If the scale is numbered as 1 to 7 and respondents are free to make their numbered continuously from 1 to 7, for example 1.54, 1.65, 2.21 Ets, We can say that the ordinal problem is solved and the scale is interval. Is it possible?
Many researcher will consider it as continuous. but the problem is that arithmetic procedure might not have much value. While Parametric tests rely on arithmetic.
@ Alireza
I had a similar doubt and raised a question (linked) on qualitative parameters...
In Likerts scale the 5 points has responses better than a dicotomous (Yes / No) answer but are affected by external locus of control (ELOC) to the decision formulated from the total outcomes of analysing a questionnaire.
UNDECIDED - Participants on two occasions say this, when
- They do not want to get into problems, by answering this
- They do not really know the answer
NOT APPLICABLE - Participants say this, when they are not within the ambit of exposure or experience to a particular question.
https://www.researchgate.net/post/Is_the_Universal_use_of_the_Likert_scale_in_evaluating_responses_sufficient_when_working_with_evaluative_statements_in_a_questionnaire
This is an interesting question. I would argue that it is a continuous variable with a bounded upper limit (typically evaluated with a censored regression). Indeed, one can argue that most continuous variables have an upper limit (with the highest value representing the ceiling). The difference here is that the upper bound is set by you (the researcher).
To be an interval variable, each change from one point to the next must mean the same thing. That is, the difference between 1 and 2 must mean exactly the same thing as the difference between 6 and 7, etc. Simply adding more precise numbering will not accomplish that.
Here is a source that discusses a number of issues related to choices in creating Liker-style items (it is by two well-recognized experts in the field):
http://www.sagepub.com/kumar4e/study/Chapter%209/Questionnaires.pdf
Dear Alireza Ghasemizad. i am attached the article you read it.
Although you will find a number of people who believe as Chalamalla Srinivas does that you cannot combine Likert-scored items into interval-level scales, you will find equally many or more who argue that it is possible to combine multiple ordinal items into a single interval scale.
Rather than falling into an argument between purists and those with a more practical orientation, I recommend that you follow the common practices in your own field. In particular, if you find that people routinely report Cronbach's alpha prior to using Likert-scored items as a scale, then do that. Alternatively, if people largely use multi-variate ordinal statistics to deal with them items, then good luck pursuing that path.
It has also been recommended that the more simple use of a bipolar scale item (Strongly disagree - Strongly agree; False - True; etc.) WITHOUT including intervals between the two poles, but instead simply mapping the indicated scale location against an underlying percentage (0 - 10 or 0 - 100 for greater precision) can be used to eliminate the assumption of "interval" altogether. This may provide one way to still use a Likert-type measure and be able to analyse this as interval data. This would be possible because the underlying percentage mapping is based not on specified intervals, but rather on an arithmetic progression that is inherently more consistent.
I do not believe that marking a point on a line between two verbally described endpoints converts a personal judgement into interval data. If a measure does not have an inherent countable score, then you need to work with multiple indicators to get the equivalent of an interval scale.
The endpoints are both mathematical, because you are using a continuous (%) scale to capture the actual response. The conceptual identifiers for these endpoints are alphabet based, but this only provides the conceptual framework for responding, it does not provide the measure, and it does not contain intervals. I heartily agree, however, with the practice of using multiple measures by which to triangulate the data. This is always a good idea, and particularly important when using a contested data analysis strategy.
If the question can logically be answered as a percentage of something, that is one thing. If you are simply asking for a response in the form of a percentage, that is just another analogy, not a genuine continuum.
Hi Alireza Ghasemizad . Based on your question it seem like you have only on likert type item. I agree with David L Morgan that it is of no help just to allow decimals on such scales. However, be aware that likert SCALES (consisting of several likert type items), are often analyzed with parametric statistics.
https://measuringu.com/likert-scales/
BestJohn-kåre
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication