I am not sure what you would classify as an outlier with a Likert scale variable? By the very nature of the variable, you have both a floor and ceiling (ie., 1 and 5, or 1, and 7, etc.), and this type of variable is ordinal. Thus, it seems imprudent to discard individuals because they responded at either the low or high end of this narrow spectrum. 

Likert scale: ordinal or continues?  This has been debated for years and I assume the debate will continue for years to come.  Many of us in the social sciences consider variables built from  Likert scales continuous in nature.  If the case is that you sum the answers to of Likert scale question to form scales then you can test for outliers.  However the approach I would take would be to test your scales for normality and normalize (perform a log conversion of) skewed scales so that your data is normally distributed and thus you no longer have to worry about outliers.    

Dear Ariel and Peter,

Thanks for the answers. As Peter suggested, I browse and read many articles about Likert scale debate, and  managed to normalize my dataset. I also found that using group Likert-scale data (instead of Likert-item) could be much of help, considering my relatively small samples. That solved my problem. Thanks again...

Hi Rizki

Agree with Peter and Ariel. To add further into this valuable discussion, youmay want to consider these suggestions:

• Analyze > Descriptive Statistics > Explore...
• Select variable (items) > move to Dependent box.
• Click Statistics... > tick Outliers > Continue... > OK.
• In Output window: Go to Boxplot > Look at circles and *. These are potential outliers. If there's none, then there's no potential outliers in your dataset.
• If there are circles or *, then there are potential outliers in your dataset. 
• To check if the outliers affect your data: In the output window: Look at Descriptive table > Compare 5% trim mean and mean values. If there's a large difference between these values, then there's huge possibility that your further analyses, e.g. correlation and  regression, will be affected. 
• 5% trimmed mean is the mean that slashes out 5% of the extreme ends (both lower and higher ends) of your dataset.

You may want to check this booklet for detail. 

All the best.

Saiyidi

note: Vote up my answer if you find this is useful. Cheers.

Article Introduction to SPSS

Thank you for your valuable and helpful answer, Saiyidi.

Nice booklet too :)

Hi Rizki, 

I do not agreed with this afirmation" As mentioned in Hair, et al (2011), we have to identify outliers and remove them from our dataset" because in statistics it is better to have a model with poor quality but adapts to reality than a better model with no aderence to reality.

is the statician that has to decide to maiintain or delete the outliers. 

Hi Helena,

Well, I do decide not to remove the outliers from my dataset. As you said, the model may have poor quality but it reflects the reality. My model actually is not very poor at final result, it works as I expected. Thanks :)

Dear Rizki

I am glad for you.

Have a nice time

Helena

NEVER remove outliers. you may wish to check our new paper

Article TO DETERMINE SKEWNESS, MEAN AND DEVIATION WITH A NEW APPROAC...

Good day to All, i am a new researcher. the multivariate detection of outliers can be done in SmartPLS?

Mehmet Guven Gunver

Hoe to run your approach using spss?

Greetings Dicky Wiwittan ,

unfortunately, it has not been implemented to spss yet. If you want, you can use the matlab code.

sincerely

In fact, you need to remove records that contain only one item of the questionnaire as an answer. This answer is wrong.