DIF= specifies the part of the person label which is to be used for classifying persons in order to identify Differential Item Function (DIF) - uniform or non-uniform - using the column selection rules. See also DIF Table and DIF and DPF considerations.

 

 

DIF= location is usually column number within person label field. DIF=1 means "DIF selection character is first character of person label."

 

Example 1: Columns 18-20 of the person label (in columns 118-120 of the data record) contain a district code:

 NAME1=101  ; person label starts in column 101

 NAMELENGTH = 20 ; person label must include all the DIF codes

 DIF = $S18W3  ; district starts in column 18 of person label with a width of 3

or

 @district = 18W3 ; district starts in column 18 of person label with a width of 3

 DIF = @district  ; DIF classifier

 tfile=*

 30   Table 30 for the DIF report (or use Output Tables menu)

 *

 

Example 2: DIF by Gender+Grade: Column 1 of the person labels contains gender (M or F) and Columns 7-8 contain grade level (K-12).

 DIF = 1W1 + 7W2

 

Example 3: I have tab-separated data and my DIF indicator is in a separate field from the Person Iabel.

 Solution: for the DIF analysis, do a separate run of Winsteps. At the "Extra Specifications" prompt:

 NAME1=(location of DIF indicator) DIF=$S1W1

 

Example 4: Columns 18-20 of the person label (in columns 118-120 of the data record) contain a district code. Column 21 of the person label (in column 121 of the data record) has a gender code. Two independent DIF analyses are needed:

 NAME1=101  ; person label starts in column 101

 NAMELENGTH = 21 ; person label must include all the DIF codes

 DIF = *

 $S18W3  ; start in person label column 18 with a width of 3 - district

 $S21W1  ; start in person label column 21 with a width of 1 - gender

 *

 tfile=*

 30   Table 30 for the DIF report (or use Output Tables menu)

 *

 

Example 5: An investigation of non-uniform DIF with high-low ability classification for the KCT data.

  ; action the following with the Specification pull-down menu

 @SEX = $S9W1 ; the sex of participants is in column 9 of the person label

 DIF = @SEX + MA2 ; look for non-uniform DIF (gender + two ability strata): MA2

 PSUBTOT = @SEX + MA2 ; summary statistics by gender and ability strata

 Tfile=*   ; This is more easily actioned through the Output Tables Menu

 30   ; Table 30 - DIF report

 28   ; Table 28 - Person subtotals for DIF classifications

 *

Table 30: DIF specification is: DIF=@SEX+MA2

+-----------------------------------------------------------------------------------------+

| KID     DIF   DIF   KID     DIF   DIF      DIF    JOINT          TAP                    |

| CLASS MEASURE S.E.  CLASS MEASURE S.E.  CONTRAST  S.E.   t  d.f. Number  Name           |

|-----------------------------------------------------------------------------------------|

| F1     -1.86  1.09  M1     -4.54  1.15      2.68  1.59  1.69   8      6 6= 3-4-1        |

 

Table 28: Subtotal specification is: PSUBTOTAL=@SEX+MA2

+---------------------------------------------------------+

| KID     MEAN     S.E.  OBSERVED  MEDIAN   REAL          |

| COUNT  MEASURE   MEAN    S.D.          SEPARATION  CODE |

|---------------------------------------------------------|

|      4   -2.08     .90     .89             .00     F1   | <- Non-extreme

|      6   -2.82     .41     .91   -2.86     .32     M1   |

 

Example 6: With Example0.txt (the Liking for Science rating scale data) you want to see if any items were biased against names starting with any letter of the alphabet, then:

 

run example0.txt

request the DIF Table (Table 30) from the Output Tables menu

specify: $S1W1

a DIF table is produced.

 

The equivalent DIF specification is: DIF=$S1W1

 

Positive DIF size is higher ACT difficulty measure

+------------------------------------------------------------------------------------+

| KID     DIF   DIF   KID     DIF   DIF      DIF    JOINT          ACT               |

| CLASS MEASURE S.E.  CLASS MEASURE S.E.  CONTRAST  S.E.   t  d.f. Number  Name      |

|------------------------------------------------------------------------------------|

| R       -.06   .54  W        .89> 2.05      -.95  2.12  -.45   8     1 WATCH BIRDS |

| R       -.06   .54  L       -.65   .75       .59   .92   .64  12     1 WATCH BIRDS |

| R       -.06   .54  S       -.42   .57       .36   .78   .46  18     1 WATCH BIRDS |

| R       -.06   .54  H      -1.63  1.13      1.57  1.25  1.26  11     1 WATCH BIRDS |

| R       -.06   .54  D        .12   .86      -.18  1.01  -.18  11     1 WATCH BIRDS |

.......

 

Example 7: The DIF demographic column is not part of the person label. Use the FORMAT= instruction to rearrange the record.

Was:

 0        1         2         3

 123456789012345678901234567890

 AxxxxxxxxxxxxxxxxxxxBCyyyyyyyy ; original record layout

 ITEM1=2

 NI=19  ; the x's

 NAME1=21 ; starts with B

 

To move column 1 after column 21 with XWIDE=1:

 xxxxxxxxxxxxxxxxxxxBACyyyyyyyy ; after internal reformatting

 FORMAT = (T2,20A,T1,1A,T22,100A)

 ITEM1 = 1 ; references position in the reformatted record.

 NI = 19 ; the x's

 NAME1 = 20 ; starts with B - references position in the reformatted record. 

 DIF = S2W1 ; code A

 

The first reformatted record is shown on your screen so you can check that it is how you want it.

 

Example 8: We only want DIF between person classification groups A and B, not the other person groups. Classification group code is in column 3 of the person label.

In the Specification pull down menu

PSELECT=??{AB}

Output Table menu: Table 30.