i have drafted 51 items that are supposed to measure 9 distinct but related constructs, and have collected data on this questionnaire from 530 people. ultimately, i want to use these data for a sem.
however, given that i devised the items myself, they seem to be far away from being perfect indicators of the constructs / latent variables that i intended to measure. therefore, it seems to me that i should make a selection of the 51 items before i proceed. all the criteria for item selection that i found in the literature (e.g., item-total correlation, item difficulty, item variance) seem to help only for choosing items for a unifactorial model, but not for the 9-factorial model that i have in mind, because they neglect the items' relationships with related concepts that would result in high cross-loadings.
therefore i have looked at a table of bivariate correlations among all items and have selected those items that show a) high correlations with the other items from their own hypothesized factor, and b) low correlations with items hypothesized to "belong" to other factors. this resulted in a list of 35 items that will be used for further analyses.
do you have any objections to this strategy?