LECTURE 4

ATTITUDE SCALING
THURSTONE SCALING
Post WWI work on interests and attitudes
during 1920s- both industrial and psychological
research interests
Thurstone proposed model based on jnd
Assumed interval scale possible for respondents to
respond to different stimuli
Developed procedure to generate about 20
statements that a respondent would agree or disagree
with, sum the items based on their interval scale value
(positive or negative items, scored appropriately)
THURSTONE SCALING
Example: Attitude toward Abortion
10 statements positive toward abortion
10 statements negative toward abortion
Respondent agreeing with positive
statement receives 1 point for each
Respondent disagreeing with negative
statement receives 1 point each
Possible range 0 - 20
THURSTONE SCALING
A. Select single concept, idea, or
construct for scaling
eg. War, marriage, abortion, mathematics
B. Collect 100-200 statements about the
concept: non-factual, opinion-oriented
C. Select about 80-100 for analysis. eg:
I like arithmetic most of the time.
Abortions should never be performed under any
circumstances.
War is usually a good thing, everything considered.
THURSTONE SCALING
D. Place statements along 11 point
continuum from (-) 1= most negative
statement to (+) 11= most positive
statement, with 6= neutral or
nonjudgmental statement.
- use 50-200 subjects to do
placement
- evaluate distribution of each
statement:

THURSTONE SCALING
Median of item as scale value:
for example, statement: Abortions should never be
performed.

1 2 3 4 5 6 ...
n: 150 50 0 0 0 0

%ile 50
score 1.17


THURSTONE SCALING
for example, statement: Abortions should be performed only to
save the life of the mother.

1 2 3 4 5 6...
n: 10 40 90 40 20 0

%ile 50
score 3.05

THURSTONE SCALING
Variability:
Eliminate items with ranges > 6 or 7
Examine conditional distributions of
adjacent or close items:
Give items to 200-300 respondents to
endorse each statement (+) agree or (-)
disagree
Examine joint endorsements of one item (a)
with another (b), using an index such as
I
a,b
= n
ab
/ n
b

THURSTONE SCALING
The distribution of the
a
I
b
s should
decrease around item a on either side
of its scale value;
that is, items with similar scale values
should have a high similarity index,
while items further away on the scale
should have scale values that drop
away with distance.
Throw out items with poor characteristics.
1 2 3 4 5 6 7 8 9 10 11
I
n
d
e
x

of

S
i
m
i
l
a
r
I
t
y
ItemScale
Value = 5.5
LIKERT SCALING
Renses Likert (1930s) researched the
Thurstone procedure
- Placement, scale valuation procedure is
cumbersome
- Likert replaced it with:
1=strongly disagree
2=disagree
3=uncertain
4=agree
5=strongly agree

LIKERT SCALING
Give items to sample of target population
Use classical techniques to select items:
item mean, SD, interitem correlation
Theoretical justification: items are
samples of the normal distribution
shifted along the 1-5 continuum so that
the mean is at the scale value
1 2 3 4 5
Item with scale value 1.5
1 2 3 4 5
Item with scale value 3.0
RATING SCALES
Derivatives of Likert scaling
Alternative adjective set (1..5)
Requires distributional validation
Even # points is problematic
Less well investigated
RATING SCALE VALUES
Number of scale values:
1 to 5 based on Likert
7, 9, or 11 can be useful for finer
discriminations
Dependent on population, concept
being assessed
SEMANTIC DIFFERENTIAL
Osgood, Suci, & Tannenbaum
Bipolar adjective pairs
Represent:
Strength: strong-weak, heavy-light
Value: good-bad, useful-useless
Activity: fast-slow, hot-cold
Two factors usually found: Evaluation and
Activity/Strength

SOME ADDITIONAL IDEAS
NON-NORMAL DISTRIBUTIONS:
POISSON AND COUNT DATA
SCORE
F
R
E
Q
U
E
N
C
Y
SOME ADDITIONAL IDEAS
NON-NORMAL
DISTRIBUTIONS:
POISSON AND
COUNT DATA
0 5
ATTITUDE TO PARENTS
0
500
1,000
1,500
2,000
2,500
F
r
e
q
u
e
n
c
y
Mean = 6.51
Std. Dev. = 1.899
N = 5,443
20 30 40 50
ATTITUDE TO PARENTS
0
500
1000
1500
2000
C
o
u
n
t
DISTRIBUTIONAL ISSUES
CENSORED: PART OF THE NORMAL
(eg. LIKERT ITEMS)- SUMMING
SHOULD CREATE NORMAL
INDIVIDUAL ITEMS CAN BE
ANALYZED AS CENSORED NORMAL
SCORES
OF INTEREST IN RELIABILITY AND
VALIDITY STUDIES