Based on your analysis, your independent variable have three dimensions.

I am suggesting you to also evaluate factor correlation matrix, and check the correlation among your three factors. 

For detail, kindly have look on paper entitle 'An easy approach to exploratory factor analysis.........

the minimum items needed for a factor are 3. Your both factors contain 3 items, so you can use this scale as multidimensional scale and check its validity through CFA

Based on your analysis, your independent variable have three dimensions.

I am suggesting you to also evaluate factor correlation matrix, and check the correlation among your three factors. 

For detail, kindly have look on paper entitle 'An easy approach to exploratory factor analysis.........

In EFA:

The criteria used for factor extraction :

1. Eigenvalue should be greater than one

2. Variance should be 50-75% explained

3. Retaining the factors above the elbow in scree test

4. At least three items per factor with significant factor loading

i.e.>0.30.

If these are satisfied then go for the validity test by CFA.

Dear Fawad,

Although statistically you can proceed splitting your scale into 2 factors, check the eigenvalues of each factor (they should be greater than 1), and if these 2 factors have theoretically meaning (that is, if the items that are loading in factor 1 have a different content than items that are loading factor 2). If yes, you can proceed naming both factors and saying that your IV is composed  by 2 factors (F1 – xxxxname; F2 – name)

Good luck with your research!

Fawad,

I do a lot of EFA/CFA, specifically for scale creation and evaluation.  A few questions: 

1) What size was the sample you did the EFA in?

2) Did you do a true EFA or did you use a PCA?  If EFA, what type (e.g., Principal Factor) and did you use a rotation (e.g., varimax; direct oblim)

3) How correlated are the two factors?

4) How much variance did the two factors explain?

5) Prior to running the EFA, did you expect the seven items to form a unidimensional scale (i.e. did we expect all items to load onto a single factor) or did we expect that two factors would emerge?  

As stated above, you certainly need a minimum of 3 items per factor. 

If the measure was intended to measure a single thing, then given the scale has only 7 items, i'd be strongly tempted to look at some of the classic psychometric indices before I did much of anything.  With a scale that short, a single item could be having a large impact.  First, using all seven items as a single scale (or at least the six items that load), I'd look at the "Change in Alpha if removed" and corrected item-to-scale correlations.  Additionally, if alpha is okay (at least .70 and ideally .80 or greater) and the corrected item to scale correlations are around .40 or higher, then the scale is probably functioning unidimensionally even if EFA is giving you a two-factor solution.  The two-factor solution may indicate sub-scales (vs. two different scales; examining how correlated the factors are will help you differentiate that).

If you have several items with corrected item-to-scale correlations below .40, then you need to more strongly consider what your EFA is telling you (i.e., the scale is likely not acting in a unidimensional manner in you sample).  

There are a number of good online papers that help you determine the number of factors you should extract (e.g., Here's one by Preacher; http://quantpsy.org/pubs/preacher_zhang_kim_mels_2013.pdf).  Also, if this is for a publication/dissertation, you might want to consider doing a MAP test or Parallel Analysis to support your decision to take out two factors.  Here is a useful resources for those:

https://www.youtube.com/watch?v=T908yGVgjPk

Yes you can. Others have already answered your question. But with 3 items you cannot use CFA since the model will be fully specified and you require at least 4 items for CFA

Why wasn't the 7th item assigned to one of the 2 retained factors?