Introduction to Human Nutrition

306 Introduction to Human Nutrition

it is commonly estimated as the difference between
the reported result and the actual value (Figure
13.1).

Reliability

Reliability or reproducibility refers to the consistency
or repeatability of a measure. Reliability does not
imply validity. A reliable measure is measuring some-
thing consistently, but not necessarily estimating its
true value. If a measurement error occurs in two sepa-
rate measurements with exactly the same magnitude
and direction, this measurement may be fully reliable
but invalid. The kappa inter-rate agreement statistic
(for categorical variables) and the intraclass cor-
relation coeffi cient are frequently used to assess
reliability.

Precision

Precision is described as the quality of being sharply
defi ned or stated; thus, sometimes precision is indi-
cated by the number of signifi cant digits in the
measurement.
In a more restricted statistical sense, precision
refers to the reduction in random error. It can be
improved either by increasing the size of a study or
by using a design with higher effi ciency. For example,
a better balance in the allocation of exposed and
unexposed subjects, or a closer matching in a case–
control study usually obtains a higher precision
without increasing the size of the study.

Sensitivity and specifi city

Measures of sensitivity and specifi city relate to the
validity of a value. Sensitivity is the proportion of

subjects with the condition who are correctly

classifi ed as having the condition. Specifi city is the

proportion of persons without the condition who are

correctly classifi ed as being free of the condition by

the test or criteria. Sensitivity refl ects the proportion

of affected individuals who test positive, while speci-

fi city refers to the proportion of nonaffected individ-

uals who test negative (Table 13.1).

Data description

Statistics may have either a descriptive or an inferen-

tial role in nutrition research. Descriptive statistical

methods are a powerful tool to summarize large

amounts of data. These descriptive purposes are

served either by calculating statistical indices, such as

the mean, median, and standard deviation, or by

using graphical procedures, such as histograms, box

plots, and scatter plots. Some errors in the data col-

lection are most easily detected graphically with the

histogram plot or with the box-plot chart (box-and-

whisker plot). These two graphs are useful for describ-

ing the distribution of a quantitative variable. Nominal

variables, such as gender, and ordinal variables,

such as educational level, can be presented simply

Figure 13.1 Accuracy and precision.

Table 13.1 Estimation of sensitivity and specifi city

True condition or outcome

present

True condition or outcome

absent

Test + A B

Test − C D

Sensitivity= Specificity

+

=

+

A

AC

D

BD