Why have my items loaded under "wrong" (non-hypothesized) components in PCA/FA? I have conducted PCA as a variant of EFA to develop a scale. I'm a student of applied linguistics and the method of rotation was promax. A questionnare of 60 items has been answered by 200 participants. The items of the questionnaire were prepared based on 15 components extracted from interviews and literature reviews.
There are two problems. 


*

*First, some items have loaded under wrong components. So I prepared the item for component one but it is loaded under component 3. What do I do with the chaotic findings?  

*Second, there are 4 components with only 2 loaded factors. Should I discard that component?  Now that the items appear to be disorganized how can I make sure that component 3 on SPSS output is component 3 on my own list of themes? 
 A: Welcome to the site Mahdieh. 
It seems like your two problems are stemming from having selected an analytic approach (PCA/EFA) that is ill-suited to your goals of testing your theory of construct measurement. 
EFA, though a very popular analytic approach for the purpose of scale construction, is better thought of as a statistical tool for building a theory of construct measurement when you don't already have one. It's a gross oversimplification of the process, but in a nutshell, you plug your variables into an EFA with a  few details (e.g., how many factor to keep; rotation and estimation methods), and the EFA spits out the configuration of which variables load onto which factors. 
But in your case, you conducted interviews and carried out an extensive literature review, and in doing so, you have come into the process of scale-development with an already-developed theory of construct measurement in mind--you think you have a pretty good idea of how many factors you need, and which items go with which factors. Simply stated, you therefore don't need an analysis like EFA to give you a theory of construct measurement; you already have one, and you just need to test it to determine how plausible your theory of measurement is.
To accomplish the goal of testing a theory of construct measurement, you would be much better off using a confirmatory factor analysis (CFA). Whereas in EFA, the statistical program tells you which items go with which factors, the process is reversed in CFA; you tell the program which items you think load onto which factors. Then, your statistical program will fit your implied model to your data, and, in addition to estimates of factor loadings/correlations between factors, will give you a variety of indexes quantifying how well your model fits the data. 
I've been deliberately avoiding describing technical details because...well, they are technical. Having proclaimed yourself as a novice, I don't think details will be especially helpful for you in an answer to your question. But what I will say is this: I think you ought to consider trying to familiarize yourself with the basics of CFA. I'd recommend Beaujean's (2014) text, which does a pretty good job covering the basics in a very accessible way. It also helps you learn how to conduct CFA using the lavaanpackage for R, which is among the most user-friendly (IMO) options out there for people new to CFA. 
