Downplaying the Harpp-Hogan index
Having had a thorough look at three data sets, I cannot recommend that the H-H index be relied on to identify student pairs who might be suspected of cheating. My findings, as reported above, do not support those found in Harpp, Hogan, & Jennings (1996).
Nonetheless I may continue to push data sets through Lertap, looking for patterns in the H-H values it produces. It is apparent that really extreme H-H values, such as the 26.00 and 7.50 figures discussed above, probably indicate cheating. It might also be noted that Harpp, Hogan, & Jennings (1996) suggest caution whenever the exam scores of the paired students are high. Their suggestion is to raise the caution flag whenever test scores are in the 90% and above range. If I brought this down to 80% and above, the lowest three of the five Table 1 pairs would drop out, leaving only those H-H values corresponding to significant SCheck outcomes; 24 of the 40 cases from the first test center B data set would go missing; and (remarkably) all but one of the five cases in Table 3 data set would disappear, leaving just the single H-H value with a significant SCheck outcome (H-H=1.19).
Downplaying the Harpp-Hogan indexHaving had a thorough look at three data sets, I cannot recommend that the H-H index be relied on to identify student pairs who might be suspected of cheating. My findings, as reported above, do not support those found in Harpp, Hogan, & Jennings (1996).Nonetheless I may continue to push data sets through Lertap, looking for patterns in the H-H values it produces. It is apparent that really extreme H-H values, such as the 26.00 and 7.50 figures discussed above, probably indicate cheating. It might also be noted that Harpp, Hogan, & Jennings (1996) suggest caution whenever the exam scores of the paired students are high. Their suggestion is to raise the caution flag whenever test scores are in the 90% and above range. If I brought this down to 80% and above, the lowest three of the five Table 1 pairs would drop out, leaving only those H-H values corresponding to significant SCheck outcomes; 24 of the 40 cases from the first test center B data set would go missing; and (remarkably) all but one of the five cases in Table 3 data set would disappear, leaving just the single H-H value with a significant SCheck outcome (H-H=1.19).
การแปล กรุณารอสักครู่..