or younger
properties. Once scores are assigned, they can attribute differences in scores obtained during
repeated observations to two sources. One source is the extent to which the variables exhibit real
differences in the properties being measured. the other source of difference in the scores is the
extent to which the measure itself or the setting in which the measurement takes place , influence
the scores. In this case, the measures reveal the illusory differences. Perfect measures reveal only
real differences between properties. However measurements are seldom perfect and often
7|P ag e
indicate not only real differences but also artifact differences variation produced by the
measuring procedure itself. Differences in measurements scores that are due to anything other
There are several common sources of measurement errors. First, the scores obtained may be
related to an associated attribute, that is, an attribute that the researcher did not intend to
measure. For example, respondents may require a certain level of intelligence and social
awareness to interpret and answer a question measuring moral development. The responses of
individuals to this question will in effect reflect real differences in moral development but also
the effect of differences in intelligence and social awareness. The influence of associated
temporary conditions such as health or mood that may affect a person‟s response to a
questionnaire or a person‟s behavior. Third differences in the setting in which the measure is
used contribute to measurement errors. For example the age race and gender of interviewer
influences the answer of the survey respondents. Fourth, differences in administration of the
measuring instrument (poor lighting, noise, tired interviewers) can lead to measurement errors
.Fifth measurement errors can also result from different people interpreting the measuring
The errors that arise from these sources are either systematic or random errors . Systematic errors
are produced whenever measuring instrument is used and they are constant between cases and
studies. They consistently introduce a measure of invalidity to the findings. Random error , by
contrast affect each usage of the measuring instrument in a different way . The seriousness of the
issue of validity and reliability are issues that prompted the introduction of techniques for
8|P ag e
Any good scientific study should be precise and unambiguous. However, some errors can occur
in the process of measurements. There are four main sources of measurement errors.
1) The respondent can be a source of measurement error. This may occur if the respondent is
reluctant to express strong negative feelings or it is just possible that he may have little
knowledge but may not admit his ignorance on the subject of study. Other respondent related
errors may occur due to fatigue, boredom, anxiety etc. and may limit the ability of respondent to
2) Situation: situational factors may also come in the way of correct measurement. Any condition
that places a strain on interview can have serious effects on the interviewer-respondent rapport.
For example, if some one else is present during the interview, the respondent may feel shy to
3) Measurer: The interview can be a source of error if they distort responses by rewording or re-
ordering the questions. The interviewer‟s behaviour, style, or looks may encourage or discourage
certain replies from the respondents. The sources may relate to incorrect coding, faulty
4) Instruments: Defective measuring instruments may cause measurement errors. For example
when a person uses complex words beyond the comprehension of the respondents, ambiguous
meanings, poor printing, inadequate space for replies, response choice omission and so on.
5) Other problems encountered include a misplaced belief in precision. It is not usually necessary
for example to measure annual income in dollars and cents. Another problem is measures that go
against social convections. It is often easier to ask people to check of categories than to supply
specific information for example with regard to age, income, education e.t.c. It is a trade off
9|P ag e
between gathering higher level (interval or ratio ) data and having a higher questionnaire
6) When the operational definition does not correspond to the conceptual definition it may be easier
to measure the number of students suspended from school than measure the concept of school
violence.
7) The last problem is when the researcher becomes addicted to certain statistics and gathers only
10 | P a g e
4.0 VALIDITY
Validity is concern with the question “I am I measuring what I intent to measure ?‟ The problem
of validity arises because measurement in the social sciences, is, with very few exceptions,
indirect. Under such circumstances, researchers are never completely certain that they are
measuring the variable for which they designed their measurement procedure. For example does
the voter turn out truly measure political development?. If a respondent agrees with the statement
“this world is run by few people in power , and there is not much the little guy can do about it “ .
Is his or her response a genuine indicator of the presence of the variable „alienation‟? To answer
such questions the researcher must provide supportive evidence that a measuring instrument does
We can distinguish three basic kinds of validity each of which is concerned with a different
aspect of measurement situation: content validity, empirical validity and construct validity.
There are two common varieties of content validity, face validity and sampling validity. Face
validity rests wit the investigator‟s subjective evaluation of the validity of measuring instrument.
In practice face validity does not relate to question of whether an instrument measures what the
researcher wishes to measure rather it is concerns the extent to which the researcher believes that
the instrument is appropriate .The main problem with the face validity is that there are no
difficult to precisely repeat evaluation procedure, the researcher has to entirely rely on subjective
judgment.
11 | P a g e
The primary concern of Sampling Validity is whether a given population (i.e. the total set of
cases in the real world) is adequately sampled by the measuring instruments in question. I other
words do, the statements, questions or indicators – the content of the instrument –adequately
represent the property being measured. The underlying assumption of sampling validity is that
every variable has a content population consisting of a large number of items (which can be
expressed as statement questions or indicators) and that highly valid instruments constitute a
representative sample of these items. In practice problems arise with the definition of a content
population, for this a theoretical and not empirical issue. These problems impair the effectiveness
serves an important function it necessitates familiarity with all items of the content population .It
follows that sampling validity is especially useful in exploratory research where investigators
attempts to construct instruments and employ them for the first time . After their initial use of the
Empirical Validity is concerned with the relationship between measuring instrument and
measuring outcome. Scientists assume that if a measuring instrument is valid, there should be a
strong relationship the result produced by applying the instrument and the real relationship
framework in order to determine whether the instrument is tied to the concepts and teoraticqal
assumption they are employing .Lee J. Cronbach an early proponent of construct validity ,
12 | P a g e
observed that “ whenever a tester asks what a score means pychologically or what causes a
person to get a certain test score means pychologically or what causes a person to get a certain
test score, he is asking what concept may be properly be used to interpret the test performance .
Theoratical expectations about the variable being measured lead the invesigator to postulate
various kinds and degrees of relationships between the particular variable and other specified
show that these relationships do infact hold . If the predictaions fails the explanation lies in
atleast one of the three possibilities (1) the instrument does not measure properly (2) the
theoratical frmawork that generated the predictions is flawed (3) the research design failed to test
Campell and Fiske suggested another method of construct validation involving correlation
method matrix techniques . This method is derived from the idea that different methods of
measuring te same property should yield similar results, whereas difffernt properties should yield
different measurement regardless of the measuring instrument . operationally this means that
correlation coeffient among scores for a given property measured by diffferent instrument should
be higher than correlation coefficiency among scores for different properties measured by similar
must make use of both convergent principle –two measures of te same property should correlate
highly with each other even though they represent different methods – and a discriminant
principle – two measures of difffernt properties should not correlate highly with each other even
13 | P a g e
In view of the distintinctions among the three types of validity it is recommended that a
researchers begin to construct a measurement instrument , they first evaluate theories that could
serve as foundation for the instrument ( construction validity ) : next , they define a content
population of items from which a representative sample is to be drawn (content validity ) ;finally
, they assess the predictive (empirical) validity of the instrument by correlating it with an exernal
criterion.
6.0 RELIABILITY
Reliability is of concern to social scientist because the measuring instruments they employ are
rarely completely valid .In many cases , evidence of validity is alsmost entirely lacking , instead
the researcher has to evaluate the measuring instrument with respect to other characteristics and
assume its valaidity . A method frequently used by social scientists for evalautig and instrument
Realiablity refers to the extent to which measuring instruments contain various errors that is
errors that appear inconsistently from observation to observation during one measurement
attempt or that vary eac time a given unit is measured by the same instrument . For example
when you measure the length of a desk at two points in time with the same instrument –say a
ruler – and get slightly different results the instrument contains varaible error. Because
measuring in the social sciences is primarily indirect the numbe rof errors tends to be greater
than when physical variables are measured . Factors such as respondant‟s momentarily
like a pencil breaking while the respondant is filling in a questinnaire ) may cause the
14 | P a g e
components ; a true component and error component . Realiabilty can therefore be defined as as
the ratio of true score varaiance to the total varaiance in the scores as measured ( the varaiance is
Reliability measure varies on a scale of 0 to 1 having the former value when the measurement
- where s is the normal score for an observation, r is the rank for that observation, n is the sample
size and (p) is the pth quantile from the standard normal distribution.
- where s is the normal score for an observation, r is the rank for that observation, n is the sample
size and (p) is the pth quantile from the standard normal distribution.
3. Expected normal order scores (David, 1981; Royston, 1982; Harter, 1961)
15 | P a g e
where s is the normal score for an observation, r is the rank for that observation, n is the sample
size, (p) is the standard normal density for p and (p) is the pth quantile from the standard normal
order scores is not practical for very large samples, n of 2500 is the maximum permitted in
StatsDirect.
Requires the respondent to estimate the magnitude of a quality that an object possesses. Scoring
Dichotomous
Likert scale
Graphic scale
Staple Scale
Requires that the respondents rank order a small number of activities, events or objects on the
Paired comparison
Forced choice
16 | P a g e
Comparative scale
.The most significant concept in this definition is rules. The function of a rule is to tie the
numerical structure and the structure of the variable being measured. If they can establish
isomorphism researchers can perform quantitative analyses with the numerals that stand for
the properties
2. Isomorphism between numerical system and empirical properties enables the researcher to
distinguish between four levels of measurement: nominal, ordinal, interval and ratio. In
general the level of measurement determine which qualitative analyses can be performed on
3. Measurement procedures are highly sensitive to data transformation and measurement error.
Properties that can be measured at a higher level of precision can also be measured at a lower
level, but not vice versa. That is some data can be transformed from ratio level to the nominal
level, but not all data can be transformed from nominal to ratio.
a. Measurement errors refers to the accuracy and consequent consistency of the measuring
instrument itself .The source of the error may lie in the misunderstanding of what is
actually being measured ( e.g intelligence rather than attitudes) or in the measure‟s
noisy testing room) . In any case the error reflects problems of measurement and not real
17 | P a g e
b. The concept of validity and reliability are inseparable from measurement. They underlie
the sources of measurement error .Validity is concerned with the question whether
researchers are measuring what they think they are measuring .traditionally three types of
validity have been distinguished each of which relates to different aspect of measurement
certain measuring instrument, the researcher must look for information geared to each of
three types.
c. Reliability indicates the extent to which measure contains variable errors operationally it
is assumed that any measure consists of true component and a error component, the
proportion of the amount of variation in the true component to the total variation
indicates the measure reliability. Researchers estimate reliability by one or more of the
following methods: test re-test, parallel forms and split half . The notion of
generalizability implies that the main concern of reliability is with the extent to which a
set of measurements is similar to other sets of measurements that might have been drawn
Reference:
Achen , Christopher H, “ Towards Theories of data : The state of political Methodology “.
In political science: The state of Discipline, No.II Rev.ed. Ada Finifier. Washington,DC : American
Political Science Association, 1993.
Allen Mary J. Introduction to measurement Theory. Pacific Grove , Calif : Brooks/Cole 1979
18 | P a g e
Newbury park ,Calif ; Sage, 1982.
Carley , Michael , Social measurement and social indicators . Boston : Allen & Unw 1981
www.simon.cs.vt.ed/measurement scale
www.fao.org/docrep/w324ie,
Zeller , Richard A, and Edward G. Carmines . Measurement in social science. New York:
19 | P a g e