You must put in the reliability option the items of each factor separately and then perform the general reliability introducing all the items of the instrument.
From analyze option you can enter all items of each component separately in box and click OK after that enter all items (19). Based on spss output you can decide degree of reliability if alpha less than 0.7 you should option of the item deleted
The answer depends on several things. First, every variable typically ihas a non-zero weight on each of the k components as extracted. How do you arbitrate which variables should be considered as contributing to a given component? That usually implies some threshold for a variable to be considered "salient" to the identification and constitution of component scores. While arbitrary, these thresholds should not be capriciously selected.
Second, did you rotate the final component solution?
Third, what do you do if a variable has a salient relationship with more than one component?
Last, how did you decide on the final number of components to extract?
At the end of the process, any attempt to estimate component reliability is likely to be biased upward, since you're evaluating the consistency of scores across variables in a component structure that is optimized for your sample. So, validation with other/additional data is a good idea.
Thank you David. I was using PCA for numerical data and began using Categorical PCA because my data is an evaluation using a Likert scale.
The 6 components extracted using as criteria an eigenvalue greater than 1 have alpha cronbach greater than 0.70 except two which have a value for alpha Cronbach of 0.50. Also the alpha Cronbach for the whole scale improved to 0.90.
Your suggestion of validation with additional data is a good idea