PDF
Citation
Print
INTRODUCTION
Today’s physician needs a strong foundation of medical knowledge coupled with non-cognitive skills. These skills include effective communication, teamwork, ethical behavior, and displaying the highest level of professionalism when treating patients. The use of the Medical College Admission Test (MCAT) is pervasive in the admissions process of all United States (US) medical schools. However, this test only focuses on medical knowledge, and neglects the aforementioned non-cognitive areas above. Could there be a way to assess non-cognitive domains? For years, admission committees have strived to assess non-cognitive skills in their candidates using standard interview day experiences, but have not identified an ideal assessment tool. The most recently proposed non-cognitive assessment tool is the multiple mini interview (MMI), which has some evidence of validity regarding the interpretation of scores [1]. However, the MMI is time- and resource-intensive and, with almost 50,000 students applying to multiple US medical schools [2], it is not feasible for every school to have all of their candidates complete a MMI. In Europe, medical schools and residency programs have used the situational judgment test (SJT) to assess the non-cognitive skills of applicants. They recognize that superior medical knowledge is not enough to become a physician, and the SJT likewise focuses on areas such as ethics, communication skills, teamwork, and professionalism. The interpretation of SJT scores has shown promise in terms of validity evidence for medical schools in the United Kingdom and Belgium [3-5]. To date, no study has been published on the use of a SJT in US medical school admissions. Since the admissions processes and education levels of students applying to medical school are different in the US and Europe, validity evidence is needed for the interpretation of SJT scores in US applicants. Thus, the purpose of this study was to develop a SJT blueprint, drawing upon Kane’s validity argument framework.
METHODS
Participants
The participants were 489 applicants selected for assessment/ interview day at the University of Utah School of Medicine during the 2014-2015 admissions cycle. Selection for assessment/interview day is based on achieving a minimum level of performance in seven areas: undergraduate grade point average, MCAT score, community/volunteer service, leadership, research, physician shadowing, and patient exposure. A total of 41 assessment days were scheduled, and one of five SJTs was used on each assessment day, such that 84-109 applicants completed each SJT. Applicants also completed a MMI and a traditional interview on assessment day.
Situational judgment test
A SJT includes a series of dilemmas. For each dilemma, examinees are asked to rank five responses, from the most appropriate to least appropriate action, or to select the three most appropriate actions (out of eight) to the dilemma. We piloted five SJTs and each test assessed one of five non-cognitive domains: professionalism, coping with pressure, communication, patient focus, and teamwork. These domains are commonly used in European medical school and residency selection decisions. Since this was a pilot, we wanted to see how each domain performed alone with as many items as possible, which is why we did not require applicants to answer questions belonging to multiple domains. The elements of each domain are presented in Appendix 1. Each SJT included nine select-best-three and nine rank-order items. The items were constructed by the Office of Admissions, in conjunction with the Dean of Admissions and members of the Admissions Committee. The overall structure of the SJT was adapted from European reference books [6-8]. An educational expert and admissions committee member who also directs medical school courses proofread all 90 items and five non-medical experts each reviewed 18 items for clarity and comprehensibility.
Two types of SJT items were used (select-best-three and rank-order), and for all items applicants read a dilemma (item stem) with no clear single solution. For select-best-three items, applicants selected the best three options for dealing with the dilemma from a list of eight options. For rank-order items applicants ranked five options based on their appropriateness to the situation, from 1 being the most appropriate to 5 being the least appropriate. Appendix 2 contains an example of each SJT item type. Applicants had 36 minutes (two minutes per item) to complete the SJT. All examinations were administered on iPads using SoftTest. Select-best-three items were worth 12 points total, with 4 points given for each correct option selected. Rank-order items were worth 20 points if all options were in the correct order, 17 points if only the first and last options were in the correct order, 14 points if either the first or last option was in the correct order, and 10 points if the first and last options were off by no more than 3 ranks. For example, if the correct order was ABCDE and an applicant answered ABCDE, he or she would receive 20 points; if an applicant answered ACBDE, he or she would receive 17 points; if an applicant answered BCDAE, he or she would receive 14 points; and if an applicant answered CAEBD, he or she would receive 14 points.
Kane’s validity argument framework
We selected Kane’s validity argument framework because it allowed us to determine which types of validity were most important and the order in which validity evidence should be collected, which earlier frameworks were not able to do [9]. In Kane’s validity argument framework, the first step is to determine the use and interpretation of scores from an assessment. The next step is to derive assumptions from the use and interpretation, much like creating hypotheses. The third step is to test the weakest assumptions of the validity argument by collecting evidence in a step-wise fashion for four areas: (1) scoring, (2) generalization, (3) extrapolation, and (4) implication inferences. Scoring inference deals with the quality of assessment items, raters, and how an actual assessment score is computed. Generalization inference deals with how well the sample of assessment items adequately represents the domain(s) of interest. Extrapolation inference deals with how well the assessment measures real world performance. Implication inference deals with the effect of scores on an assessment (e.g., 70% equals passing) and therefore the consequences of those decisions (e.g., a student with a 68% requires remediation). Depending on the presence of gaps between evidence and assumptions, the assessment may need to be revised or may not be recommended for further use. Thus, it is important to review evidence at each stage in Kane’s framework and use that information to drive future evidence collection or to make revisions, rather than gathering all possible evidence at once.
For step one in Kane’s validity framework, the use of the SJT is to assess non-cognitive skills in applicants to medical school and the interpretation is that higher scores would indicate better non-cognitive skills. Therefore, the assumption (step two) is that applicants with higher SJT scores would turn out to be better practicing physicians because they have effective interpersonal, communication, teamwork, ethics, and professionalism skills applicable to patient care. We developed a pool of SJT items and conducted this study to gather evidence for scoring inferences and generalization inferences (step three), so that a test blueprint could be constructed for future use in gathering evidence for extrapolation and implication inferences.
Data analysis
All data were analyzed with IBM SPSS ver. 21.0 (IBM Co., Armonk, NY, USA). For scoring inference, we investigated how each domain performed in terms of item difficulty (average point value) and discrimination (corrected point-biserial). Item difficulty ranges from 0 (no one answered the item correctly) to 1.00 (everyone answered the item correctly). An item should have a difficulty index between 0.30 and 0.80, meaning that 30%-80% students answered the item correctly, in order to be considered good [10]. Values below 0.30 suggest the item is too difficult, and values above 0.80 suggest that the item is too easy. In either case, the item would need to be revised for future use. We also determined if any item’s difficulty value was two standard deviations above or below the mean. Discrimination indexes range from −1.00 to 1.00. A good discrimination value for corrected point-biserial computation is 0.30 or higher [11]. Values of 0.11-0.29 are fair, while values below 0.11 are poor and suggest that the item should be omitted or is in need of major revision. For generalization inference, we investigated the internal consistency of item scores in each domain with Cronbach’s alpha and measured the concordance of two education experts in the categorization of items by each domain. A Cronbach’s alpha coefficient of at least 0.70 is considered good [12], while a coefficient below 0.70 suggests that major revision is necessary or that more assessment methods should be used in conjunction with the examination. Percent agreement values were provided as a descriptive measure, so we did not specify a cut-off for what was constituted good concordance between the experts.
Scoring and generalization evidence was used to select the best items from the pool of 90 multiple-domain SJTs. Specifically, items with difficulty values two standard deviations or more above or below the mean were omitted from the question pool. Once the best items were selected and/or constructed (if necessary, based on how many items were omitted), two educational experts categorized the items to determine how well the pre-determined domain matched their categorization and if items assessed multiple domains. The percent agreement was computed between the educational experts for each of the five domains. Specifically, the number of items that the experts agreed upon was summed and divided by the number of all items in each category. The educational experts used the descriptions in Appendix 1 as a rubric for categorization. Items that assessed multiple categories or items that the educational experts could not reach agreement on were revised or omitted from the question pool. Since each item that the experts disagreed on was discussed, we only provided percent agreement as a descriptive measure and did not compute Cohen’s kappa. The final goal was to have 60 items for a two-hour test.
RESULTS
The average difficulty and discrimination index for each SJT domain are presented in Table 1. The difficulty index ranged from 8.27 (effective teamwork) to 9.94 (coping with pressure) out of 12 points total for the select-best-three items and 13.72 (commitment to professionalism) to 15.55 (effective communication) for the rank-order items. Two rank-order items had a difficulty index two standard deviations below the mean of all rank-order items. One select-best-three item had a difficulty value two standard deviations below the mean, and one select-best-three item (effective communication) had a difficulty value two standard deviations above the mean for all select-best-three items. All four of these items were omitted from the question pool. The average discrimination was fair (0.11 to 0.20) for rank-order items in the coping with pressure, communication, and patient focus domains and also for select-best-three items in the communication domain. All other average discrimination values were poor (<0.10). Ten rank-order and seven select-best-three items had negative or zero discrimination values, and these 17 items were omitted from the pool.
Cronbach’s alpha results for the 69 items (33 rank-order and 36 select-best-three) selected out of 90 items suggested that the items in each domain had low internal consistency (coefficients <0.60). Additionally, the majority of the items (63%) categorized by the two education experts appeared to be assessing multiple domains (e.g., communication and teamwork). Specifically, 80% of patient focus items, 72% of teamwork items, 71% of communication items, 67% of coping with pressure items, and 33% of professionalism items assessed multiple domains. The concordance of categorization between the two educational experts was higher for the items in the teamwork (72% agreement), professionalism (78% agreement), coping with pressure (76% agreement) and patient focus (70%) domains than in the communication domain (24%).
Final testing blueprint
Based on the generalization inference evidence, we decided to omit the patient focus category from the final testing blueprint. Initially, patient-focused care was felt to be extremely important to the staff and faculty at the University of Utah School of Medicine. However, the dilemmas for these items took place in a hospital environment and may unfairly advantage applicants with prior medical experience. Communication appeared to be present and required for handling many of the SJT dilemmas. Thus, we decided to tighten the description of the communication category to focus specifically on how information should be delivered and received in the context of others. Therefore, we used four domains for the testing blueprint: communication, teamwork, professionalism, and coping with pressure. We did note that rank-order items required more time (and thus cognitive load) for the applicants to complete. These items were also difficult to score with our current testing software. Therefore, we decided to include 10 select-best-three and five rank-order items for each of the four domains, ultimately obtaining a final SJT with 60 items.
DISCUSSION
This is the first study to report the development of a SJT for use in the admissions process to a medical school in the US. Based on Kane’s validity argument framework, we were able to create a testing blueprint for future use. The generalization evidence suggested that SJT domains do not have good internal consistency, and two raters found that multiple domains were assessed by many items. Therefore, more generalization evidence is needed for the final 60-item test before moving to collecting extrapolation and implication evidence according to Kane’s validity argument framework. Few studies in medical education have used Kane’s validity argument framework for developing and validating assessment scores, so this study will provide an example to other health professions educators and administrators.
Several limitations of this study should be considered. First, we sampled a low number of items for each applicant, which was reflected in the low discrimination and reliability values. Since this was a pilot, we wanted to ensure that fatigue did not factor into applicants’ selection of answers to the items and also wanted to be able to assess a larger number of items. Second, the SJT questions could have been circulated among applicants, since the assessment dates spanned many months. On the morning of their assessment or interview day, applicants were asked to sign a confidentiality statement in which they agreed not to reveal the SJT scenarios to anyone. The admissions staff monitored known pre-medical student websites and did not see any posts of sensitive SJT material. However, the possibility still exists that this information could have been leaked. Third, this study was conducted among applicants to one medical school. Thus, future validity evidence collection needs to focus on applicants at multiple medical schools. It should also be noted that the majority of the SJT items were found to assess multiple domains and the reliability was low for each domain. However, this finding was based on only two experts’ opinions about categorization. Thus, more evidence is needed for generalizability, and more experts may be needed to categorize the items.
The results of this study will help medical school administrators determine how to begin constructing a SJT, and we identified important factors to consider based on Kane’s validity argument framework. Specifically, we will continue to build our SJT question bank with more of a focus on dilemmas that assess a single non-cognitive domain. Further testing of the validity argument framework with a more representative sample, perhaps including multiple institutions, is needed to determine if the SJT is the optimal assessment tool for measuring the non-cognitive skills of medical school applicants. The SJT is one of many possible ways to assess non-cognitive skills. Data from such testing will need to be followed to determine whether the SJT has long-term predictive value. If successful, the SJT could be included as a portion of the initial assessment of applicants, in either a testing format such as the MCAT, or perhaps as part of an online application prior to granting an interview.
XML Download