If your items are Likert scales, that would normally be considered ordinal and discrete (requiring a polychoric correlation matrix) rather than continuous. Stas Kolenikov and Gustav Angeles wrote a paper a few years ago that appeared in Review of Income and Wealth on the problem of treating ordinal scales (like Likert scales) as continuous in data reduction methods. Below are links to the paper and a presentation based upon it. I believe they focused on PCA but I would imagine that the message would be similar for EFA. 

Presentation :

http://www.unc.edu/~skolenik/talks/Gustavo-Stas-PCA-generic.pdf

Paper: 

http://web.missouri.edu/kolenikovs/papers/polychoric-technical-4.pdf

The answer most definitely depends on the pragmatic goals of your work. Of course you want to adhere to Groner's application of parsimony as much as is reasonably possible. But I would feel uncomfortable saying much more than that without having a bit more context as to the specific model(s) that you are basing your aims from.

Usually, polychoric correlations are for ordinal variables, but for 7-point Like Scale, the differences are negligible.

Bang Quan Zheng That is very interesting! Which book or article did you find that out from? This is very relevant to a study that I am doing now.

@ Nicole Moskowitz Depends on what you want to do. If you just do EFA, just read EQS manual. It has very good explanations. Here is the link to EQS manual. https://www3.nd.edu/~kyuan/courses/sem/EQS-Manual6.pdf