Faten Amer, I appreciate your concern because ideally variables are measured by at least three items in contexts that involve EFA.

However, although single-item variables might be criticised for not capturing the richness of a construct, they have been defended by Willets, Theodori, and Luloff (2016), and I think they can work effectively at times. I recommend you consider your single items carefully, and, if you think their use is justified, go ahead and use them.

Willits, F. K., Theodori, G. L., & Luloff, A. E. (2016). Another look at Likert scales. Journal of Rural Social Sciences, 31(3), 126–139. https://egrove.olemiss.edu/jrss/vol31/iss3/6

Faten Amer If the item is under one construct (more than 3 items construct), then you should remove the item if it did not load in EFA. For those variables with only one item, you do not have to include them in EFA.

Faten Amer, Usually items not loading on their respective constructs are discarded. However, at times discarded items can be combined to make a new construct.

Robert has already given you advice regarding single items.

Can you share a little more information regarding the EFA, so that we could together see what you have done and how could we help you further? Information such as:

- Which extraction method are you using?

- How the number of factors was identified?

- and, which rotation method was employed?

There is a common misconception between PCA and EFA, and at times right parameters (above) not selected. Consider reading the following article, if not already done:

Costello, A. B., & Osborne, J. (2005). Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Practical assessment, research, and evaluation, 10(1), 7.

Robert Trevethan thank you

it was helpful

@Ali Farooq

I used the Principal component method with Promax rotation.

the number of items is 45

thank you for the articles I will read them.

Faten Amer, please be aware that principal components analysis (PCA) is not the same as exploratory factor analysis (EFA) - although both will often produce much the same outcome.

You have not answered all of the questions that Ali Farooq asked (e.g., how you are determining the number of factors/components in your data), and it might be important to do so. In addition, I think those of us who are trying to help you will be interested in the number of people in your sample.

Please feel free to get back if you think we might be able to help you more.

Incidentally, using a promax rotation is probably quite a good idea.

Robert Trevethan Thank you for your reply and questions.

So which method should I choose instead, is it principal axis factoring?

I chose the number of components based on Eigenvalues higher than 1

I suppressed loadings less than 0.5

My total sample is 550, I split equally between EFA and CFA

Faten Amer, you're welcome.

From a distance, and not being familiar with the nature of your data, I think that principal axis factoring is a good choice for extraction method.

Although many people use the Kaiser criterion (eigenvalues > 1) for determining the number of factors in a data set, that method has been criticised for more than 20 years. (I think that many researchers keep following each other like sheep.) The scree test is often more enlightening, and parallel analysis is recommended as being even better. I have found that the Kaiser criterion is often VERY misleading and that a combination of the scree plot and parallel analysis ends up producing the cleanest and most interpretable solutions.

Parallel analysis can be conducted quite easily. Feel free to ask me for help if you're not sure about it.

I guess you're using about half your sample for EFA and then the other half for CFA. That's not a bad strategy, but I hope you have enough participants in each case. Working out whether you have enough participants can be a bit tricky, but there are some quite sophisticated and helpful guidelines around.

Robert Trevethan Thank you again for the valuable information

Do you mean the rule of thumb and KMO adequacy?

Actually, I have further 2 questions, please

1- if the rule of thumb is not met but KMO is higher than 0.8 does this means that the data is adequate?

2-Can I use PCA if I want to validate a new scale or should I use EFA?

Faten, firstly you will need to identify a number of factors in the data. As David mentioned, the Eigenvalues method is criticized, consider using parallel analysis. The following video provides an excellent guide on running parallel analysis:

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

The rule of thumb for item loadings and KMO adequacy is in the next step, when you know the number of factors and check their item loadings and KMO adequacy.

2ndly, if the purpose is to create a new scale, use EFA and not PCA. PCA can be useful if you want to prepare a shorter version of a big scale.

I once again suggest reading the paper by Costello & Osborne (2005). It provides guidelines for running EFA. Moreover, if you explore the How2Stat channel (the above video is from the same channel), it has a step-by-step guidelines for running EFA.

Faten Amer, first up, sorry, but I don't know what you are referring to when you mention "rule of thumb" - and please accept my apologies that I didn't draw attention to that in my most recent post above here. Feel free to explain, please.

As Ali has recommended in the post immediately above here, EFA is more appropriate if you want to create a new scale. PCA is probably more appropriate if you want to create an index.

The appropriateness of your data for conducting EFA with is indicated by the KMO index and Bartlett's test of sphericity. If your KMO index is > .80 (as you suggest), that's certainly adequate, and the Bartlett's test needs to be significant - which, in my experience it pretty well always is, so to have it significant at < .001 is desirable.

Please do remember that the scree test and parallel analysis, in combination, are likely to be better than the Kaiser criterion for indicating the number of factors in your data.

As I mentioned above, if you need help with parallel analysis, feel free to let me know. I could do that for you.

Robert Trevethan, I mean by the rule of the thumb (calculating the required sample as per the rule of the required respondents' number per item 5/1, 10/1). Thank you for offering help. I will try to do it myself because I want to learn this. If I could not manage, it will honor me to ask for help.

The information you and Ali Farooq provided to me was really beneficial. I re-run the EFA as you recommended.

1- I split the data. KMO was 0.8. However, 3 components did not have logically related items. So, I tried to run EFA for the whole sample; the constructs became much better. All of them were logically meaningful. However, a few items did not load at all, some of them I can discard. But, some are really important to me. Can I use them as single items, specifically that each represents a direct question to healthcare workers? For example, one is a question assessing directly patients' satisfaction rate, the other is evaluating health worker teamwork, the third is assessing their knowledge to a particular aspect?

2-If I used the whole sample for EFA (because it gave better components) and I cannot enlarge my sample further. Should I perform CFA using the same sample and report this in the limitations of my scale development or should I perform EFA only?

Thank you so much for your guidance.

Faten Amer, I'm sorry, but I don't know what you mean by "3 components did not have logically related items". Might you mean, perhaps, that, after determining the number of factors in your data (hopefully as suggesed by the scree plot and parallel analysis, NOT by the Kaiser criterion), those factors each had a distinctive set of items? If so, that's exactly what a successful EFA will produce.

As a side note, I think it's important, in the environment of EFA and PCA (and also CFA), to use words such as factors, components, and constructs very carefully or other people won't be sure what you're referring to.

With regard to using single items as variables, please go back up to the first post I made to this thread.

With regard to the final question in your most recent post, you should not use the same sample for EFA and then for CFA. Doing so is essentially engaging in a self-fulfilling prophesy. It is permissible, however, to use the same data to go from a CFA that didn't work and use those data to conduct EFAs to see why the CFA didn't work.

I hope that makes sense.

According to my knowledge the term component is the right one to use for EfA not construct which is used for CFA

What i meant previously is that the items at 3 components out if 14 were not making sense

for example one of the 3 components had cleanliness item with patient communication item loading together

however using the whole sample solved this problem and items loaded in a meaningful and logical way

my question is : can i use EFA only in validation of a new scale? Of course In addition to the inter item correlation, corrected total item correlation, Cronbach’s Alpha

convergent, discriminant and divergent validity

but not using CFA since the items are too many and the sample is sufficient only to perform EFA

or this will be criticized too?

Here is my suggestion: https://www.researchgate.net/publication/355207581_Principal_Component_Analysis_versus_Factor_Analysis

Here is my suggestion: https://www.researchgate.net/publication/355207581_Principal_Component_Analysis_versus_Factor_Analysis

Faten Amer EFA is not sufficient to validate a new scale. Have you thought of mixing the data, making two new sub-sets, and running EFA on one and CFA on the other? It may solve your problem

Ali Farooq, I tried that, but the results were not good. I have many items so I think my sample will be only sufficient to perform EFA

I found recent scale development papers published in high-ranking journals which used only EFA without CFA

Article The development and validation of a scale to explore staff e...

https://academic.oup.com/intqhc/article/25/4/443/1798960

or only PCA

http://www.hqlo.com/content/7/1/21

Faten Amer, you were asking us the right procedure of scale development and analysis, and we're telling you so. Publishing a paper is altogether a different thing. You may be able to publish a paper with just an EFA.