The concept of measurement and attitude scales

Measurement is the process of assigning numbers
or labels to persons, objects or events in
accordance with specific rules for representing
quantities or qualities of attributes.
Rule is a guide, method or command that tells a
researcher what to do.
Scale is a set of symbol or numbers so constructed
that the symbols or numbers can be assigned by a
rule to the individuals to whom the scale is applied.
Nominal scale is a discrete classification of data, in
which data are neither measured nor ordered but
subjects are merely allocated to distinct categories
(male/female, buyer/non-buyer).
Ordinal scale is a scale on which data is shown
simply in order of magnitude since there is no
standard of measurement of differences.
Ordinal numbers are used strictly to indicate rank
Interval scale is a scale of measurement of data
according to which the differences between values
can be quantified in absolute but not relative terms
and for which any zero is merely arbitrary.
Ex : Temperature in Celsius or Fahrenheit.
Ratio scale is a scale of measurement of data which
permits the comparison of differences of values; a
scale having a fixed zero value.
Ex : The distances travelled by a projectile.
Measurement = Accurate data + Error
Random error ?
Special error ?
Personality, value, intelligence
Temporary mood, fatigue
Variations in the interviewer
Sampling of items
Lack of clarity
Mechanical error
The reliability is the degree to which measures are
free from random error and, therefore, provide
consistent data.
The less error there is, the more reliable the
observation is.
The key question about reliability is : “If we
measure some phenomenon over and over again
with the same measurement device, will we get the
same or highly similar results ?
An affirmative answer means that the device is
Test-retest reliability is obtained by repeating the
measurement with the same instrument,
approximating the original conditions as closely as
Equivalent form reliability is the ability of two very
similar forms of an instrument to produce closely
correlated results.
Internal consistency reliability is the ability of an
instrument to produce similar results when used on
different samples during the same time period to
measure phenomenon.
Validity is the degree to which what the researcher
was trying to measure was actually measured.
When Pontiac brought out the Aztec, research told
them that the car would sell between 50,000 and
70,000 units annually despite the controversial
After selling only 27,000 cars per year, the model
was discontinued in 2005.
The research measuring instrument was not valid.
Face validity is the weakest form of validity. A test
can be said to have face validity if it "looks like" it
is going to measure what it is supposed to
Content validity is the representativeness of the
content of the measurement instrument. In other
word, an interview which is not speaking about the
main subject or main objective has a lack of
content validity.
Measurement of attitudes relies on less precise
scales than those found in the physical sciences.
Indeed the mind of the consumer is not directly
observable unlike the weight.
Unidimensional scales measure only one attribute
of a concept, respondent or object.
Multidimensional scales measure several dimension
of a concept, respondent or object.
Graphic rating scale is a measurement scale that
include a graphic continuum, anchored by 2
There are simply to use and considered as interval
But one disadvantage of this scale is that the
extreme anchors tend to force the respondents
toward the middle of scale.
Itemized rating scale is a measurement scale in
which the respondent selects an answer from a
limited number or ordered categories.
It is also simply to construct and administer.
Rank-order scale is a measurement scale in which the
respondent compares several items and ranks them.
Itemized and graphic scales are considered to be noncomparative scales, in other world the respondent
makes a judgment without reference to another object.
Place a 1 next to the brand you prefer, then a 2 for the
second and so on
_ Chevrolet
_ Toyota
_ Ford
Paired comparison scale is a measurement scale that
ask the respondent to pick one of two objects in set,
based on some started criteria.
Problem : the respondent select only one object and
does not rank.
Constant sum scale ask to respondent to divide a given
number (100) among several attributes based on their
importance to him or her.
It is preferred to paired comparison scale but it is
preferable to have 10 items or less because the
respondent can have difficulties allocating points if
there are too many characteristics.
The construction of a semantic differential scale begins
with determination of a concept to be rated, such as the
image of a company, brand or store. The researcher
selects opposite pairs of words that could be used to
describe the concept. Respondents then rate the
concept on a scale (usually 1 to 7)
Very useful, reliable and valid for decision making and
prediction, it has proved to be statistically robust but it
has to be adapted (more researched).
Be careful with the halo effect!
The semantic differential scale. (BIS)
Be careful with the Halo effect ! Indeed the rating of
a specific image component may be dominated by
the interview’s overall impression of the concept,
thus you rate every time the good or bad answer.
To counteract the halo effect we should randomly
reverse scale adjectives so that all the “good” ones
are not placed on one side of the scale.
The purchase intent scales are used to measure a
respondent’s intention to buy or not buy a product. The
ultimate issue for the marketing manager is “Will they
buy the product or not ?”
Purchase intent is evaluated at each stage of
development and demand estimates are refined.
It is a good predictor of consumer choice, very easy to
construct and we need just a subjective judgment.
Example : If a set of final exam answer is available
would you :
1/Definitely buy
2/Probably buy
3/Probably not buy
4/Definitely not buy
The purchase intent scale. (BIS)
For example, based on historical follow-up studies we learnt that :
63% of the “definitely we will buy” purchase it.
28% of the “probably we will buy” purchase it.
12% of the “probably we will not buy” purchase it.
3% of the “definitely we will not buy” purchase it.
And after our own study we have the result in the following :
“definitely we will buy”
“probably we will buy”
“probably we will not buy”
“definitely we will not buy”
So assuming that the sample is representative of the target market
we will have :
0,4 x (63%) + 0,2 x (28%) + 0,3 x (12%) + 0,1 x (3%) = 34,7% of the
market share
Type of scale
Most commercial researchers use telephone or internet to
save expense. So a rank-order scale can be quickly created
whereas developing semantic differential scale is often long
and tedious process.
We can also consider the respondent preference who
usually prefer nominal and ordinal scale because of their
Balanced VS non-balanced scale
The balanced has the same number of positive and
negative categories whereas the non-balanced is oriented
toward one end.
If past research has determined that most opinions are
positive, we will use more positive gradients than negative.
Number of scale categories
If the number of categories is too small (good, fair
and poor) the scale is crude and lacks richness. After
if we have too many categories we can have a kind of
discrimination of some ones.
Research has shown that rating scales with either 5 or
7 points are the most reliable.
Forced VS non-forced choice
If a neutral category is included, it typically attract
those who are neutral or those who lack adequate
Adding a “Don’t know” response is the solution.
However this option can attract lazy respondent.

similar documents