Dear All:
Is it necessary to evaluate the outlier in PLS?
If necessary, how to test it?
How to evaluate the Outlier?
-
christian.nitzl
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi,
When we use Likert scale or similar scales, usually we do not have a normal distribution of the frequencies, then to use this kind of distribution to identify outliers (like z greater than 3 or 4, or even Mahalanobis distances) do not make sense. In these cases I prefer the approach used by ESS EduNet (bellow).
..., it is better to drop those who did not respond to many value items or who did not discriminate in their responses to the value items.
[…] and those who have given the same answer to more than 16 of the 21 value items. Delete these respondents from the file.
16 / 21 = 76%
The European Social Survey Education Net. First round of preparation, cleaning and recoding. Available at: <http://essedunet.nsd.uib.no/cms/topics/1/4/2.html>.
Best regards,
Bido
When we use Likert scale or similar scales, usually we do not have a normal distribution of the frequencies, then to use this kind of distribution to identify outliers (like z greater than 3 or 4, or even Mahalanobis distances) do not make sense. In these cases I prefer the approach used by ESS EduNet (bellow).
..., it is better to drop those who did not respond to many value items or who did not discriminate in their responses to the value items.
[…] and those who have given the same answer to more than 16 of the 21 value items. Delete these respondents from the file.
16 / 21 = 76%
The European Social Survey Education Net. First round of preparation, cleaning and recoding. Available at: <http://essedunet.nsd.uib.no/cms/topics/1/4/2.html>.
Best regards,
Bido
Outliers
Hello, I am new in pls and smartpls and I´ve go a doubt. You have diffetent latent variables which are composed by different observable variables, how can you find an outlier? in a lineal regression is easy but here some people can be outliers in one observable variable related to one latent variable but the people´s responses are "normal" data in others observable variables. Thank you
Please, Would you give me a reference paper or some in regard to quoted remarks?Hi,
When we use Likert scale or similar scales, usually we do not have a normal distribution of the frequencies, then to use this kind of distribution to identify outliers (like z greater than 3 or 4, or even Mahalanobis distances) do not make sense. In these cases I prefer the approach used by ESS EduNet (bellow).
..., it is better to drop those who did not respond to many value items or who did not discriminate in their responses to the value items.
[…] and those who have given the same answer to more than 16 of the 21 value items. Delete these respondents from the file.
16 / 21 = 76%
The European Social Survey Education Net. First round of preparation, cleaning and recoding. Available at: <http://essedunet.nsd.uib.no/cms/topics/1/4/2.html>.
Best regards,
Bido
Thank you in advance
Regards,
Ken
Re: How to evaluate the Outlier?
Dear Dr.Diogenes,
I am a doctoral student from India.
Please could you point me to a reference article on this one.
"When we use Likert scale or similar scales, usually we do not have a normal distribution of the frequencies, then to use this kind of distribution to identify outliers (like z greater than 3 or 4, or even Mahalanobis distances) do not make sense. In these cases I prefer the approach used by ESS EduNet (bellow)."
As you rightly said, "usually we do not have a normal distribution of frequencies". My data (got 265 samples) is completely non-normal (all 45 indicators are non-normal based on Kolmogorov-Smirnov/Shapiro-Wilk tests), highly skewed, leptokurtic and I see a lot of outliers (all my data is on a 1-7 Likert scale).
Definitely, answering 1 or 7 on the Likert scale cannot be considered as outliers and they are valid responses.
Thanking you in advance for your help.
Thanks & regards,
Arun
I am a doctoral student from India.
Please could you point me to a reference article on this one.
"When we use Likert scale or similar scales, usually we do not have a normal distribution of the frequencies, then to use this kind of distribution to identify outliers (like z greater than 3 or 4, or even Mahalanobis distances) do not make sense. In these cases I prefer the approach used by ESS EduNet (bellow)."
As you rightly said, "usually we do not have a normal distribution of frequencies". My data (got 265 samples) is completely non-normal (all 45 indicators are non-normal based on Kolmogorov-Smirnov/Shapiro-Wilk tests), highly skewed, leptokurtic and I see a lot of outliers (all my data is on a 1-7 Likert scale).
Definitely, answering 1 or 7 on the Likert scale cannot be considered as outliers and they are valid responses.
Thanking you in advance for your help.
Thanks & regards,
Arun