In decades of research and supervision of Likert type scales, comprehensive investigation of many performance indicators reveals:

Four ordered categories were often enough;

Five worked sometimes;

Six worked once.

Of course, with less stringent criteria, any number of response categories might work.

A simple floor and ceiling effects will reveal a lot. The aim is to have minimal effects for both. Studies have shown that the more the options, the minimal (better) the floor and ceiling effects.

But to have a full picture, consider running reliability and validity tests to compare the features.

- Best wishes

- Alhaji

Dear Riangga Novrianto ,

7 point is better because it yields richer data. However, 9 or 10 points may not be as good because the human brain can only differentiate between about 7 pieces of information at once. The more choices you give the greater the tendency to simply the choice process.

In the science of measurement, for short scales such as the one which you are presenting [5 items] it is better to use more steps in Likert scale, and vice versa.

You can read in the textbook of measurement for more details.

Waltz C., Strickland O., Lenz E. (2017; 5th ed.). Measurement in Nursing and Health Research Fourth Edition. Springer; New York.

Dear Riangga

You should consider the multi-trait multi-method model. This will allow you to assess the effects of different types of measurement of the same construct.

Article A Multitrait–Multimethod Analysis of the Construct Validity ...

Dear Riangga

First of all congratulations for your scholarly effort. My suggestion is that you can collect data from the same respondents by using both point scale and then go for t-test for dofferences. Initially you will be able to find is there any significant difference exits or not. If so, then you can test the reliability (alpha & CR) and validity (AVE, nomological, & htmt) of the scale for both scale points. Then finally ascertain which one produce good result. But keep in mind that the scale point is sensetive to reliability and validity subject to the level of education (in fact in broader sense, culture) of the respondents.

If you want compare both results, I Think that you should recathegorize the outcomes of 7 Points Scale to 5 points scale with the purpose of standarize your findings and conclusions. One of the problems of Likert Scale is the selection and perception bias of respondents. In this regard, the Likert scales in multiple samples should be standardized in a single measure in order to further reduce biases.

It is impossible to give an answer to this in isolation. A general answer is that the more points a scale has where each point is receiving a reasonable percentage of the responses, the better. In other words, the more selected options an intem has, the variability of the item increases, and the better the item. That means that in a given situation, you must pilot test different numbers of response options to find which is the best. The general theory is that that the more spread that you get from the response set, the better. However, as you add more options, you get fewer and fewer respondents selecting each option. When options start receiving very few selections or no selections at all, it begins to reduce the overall variability. The extreme options are the ones that usually begin to not being selected first. Over the years, research has shown that, in general, five and seven options tend to maximize the variability.

Thank you Prof. James E. McLean for detailed explanation.

Thank you for the imprtant details James E. McLean

Hello Riangga,

James McLean up above has given some good advice. The main reason for providing more response options is that you might obtain greater variability as a result - and statistics thrive on variability. However, that variability needs to be valid. A colleague and I have recently done some research in which we were able to compare responses on 7-point option scales with responses on 9-point option scales. We found that there wasn't necessarily greater variability when respondents were offered 9 options, and that variability seemed to be dependent on the particular sample being investigated.

To come more directly to your question, I would consider carefully whether you need (or rather, whether your respondents would need) all response options that you might offer. And I would compare responses from two similar groups of people and also, perhaps, in a piloting activity talk with some respondents to see whether they were "comfortable" with the response options provided for them. Peter Samuels' comments above relate to that kind of thing.

All the best with your research.

Robert

You can use some method:

1. Use a scaling analysis with RASCH model, and see the Andrich Treshold results.. You can find the problem of the scaling with this method..

Or

2. Use an expert judgment method..

Increasing the number of points on the scale increases the potential variance, and up to a point it also increases the precision of the ratings. But you run into diminishing returns as you approach the human mind's limits as an information processor. Remember "the magic number 7, plus or minus 2?" While not perfectly accurate, it's a pretty good approximation of how to optimize a likert-type scale. The main condition under which a 5-point scale would be preferable would be if you are studying something that your raters can't discriminate at the higher level of precision. I'm not quite sure why you want to bother with this study, to be honest; chances are there will be very little difference, and what you get will slightly favor the 7-point scale. But if you want to proceed with it, then the key is for you to have some kind of criterion measure(s) against which to validate your ratings. These could include objective measures (such as income, since you are in a relatively poor country), as well as other subjective measures, such as an affect-balance report or the PANAS. Simply correlate the SWL scale scores with those on your criterion measure(s) and compare the correlation coefficients across versions. If the 7 (or 5) point scale is significantly more strongly correlated with the criterion, then it is better. Warning: It is difficult to establish significant differences across correlation coefficients; you may need a large sample. Since the two SWL scales will be very strongly correlated with each other, it would be better statistically to have the same people complete both versions. But you could use separate samples as well.

In decades of research and supervision of Likert type scales, comprehensive investigation of many performance indicators reveals:

Four ordered categories were often enough;

Five worked sometimes;

Six worked once.

Of course, with less stringent criteria, any number of response categories might work.

The standard way to do this sort of comparison is to randomly divide the sample in half, and to use one question format in each. Then you can assess whether one format or the other produces higher correlations, etc.

But my question would be: why bother? Is this so important in the Indonesian context that you would risk losing the half of your sample that uses the non-standard format?

I have attached a review article that reviews the literature on various formats for Likert response scoring, which generally favors odd numbered responses over even numbered and 7-point over 5-point scoring.

I recommend that neutral option should be avoided for any meaningful results when applying Likert categorical scales in a survey.

In my opinion neither 5-point likert scale nor 7-point likert scale is better for a research. This method of data collection doesn't provide authentic result in research. People just tic the questions even without reading it. I think this is a weak method.

Dear Subir,

I'm agree with you in part, due to we have empirical evidence that linkert scale has bias in multiples ways ( Flaskerud, J. 1988; Voutilainen, A., Pitkäaho, T., Kvist, T., & Vehviläinen‐Julkunen, 2016; Dodds, P. S., Clark, E. M., Desu, S., Frank, M. R., Reagan, A. J., Williams, J. R., ... & Megerdoomian, K. 2015). However is a practical tool for qualitative approches that should be combined with other instruments such as deep interviews with the purpose of avoiding selection and perception's bias.

Riangga,

The 5 point Likert scale has greater likelihood of central tendency error with little variation whereas 7 point scale will offer a better range. Of course it would require use of principles of psychometric testing in deciding your response categories.

You may refer to:

Miller, LA; Lovler, RL & Mclntire: Psychological Testing: A practical Approach: Sage

I agree with David Morgan regarding the approach you can employ when conducting the test. The 7 point and 5 point Likert scale are both odd numbers and they are preferred to even numbers. The former gives raters the chance to choose more options, thus increasing the variability and precision. However, you shouldn't expect much difference after comparing both Likert scale, except if a difference of 2 score (7 vs. 5) will offer the raters a higher level of precision (depending on the measured variable). Correlation coefficient can be applied to see which of the scale correlate with the measured variable (after creating a criterion).