I have some converted Likert-scale to a metric scale as dependent variable (after testing Internal Validity and Reliability of generated scale I combined 6 ordinal Likert-scale into a single score for the scale) with nonmoral distribution and I have read "outliers do not really exist in Likert-scales", even if you remove them, more will show up, it is like a slippery slope, as my sample size is big enough (more than 200) should I consider its distribution as normal and conduct parametric tests?
Also check https://core.ac.uk/download/pdf/195151451.pdf
Hello Mina,
Be cautious in deciding to censor, exclude, or Winsorize extreme scores. These can be legitimate, albeit "rare" (for your sample) values. That's not to say that you should blithely accept "outright liar" cases, as might be true if you had evidence that someone had considered the response scale to have been reversed from its true orientation.
You can always opt for resampling/bootstrap tests if you are concerned about distributional shape.
Finally, if you do decide to discard cases, it would probably be worth reporting the results both ways (with and without your "outliers").
Good luck with your work.
Dear Mina Sufineyestani ,
By virtue of the discreet nature of Likert scale data bounded on an interval of two extreme values (left and right), such data then cannot be normally distributed. Normality test methods such as the Kolmogorov-Smirnov test (which performs badly on data with single outliers, 10% outliers, and skewed data at sample sizes
@Mina I agree with @Elvis that absolute normal can not be achieved for Likert scale data. For normality assumption, you can use Skewness and Kurtosis for outliers. Refer to the link below for further details on data screening.
http://statwiki.kolobkreations.com/index.php?title=Data_screening
Check https://statisticsbyjim.com/hypothesis-testing/analyze-likert-scale-data/
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size