Table 5.

Differential item functioning in the original and the reduced GSE versions

Differential item functioning	Results	Results
(DIF): Are item difficulty calibrations stable in relation to the following demographic variables? (generalizability validity)	Original 10-item GSE	Reduced 7-item GSE (omits items with poor fit)a
Age	Item 1: easier to agree with for people < 40 (p = .045)	• No DIF
Gender	• No DIF	• No DIF
Work	• Item 7: easier to agree for workers (p = .012)	• Item 7: easier to agree for workers (p = .003)
Education	• Item 2: easier to agree for persons with higher education (p = .045)	• Item 8: easier to agree for higher education (p = .046)
	• Item 4: easier to agree for persons with lower education (p = .024)
Relationship	• No DIF	• No DIF

Note. One item with DIF out of 20 can be expected to occur by chance and is deemed acceptable. Thus, the criterion for differential item function was a Mantel-Haenszel statistic [49] with p < 0.01 after Bonferroni correction [51]. Using an uncorrected p-value of < 0.05 is not common, but minimizes the risk of underestimating item bias.

^aItems #1, 2, and 3 removed.