Dear Vaios,

Yeah Absolutely, You can use likert scale (which is obviously linear in nature) data as input for Factor Analysis.

The whole industry of Market Research using FA on this kind of data. Just take care of outliers/cleaning and other anomalies.

Which kind of factor analysis you are doing, Exploratory FA or CFA?

Do let me know in case of any queries.

Regards,Abhay

Dear researcher,

This topic have been focused before.

Please open this link to benefit from the previous answers:

https://www.google.com/amp/s/www.researchgate.net/post/How_can_I_apply_factor_scores_to_likert_scale_data/amp

Hello Vaios,

You can apply Factor Analysis techniques (EFA and CFA) to analyze scales based on ordinal items (e.g., Likert scales). 

Because your items have seven categories, you can analyze your data using the estimator "Robust Maximum Likelihood." However, if you are not sure about the assumption of multivariate normality, you can try the estimators:

- "Weighted Least Squares Means and Variance adjusted," or - "Unweighted Least Squares Means and Variance adjusted."

The software packages Mplus, EQS, and Lisrel offer these estimators.

Roberto Melipillán

Dear Valos:

In my opinion, a scale of 7 points is perfect to use any multivariate method. Sufficient points to simulate the continuity (and maybe multivariate normality) assumption and not too many to affect the reliability of the subjects' responses.

If you want to be more sure of fulfilling the assumptions, you can transform your items (variables) into a range scale.

In R that is done with the following function:

master3 $ Alum_g_r

Dear All

thank you very much for the insights you have provided me.

However, my main concern is that "the Methodological and statistical texts are clear that for ordinal data (like likert scale) one should employ the median or mode as the measure of central tendency" and not the mean.

REF: Likert scales: how to (ab)use them, Susan Jamieson, Blackwell Publishing Ltd MEDICAL EDUCATION 2004; 38: 1212–1218

So my Guess is what to do with my observations of each variable before I perform a multivariate analysis..... (with the standard way)

Any Idea? is it appropriate to use Multivariate statistics using as measure of central tedency the mean and not the mode or the median? What I sould do witht the data before I perform a Factor Analysis? Or is it Ok to go on with the analysis?

Best regards

Hello Vaios,

As I wrote in a previous comment if your research objective requires conducting a Factor Analysis (EFA or CFA) based on your seven points Likert items, you don't need to modify them before using any of the estimators that I mentioned before.

Regarding the use of Likert scales in other multivariate analysis, the response will depend on the specific research objective that you want to achieve. Therefore, if you can provide more info about your research questions, I could offer more specific suggestions.

Roberto

Thank you Roberto.

There's only SPSS (version 23) available for use, since it is the only available software provided by my institution.

Your answer is noticed and I might come back later if I need further assistance.

Best regards

Dear Vaios:

In effect, we should develop a theory of multivariate analysis that uses the median as a measure of central tendency; It's something that I'd like to do myself or with a PhD student, but sometimes life calls you to more pressing issues. But a good option is to previously transform your items (variables) into a range scale; I have not proofed it mathematically, but I suspect that it is equivalent (under certain conditions that will have to be determined) to a multivariate multivariate analysis that uses the median as a measure of central tendency. (several of the sentences in this paragraph must be understood in quotes).

Regards,

Reynaldo Rocha

Dear Vaios:

Any measurement is a map between a real phenomena that lies behind researcher purpose and an abstract scale of measurement that researcher chose in order to make a specific measure.

I agree regarding to use the median instead of the mean to analyze the ordinal scales within an strict theoretical and methodological framework, but it is not collide with the use of an ordinal scale into other statistical analysis once you provide arguments to show that your ordinal scale is a suitable representation of a continuous phenomena. I strongly recommend you to read David Hand book in order to support your arguments:

Hand, D. J. (2004). Measurement Theory and Practice. The World Trough Quantification. Chichester, West Sussex, United Kingdom: John Wiley & Sons.

The Susan Jamieson article is correct. Means and standard deviations are based on the property of *distance* - a property that ordinal data do not possess. Use medians and nonparametric tests, frequency/percentage plots of the ratings.

Dear all,

thank you very much for your contribution about what to do with the data.

Best regards