What's the recognized criteria in statistics world?


Your question is closely linked to the one found here relating to the magnitude of slopes in order to determine if an item contributes a meaningful amount of information (I assume you meant to say standardized factor loadings rather than Cronbach's alpha < .3). Basically, if an item correlation is too low with the latent trait being measured then it's likely of little use, and if an item has too high of a guessing parameter then the same occurs as well (too many low-ability individuals can answer an item correctly seemingly just by chance).

  • $\begingroup$ I just need a recognized criteria, maybe citation paper. $\endgroup$ – WhiteGirl Jul 22 '17 at 6:53
  • $\begingroup$ +1. However note that we can't be sure the OP meant factor loadings. They were speaking about alpha anyway. The very question was quite unclear to me. $\endgroup$ – ttnphns Jul 25 '17 at 7:53
  • $\begingroup$ @WhiteGirl: last time I checked, there is none -- especially statistically speaking. Closest to what you're looking for, from a test development point of view, is the TIMMS2007 technical report. Otherwise consult Handbook of test development (2nd Ed), chapter 20, Item Analysis for Selected-Response Test Items. But AFAIK both are not discussing benchmarks for guessing parameters. $\endgroup$ – Tom Jul 26 '17 at 6:25
  • $\begingroup$ @Tom,it's so strange, why not recognized standard in statistics world?? $\endgroup$ – WhiteGirl Jul 27 '17 at 16:14
  • 2
    $\begingroup$ Because there is no single 'statistics world', there is just statistics applied in various disciplines/domains? Because rules of thumb aren't particularly strong arguments for choosing courses of action? $\endgroup$ – russellpierce Aug 3 '17 at 9:38

Not the answer you're looking for? Browse other questions tagged or ask your own question.