Julius Daugbjerg Bjerrekær edited section_Abbreviating_ICAR_label_sec__.tex  over 8 years ago

Commit id: b7ce90670fefe60b3780d5f908ce71825b7e0e7a

deletions | additions      

       

Several methods can be used to pick out and determine which items should be included in an abbreviated ICAR test. One way would be, as done in a recent study (\textbf{cite}), to use a evolutionary algorithm to search the composition space of desired item combinations. The desired item combinations can then be measured for their correlations between it's total test result and the total test results of ICAR16. A evolutionary algorithm will not try all available combinations, but try a few and systematically change and improve those. Unfortunatly this method will only provide solution as a local maximum. To find the best combination one can use exhaustive search - though this is far more resource intensive, it's doable since finding 5 combinations out of 16 possibilities is a relative small task. \par  Something which haven't been touch on yet is which results from ICAR16 of which the ICAR5 is drawn from. Sicne ICAR5 was to be used on Danish pupils, the first attempt was to use results from two Danish tested versions of ICAR16 - both of which had modest sample sizes of \textbf{X} and \textbf{Y}. An algorithm was written to try all possible combinations from the 16 items to find the 5 items which were to be used in ICAR5. The correlation was a mere \textbf{0.ZZ}, which made us look elsewhere for a possible higher correlation. A new set of data from \textbf{QQQ}, which was built in the Psych packages in R studio. TEST. studio, which the the same algorithm was applied to. \par  \subsection{old shit}  studies using the ICAR16 revealed that many subjects thought that the test took too long. Especially the rotation items required time to complete. This resulted in fewer and smaller samples which is highly problematical for social science and has been so for decades. As in most cases, there is a trade-off to be made between speed and accuracy. We estimate under these conditions that it's better to use abbreviated scales to boost samples at the expense of construct validity.\\