i have been using Likert scales as an interval for my previous papers, however I had two workshops last two months, considering Likert scales as Ordinal and the other ratio. So confused !
I do understand your confusion as quite a few research methods textbooks give incorrect definitions and many users of likert and likert like scales ignore the question and use inappropriate analytical techniques.
One way of thinking about it would be that both interval or ratio scales have a consistent and equal distance between each item on the scale. This is clearly not the case in any likert or likert like scale; for example the difference between (5) strongly disapprove - (4) disapprove and (4) disapprove - (3) neither approve nor disapprove cannot be seen as measurably similar. the former simply seeks to differentiate between levels of an attitude (disapproval), while the latter distinguishes between either disapproving or not disapproving, two different states of mind. You could also ask yourself precisely what 3.5 or 4.5 means. I think you’d agree that neither has any real meaning. Ordinal scales use numbers to represent a single subjective judgement (such as those above), so only whole numbers have any real meaning. The scales are certainly in some sort of order (ordinal) but the interval between each number is neither equal nor even. A key implication is that analysis based on arithmetic means (eg t tests or anova) is poor methodological practice.
This question generates a lot of discussion and conflicting opinions.
In general, we often treat the results of single Likert-type items (individual questions) as ordinal. But if the results of several questions are combined into a scale, as is often done in psychology and sociology, the results are considered interval.
I would be interested to hear a full explanation from the author of the third answer how It is possible to convert ordinal data into interval simply due to the number of items. I realise this is common practice in some traditions of research, but I would never expect common practice to be a convincing scholarly rationale. I would be genuinely interested to hear this as no-one has been able to explain it to me throughout my career and few have tried. It is worth pointing out that analyses tend to go item by item anyway, so incorrect arithmetic analysis is still an issue where item means are generated from ordinal data.
Peter Sandiford , It's not that it's possible to convert ordinal data to interval *. It's that if someone is summing or averaging individual Likert type items to compose a scale, the data are already being treated as interval.
________
* Except by making assumptions about the spacing of the response categories.
Sal Mangiafico hmm, this doesn't seem to be an explanation; treating something as interval in nature is not the same as it being interval in nature; the whole point of my argument is that individual likert like items cannot be averaged (mean based analysis cannot work for ordinal data). I repeat that I would be genuinely interested in what the methodological argument could be for doing so. I recognise that summing can be justified as a way of making sense of this type of data (although I remain unconvinced of the value of doing so), though it does not de-intervalise it; the labels remain absolute (disapprove etc) as the numbers continue to represent these descriptors. Regarding your other point, relying on unquestioned assumptions is not very scholarly. I recall a little fable about donkeys, warning against such assumptions.
There seem a lot of alternatives to Stevens. I was just preparing my lecture on Mosteller & Tukey's ladder of re-expressions. Does anyone still teach just Stevens LoM without pointing out alternatives/criticisms/complements?
Few discussion above mentioned based on my Ques. however need more time to study carful ... I really thank you all Sal Mangiafico Peter Sandiford Addisu Damtew Atnafe for your relevant and eye-opening discussions.
Dawit Negussie you're welcome; it is always a pleasure to engage in a little debate with colleagues and this sort of question can stimulate just that. I did present a paper on this subject a while ago, available from :Conference Paper Important or Not? A Critical Discussion of Likert Scales and...
this was presented a long time ago, but I present the argument in a little more detail if you are interested (please note my added comment embedded in the original paper).
Developed in 1932 by Rensis Likert1 to measure attitudes, the typical Likert scale is a 5- or 7-point ordinal scale used by respondents to rate the degree to which they agree or disagree with a statement (table).