Developing and Improving Measurement Methods - Evaluation Methods in Biomedical Informatics - page 151

TABLE 6.1. “Typical measurement result” with cor-

rected part-whole correlations.
Results of five observations

bject 1 2 3 4 5 otal score

A 4 5 3 5 5 22

B 3 5 5 3 4 20

C 4 4 4 4 5 21

D 3 2 3 2 3 13

Corrected
part-whole
correlation 0.62 0.83 0.11 0.77 0.90

“corrected part-whole” correlation coefficients for each observation. A
corrected part-whole correlation is computed for each observation, and is
the correlation between that observation and the object's total score after
excluding that observation . The process for computing this correlation is
described below. Using corrected part-whole correlations, only N correla-
tions need be inspected for N observations to determine if the observations
are well behaved. Typically, an observation exhibiting a corrected
part-whole correlation below 0.4 should be modified or deleted. Always
keep in mind that modifying an observation or eliminating it from a set
can change what the scale is assessing and can affect the validity of the
measurement.
To see how this works computationally, examine Table 6.1, which
portrays the “typical measurement result” previously shown in Table 5.2.
Recall that the reliability coefficient for these results is 0.81.
To compute the corrected part-whole correlation for each observation,
it is necessary to create ordered pairs of numbers representing the score
for each observation and the total score for each object, excluding that
observation. Table 6.2 illustrates computation of the corrected part-whole
correlation for Observation 1 from Table 6.1. The ordered pairs used to
compute the correlation consist of each object's score for Observation 1
paired with the object's total score summed across all observations but
excluding observation 1. Because Object A has a total score of 22, exclud-
ing Observation 1 yields a corrected total score of 18. The correlation

TABLE 6.2. Computing the corrected part-whole corre-
lation for observation 1.
Score for

Corrected total score,

Object Observation 1 excluding Observation 1

A 4 18

B 3 17

C 4 17

D 3 10