Classical
Assumptions

Classical
assumptions for assessing internal consistency among
indicators:

 The
indicators
 contain
random errors
 these
errors are not correlated with each
other
 the
errors not correlated with true scores
 any
latent variables affect all items equally
 the
amount of error for each item is equal
 The
classical assumptions are relaxed in what's called
"Tauequivalent" tests (tau stands for T as in true
score):
 The
more liberal assumption is that the amount of error
variance associated with a given item need not equal
the error variance.
 The
last condition above of "equal errors" for each item
is relaxed.
 This
means that item means and variances may also
vary,

The
classical assumptions for scale reliability in the internal
consistency sense.

 The greater the intercorrelations
among the items, the more reliable the scale.
 The greater the number of items, the
more reliable the scale.
 For practical research
applications, one seeks
 A scale of few items with high
intercorrelations to produce reliable
measurement.
 A metric that can express the
degree of reliability.
