The likert scale is originally ordinal. But commonly, it is managed on a numerical scale. In consensus, is it valid to use one scale as an approximation of another? I would like to know your opinion and experience.
the simple answer is that Likert is always ordinal but generally it depends on how you want to look at the data and what is your approach and assumptions about the results. you can also treat it as an interval scale. however, the scale is ordinal, the variable can be assumed or treated as continuous. also if you have 5 items (1 to 5) or seven, it can change how you look at it. a 5-item Likert is ordinal but a 7-item is continuous. but there is a debate on this also. in general you can assume that the same scale is valid for another estimation. again, your question is very general. maybe you can provide more info on what you are doing. this can help us understand what is the proper approach to your subject.
Unfortunately LIKERT Is always ordinal and can't be well approximated by anything else. I am tired of writing this at least once a week. Google Ordinal logistic regression and read what it says. Alternatively, search this portion of RG and read what has been written here. If you are looking for ways to deal with Likert data see the attached notes. Please remember convenient algorithms are not always the best available. Good research depends on learning the basics and adopting new and better methods as they are developed, not wishful thinking.. David Booth
The score for a single Likert-type item is ordinal. When several items are aggregated into a total score psychologists typically treat it (the total score) as if it is interval, but it still is ordinal and it not linear. Typically, a one unit change at the upper range of the scale represents a bigger change on the latent trait than a one unit change in the middle range of the scale. Proponents of the Rasch item response model claim that (if the data fits the model), linear (interval) Rasch measures can be derived from ordinal test scores.
Gideon P. De Bruin Please read the CURRENT literature such as FranK Harrell's notes. Psychologists once did what you suggested because there was nothing else now there is. Please remember the take-home message from Numerical Analysis-Anything is an approximation to anything else the question is how good. That is why zero was added to the number system. Mathematics worked better that way. It is the same with psychology. BTW for some history look at the arguments by psychologists that all tests should be two tailed,. Circa 1950. How well did that turn out? How many still hold to that. D. Booth.
David Eugene Booth You and I both agree that Likert scores are ordinal. I am fully aware that in modern psychometrics raw ordinal scores are often logistically transformed via item response models to obtain estimates of persons' standings on a continuous latent trait. Those person measures could then be used in further analyses. This is good practice (on condition that there is satisfactory data-model fit).
I appreciate your contributions David Eugene Booth Gideon P. De Bruin Pouya Zargar . I will keep analyzing your comments. In general, I understand that at the item level, the scale is ordinal, but at the construct level (the underlying idea to be measured) they are assumed as an interval scale (continuous or discrete quantitative variable). Why not use nominal analysis? Thus, the problem of "distance" between categories is avoided.
In the analysis for example of structural equations in spite of using likert scales that are of ordinal nature it should only be considered this way with a robust estimator, if this is not used it should be considered to scale, this example can be considered considering that sometimes to obtain results its effectiveness in the respective analysis is not taken into account