Hi all,

I have a basic question about factor analysis. From what I can understand, if different people look at the same data and factor analyze it (even with the same extraction method and rotation) they can come to different conclusions, based on certain factor loading cut offs and communality values?

I have done a factor analysis and have reduced my data to two factors, where all the items load reasonably well between the two factors, however, the communalities for two items are below <.3. I have chosen to use .3 as a cut-off even though it's still quiet high, especially for my sample size. When I remove those two items with low communalities I get 1 factor.

This is what I'm expecting from my theory, and the two items I removed don't make a lot of theoretical sense. I'm wondering whether I have made the right move here, because the final two items I removed did load reasonably well on a separate factor that had a few other items on it as well (loadings of .4 and above for both factors), but I have made the arbitrary rule to cut communalities at .3. Considering the theory that the data is based on, I think this makes sense, but I'm not sure whether I should have retained the two factor structure. The data is based on a reasonably new construct, and previous studies have found only 1 or 2 factors. I'm not sure whether turning in a single factor is correct or whether it's purely about what level of loadings and communalities I'm willing to accept. In relation to my eigenvalues, with the two factors included together I have a higher % of variance explained with the two factors as opposed to the one, but only by a very very small amount, and I have a scree plot that seems to show 1 factor for my entire efa.

Thanks in advance.

I have a basic question about factor analysis. From what I can understand, if different people look at the same data and factor analyze it (even with the same extraction method and rotation) they can come to different conclusions, based on certain factor loading cut offs and communality values?

I have done a factor analysis and have reduced my data to two factors, where all the items load reasonably well between the two factors, however, the communalities for two items are below <.3. I have chosen to use .3 as a cut-off even though it's still quiet high, especially for my sample size. When I remove those two items with low communalities I get 1 factor.

This is what I'm expecting from my theory, and the two items I removed don't make a lot of theoretical sense. I'm wondering whether I have made the right move here, because the final two items I removed did load reasonably well on a separate factor that had a few other items on it as well (loadings of .4 and above for both factors), but I have made the arbitrary rule to cut communalities at .3. Considering the theory that the data is based on, I think this makes sense, but I'm not sure whether I should have retained the two factor structure. The data is based on a reasonably new construct, and previous studies have found only 1 or 2 factors. I'm not sure whether turning in a single factor is correct or whether it's purely about what level of loadings and communalities I'm willing to accept. In relation to my eigenvalues, with the two factors included together I have a higher % of variance explained with the two factors as opposed to the one, but only by a very very small amount, and I have a scree plot that seems to show 1 factor for my entire efa.

Thanks in advance.

Last edited: