Hearted Youtube comments on Pete Judo (@PeteJudo1) channel.
-
31000
-
4900
-
4500
-
3300
-
2900
-
2600
-
2500
-
2100
-
2000
-
1600
-
1600
-
1500
-
1200
-
1000
-
924
-
912
-
871
-
833
-
832
-
777
-
738
-
613
-
436
-
427
-
411
-
389
-
373
-
308
-
304
-
275
-
261
-
251
-
229
-
183
-
175
-
169
-
163
-
152
-
147
-
140
-
As I understand it, p-hacking does not involve leaving out data points that are unsupportive of the original hypothesis (this is called cherry picking), but taking a dataset and coming up with alternative hypotheses until you find one that fits the data to a certain p-value (typically 0.05). This can happen unconsciously, for instance in biology (my background), looking for correlations in gene expression and a certain phenotype can easily lead to 'finding' a gene that correlates very well, corresponding to a hypothesis that this gene is causally involved in the phenotype. In this case it is called overfitting, and is not considered fraud, just bad statistics. Knowingly trying different hypotheses until one fits a given dataset gets much closer to fraud, and can be avoided by pre-registering the hypothesis that is going to be tested. This is better science, but with a lower chance of getting a high impact publication. Next on the scale are cherry picking, and then outright data fabrication...
116
-
116
-
114
-
111
-
101
-
89
-
81
-
70
-
70
-
70