We are developing a scale with Likert items. There is an item which loaded very well on factor 1 in PCA. After determining the factor structure, I thought of observing the gender DIF patterns and found through ordinal regression that this item has significant DIF when grouped by gender. What shall I do now? Shall I remove this item from the scale and re-run the PCA or is it okay to retain an item with DIF if it is reckoned to be important for the construct theoretically?