Timeline for Distinguishing Bad Leverage Points from Vertical Outliers
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 10, 2020 at 10:47 | comment | added | user215517 | The x coefficient with all outliers omitted was 0.968 and had a standard error of 0.022. Adding in the "good leverage" point didn't bias the coefficient, with the coefficient becoming 0.999 but the standard error has now decreased to 0.009. The addition of this single point has made the null hypothesis less tenable in a disproportionate way. Also, note that whether an outlier is a "good" or a "bad" leverage point depends on what we were expecting: for an n-shaped association (with, say, a quadratic term to accommodate this), the "good" and "bad" leverage points could swap labels, for example. | |
| Jan 7, 2020 at 12:31 | comment | added | Pedro Alonso | Thank you for taking the time to help! I get it. Just would like to ask one thing: what exactly does this mean - "adding the "good outlier" will have relatively little effect on the estimated slope but will decrease the standard error (and not in a good way)"? | |
| Jan 6, 2020 at 18:40 | vote | accept | Pedro Alonso | ||
| Jan 6, 2020 at 12:04 | history | edited | user215517 | CC BY-SA 4.0 |
Added a small clarification at the end.
|
| Jan 6, 2020 at 11:55 | review | First posts | |||
| Jan 6, 2020 at 14:17 | |||||
| Jan 6, 2020 at 11:53 | history | answered | user215517 | CC BY-SA 4.0 |