Professional Documents
Culture Documents
ASSESSMENT INSTRUMENTS
LAWRENCE CALHOUN
Slippery Rock University, Slippery Rock, PA I6057
ROBERT L. SHRIGLEY
The Pennsylvania State University, University Park, PA I6802
DENNIS E. SHOWERS
State University College at Geneseo, Geneseo, New York
Introduction
Newcomb's (1986) scale, reported in the literature after this scale was devel-
oped, has similar problems.
Nuclear energy is an attitude object with emotional intensity that can be
effectively quantified by the Likert technique. The use of such an instru-
ment to test treatment effects, such as a persuasive communication delivered
by a credible source, seemed t o be worthy of the effort required to design
the scale. A long form of the scale would encourage paper-and-pencil testing
of large and diverse groups of subjects. Where time was limited, a short form
of the scale could be useful in telephone surveys at polling booths in shop-
ping malls or as a clip-out, mail-in questionnaire printed in local newspapers.
An all-pervasive purpose of the study addresses a concern of many science
educators: a greater inclusion of the affective realm within science classroom
teaching. The majority of students will not enter science as a career, but they
will vote on scientific and technological issues. Therefore, a greater emphasis
on attitude research toward such controversial issues as nuclear power plants
is in order. The beginning point for teaching and testing attitudes is the
development of a procedure for designing valid attitude measurement scales.
The ultimate purpose of this study was to design a valid and reliable
Likert-type scale to test attitudes toward the generation of electricity from
nuclear energy. More specifically, the steps of this study were the following:
(1) Fine-tune the criteria generated by Abdel-Gaid et al. (1986) and estab-
lish a sequence of tests leading t o the design of valid Likert scales.
(2) Generate a pool of 100 trial attitude statements, administer them to a
small sample, select a smaller and more manageable pool of items, and
resubmit them to a larger and more diverse sample.
( 3 ) Apply the tests generated in (l), select two sets of items, a long form
for standard research and a short one t o use where time is a constraint.
Defining the attitude object and identifying its subconcepts was the first
step in dtsigning the nuclear energy scale. A list of concepts and beliefs that
might be held by citizens regarding nuclear power was categorized into seven
subcomponents. Five judges evaluated the subcomponents: the nuclear
energy department head at a large state university, a senior reactor operator,
an energy education consultant at a large state university, a high school
teacher of social studies, and the manager of Educational Services of the
Atomic Industrial Forum.
Forty-one high school juniors and seniors, representing rural and suburban
schools in Pennsylvania, responded to the 100 trial statements. Subjects were
tested by their teachers and data were analyzed by Likert’s method, using
Kohr’s (1973) LIKERT computer program.
Using the criteria listed below, 27 of the 100 trial statements seemed
acceptable:
With an orignal goal of six trial statements for each subcomponent of the
attitude object, nine new statements were written to represent subcornpo-
nents that lacked statements.
TABLE I
-
Groups Responding to the Pool of 36 Items.
TOTAL 873
items involved the use of eight tests generated from the work of Abdel-Gaid
and her colleagues (1 986) (See Table 11). Four tests involved the selection of
items, and four additional tests sought reliability and validity data on the 20
items as a scale.
TABLE I1
Selecting andverifying the 20 Items for the Long Form (N=829).
’Test Purpose Results on Nuclear A t t i t u d e Scale
6. C o e f f i c i e n t A t e s t confirming r - v a l u e o f 0.93
a1 pha homogeniety o f a pool
of iterns.
The 20 items selected for this scale generated mean scores ranging from
2.37 to 3.87; two statements were outside the range of means. The range of
standard deviations was 0.93 to 1.23. With a range from 19% to 4176, several
of the items generated neutral responses above the suggested minimum of
2 5%.
Thirty or forty percent of undecided responses is a yellow flag - not a red
one. An intervention that modifies attitudes could swing neutral responses t o
the right and left of center. In other words, one might expect a higher per-
cent of neutral responses on pretests than tests following treatment. A
Chernobyl incident or the inclusion of data from a more heterogenous popu-
lation, e.g., including antinuclear groups, could also lower the number of
neutral responses. Although future data on neutral responses generated by
this scale should be monitored, we suggest that the 20 items are adequately
evaluative.
All six subcomponents of the attitude object are represented. The number
of items representing the subcomponents ranged from two t o five.
Test-retest, split-half, and equivalent forms are common tests for the
internal consistency of an attitude scale. With the limitations of the three
techniques as cited by Bohrnstedt (1970) and Cronbach (195 l), the investi-
gators followed Crano and Brewer’s (1973) advice and used the coefficient
alpha to test internal consistency of the nuclear scale. The split-half correla-
tion technique taken to its logical extreme, Cronbach’s alpha assumes each
item on a scale to be a subtest. An average intercorrelation of the 20 sub-
tests-one for each item-was completed. The scale generated a coefficient
alpha of 0.93, well. above a minimum of 0.80 recommended by Crano and
Brewer .
Converging the interitem correlation matrix into two, three or more clus-
ters, factor analysis serves as a psychometric tool for testing the judgments
made earlier by the jury. Warned that one can easily attribute reality to a
nonexistent factor (Kerlinger, 1965)’ we approached the interpretation of
factor analysis data with caution. The primary purpose for generating six
original subcomponents of nuclear attitudes was to adequately sample the
universe of items that exist under our attitude object. But it also seemed
prudent to analyze the data using a six factor format.
A principal components analysis was used to extract factors that might
account for the variance by the subjects. Our N of 829 subjects was well
beyond the minimum of 10 subjects per variable -in this case 20 statements
-a practice recommended by Nunnally (1978). Table IV shows the eigen-
values and the percent of variance generated by the responses of 829 sub-
jects. The six factors accounted for 67.84% of the variance. The varimax
166 CALHOUN, SHRIGLEY, AND SHOWERS
TABLE IV
Eigenvalue and Percent of Variance for the First Six Factors
1 8.60 42.63
2 1.35 6.80
49.43 Total for two-factor analysis
3 1.09 5.44
4 0.99 4.99
59.86 Total for four-factor analysis
5 0.83 4.16
6 0.76 3.82
67.84 Total for six-factor analysis
rotation procedure was chosen because of Comrey’s (1 973) advice that it has
high utility with factors that are orthogonally rotated. Table V shows the
results of the six factor rotation. T o be assigned t o a factor, we chose the
following parameters: items had to correlate with an r-value of 0.50 or
greater with that factor and 0.40 or less with the five other factors, i.e.,
Comrey’s ( 1973) factorial validity test in which absolute values are utilized.
Items 3, 6, 7, 9, 10, 1 1, and 12 clustered on Factor 1. (See Table I11 for a
listing of the statements.) The seven items represented not one but four of
the original attitude subcomponents. However, usefulness of nuclear energy
weaves its way throughout the wording of the seven statements. Factor
analysis should probably not be expected t o differentiate such subtle shades
of meaning within a test having six components. Also, the interrelatedness
of the seven items under the rubric of usefulness of nuclear energy suggests
some degree of validity for Factor 1.
Items 14, 15, 16, and 20 clustered under Factor 11. The four items repre-
sented three of the original subcomponents. Here again, a close examination
of the statements reveals safety as the common element. Safety was one of
the original subcomponents and two of the four safety items on the scale
clustered here.
Item 8, another safety statement, factored out under Factor 111. Item 18
factored out under Factor IV, a statement originally written to represent
international use of nuclear energy. Factor V is represented by one item,
No. 4, a statement originally written to represent radiation. Factor VI has
one statement, representing domestic use of nucIear energy. In summary,
15 of the 20 items reached criterion; five did not, and therefore, are not
clearly interpretable by this data.
We rotated six factors in the original analysis. However, there were other
choices. Kim and Mueller (1978) suggest the use of a minimum eigenvalue of
TABLE V
A Varimax Rotation With Six,Four. and Two Factors (N=829)
Two-Fac tor
Item Original Six-Factor Analysis Four-Factor Analysis Analysis
No. Subcomponents I 11 I11 IV V VI I I1 I11 IV 1 I1
1 Relative risks (3) -.47 .27 -.08 .05 .48 .25 .01 .37 -.62
- .06 -.35 .41
2 Future
(domestic (5) .25 -.15 -.03 .12 .10 -.82
- -.20 -.24 .56
- .3 1 .54
- - .20
3 Future
(domestic) (5) .71
I
-.21 -.36 .18 .02 -.14 .44 -.15 .63 .27 -.69 -.39
4 Radiation (1) - .08 .15 .24 -.11 .86
- .04 -.23 .46 -.24 .12 - .08 .56
5 Societal benefits (4) .4 9 -.27 -.18 .25 .02 - .42 .15 - .25 .58
- .39 .6 7
- -.33
6 Future
(international) (6) - .60
- .3 9 .14 -.02 .22 .23 -.21 .40 -.63 - .09 -.51 .55
7 Future
(domestic) (5) .70
- -.20 -.35 .22 -.01 -.21 .43 -.15 .65 .32 ~
.74 -.38
8 Safety (2) .33 -.12 -3 -.02 -.16 -.11 .I6 -.22 .27 .03 .33 - .55
~
8
7
m 6
al
7
Fi
; 5
C
al
M
$ 4
3
L
1 2 3 4 5 6
F a c t o r Number
Figure 1. The Scree Test
1.0 as a criterion for choosing the number of factors to rotate. In this case,
we have the choice of a four factor rotation, a choice that would have ac-
counted for 59.86% of the variance (See Table IV). The scree test suggests
that we choose the number of variables to rotate by graphing the eigenvalues
(Cattell, 1965). As shown in Figure 1, the diagonal-to-horizontal shift of our
data occurs at the second factor, suggesting that we rotate only two factors.
The percent of variance accounted for in this case would be 49.43.
The results of the two and four factor analysis are shown in Table IV. The
former accounted for 14 items; the latter, 13. A detailed analysis will not be
described here, but it is important to note that the same two elements, use-
fulness of nuclear energy and its safety, seem to influence the grouping of
statements. In the two factor analysis, Factor 1 consisted of positively-
worded statements and all but one of the negatively-worded statements clus-
tered under Factor 11. Clustering here may be due to statement polarity-a
rather trivial criterion.
In summary, the juries’ subcomponents were too narrow in scope to be
supported by factor analysis. Clusters resulting from factor analyses were
broad in scope: usefulness and safety. The three rotations suggest that 13,
14 or 15 of the items represent a fair share of the variance in one broad
170 CALHOUN, SHRIGLEY, AND SHOWERS
* A t this point in the study, the scale consisted of 36 trial items. Therefore, the mean scores were
affected accordingly. Data for the 20-item scale for the nuclear engineering students were inadvertent-
ly erased. However, the correlation of the 36 and 20 item pools generated an r-value of 0.98 -suggest-
ing a high degree of relatedness and providing some justification for including these data as a part of
the validation procedure.
NUCLEAR ENERGY ATTITUDE SCALE 171
Four criteria were used for selecting six statements from the 20-item
form :
Table I1 summarizes the eight tests for selecting Likert statements for a
scale. The 20-item pool seems to represent one domain: attitudes toward
the use of nuclear energy to generate electricity. Factor analysis did not
support the six subcomponents, but two logical subcomponents, usefulness
and safety, did factor from the data.
The distribution of data on each of the 20 items suggests that the items
have evaluative quality. The scale is reliable and homogenous as indicated by
1) a coefficient alpha of 0.93, 2) positive interitem correlations ranging from
0.15 to 0.73, and 3) adjusted item-total correlations ranging from 0.46 to
0.80.
The scale passed three more validity tests: 1) Subjects living farther from
nuclear power plants generated more positive mean scores than those living
closer, 2) nuclear engineering students scored higher than subjects in an
anti-nuclear citizens’ action group and 3) males scored higher than females.
The six items of the short form were drawn from the 20-item pool. There-
fore, some of its tests for validation are embedded within the long form.
The short form represents the six subcomponents of nuclear attitudes. It has
a coefficient alpha of 0.82, positive interitem correlations ranging from 0.23
to 0.69, and adjusted item-total correlations ranging from 0.51 to 0.72.
The eight tests used in this model fine-tuned the model of Abdel-Gaid and
her colleagues (1 986) by adding a jury of judges to better define the attitude
object. Unidimensionality was explored and used in the Abdel-Gaid study; at
172 CALHOUN, SHRIGLEY, AND SHOWERS
Strongly Strongly
Agree Agree Undecided Disagree Disagree
Negative
statements 1 pt. 2 3 4 5
Positive
statements 5 pts. 4 3 2 1
N U C L E A R ENERGY ATTITUDE SCALE 173
References