First, IRT analyses compute the standard error of
measurement (SEM) at each level of the latent trait,
which indicates the extent of measurement preciseness
at each level of the trait. For instance, it may be the
case that the UWES may be more precise at particular
levels (high vs. low) of work engagement. Second, IRT
analyses compute the amount of psychometric “information”
about the latent trait at each level of the trait that
is provided by each item, as well as the entire measure,
using the item information functions (IIFs) and the test
information function (TIF), respectively. The IIFs and
TIF are particularly useful because they indicate which
items, and which levels of the latent trait, provide substantial
information. For instance, it may be that some
items or particular levels of the trait (e.g., high vs. low
levels of work engagement) provide less information.
Taken together, IRT can be used to evaluate measures
in terms of how well the items and the entire measure
assess a trait at different levels on the continuum for
that trait [46].
By using IRT, we [47] investigated (1) the measurement
accuracy of the Japanese and the original Dutch
version of the 9-item short Utrecht Work Engagement
Scale and (2) the comparability of the scale between
Japan (N = 2,339) and the Netherlands (N = 13,406).
Figure 2 and 3 show the results of TIF and SEM among
Japanese and Dutch samples, respectively (please note
that SEM equals the root square of 1/TIF), whereby the
x-axis indicates the latent trait of the scale and the
y-axis indicates measurement precision conditional on
latent trait for the whole scale.