Remember confusing mass of stats, lets look again (briefly)

Can't Find This Yet : CORRESPONDING QUESTION PRINTOUT

• total = total number of students

• high, mid, low --> number in each group
  
  • changes for each question though (omits)
    
  • bit more complicated in how divided up

• omit = how many didn't write this one
  
• NF = not finished, omitted everything here on
  
• NR = no response = omit broken down by hi/mid/low group
  
  

• total across the top = % that chose that alternative
  
• test score means = average on 60 item test that students choosing this alternative got
  
• discriminating power -->that's what our "d" means:
  
• high group minus low group

• RPB = point biserial correlation (R= correlation)

• CRPB = corrected point biserial is point biserial, not counting this question
  (want these around .200 --> but have to look at each item
  --> may be valid but different sort of topic)

• IRI= Item Reliability Index ((combines CRPB and Difficulty into one number, but since we look at CRPB and Difficulty directly, don't need it))

• Standard Error of Discriminating Power = standard deviation of the error column of discrimination --> don't need to know about? -->zero because such huge numbers to play with

• Con= if did this infinite number of times would get a frequency distribution of point biserial; 95% of corrected point biserials would be within this range)

Can Do item analysis on students responses to PROVINCIAL EXAMS

• doesn't just have to be test questions you wrote
  
  
• after every exam, dept. publishes item analysis for that exam