The more, the best. With less Likert-like options there is possibility to loose some information. With more options it would be possible to better discriminate on the measured concept . Psychological constructs are mostly measured on a 5-point scale. 

Less options are better for children and older, or in the case you want to force participants to answer in a more limited way. 

Dear Atiq,

Of the three options about which you have enquired I would definitely recommend 7 point scales because they provide richer data whilst still remaining valid because working memory studies suggest that people are able to discriminate between up to 7 items at once.

Another important issue is whether you are going to anchor all the point values with an interpretation. We recommend only anchoring then end values (e.g. 1 = strongly disagree and 7 = strongly agree). If you do this then you can treat the data as scale and calculate scale statistics such as a mean and a standard deviation. If you anchor all the point values to an interpretation then your data type is ordinal which means you shouldn't do this.

 Dear Dejan Stevanovic and Peter Samuels, thanks for your comments

Try to use 5 point Likert scale and even better if you use 7 point Likert scale. If you use 3 point, then there are chances that you lose some information

One thing to think about when using such a scale with odd numbers is that there is always a neutral point. For example, on a 5 point scale you might have strongly disagree, disagree,  neither agree or disagree,  agree and strongly agree. Some participants will give a higher proportion of neutral responses because they don't want to say one way or another, and this can lead to response biases. 

Whenever you are using an odd numbered Likert scale it is important to keep in mind that you need to have a balanced scale with the middle response position representing an appropriate response.  For example, if you have a five point Strongly Agree to Strongly Disagree scale the appropriate middle response position is Neither Agree Nor Disagree.  The reason for this is that without a balanced and appropriate response scale you cannot assume that you have a quasi-interval scale and, therefore, cannot use parametric statistics to analyze your data.  Secondly, you need to use response scale abbreviations for the respondees to use as they will attribute a different set of values to a numerical scale.  For example, using AS, A, NDNA, D, and DS is interpreted differently than 1, 2, 3, 4 and 5, unless each numerical value is specifically defined.  That being said, it is also important to define every point on the scale since not doing so will result in an unreliable response.

Of course, the more the better as a wider range of choices are offered to respondents. Also more accurate data is possible with a 7-point scale, for example. But the Likert Scale, no matter how many point scale it is, needs to be triangulated via for e.g. interviews as the data may not be diverse or rich enough especially in exploratory studies. 

Ahmad Muhammd Shakil, Erin Bailey , Lawrence M. Aleamoni and Debbie G. E. Ho thanks for your comments

Like all of them said, the more the better. But the problem with likert-scale is that the relevance of the response categories/options, through the sample assessed can't be investigated. Sometimes, researchers realize that subjects could not even discriminate three-level of likert scale when using Rasch model to investigate the discrmination of the response categories/options. From my little experience and through litterature, people are more able to disciminate between 4 and 5 levels. So I may advise you 5-level, but I think you'd better use Rasch analysis to analyse the discrimination of the five levels after data collection.

Thanks!

Dear Segnon Sogbossi, thanks for your comments.

If you want a Likert scale then it needs to be five points.  A 3-point and a 7-point scale will have different characteristics.  A Likert scale must have symmetry around a neutral point and asks the respondent to evaluate a set of statements using the scale.  The values are then summed to provide a scale score.  You can appropriately apply parametric statistics to this suumative score.

Several comments are very good: labeling response categories, etc.  The traditional research in psychometric argues for a five-point scale to maximize discriminability.    See the following articles by myself and co-authors:                                                     Response Alternatives in Likert Scaling, Educational and Psychological Measurement,  v44, n3, pp 563-67. Aut 1984 .                                                                                                   Two Methods of Scoring Discrepancy Format Needs Assessment Instruments, The Journal of Experimental Education, v55, No.1, pp 39-42, 1986.

Hello there, 

I generally 'vote' from more answer categories regarding Likert type format of answers. I just gives more opportunities to respondents who may be reluctant to answer the question with 3 categories only. Also, in analysis like SEM, you may treat your variables then as either categorical or continuous.

Regards, Witold

In 3 point scale one has very less scope in answering and we want to measure social or personal issues which can be definitely rated with more no.rating scale.As these are all minimum to maximum observations and its interpretation as per the analysis.

Dear Rajashree Kapure, Witold Orlik, William L. Johnson and Joyce Morris thanks for your comments.