Karima Chelihi, unfortunately, your questions are a little bit too vague for you to obtain really useful advice. However, for a start, I would recommend that you delete items that do not load on any factors, then probably run the exploratory factor analysis (EFA) again to see which items load on your remaining factors.

I think that, for the items you retain, you are referring to how you should present your results. I think it would be best to sequence the items from those with the highest loadings down to those with the lowest loadings for each factor. I hope that advice is appropriate, but I'm not really sure what your question is in relation to that.

Apart from that, having eight factors seems to be a large number. It makes me wonder whether you conducted your EFA properly - but that's another matter.

Do we need more detail here on the correlation coefficients and information on whether the factor solution the question refers to is an unrotated solution or rotated in an informed way?

Dear Karima Chelihi ,

Just to add to Robert Trevethan 's useful advice, there are essentially two ways of carrying out an EFA: to create an instrument which might be used in the future; and to analyse a data set based around factors. If your purpose is the former then the non-loading items are less useful. If it is the latter then you might still wish to use individual items that don't load sufficiently on any factor as part of your analysis.

What are you going to do with the results of your factor analysis? Other than being told to do it, what is the purpose of it? And as everyone above says, you haven't provided enough information for anyone to provide detailed suggestions.

Robert Trevethan & Peter Samuels I Apologize for the unclear question,

My Thesis consists of 8 Elements studying the impact on productivity within the workplace environment. The questionnaire was taken from previous research conducted by another researcher consists of 41 questions divided into the elements

Each of the Elements "before using factor analysis" has a specific number of questions. Now the factor analysis is deleting some of the 41 because of some loads on two factors and some load with.30 or less and the sixth and seventh has only two items loaded on them

my questions are: for example Question number 29, 36, 37" questions regarding office lighting" are loading on three factors, should I delete it? like from the whole thesis ? or I just delete it from factor analysis to make it fit?

The second question is: having eight elements and factor analysis gives 7 factors does that mean I need to change the order of my elements?

I would like to inform you I am a beginner in SPSS and sorry

Daniel Wright sorry i tried to explain more in my last comment, also I conducted factor analysis because I was told to do so, I believe it I will NOT get the same factor structure as the scale author has advertised in the Questionnaire from which I've taken since I didn't write this questionnaire myself. and indeed it didn't questions are loading randomly among 7 factors when the thesis consists of 8 elements

Peter Samuels I am analyzing a data set based around 8 factors, should my factors be effected by the Factor analysis? and cause some of the items to be removed or moving questions from a factor to another???

Karima Chelihi, from what you have written, I think you should probably conduct confirmatory factor analysis (CFA) first, then EFA if that does not "work out". One problem with that, however, is CFA requiring an add-on to SPSS and you might not feel confident to proceed with doing a CFA.

I have a question: Has your thesis been examined, and you are now being required to conduct factor analysis, or have your supervisors asked you do conduct it? Either way, maybe doing an EFA, rather than a CFA, would keep the others happy.

Also, do you have enough participants to run a "decent" factor analysis? There are no definite rules about sample size, but one rule of thumb is that you should have at least 10 times the number of people as you have items. In your case, that would mean at least 410 participants. Is your sample large enough?

Also, have you done foundation work such as checking the KMO index and Bartlett's test of sphericity, and did you choose an appropriate extraction method (e.g., principal axis factoring) and did you rotate your factors appropriately (e.g., with an oblique rotation)? And, if you used an oblique rotation, did you examine the pattern matrix?

I ask all of these questions because, if you didn't do the right things, you could well be just muddling around in fuzz.

If you obtain a factor structure that is different from what you had before, I think that should "reverberate" throughout your whole thesis.

Dear Karima

Getting clean Factor Structure on EFA is often complicated however based on the your explanation if"29, 36, 37" questions regarding office lighting" are cross-loadings between factors ( primary loading should be at least 0.200 larger than secondary loading) then you need to remove one by one and check the results.

In addition if you are using EFA for " convergent and discriminant validity " then you can use CFA directly and the results of CFA is acceptable.

Dear Robert Trevethan I am not quite confidant in how to conduct CFA but I got the AMOS and I am learning from what I was reading on the internment this might be helpful, the thesis was examined and the supervisor didn't ask for the Factor but the jury and they requested a Factor analysis be it EFA or CFA, number of participants is 322 employees and yes I checked the KMO index and Bartlett's test it was significant and KMO = 0.927 and I believe the rotation was done properly.

but the results Order is different from the Original Questionaire from which I have taken and the factor structure is indeed different from what I had before, and I'm not sure how can I fix this issue

Can I use any other test or analysis?

Dear Peter Samuels I Will take your advice and try the CFA as the EFA structure is different from the main structure of the Questionnaire

thank you so much.

Alireza shabani shojaei i will thank you so much for your help

Karima Chelihi, unfortunately, it appears that you will have to go back to "basics" in your research in that you'll need to determine the factor structure of your data and then conduct you subsequent analyses in light of that.

I suspect that your CFA will not be "successful" in the sense of confirming the earlier / assumed structure, and that you will therefore need EFA to identify the structure of your data.

It would appear from the KMO index and Bartlett's test that your data are amenable to EFA. So, I would eliminate cross-loading items and items with low loadings (perhaps not eliminating them all at once, but getting rid of the worst ones first and seeing what happens as you successively remove unsatisfactory items.

Without knowing much about your project and what your analyses are producing, I suspect that EFA is going to be the most useful fundamental analysis at this point.

Karima Chelihi, I apologise for not thinking of this earlier, but eight seems to be a large number of factors. I'm not sure how that number was determined, but the most highly recommended way to determine the number of factors in data is to conduct parallel analysis. It is really quite easy - but, in my experience, most researchers don't use it. Instead, they often use the Kaiser method (number of eigenvalues > 1), which has been criticised for about 40 years.

If you would like me to provide information about parallel analysis, feel free to ask and I'll put something together for you.

You should consider your theoretical model, to evaluate CFA instead of EFA

Robert Trevethan Thank you so much for your suggestions and you help if you believe that parallel analysis would be helpful and you have some time please do guide me through how is it used, thank you so much

Cristian Ramos-Vera yes right now i will try the evaluation through CFA

Karima Chelihi, I am attaching something that is based on some imaginary data from a person I've called John. In essence, you need to obtain "random" eigenvalues from the website I have shown you, and also the eigenvalues from your own exploratory factor analysis.

You enter those values into a table of the kind I have created and check to see where your own eigenvalues are no longer greater than the random ones. At that point you have an idea about how many "real" (as opposed to random) factors there are in your data.

I hope that helps.

Robert Trevethan i will check it right away thank you so much for your help

Karima Chelihi, let me know how you go. I anticipate you'll get fewer factors than were suggested by the Kaiser criterion.

Here is the reference for the parallel analysis calculator I sent you:

Patil, V. H, Surendra N. Singh, S. N., Mishra, S., & Donavan, D. T. (2017). Parallel analysis engine to aid in determining number of factors to retain using R. https://analytics.gonzaga.edu/parallelengine/

Robert Trevethan thank you i will let you know how it goes thx again ^^

Karima Chelihi ,

Factor analysis, EFA or CFA, is one of many measures assessing the validity and reliability of instruments used in research. I suggest that you look at our recent publications where these issues are discussed.

Article Developing and Validating a New Multi-Dimensional Scale for ...

Best,

Ali

I may not be so much accurate on your question but I am suggesting what I think could be appropriate. First, the study contexts that tested the tool you are using might not be the same as yours. Similarly, the study objectives and the hypotheses or models tested could be different. Therefore, sample size, study contexts (e.g., nature of the population) or any other factors can influence the differences, factor structure being one of them. If for instance, the original structure had 8 dimensions, but your analysis reveals only three or otherwise, report what you have seen. Items that do not meet the required psychometric properties should be treated with caution if you want to retain them. Items may work in one context but may not work properly in another due to different reasons which you have to consider in your discussion. For example, language, maturity, nature of work could be some of the reasons. Regarding the loading value less than .30, I think you have to remove the items, normally it has to be .30, .40, sometimes .50 can also be considered as the minimum loading value. Consider the values of Eigenvalue (both observed and calculated from Montecarlo parallel analysis) to make a correct decision. Then, check the internal consistency of each dimension separately and of the whole scale. Do not forget to assess the inter-item correlations. Good luck.

Dear Karima Chelihi ,

From your question I guess you have adapted the measurements from prior literature (necessarily from well reputed peer reviewed sources, of course). If this is the case, you do not need to perform EFA but only opt for CFA to validate the constructs in your context, as EFA is a data reduction technique and is used in the process of new scale development. Furthermore, deleting items due to poor loadings and poor fit during CFA is also not an issue if deleted items do not acceed from around 20% of the total items. Thank you.