Factor loading and significant loading value?

Significant loading in factor analysis
When carrying out factor analysis (Exploratory factor analysis) using SPSS, I found that people supress the value under certain values in factor loading and only take values above certain point are considered as significant loading.
Many suggest this value to be 0.30 (and load only value above 0.30 as significant factor)
Whereas some other consider 0.50 as value for significant loading.
I found the if value is high, then only high loading factors are in each factor which looks good but it tends to ignore the high loading cross loading. For example, if any item is cross loading in 2 factors may be with 0.60 in one and 0.48 in another, that 0.48 will be supressed or ignored or cross loading won’t be visible.
If we take 0.30 as boundary to decide significant loading, then even low value (i.e 0.31 or 0.35) may be loaded in each factor and there may be many cross loading.
So, I am bit confused. High value is good to take significant loading or low value?
0.50 is better or 0.30
In my case, the factors are loading same for both values which left me wondering which value should I take?


King of all Drama
Rules of thumb like these ones you are describing (e.g. "only interpret factor loadings greater than X or Y value") are basic heuristics and have very little justification in statistical theory. That's why you can easily find yourself in situations like this one where the heuristic either contradicts itself or makes the interpretation difficult.

You'd need to switch your model of analysis to something like Confirmatory Factor Analysis (CFA) or a form of Exploratory Factor Analysis (EFA) that provides significance testing and p-values for the factor loadings.
So,in my case 0.30 would be better or 0.50 for significant loading , which one would be better? (no loading or cross loading between 0.30 and 0.60).
What do you suggest?
(doing research as a part of academic degree).


King of all Drama
None. Each heuristic has its own drawbacks and they come from a book by S. Stevens in the 1970s which was published way before we had the software and statistical theory needed to evaluate our models. It's time to abandon heuristics that do not aid in our analyses.