Patient_Reported_Outcome_Measures_in_Rheumatic_Diseases

(ff) #1
49

scale construction. Local independence can be evaluated by examining the residual
correlation matrices for systematic error among item clusters that may indicate vio-
lation of the assumption.
Model fit can be examined at both the item and person level to determine whether
the estimated item and person parameters can reproduce the observed item
responses. Since IRT is probabilistic in nature, most fit indices measure deviations
between predicted and observed response frequencies. Many types of residual anal-
ysis can be used to evaluate model fit.


RaschModelt()he 1 PL=OneParameter LogisticModel

The Rasch model includes two determinants of an item response—the respondent’s
trait level and the items’ difficulty level.
The initial estimates of trait levels can be seen as a two-step process. First, we
determine the proportion of items that each respondent answered correctly. For a
respondent, the proportion correct is simply the number of items answered cor-
rectly, divided by the total number of items that were answered. To obtain estimates
of trait levels, we next take the natural log of a ratio of proportion correct to propor-
tion incorrect:


q 5
1

=





æ
è

ç

ö
ø

LN ÷
s
s

P
P

where Ps is the proportion of items answered correctly by Respondent 5 (a specific
respondent).
The initial estimates of item difficulties also can be seen as a two-step process.
First, we determine the proportion of correct responses for each item. For an item,
the proportion of correct responses is the number of respondents who answered the
item correctly, divided by the total number of respondents who answered the item.
To obtain estimates of item difficulty, we compute the natural log of the ratio of the
proportion of incorrect responses to the proportion of correct responses:


bi i
i

=LN

æ -
è

ç

ö
ø

÷

1 P
P

where Pi is the proportion of correct responses for item i.


Item and Test Information


As a psychometric approach, IRT provides information about items and about tests.
In an IRT analysis, item characteristics are combined in order to reflect characteris-
tics of the test as a whole. In this way, item characteristics such as difficulty and


2 A Guide to PROMs Methodology and Selection Criteria

Free download pdf