Understanding Differential Item Functioning (DIF) in Item Response Theory (IRT), Study notes of Statistics for Psychologists

The concept of differential item functioning (dif) in item response theory (irt), focusing on the definitions, identification through item parameter comparison and areas between two irt curves. It also discusses the significance of the b-parameter and the need for research on sample size.

Typology: Study notes

2011/2012

Uploaded on 11/21/2012

ashakiran
ashakiran 🇮🇳

4.5

(27)

261 documents

1 / 2

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
Ch. 8: Differential Item Functioning (DIF)
I. Introduction
A. Definition 1:
An item shows DIF if two groups are different in their mean performance on the
item (p-value in CTT). Does not consider a real difference between the two
groups.
B. Definition 2:
An item shows DIF if individuals with the same ability in different groups have
different probability of getting the item correct (P( ) in IRT). An item shows
DIF if the item response functions across different sub-groups are not identical
over the same ability range.
II. DIF in IRT
A. Item parameter comparison
1. Testing Ho: b1 = b2, Ho: a1 = a2, Ho: c1 = c2.
2. If item parameters are identical, ICCs should be identical over all points of
ability scale for two different groups. No DIF.
3. If Ho’s are rejected, DIF is present for the item.
4. Test statistics (
2
)
2
= (ad bd cd)’
1
(ad bd cd)
where
ad = a2 a1,
bd = b2 b1,
cd = c2 c1.
= covariance matrix of ad, bd, and cd.
ad bd cd
=
2
2
2
ccbca
bcbba
acaba
,
1
= inverse of .
5. Test statistic is asymptotically
2
-distribution with df = p
(p= # of parameters).
Docsity.com
pf2

Partial preview of the text

Download Understanding Differential Item Functioning (DIF) in Item Response Theory (IRT) and more Study notes Statistics for Psychologists in PDF only on Docsity!

Ch. 8: Differential Item Functioning (DIF)

I. Introduction A. Definition 1: An item shows DIF if two groups are different in their mean performance on the item (p-value in CTT).  Does not consider a real difference between the two groups. B. Definition 2: An item shows DIF if individuals with the same ability in different groups have different probability of getting the item correct (P( ) in IRT).  An item shows DIF if the item response functions across different sub-groups are not identical over the same ability range. II. DIF in IRT A. Item parameter comparison

  1. Testing Ho: b 1 = b 2 , Ho: a 1 = a 2 , Ho: c 1 = c 2.
  2. If item parameters are identical, ICCs should be identical over all points of ability scale for two different groups.  No DIF.
  3. If Ho’s are rejected, DIF is present for the item.
  4. Test statistics ( 2 ) (^2) = ( ad bd cd )’ 1 ( ad bd cd ) where ad = a 2 – a 1 , bd = b 2 – b 1 , cd = c 2 – c 1. = covariance matrix of ad, bd, and cd.

ad bd cd

2

2

2

ca cb c

ba b bc

a ab ac ,

(^1) = inverse of.

  1. Test statistic is asymptotically 2 -distribution with df = p (p= # of parameters).

Docsity.com

  1. When p=1 (Rasch model with only the b-parameter), (^2) = (^2) ( 1 ) (^2) ( 2 )

2

s b s b

bd ,

where s^2 (b 1 ) and s^2 (b 2 ) are variance of the b-parameter estimates,

s (b 1 ) = ( )

Ib 1

  1. If a and b are different, the ICC will be different regardless of the c-parameter.
  2. Again, the b-parameter should match the -range.
  3. Research is needed for the sample size associated with parameter estimation (Kim & Nicewander, 1993). B. Comparison of areas between two ICCs
  4. Numerical procedure a) Divide the -range into k-intervals. b) Make rectangles centered at the mid-point of each interval. c) Compute two ICCs (P( )) at the mid-point of each interval. d) Compute the difference between two ICCs and take the absolute value of the difference. e) Multiply the difference by the interval width and sum the quantity across the intervals.

Aj =

k

k

Pi Pi 1

where = interval width (e.g., .01, or .03), k = # of intervals ( 3 k 3 ).

  1. Exact calculation for the area (Raju’s 1988 article).

Docsity.com