I think you have all the necessary information to make your decision. Depending of the rotation you're using (varimax for example), a solution with more factors will be always presented a better solution for the program, as its explain more of the variability of the variables. But, more components not always mean a better solution, simple because maybe this 5 factors/components doesn't  make any sense theoretically. So the decision from now on, in my opinion, depends of your theory. If you want a metric justification of using a 2 factor solution vs a 5 factor solution, I suggest running a CFA separately for both, and test for a chi-squared difference. This will tell of a solution of 5 factors is significantly better than a solution of 2 factors :).

I agree with Daniel that you should go with a CFA theroy based approach. In addition, I would not use a prinicipal component analysis since its actually no factor analysis in the classical sense. Furthermore, I think the adult hope scale is pretty short scale, isn't it (12 items)? I would be worried about an over-factorization with to many factors with only a few item (less than three) per factor. 

If you are using a centered and standardized PCA (each variable centered on its mean and divided by its standard error: a default option in several PCA programs), the overall variance is 5. So, the sum of the first and second eigen values divided by 5 provides you with the proportion of variance explained by the first two principal components. This simple procedure, together with the bar plot of eigen values, may help you to decide whether most of information in your dataset is borne by these components.

A practical approach to deciding how many factors to retain is to stop when you are unable at some point to assign a meaningful interpretation to a factor. The other useful measure is the percent explained (out of the total variability). Use these two metrics combined, and you will have a reasonable and defensible decision on how many factors to retain.

Make sure to consider rotation of the factors.  Try oblique rotations for maximum impact, and try orthogonal rotations to  keep any subsequent analysis of factor scores at the univariate level.

I would not worry that your results look "different". Document why they look different. Compare sample sizes and data sources. 

In using Factor Analysis I suggest using Oblique rotation  to construct the underlying dimensions involved in the measuring instrument   and to have  eigenvalues higher than 1.0to   control the number of extracted factors.  Moreover, in using factor analysis I recommend that first of all you should use    Keyser–Meyer–Olkin (KMO) measure of sampling adequacy (KMO) and the Bartlett test of sphericity (B =  , p = 0.000)   because they will suggest if your data is suitable for factor analysis.  That said then, you don’t have to force your extraction  and if you end with 5 factor instead of 2 then, I would prefer reporting that because it is considered a theoretical contribution because evidently you’re the hypothetical dimension in your study is composed of sub dimensions

Factor analysis depends strongly on the assumption of normality. Was this checked, e.g. by a Q-Q plot. Any strong outliers?

Then: How large is the sample size N serving for calculations? If you have  reasonable large data, it might be splitted into two parts.  If both parts would point to the same number of factors (e.g. 5), and the same interpretation of factor scores, I would decidedly take that number of factors  for further analysis.

Anna

It depends what you call a factor analysis. In her initial question, Sandy mentionned principal component analysis (PCA) which does not rely on normality assumption (at all !).