In AMOS, deleting an item from a measurement model or a structural model should be done carefully and based on certain criteria. Here are some common guidelines to consider when deciding whether to delete an item:
1. Item-Total Correlation: Check the correlation between each item and the total score of the scale. If an item has a low correlation with the total score compared to other items, it may be a candidate for deletion. A commonly used threshold is a correlation below 0.3.
2. Factor Loading: Assess the factor loadings of each item. Items with low factor loadings (below 0.4 or 0.5) on their respective latent variables may indicate poor measurement quality and could be considered for deletion.
3. Modification Indices: Examine the modification indices, which indicate how much the model fit would improve if a specific parameter were freely estimated. High modification indices suggest that an item's removal could improve the model fit, although caution should be exercised as it is possible to overfit the model.
4. Theoretical Grounding: Consider the theoretical relevance and conceptual basis of each item. Ensure that removing an item does not compromise the underlying construct being measured or the theoretical framework of your study.
5. Substantial Change: Evaluate the impact of removing an item on the reliability and validity of the scale. Removing an item should not substantially alter the psychometric properties or the overall meaning of the construct being measured.
6. Content Analysis: Conduct a thorough content analysis of the item and seek expert opinions. Determine if the item is clear, relevant, and adequately represents the construct of interest.
It is important to note that the decision to delete an item should be made judiciously and in consultation with statistical and subject matter experts. Deleting an item can affect the validity and reliability of your measurement model, so it is crucial to carefully evaluate the implications before making any modifications.