0

As I understand it, it depends on the test to be used whether the screening for outliers and the test for normality is performed within individual groups or with the total sample. If groups are to be compared in subsequent analyses (e.g. 2 groups with a t-test), the screening is performed within each of the groups. In contrast, data screening for linear regression would be performed with the total sample.

If I run a linar regression with one dichotomous predictor (group), the result is the same as with a t-test. Yet, if I were to follow the above-mentioned rule, I would go about data screening differently depending on the type of test I use.

Here are my questions:

  1. Is the scenario described above correct in principle (i.e. data screening would take place within groups for one test and overall for another)?
  2. If yes, could somebody please explain (in not too technical terms) why the approaches differ although statistically I would get the same p-value from the two approaches if the same dataset is used? OR If not, is there a different rule that could be applied and would not result in the above-mentioned scenario?
grey
  • 625
  • 1
  • 4
  • 16
  • Similar Qs (testing for normality) https://stats.stackexchange.com/questions/224673/difference-between-normality-of-residuals-vs-normality-in-each-group, https://stats.stackexchange.com/questions/139440/anova-testing-assumption-of-normality-for-many-groups-with-few-samples-per-grou. For the outlier Q you must give more details, please. – kjetil b halvorsen Aug 11 '20 at 03:37
  • I am not quite sure I understand how the linked question answers the same issue as the one I raised; if it does, a hint would be appreciated. RE the outlier Q: I suppose it is possible, for instance, that a score is considered an outlier when analysing the groups separately but not in the overall sample. If this outlier were say windsorised (or whatever approach is taken to deal with it), the numbers we work with would differ depending on whether outliers are identified in the separate groups or the overall sample. – grey Aug 11 '20 at 07:41

0 Answers0