Specifies the number of reported user-scaled units per logit. When USCALE=1 (or USCALE= is omitted) then all measures are in logits.
Negative USCALE= values, such as USCALE= -10
Higher row score -> lower row measure
Higher column score -> higher column measure
USCALE= does not change the direction of the Andrich thresholds, so
with positive USCALE=, Andrich thresholds are in the direction of items (columns)
with negative USCALE=, Andrich thresholds are in the direction of persons (rows)
Table 20 gives the UMEAN= and USCALE= values for a conversion that gives the measures a range of 0-100.
Example 1: You want to user-rescale 1 logit into 45.5 units, so that differences of -100, -50, 0, +50, +100 correspond to success rates of 10%, 25%, 50%, 75%, 90%:
USCALE = 45.5
Example 2: You want to reverse the measurement directions, because
a) the data matrix is transposed so that the 'items' are examinees and the 'persons' are test questions:
b) the data are scored "backwards", such as rank orders, where lower score value -> higher ability
USCALE = -1
KEYn=, RESCORE=, ISGROUPS= will still apply to the columns, not the rows, of the data matrix. Centering will still be on the column measures.
Example 3: You want to approximate the "probit" measures used in many statistical procedures.
UPMEAN = 0 ; set the person sample mean to zero
USCALE = 0.59 ; probits * 1.7 = logits
but if you want your person sample to approximate N(0,1):
1. first analysis:
UPMEAN=0
USCALE=1
look at the person S.D. in Table 3.1
2. second analysis:
UPMEAN=0
USCALE=1/person S.D. from 1.
Example 4: You want to report measures in "log-odds units to the base 10" (lods) instead of the standard "logits to the base e".
USCALE=0.434294482 ; the conversion between "natural" and base-10 logarithms.
For more examples, and how to compute this by hand, see User-friendly rescaling
