Timeline for Bad sampling or just bad correlation?
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 19, 2022 at 10:17 | comment | added | Mive | If such low correlations could have significance, how could you filter potential variables? Wouldn't the resulting coefficients of a multilinear regression model highlight a variable's significance more than the correlation coefficient? | |
| Jul 18, 2022 at 2:46 | comment | added | Tanner Phillips | Correlation by itself does not imply correlation. If you're sample size is large enough, a correlation of 0.001 could be significant. As to your point about normality, depending which derivation of the Linear Model you use (MLE vs LSE), the assumption of normality is.... lose at best. Yeah if your data is wildly skewed you might have issues, but your data is plenty "normal enough" that it isn't skewing your results. | |
| Jul 16, 2022 at 7:37 | comment | added | Mive | The registry has the following definition of "surface": "The usable area of a residential object". I can distinguish between "building purpose", e.g. residential, school, sports, shop, industrial... Current plots include all buildings regardless of the type os use. | |
| Jul 15, 2022 at 23:43 | comment | added | dipetkov | Does "surface" account for different number of floors? Do you distinguish between commercial and residential buildings? | |
| Jul 15, 2022 at 21:04 | history | edited | kjetil b halvorsen♦ | CC BY-SA 4.0 |
appended answer 582123 as supplemental
|
| Jul 15, 2022 at 19:30 | comment | added | Mive | I agree. I tried it under the assumption that some oddly specific values might dominate the data (surface around 40 and 60m2) which show both in the scatter plot and histogram. I felt that missing data in between might skew correlation results and the filtered data looked way more "natural". It felt wrong though, hence my question. | |
| Jul 15, 2022 at 15:29 | answer | added | Tanner Phillips | timeline score: 0 | |
| Jul 15, 2022 at 15:18 | answer | added | Ben | timeline score: 0 | |
| Jul 15, 2022 at 15:16 | comment | added | Eoin | Expecting to find a strong correlation and then changing your analysis until you find what you expect might be common, but it's very very bad practice. If the results don't match your expectations, consider that your expectations might be wrong. | |
| S Jul 15, 2022 at 14:59 | review | First questions | |||
| Jul 15, 2022 at 15:31 | |||||
| S Jul 15, 2022 at 14:59 | history | asked | Mive | CC BY-SA 4.0 |