The Study Of Multiple Choice Questions In Mathematic Of National Entrance Examination

Objective: Make a comprehensive analysis of the multiple choice questions in Hunan Mathematical Paper of National Entrance Examination in 2007, to examine this type of items' adaptability.Method: We used three fact Logistic Model of Item Response Theory to analyze the difficulty,discrimination and the degree of guess of the paper, used three different methods to analyze the attraction of the distracters, including High-low group method,point biserial correlation and a new point biserial correlation method. To analyze the reliability of the multiple choice items,the un-multiple-choice items and the whole test reliability, we used General Theory and wrote a mGNOVA program.Result: Through item analysis based on IRT, we found that the difficulty of most items were between (-2, +2), only one item (NO.3 in liberal art paper) was lower than -2.5. The discrimination of most items were above 0.3, only two of them (NO.7 and NO. 10 in liberal art paper) were lower than 0.3, need modified. All items'degree of guess were lower than 0.3. The analyze of the attraction of the distracters showed that the distracter B of NO.7 item and the distracters A,C of NO. 10 item in liberal art paper had big problems, because their distract coefficient were positive. The Generalizability Coefficient of multiple choice items and un-multiple-choice items were 0.57476 and 0.84164(science paper: 0.57571 and 0.79014), the Composite Generalizability Coefficient was 0.83682(science paper: 0.81651).Conclusion: The multiple choice questions in Hunan Mathematical Paper of National Entrance Examination in 2007 showed an acceptable difficulty,discrimination and degree of guess. The result of the attraction of the distracters was not as good as expected, three of them had big problems, need modify or delete. The GT analyze showed that the reliability of multiple choice items were not as good as un-multiple choice items, but both of them were had high general coefficient. Differ the number of items, the result showed that more items had higher reliability.