SETTING & MAINTAINING EXAM STANDARDS

Raja C. Bandaranayake

DEFINITIONS
Standard setting is a process of determining how much is good enough. The standard or criterion level of performance is a point on the scale of measurement at which separation of competence and incompetence occurs. Cut-score, cut-off score or passing score represents this standard on a given test for making decisions pertaining to the purpose for which the test was conducted, e.g., to certify competence.

ERROR IN MEASUREMENT
True score is a conceptual measure indicating true extent of competence in a given subject, e.g., Anatomy.
Observed score is the score assigned as a result of taking a test, say in Anatomy. The difference between true and observed scores is indicative of the amount of error in the measurement. The reliability of a test and the associated standard error of measurement are estimates of the amount of error in the measurement.

DECISION ERRORS
False positive: passing an incompetent examinee False negative: failing a competent examinee

NORM- & CRITERION-REFERENCED STANDARDS NORM-REFERENCED Relative CRITERIONREFERENCED Absolute Not related to peer performance Standard set prior to exam Referenced to a defined level of performance

Based on peerperformance
Varies with each group Cut-off point not related to competence

METHODS OF STANDARD SETTING 1. Test-centred methods

Standards derived from hypothetical decisions based on test content before the test is answered. Standards derived from reviewing examinees performance before deciding cut-off score.

2. Examinee-centred methods

3. Compromise methods

Provide flexibility for adjusting the standard based on the examinees performance on the test.

NEDELSKY (1954) METHOD:

Example

Consider N judges and n MCQ items of 1 in 5 type Judge A identifies 2 options in item 1 as those which a minimally competent examinee should eliminate as incorrect. MPL for that item for Judge A [MPLA1] = 1/(5-2) = 1/3 Similarly, in item 2 he identifies 3 options, giving an MPLA2 = 1/(5-3) = 1/2 He repeats this process for each item. The exam MPL for Judge A [MPLA] = MPLA1 +MPLA2 + MPLA3 + .MPLAn

Similarly, Judge Bs MPL [MPLB] is determined

The MPL for the exam (= cut-off score) is: (MPLA + MPLB + MPLC +.MPLN) / N

ANGOFF (1971) METHOD

Example
N judges consider 100 minimally competent examinees taking an MCQ exam of n items. Judge A estimates that, of these examinees, 50 should answer item 1 correctly, 20 item 2 correctly, 70 item 3 correctly, and so on to item n. The MPL for Judge A [MPLA] = (0.5 + 0.2 + 0.7 + . xn) / n X 100 = (say) A%. Similarly, for Judges B, C, D, E, ..N, the MPLs would be B%, C%, D%, E% N%, respectively. The MPL (cut-off score) for the exam is: (A% + B% + C% + D% + E% +....N%) / N

EBEL (1972) METHOD

Example
Assume that Judge A assigns items in a 200-item MCQ test to the cells of a relevance-by-difficulty matrix, as follows. He then estimates the percentage of items in each cell of the matrix that a minimally competent examinee should be able to answer correctly (as indicated within the cell). Each cell also includes the products of these two values. EASY 15 x 100% = 1500 20 x 80% = 1600 10 x 50% = 500 MEDIUM 25 x 80% =2000 40 x 60% =2400 25 x 40% = 1000 15 x 20% = 300 HARD 10 x 60% = 600 20 x 50% =1000 5 x 10% = 50 5 x 0% = 0

ESSENTIAL IMPORTANT ACCEPTABLE

QUESTIONABLE 10 x 30% = 300

EBEL (1972) METHOD - contd.

Example

The MPL for Judge A [MPLA] is then: (1500 + 1600 + 500 + 300 + 2000 + 1000 + 300 + 600 + 1000 + 50 + 0) / 200 = 56.25 % Similarly, the MPL for Judges B [MPLB], C [MPLc], D [MPLD] ..N [MPLN] are determined. The MPL for the exam (cut-off score) is: (MPLA+ MPLB+ MPLc+ MPLD + ..MPLN) / N

PROPOSED EBEL MODIFICATION

EASY MEDIUM HARD

ESSENT.

6x 100% = 600

12 x 80% = 960

7 x 50% = 350

IMPORT.

12 x 80% = 960

24 x 60% = 1440

19 x 40% = 760

ACCEPT.

5 x 60% = 300

12 x 50% = 600

3 x 10% = 30

MPL:

=600 + 960 + 350 + 960 + 1440 + 760 + 300 + 600 + 30 =6000/100 = 60

HOFSTEE METHOD

fmax
Failure Rate%

20

15

fmin

10

35

40

45

50

Cut-off score(%)

cmin

cmax

Example

HOFSTEE METHOD

A plot of cut-off scores for a given exam against resulting failure rates is given

cmin = 40% cmax = 45% fmin = 10% fmax = 20%

A = point representing cmin,fmax B = point representing cmax,fmin Line AB intersects the curve at a cut-off point of 42.5% Thus, operational cut-off score = 42.5%

CUT-OFF SCORE FOR 1 IN 5 MCQ [FRACS PART 1] Probability of guessing (=1 in 5) = 20%

Total ignorance score

Maximum possible score Effective range of scores 100% Mid-point of this range

= 20%
=100% = 20% to = 60%

Additional factor (as PG exam)

= 5%

Nominal cut-off score (60%+5%) = 65%

CUT-OFF SCORES: MARKER QUESTIONS

1. Comparison of exam scores

Mean score in this exam:

56.7%

Average exam mean score over last 4 years: 59.4% Thus mean score in this exam is: Assuming this candidate group is of same standard as in last 4 yrs, this exam is: 2.7% lower

2.7% harder

CUT-OFF SCORES: MARKER QUESTIONS - contd.

2. Comparison of marker scores

Mean score in this exam on previously used questions (N=162): Mean score on same questions when they were each last used: Thus, when compared with previous candidates, this group of candidates, on these items, scored (62.5-60.5)% = 62.5% 60.5%

2.0% higher

Thus this group of candidates is: than previous groups

2.0% better

CUT-OFF SCORES: MARKER QUESTIONS contd.

3. Estimating examination difficulty

Thus it is expected that their mean score in this exam would be: 2.0% higher But their mean score in this exam is: 2.7% lower Thus this exam is really: 4.7% harder

CUT-OFF SCORES: MARKER QUESTIONS contd. 4. Determining cut-off score The cut-off level for an average exam is: Thus the cut-off level for this exam should be (65 4.7)% = 65.0%

60.3%

Cut-off score = 60.3%

HOFSTEE CURVE

20

Failure Rate%

15

10

55

60

65

70

Cut-off score(%)