Why do you want to set a cut-off point? This is always problematic because it results in a loss of information. In particular, converting to a binary variable (above or below the cut-off) limits the kinds of analyses that you can perform, compared to using the full scale as an interval level variable.

I would suggest reviewing the original publication that created this measure. Was there a single scale? Did they report the reliability of the scale(s) they created? What kind of analyses did they do?

Dear Edmond Dufatanye ,

It is best practice to use about 7 response value in a Likert style question as this will yield the richest valid data. Also, it is common practice to combine several Likert style questions together using exploratory factor analysis or scale reliability analysis in order to create a scale variable.

I am not familiar with cut off values but I agree with David L Morgan 's general point of not prematurely reducing the quality of your data before analysis.

I agree with David - what is the purpose of creating a cut off value? Many researchers want to dichotomize scale values, but this discards potentially useful information present in the original ordered categories. If you need a cut off because you will be using the scale scores to make a binary decision, how would you validate the different possible cut offs? (E.g. do you have a criterion or 'gold standard/'truth'? If so is it the same criterion for all 11 questions?) This is why I also agree with David re: reading what the originators of the scale said. For example, do the 11 questions combine to form a total score of (11 x 4)? In that case, again why create a cut off?

Thanks to you all for your valuable advice and insightful guidance.

Edmond

To assess the prevalence of stigma among patients with COVID-19 in a city in Bangladesh, I have used the COVID-19-related stigma scale which was adapted from Berger's HIV stigma scale. This is a 15-item scale. Each item is rated on a 4-point Likert scale (0= strongly disagree, 1= disagree, 2= agree, 3= strongly agree), with total scores ranging from 0 to 45.

Now I want to set a cut-off score for those who experienced stigma. For example, those having a cut-off score of ≥16 experienced stigma, whereas those having a cut-off score of

As noted above, converting a measure that ranges from 0 to 45 into a dichotomous 0 and 1 looses a great deal of information. I think you should interpret the full scale as the degree of stigma experienced.

Dear Prof. David L Morgan thank you so much for your valuable suggestion.

To anynlse Likert like data we can use mode as then would know what the majority agrees with . Interpreting the findings as they are useful . Sometimes as in a stigma scale its how one perceive the situation and therefore no right or wrong answer to this . Hence interpretation is more important .

David L Morgan , good point as usual. I have a similar dilemma. My research aim to evaluate conection between selfesteem knoweldge and selfesteem capable to act. Both are measure using liker like scale. Knowledge from noknoweldge to expert knowelge. Willingness from strongly disagree to strongly agree. I would use ordinary log regression. However, should I record knowledge into two dichotomous variable and how (nonknowlede, minimum, basic, elementary and perfect). Since, my poulation are health workers, I suppose that most of them have knowedge, so distribution of answers is right-modal. In additiona, I have 7 areas of action with a question for each (diagnose, initial tretment, motivation, looking for help...). Therefore, I need to do, 7 regressions and five areas of knowldge (risk factor, preventions, urgent help, physical aspect of dependence...). If I recored knowlege into W/O knowledge, it will be mush easier for me and readers. However, I will lose some information.

I agree with Prof. Morgan that dichotomizing a continuous scale discards potentially informative differences - for example, cutting at >

Edmond Dufatanye Angelina Inesia Forde

I am not interesed in ''sometimes''/''often'' part, since it will sray in dependent variable.

I have additional responsed in Yes/No form. Have you every had an education about... for instance... prevention. However, since the population are medical professionals I expect that all of them had some kind of education about prevention, treatment... which are integral parts of curriculum. It seems that the best solution is create a table for each question and each level of independet variable. Which means 7 tables. Pretty large amaout of figures for a paper?!

Edmond Dufatanye, I totally agree with you. However, for my perspective, one who has a basic knolegde of some issues is in the same position as some without knowledge, since none of them can adequatly use that knowledge. Sometimes it is better to do nothing, than improvise.

I have used likert scale questions and cut them during analysis based on mean score and principal component analysis

Aleksandar P. Medarevic You seem to assume that your individual variables need to have a normal distribution in order to be included in a scale, but there is not such limitation. The usual reason for collecting 7 variables that all measure the same underlying concept is to create a single variable that it is more reliable, rather than separately analyzing each of a set of weaker measures.

David L Morgan Thank you professor Morgan. I can not agree more. Hoewever, my idea is not to create scale for measuring, but more realistic. Do you mean that analysing 7 responses are not meaningful? In my opinion, creating scale is extremely hard, and it is out of my scope. I try to estimate healthcare workers selfesteem capable to manage different issues of their jobs (diagnosis, treatment, providing essential help, motivation... to specific population). Nothing more. My hypothesis is related to that connection between experience/knowledge and capacity to provide care.

I mean that analysing 7 separate tests makes the results harder to interpret, and less likely to produce significant results due to the loss of reliability.

In terms of creating a scale, here is a link to a set of resources I have put together

https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items

I think we can set cut-off point for Likert scale studies by mean score