I want to detect multivariate outliers in my dataset, which contains participant responses to various questionnaires such as DASS-21, PSWQ etc. Should I compute Mahalanobis distance using total scores for constructs such as depression, anxiety and worry or should I use item-level data from the questionnaires before aggregating them into their total scores? When using item-level data, 25 participants are detected as outliers among 370 participants, but when using the total scores, only 1 participant is detected as multivariate outlier.
Hi,
In general, when using the Mahalanobis distance method to detect multivariate outliers, it is recommended to use item-level data rather than total scores. The reason for this is that using total scores can sometimes mask the outliers present in the individual items, as in your case where using total scores resulted in only 1 participant being detected as an outlier compared to 25 participants when using item-level data.
Using item-level data, you can detect outliers at a more granular level and potentially identify participants who exhibit unusual patterns of responses across multiple items. This can provide more insights into the outliers' nature and help understand the underlying reasons for their unusual responses.
Therefore, in your case, it would be more appropriate to use item-level data from the questionnaires before aggregating them into total scores when computing Mahalanobis distance to detect multivariate outliers in your dataset.
@Jennifer Santos , As I did not use chat GPT, I do not have access to this platform at all. Because it is banned and sanctioned in our country. But I want to know what is your problem?
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude