Construction, administration and quality assurance of student assessment in HPE - Part II

Kathmandu, Nepal

In the last post, I have described the rationale and steps followed for creating SISAQ or PBQ. This post deals with the cut-off score (pass mark) determination of the SISAQ. Item analysis of SISAQ will be discussed in the next post.

SISAQ mostly test the Higher Order Thinking Skills (HOTS) and it will be usually difficult for the minimally competent "borderline" candidates. Since it does not allow the test-takers to see the options as in MCQ, it eliminates the chance of "guessing". Further, SISAQ questions are NOT direct so the test-takers should extract the answers based on the clinical/lab/public health based scenario and vignettes, which is again a challenging task for the borderline students.

Unlike OSPE/OSCE where students are observed for 5 - 8 minutes while doing the task and then awarding a holistic "global" rating at the end to facilitate the calculation of the pass-mark, SISAQ cannot be observed for such as short span of time for each students. Thus, SISAQ pass mark is determined based on the compromised method using standard setting done a priori of the exam by the concerned faculty and performance of students (marks obtained) in the same SISAQ.

SISAQ standard setting of faculty:

 

Once the SISAQ construction is done as per the rules laid down in the previous blog post in this thread, group of faculty responsible for the particular SISAQ decides on the range of the expected marks for five mutually exclusive groups using required competencies.

 

The five groups are as follows:

Very Poor:

Poor:

Borderline:

Good:

Very Good:   

The most important group is the borderline group as it is used to determine the pass mark. Thus, it is very important to assign the range of marks for this group as it is based on difficulty level of scenario, vignettes and questions from each disciplines. Faculty trained in criterion-referenced student assessment method also know that they have to think of the "borderline" group of students as those who could either pass or fail the SISAQ while referring to the students in this range.

Once the SISAQ is administered and marked by the faculty then a compromised method known as "borderline regression method" is used to calculate the pass mark. Actually, it is a simple linear regression analysis where marks obtained by the students in the SISAQ are taken as the dependent variable (y) and five mutually exclusive groups for the SISAQ are taken as independent variable (x). As the five standard setting groups are ordinal in nature, they are coded from 1 to 5 for data analysis. Code of the borderline group (i.e. 3) is then used to estimate the pass mark using the regression equation obtained.

 

Illustration:

 

SISAQ # 2

 

Full marks: 15 marks

Total time: 25 minutes

Criterion-referenced standard setting:

1. Very Poor:    0 - 3

2. Poor:             3 - 5

3. Borderline:   5 - 7

4. Good:           7 - 11

5. Very Good:  11 - 15

I found Google docs very easy for running the simple linear regression using the in-built "forecast" function to determine the pass-mark of SISAQ. You can see the details following this link:

https://docs.google.com/spreadsheet/ccc?key=0AoJkXbvD07ECdHh0NmUyUkFjNnA4aXozWnlJajM4QWc&usp=sharing

The borderline-regression (BLR) method reveals the pass-mark of this hypothetical SISAQ as 6.242 (41.61%) and based on this pass-mark there are 4 students (20%) fail and 16 (80%) pass it. The pass-mark (6.242) was obtained from using performance of the students in the exam and it lies between 5 and 7 which are the minimum and maximum criteria set for passing the SISAQ by faculty.

 

The Google Spreadsheet link above also shows the outcome for this SISAQ if the conventional 50% pass mark is used. It reveals that only 60% of the students would pass and 40% of them will fail it, which is different than the one obtained from using BLR method above. This happened because the "fixed" criteria of 50% does not take account of the "difficulty" level of the SISAQ in reference to the "borderline" group of students and thus is "biased". We all know that every SISAQ cannot have same level of difficulty. Thus, it is "injustice" to have a "fixed" pass-mark for each and every item and tests. So, I strongly suggest to have fluctuating pass-mark based on its difficulty in terms of content and construct for SISAQs.

 

In sum, theory paper should have both MCQ and SISAQ to test the knowledge from LOTS to HOTS. They must be based on the Angoff or Hofstee or Beuk cut-off score for MCQ and BLR cut-off score for SISAQ in order to get "holistic" and "criterion-referenced" assessment of the knowledge for any competency based curriculum. This method is defensible as they are "always" based on the decision of a group of faculty rather than a single faculty and "mostly" they also used students exam scores.

 

Most importantly, these criterion-referenced methods provide defensible and justifiable grounds for scientific and evidence based student assessment method for the undergraduate as well as post-graduate health professions education program.

 

Rest later, happy SISAQing ...

Testing Post-partum hemorrage management skills using OSCEs

Kathmandu, Nepal

Developing and establishing Reliability and Validity of OSCE stations on PPH Management skills:

 

1. Write a checklist /series of checklists that can be assessed and feasible in 5 minutes (remember OSCE is usually of 5 minutes duration). If you want to test many things then break them down to multiple checklists to assess the "holistic" assessment of skills related to the topic. This will now become "modified OSCE".

 

2. You can take advantage of the currently available tools from articles, books, websites etc. to supplement/complement you own checklists. Make the skills and checklists "mutually exclusive" i.e. don't assess same thing on different checklists.

3. Once you develop the tools then discuss it in a team (experts in PPH management) to ensure the "content validity". Remember this will be iterative process and you will need to document the discussion in each step.

4. Give the tool/tools to review by the other content and non-content experts (GPs, Emergency Medicine, Public Health including statistician, Allied Health etc.) to ensure "face validity".

5. Think wisely if you want to use Standardized patients or real patients. You need to train them beforehand to get the "standardized" outcomes. Real patients are difficult to manage and handle (we recently faced a serious problem with our Obs/Gyn OSCE during formative exam as only 2 out of 8 volunteer patients were willing and available at the end of the OSCE circuit!).

6. Train the interviews rigorously to ensure inter-rater reliability.

 

7. Standard set each OSCE station using holistic/subjective approach i.e. use global rating along with the objective checklist ratings. These scores in combined will give you the "competency" required to pass each OSCE station using Borderline, Contrast group or Borderline Regression Method. One can also use Angoffing but it will be tedious process.

8. Once it is done, then pilot test the OSCE stations with volunteer participants (students, interns etc.) to know about the feasibility, problem in checklist and reliabilities. Use at least 10-15 volunteers in this process. Pilot test will also give you the internal construct reliability commonly known as Cronbach's alpha.

 

9. Use at least 2 raters in each station to calculate inter-rater reliability.

 

10. If you use more than 1 station then you can calculate the Generalizability (G) Coefficient which shows how much reliable the whole OSCE circuit. You need special software for this but it is available free from PERD, McMaster University, Canada.

 

11. If you also want to assess the test-retest reliability of your PPH OSCEs then run the pilot test two times using same interviewers and same candidates immediately after a break of, say, 30 to 60 minutes in the same day. If possible make sure not to contaminate your test-re-test reliability estimate i.e. don't let students refer the OSCE related resources from any references and interact on the OSCEs).

12. Modify the checklists after the pilot test if required and document them properly. Re-run the pilot test if the reliability estimates are not in the acceptable range. If they are acceptable then you are ready to run the "full fledge" OSCE on PPH management.

 

13. Calculate the relevant statistics, interpret the results, document them and publish it on MedEdPortal so that others can assess it when required.

 

This will complete all the reliability assessment for your PPH OSCEs:

I. Inter-rater reliability
II. Internal construct reliability
III. Test Re-test reliability

And remember that "Rome was not built in a day" so go slowly and document all the process. Your tool development and/or pilot testing phase can be your curriculum innovation project.

Once you run the full OSCEs then you can proceed with more complex statistical analysis to establish the "construct validity", "Predictive validity" & "Criterion validity" later ...

 

N.B. -

 

a. Same process holds true for pre-validate and/or validate the OSPE/OSCE stations properly. 

 

b. If the tool needs to be used in the local language (say Nepali) then it should be translated from English to Nepali first by a professional translator/expert bilingual followed by back-translation into English using translated Nepali tool by another professional. Then original English tool and back-translated English tool should be compared by the study team and modifications should be reflected in the Nepali tool as well. This process will also be iterative one and must be documented properly. It can be resource demanding as well.

Standard Setting in Health Professions Education

Kathmandu, Nepal

Health Professions Education (HPE) is embracing the competency based curriculum, teaching and assessment cycle, which is quite different from the conventional objective driven approach.

 

The competencies laid down in the course and/or curriculum require a criterion-referenced standard setting system for an effective and defensible student assessment. Standard setting sessions require subject matter / content experts to determine the competencies for each item such as Multiple Choice Question (MCQ), Short Answer Question (SAQ), Objective Structured Practical/Clinical Examination (OSPE/OSCE).

 

One of the most widely used criterion-reference method of standard setting in HPE is Angoff method where judges review the items in terms of its content and difficulty levels and award score between 0 and 1. The underlying assumption of this score is related with the "Minimally Competent Borderline Candidate" or simple "Borderline" candidate who are those students who "either can barely pass or fail" in any examination. These are "hypothetical" students and thus a common understanding among the judges must be obtained a priori to the actual Angoff scoring session (Angoffing).

 

When the  individual scores of the judges for an item are averaged then the resulting value becomes the "cut-off score" or pass-mark for that particular item. So, if a test consist of 50 standard set items then the pass mark of the test is obtained as sum of the Angoff score of each of the items included in the test. This shows the "competency" required to pass the test.

 

Below I present a hypothetical example for a test with 5 items where each item is judged by a "mixed" panel of  six judges. With six judges, Angoff scores become reliable further adding weight in favor of its validity.

 

Example:	Judge1	Judge2	Judge3	Judge4	Judge5	Judge6	Angoff
Item1	0.45	0.5	0.55	0.4	0.6	0.55	0.51
Item2	0.4	0.35	0.45	0.4	0.45	0.4	0.41
Item3	0.5	0.65	0.6	0.55	0.6	0.5	0.57
Item4	0.7	0.75	0.8	0.65	0.7	0.65	0.71
Item5	0.65	0.55	0.75	0.7	0.55	0.6	0.63
Pass Mark	2.70	2.80	3.15	2.70	2.90	2.70	2.83
						Pass % =	56.5

 

One of the main problem of the Angoff method is the "content expert bias" which means that when standard setting is done by a group of experts belonging to the same discipline they tend to give higher score resulting the "upward bias" in the Angoff score by shifting the pass-mark higher. It has big consequences if the test is "high stake" as adjustments in the Angoff scores are not permitted. Thus, it is recommended to use a "mixed judge" panel to balance the scores in the standard setting process.

 

Another problem occurs when Angoff score is done for the very first time and there is a little discussion on the concept and meaning of the "borderline" student. As most of the HPE courses in South Asia including Nepal uses 50% cut-score, faculty here tend to give at least 50% marks to maintain this norm. I call this a "novice bias".

 

When "content expert bias" or "novice bias" occurs then a compromised methods like Hofstee and Beuk are recommended for formative examinations. These methods use the students' actual scores and Angoff scores to determine the adjusted pass-mark after correcting these bias if present. It is actually recommended to use these methods in the initial phase of the formative student assessment until a normalization effect takes place among the judges. For summative assessment, it is recommended to standard set the items with item analysis results and/or repeat Angoffing to reduce the biases.

 

If standard setting is done correctly then a test with pre dominantly easy items will have higher pass mark whereas a test with mostly difficult items will have lower pass mark. In other words, "competency" for a test is known to the students and faculty before the test is conducted.

 

At last, a test should typically be compiled based on examination blueprint as it allows the teachers/faculty to sample the contents from different sections and of varied difficulty. It also allows the faculty to choose the knowledge and skills using appropriate methods, which in turn increase the validity of the test. As knowledge can also be assessed in varying degree of difficulty, it is always better to use the "mixed bag" approach to get the sample of curricular contents, item difficulty and assessment methods for producing technically "competent" human resources for health.

 

More can be found here:

 

1. www.kumj.com.np/issue/33/3-4.pdf

 

2. www.act.org/research/researchers/reports/pdf/ACT_RR89-2.pdf

Rest later ...

So, happy Angoffing!