A long road of thorough statistical analysis of my tests and revision of some of them has come to its end. From now on, already existing tests (including IIC 16) will remain as they are with no further withdrawal or revision/changes.
In my humble opinion, high range tests should be seen as mental puzzles; however, we must get out of them as much science as it can. Creating tests with high reliability is the first and most important step. My intention is to offer quality tests with high reliability (>0.8), so as to provide testees with a really close approximation of their true potential. Inflated, overwhelming scores, are pleasant, indeed. But, being first of all honest with ourselves, are they true? Are they even close enough? The point is to create tests that, besides giving the pleasure of solving, they also give some truth. That is, tests that besides being puzzles, are, as valid and reliable as possible, psychometric tools, too.
So, after some comprehesive item analysis (Pearson Correlation Coefficient,Discrimination Indexes for higher-lower 33% and 50% and Difficulty) for all items of every test, the following revisions were made (or not) :
- NPRA : NPRA (50 items, Cronbach’s alpha 0.73) was revised to NPRA-R (32 best items of NPRA, Cronbach’s alpha 0.91). Although an alpha of 0.73 is acceptable for a psychometric tool, an alpha of 0.91 definitely provides remarkable reliability.
- NRA, NRA II and NGT-F : These 3 numerical tests included some items of high quality, but they included some of very low, too. So, best items of these 3 tests were chosen and revised into NGT-R (30 items). Preliminary estimation (N=8) shows a Cronbach’s alpha > 0.8. We should wait until test reaches 30 submissions for an official norm report and a more valid statistical analysis.
- VGT, VRA and VRA II : These 3 verbal tests show all high reliability (> 0.81), so they will remain as they are. More details will be found at their stats page. The only change is that alternative answers on 2 items were reconsidered on VGT (showing plain logic and high correlation with overall scores), so some scores will raise and norms will adapt.
- SRA-F and VSIT-R : These 2 spatial tests will remain as they are, showing high reliability (>0.81). For VSIT-R, estimations are still preliminary.
- N-VRA : Still preliminary estimation, but quite high Cronbach’s alpha of 0.86. Not expected to change much, so test will remain as it is.
- IIC 2016 and NRA-X : It’s too early to speak in statistical language about IIC 2016, but according to current submissions, items show good correlation, so IIC 2016 will remain as it is and will continue to function as a test after Contest’s end. It’s too early to speak about NRA-X, too, but things seem fine. This probably happens due to the fact that I have been working on tests’ statistics some time before publishing NRA-X and IIC 2016, so items for IIC 2016 and NRA-X were very carefully designed.
So, more details will be found at each test’s stats page. More crucial statistical values will be added, soon. I have to recheck them and this takes time.
I have to thank everyone for their support. Fun is good, but combining fun with science is even better. Good luck with my tests and have fun!