hi dears,
I got 267 survey responses (5-likert scale) asking for self-evaluating individual ability, 6 of responses all answered "1" (represent 'has no ability doing XXX) - i guess when they clicked their mouses, option '1' is the easiest for them to click (maybe they don't want to answer my survey at all, because the survey has 30 questions, there's no reason someone would score all of their individual abilities as the worst ). the mean of all items are around 3.0 to 3.8.
can i identify those 6 responses as outliners? i know some mathematical approachs such as Mahalanobis distance method, but can i identify outliners just because they are contrary to common sense?
many thanks!!!!
I think it is a better reason to exclude participants that apparently did not follow or understood the instructions than to solely rely on statistical tests. The latter do not show you how, why and where the outliers are and in large samples, I would simply expect "unsusual" or "more extreme" values, which are just a natural phenomenon.
What you could do is to calculate your analyses with and without your "outliers" and check how much the results change. I think the best strategy is to openly communicate what you have done and how the results are for both analyses, if quite different (if they are essentially the same you can just report that).
You could use the explore feature in IBM SPSS Statistics to identify outliers by creating box plots and stem-and-leaf plots that make it easier to spot outliers according to the responder number. Dealing with outliers depends on whether a researcher has a clear explanation for their existence (or lack thereof). For helpful insights, you might go through the outliers subsection (pp. 85–93) in Hair et al.’s (2019) textbook, fully cited below.
Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2019). Multivariate data analysis (8th ed.). Cengage. https://www.cengage.uk/c/multivariate-data-analysis-8e-hair-babin-anderson-black/9781473756540/?searchIsbn=978-1-4737-5654-0
Good luck,
Outliers are of two types: rogues and blunders. Rogues are observations that are unusual but real. Blunders are observations that are errors. It's remotely implausible that the six outliers are rogues. Unless they all have a massive inferiority complex! I'd drop them as nonsense data.
Patient says to psychiatrist Mine is probably nothing at all compared with some of the inferiority complexes you must have seen, doctor!
Ronán Michael Conroy hi Ronán thanks!
Mohialdeen Alotumi hi Mohialdeen thank you!
Rainer Duesing hi Rainer many thanks!
Outliers in surveys or questionnaires are very different from those that appear in instrumental measurements.
As Ronan rightly says, these 2 types behave differently in the responding subject, since there is the concept of social desirability.
For the same reason in Psychometrics, Response models based on the complexity of the topic.
In this regard appear:
a) outliers per subject:
- Ambiguous, tends to generate a high frequency at the tips (1 - 5)
- Invariability in the answer (responds by obligation)
- No answer, (selection of the subject)
b) Outliers per item
- invariability
- High frequency no response
- No answer, (item problem)
it is better at this point to evaluate by RASCH.
I hope it works for you,
Success.!!!
- Badges
- Science topic
- Similar topics
- Mathematics