We know outlier obs should be remove from factor analysis.

Attach plot of this column data Attach plot of this column data.

What kind of outlier should be removed and use which function in R?

  • 3
    $\begingroup$ To me, the histogram doesn't indicate any outliers. It's a nice right-skewed distribution. If your goal is make it more symmetric or normal, something like a log transformation might do the trick. $\endgroup$ Jul 9, 2017 at 13:28
  • 2
    $\begingroup$ But I am very disturbed by the statement that we should remove "outliers" from the data. How would the results have any meaning or integrity if you just remove data that don't fit some pre-conceived distribution?? Think about a real example. If you have a town with a few unusually wealthy people, if you remove their data, how would the resulting statistics on the weath of the town have any meaning? $\endgroup$ Jul 9, 2017 at 13:56
  • $\begingroup$ That's why I need standard in statistics world.I read textbook that outlier ob should be removed before factor analysis $\endgroup$
    – WhiteGirl
    Jul 26, 2017 at 3:21
  • $\begingroup$ I understand your frustration, but you have to understand that the world is not that simple. Some information, even in a textbook, may be less-than-ideal. In truth, anyone can write a textbook, and anyone can post something on the internet. Also, there are different practices in different sub-disciplines. A standard approach in agriculture may not be the standard approach in ecology. There is also the issue that the practice of analyzing experiments has changed a lot over the past several decades. (Cont) $\endgroup$ Jul 26, 2017 at 12:50
  • $\begingroup$ (Cont.) Before the widespread use of computers, there were limitations to how data could be analyzed. Some techniques have become obsolete, but are sometimes still taught as a standard practice. I suspect that some of this talk about "removing outliers" stems from a time when the only available techniques were really bothered by outliers in the data. Unfortunately, it is difficult to find a source that tells you everything you need to know, even about a single test. It often takes cobbling together information from several sources. And then a year later you find something you like better. $\endgroup$ Jul 26, 2017 at 13:05

1 Answer 1


When the data don't fit the model, don't change the data, change the model.

I would only remove outliers for factor analysis if the data were entered incorrectly or clearly wrong (e.g. a 7 meter tall human). Because if you remove correct data from your sample, then the factor analysis is fitting a non-random sample of your data - indeed, a deliberately biased sample of your data. What will you do with the output?

However, what remedy I would take depends on the individual case, and, since you have not told us anything about your data or what you are trying to do, there's not much more to say.

  • $\begingroup$ Before factor analysis, we may delete some items based on KMO.Is that right? $\endgroup$
    – WhiteGirl
    Jul 26, 2017 at 3:22
  • $\begingroup$ I would not remove outliers based only on the results of any statistical measure. If the outliers are real, then they should be included. But you may need to use a different method. $\endgroup$
    – Peter Flom
    Jul 26, 2017 at 11:46
  • $\begingroup$ If you would not remove outliers,the factor analysis may cannot going on.Do you agree? $\endgroup$
    – WhiteGirl
    Jul 26, 2017 at 13:05
  • $\begingroup$ No. I don't agree. But you might have to use something for ordinal data, or you might want to transform a variable (for substantive reasons). Or something else. It depends on the case. $\endgroup$
    – Peter Flom
    Jul 26, 2017 at 21:16
  • $\begingroup$ remove outliers is typical step before factor analysis, I am sure. $\endgroup$
    – WhiteGirl
    Jul 27, 2017 at 16:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.