This specification names each facet and follows it by a list of the element numbers and names to be included in the analysis.

 

Elements not included in the list are usually omitted from Facets analysis. You can deliberately omit an element from an analysis by typing a ";" before its element number, turning that specification line into a comment.

 

I. Choose the order you prefer to think about your facets, and number them in that order. The number you assign to each facet will determine the order in which the facets are reported.

If your facets are persons, judges and items, you may choose:

Facet 1: judges

Facet 2: persons

Facet 3: items

 

II. Assign each element within each facet its own integer element number in the range 1 to 2147483647 for each facet. Numbers may be skipped over or omitted. Thus,

each judge must be assigned a unique element number in the range 1 to 2147483647.

each person must be assigned a unique element number in the range 1 to 2147483647.

each item must be assigned a unique element number in the range 1 to 2147483647.

The ranges of judge, person, and item numbers may overlap, so that judge 1 may rate person 1 on item 1.

Your elements may already labelled with numbers that are too big, such as social security numbers, or with non-numeric names, such as Mary. These elements must be re-identified with unique element numbers in the range 1 - 2147483647. Facform can assist with this. It can be downloaded from www.winsteps.com/facets.htm.

 

Element 0 is the null element, indicating that the facet does not apply to the observation. But 0 can be used as a regular element number if Keepasnull= has been used to set a different element number as the null element.

 

Suggestion: Globally replacing long numbers with short numbers in a Facets control and data file usually solves the problem of numbers that are too big.

 

III. Construct a list containing all facets and elements:

a) Start with Labels=

 

b) List for yourself all the facets, in whatever order you like:

e.g., raters, persons, items

 

c) For each facet, the facet number and its identifying label are entered first.

Labels=

1, Judges ; the rater facet

 

d) For that facet, the element numbers and their labels are entered, in any order.

Labels=

1, Judges ; the rater facet

10 = Rehnquist

1 = John Jay

123456789 = Jose ; this is Jose

 

If the element numbers are in a range, which share features, then they can be summarized:

Labels=

1, Judges ; the rater facet

1-100 = Males

101-200 = Females
Specifying long ranges containing many unused number will slow down Facets, e.g,,
1 - 5000000 = Males ; when there are only 10 males
Use the Excel "unique records" feature to list the numbers used

 

If the labels are to be used for reference by Dvalues=, then the code letter can be indicated by ":".

Dvalues=

3, 1, 2, 1 ; element number for facet 3 in the data is in the for element label for facet 1 at column 2 with a length of 1.

*

Labels=

1, Persons

1, 3M Fred ; M is used by Dvalues=

2, 2F Mary ; F is used by Dvalues=

3, 2M Jose ; M is used by Dvalues=

...

*

; either

3, Gender

1, F: Girls ; F prior to : is matched to F of Mary

2, M: Boys ; M prior to : is matched to M of Fred and Jose

*

; or

3, Gender

1, Female ; F is matched to F of Mary

2, Male  ; M is matched to M of Fred and Jose

*

 

e) An "*" marks the end of each facet's elements.

Labels=

1, Judges ; the rater facet

10 = Rehnquist

1 = John Jay

*

 

f) For each further facet, repeat c) through e).

Labels=

1, Judges ; the rater facet

10 = Rehnquist

1 = John Jay

*

2, Persons

 

Example:

Labels= ; specifying that facet/element details follow:

 

1,Judges ; This labels the first facet

1=John Jay ; Judge no. 1

2=Roy Beam ; Judge no. 2

; Judge 3 to be omitted

4=Wapner ; Judge 4

| ; the other judges go here

15=Scalia ; Judge no. 15

* ; an asterisk terminates the elements for this facet

 

2,Persons ; facet 2

1=Ben ; first person

2=7021596 ; person 2 as a telephone number

|

23=Zabado ; the last person

*

 

3,Items ; facet 3

1= 2+2 ; the first item addition

2= Tennis ; second item

3= Attitude ; third item

4 ; the fourth item has no label

* ; end of labels for last facet also ends Labels=

 

When there is no description, Facets uses the facet or element number instead, e.g.,

Labels=

1, Rater

4  ; this means 4=4

*

 

IV. Establish pre-set anchor measure values

Ignore this for initial analyses.

When logit anchor values or starting values are to be provided, a comma followed by a third value is appended after each element label. These values have one of three meanings:

 

1) Starting values (the standard when values are provided)

this is useful for speeding up analyses when you have a good idea of what some of the measures will be.

Whenever the element label is followed by a logit value, this is used as the starting estimate for that element in the analysis.

Labels=

1,Persons ; no code after facet label

23=Joe,2.3 ; means Estimation starts with Joe at 2.3 logit

 

2) Anchor values: A and D

Use this when you know what you want some of the measures to be, e.g., calibrations from an item bank. Elements representing demographic facets such as gender can be anchored at 0 logits. This excludes them from estimation, but includes them for fit statistics and bias analysis.

When the facet label is followed by ",A", then the measure of each element in that facet is anchored (fixed) at the value following its label, whenever such a value is provided:

Labels=

2=Persons,A ; anchoring wanted

1=Ben ; no value - anchoring does not apply

23=Joe,2.3 ; Joe anchored at 2.3 for the entire analysis

 

When Umean= is used, then the anchor values must align with it:

Umean=50,10 ; User mean is 50, user scaling is 10 per logit.

2=Persons,A ; anchoring wanted

1=Ben ; no value - anchoring does not apply

23=Joe, 73 ; Joe anchored at 50 + 10*2.3 for the entire analysis

 

This feature is useful for dummy facets removing classification elements from the measurement. These may be used to partition fit, detect bias or select rating scale (or partial credit) models.

Labels=

2=Classifier,A ; anchoring wanted

1=Type A,0 ; anchor at 0, so doesn't affect measurement

2=Type B,0 ; anchor at 0, so doesn't affect measurement

 

or, and easier,

Labels=

2=Classifier,D ; dummy-facet anchoring wanted

1=Type A ; no anchor value required

2=Type B ; no anchor value required

 

When Umean= is used, use the Umean= value for these "dummy" elements

Umean=50,10 ; User mean is 50, user scaling is 10 per logit.

Labels=

2=Classifier,D ; dummy-facet anchoring wanted

1=Type A ; no anchor value required

2=Type B ; no anchor value required

 

3) Group anchoring: G

Use this to equate by groups of elements, e.g., to maintain the same average severity of a group judges from one test analysis to the next, or to anchor the average difficulty calibration of a set of 6th grade math items. An easy way to specify a particular group mean is to give all group elements that same value.

When the facet label is followed by ",G", groups of elements are anchored so that their mean is fixed, though they float individually relative to that mean. If several groups are to be defined, enter each element's group number as a fourth entry:

 

2,Raters, G  ; Raters are facet 2

1=Ben  ; no value, normal estimation, does not belong to a group

2=Mike, 2.7,1 ; Mike contributes 2.7 logits to group 1

3=Mary, 3.3,2 ; Mary contributes 3.3 logits to group 2

5=Anne,-1.7,1 ; Anne contributes -1.7 logits to group 1

6=Joel, 0.5,2 ; Joe contributes 0.5 logits to group 2

7=Fred,  ,2 ; Fred is reported in group 2, but he does not participate in the group anchoring.

; The double comma, ", ,", omits the anchor value.

8=Kent, 3.1 ; No group number, so 3.1 is a starting value

9=Irma,-2.8,0 ; Group 0 means "anchor at this value"

10=Jim, 1.3,0 ; Group 0 means "anchor at this value"

11=Abe, 1.5,3 ; Only one in a group anchor, so treated as if anchored, "A", but reported as a group

*

 

Mike, Anne and Joe are in group 1, so their individual estimated measures will alter, but their mean measure will be fixed at (2.0 -1.0 + 0.5)/3 = 0.5 logits. Mary is the only one anchored in group 2, so her measure will be fixed at 3.0.

 

An easy way to specify a particular group mean is to give all group elements that same value. In this example we want Groups 1 and 2 both to have average measures of 0.

Labels=

2=Persons,G ; group anchoring wanted

1=Ben,0,1 ; Ben is in group 1

2=Mike,0,1 ; Mike is in group 1

3=Mary,0,2 ; Mary is in group 2

5=Anne,0,1 ; Anne is in group 1

6=Joe,0,1 ; Joe is in group 1

7=Fred,0,2 ; Fred is in group 2

 

See Nested Designs for more examples.

 

V. Grouping

Grouping of elements is controlled by a fourth parameter. This is used for reporting as well as group anchoring. Thus, to report people classified into two groups, males and females:

Labels=

2=Person ; No anchoring specified, so values (if any) will be starting estimates. We could also specify ",A" (anchoring) or ",G" (group anchoring).

1=Ben,,1 ; in group 1, note use of double commas, ",,"

2=Mike,,1 ; in group 1

3=May,3.4,2 ; in group 2, with a starting logit value of 3.4

5=Anne,,2 ; in group 2

6=Harold,,1 ; in group 1

 

The output report will provide tables for

1) all elements of each facet with summary statistics for the facet.

2) each group of elements within each facet with summary statistics for the group.

 

Grouping simplifies analysis of subsets of elements in a facet. Boys and girls can be separately listed. The mean measures of the two groups can be directly compared. Grouping of test items by strand, and arranging the elements by measure, aids in verifying the construct validity of each strand. It also makes the strand information more available for use in curriculum and assessment decision-making.