Approaches to learning in pre-medicine: a multi-university mixed-methods study
Quantitative analysis
All statistical analysis results not presented in tabular format within the results section can be found in Supplementary file 2. One hundred and fifty-nine pre-medicine students participated in the study, of which 59 were in Bahrain, 83 in Dublin, and 17 were located in Malta, representing response rates of 49%, 52% and 100% at each site, respectively. Each site had a similar gender balance, with approximately 60% of participants being female. Prior to joining the pre-med foundation programme, 27% of students completed other foundation or preparatory programmes, 19% a Tawjihiya qualification, 19% a American High School Diploma, 16% a Canadian High School Diploma, 12% the International Baccalaureate, and 7% another curriculum. The Central European Framework of Reference for languages (CEFR) English level of all participants showed that Dublin participants, had the highest level of English language proficiency, followed by Bahrain and then Malta. Academic achievement of participants upon entry to the foundation programmes was available for participants in Bahrain and Dublin. Participants who were judged to have presented entry scores which significantly exceeded the minimum entry requirements were categorised as high achieving, while those who just met or marginally exceeded admission entry requirements were classed as low achieving. Fifty-one per cent of Bahrain participants and 80% of Dublin participants were ranked as having high academic achievement upon entry. The participant characteristics are summarised in Table 1.
ASSIST instrument reliability
The reliability of the ASSIST (18-item) short form has been reported previously as having Cronbach’s alpha values in the range of 0.65 and 0.82 [32, 33], with the Deep scale showing lower alpha values by comparison to the other scales. The alpha values for the whole study sample were shown to be Deep = 0.51, Strategic = 0.73 and Surface = 0.71. The alpha values for the Strategic and Surface scales indicate good internal consistency; however, the Deep scale falls below the accepted minimum alpha threshold value.
Predominant learning approaches
Medians and inter-quartile ranges (IQR) were calculated for the whole study sample and individual university sites separately. Whole sample median values for each scale were: Deep [23], Strategic [23] and Surface [19], Table 2. Deep and Strategic approaches to learning were identified as being more dominant compared to the surface approach at each of the sites.
Shapiro–Wilk tests indicated deviations from normality for all three ASSIST scales (Deep: W = 0.971, p = 0.002; Strategic: W = 0.965, p < 0.001; Surface: W = 0.980, p = 0.024; N = 159). Kruskal-Wallis tests showed a significant effect on the Deep scale, χ2 (2, N = 159) = 7.927, p = 0.019, and Surface scale χ2 (2, N = 159) = 16.853, p < 0.001, between the three different sites, but not on the Strategic scale χ2 (2, N = 159) = 2.391, p = 0.303. Dunn’s post-hoc tests with Bonferroni adjustment indicated that Dublin had lower Deep scores than Bahrain (z = −2.51, adj. p = 0.036, r = − 0.20). Dublin-Malta (z = − 1.90, adj. p = 0.172, r = − 0.15) and Bahrain–Malta (z = − 0.29, adj. p = 1.000, r = − 0.02) contrasts were not significant. On the Surface scale, Malta had lower scores compared to both Bahrain (z = 3.48, adj. p = 0.001, r = 0.28) and Dublin (z = 4.09, adj. p = 0.000, r = 0.324) and no significant difference was observed for Bahrain-Dublin (z = 0.767, adj. p = 1.0, r = 0.06). Fig. 1 displays a boxplot chart of Deep, Strategic and Surface scale scores by site (boxes = IQR, line = median, whiskers/outliers shown).
Boxplots of ASSIST deep, strategic, and surface scale scores by site (Dublin, Bahrain, Malta)
Preferred approach to learning
All participants were categorised as having a single or combined preferred approach to learning based on scores in the ASSIST scales using a method previously reported [38]. The preferred approaches to learning categories are: Deep, Strategic, Surface, Deep-Strategic, Deep-Surface, Strategic-Surface, No preference. The distribution of the preferred learning approach of students in the study sample is presented in Table 3. The dominant preferred approaches to learning of participants in the study sample were Deep (21%), Strategic (21%) and Deep-Strategic (29%).
Associations between demographic factors, background and approaches to learning
Considering the observed significant differences between Malta and the other sites on two of the ASSIST scales, and in addition to Malta following a different curriculum and assessment strategy than both Bahrain and Dublin, participants at the Malta site were excluded from all quantitative analyses assessing associations between student demographics, background, and approaches to learning.
Gender
Mann-Whitney U tests for the ASSIST scales across gender groups returned values > 0.05 for each of the Deep, Strategic and Surface scales, indicating that there is no difference in the distribution of any ASSIST scale according to gender (Deep: U = 2056, z = −1.48, p = 0.140; Strategic: U = 2349, z = −0.25, p = 0.805; Surface: U = 1975, z = −1.81, p = 0.070; N = 142).
English language proficiency
A Kruskal-Wallis test showed an effect of English Language CEFR level on the Deep scale, χ2 (2) = 7.56, p = 0.023, but not on the Strategic scale, χ2 (2) = 5.51, p = 0.064, or the Surface scale, χ2 (2) = 1.43, p = 0.488, (all N = 142). Dunn’s post-hoc test with Bonferroni adjustment indicated that B2 level students scored higher than C2 students (z = 2.73, adj. p = 0.019, r = 0.23). Differences between C2-C1 (z = 1.41, adj. p = 0.158, r = 0.12) and C1-B2 (z = 1.128, adj. p = 0.777, r = 0.10) were not significant (Fig. 2).
Boxplot of ASSIST deep, strategic and surface scale scores by English language CEFR levels (B2, C1, C2)
Education background
Participants were grouped according to the type of curriculum studied immediately prior to joining the pre-medical foundation programme (e.g. AHSD, CHSD, IB, Prior preparatory/foundation programme, Tawjihiya, and other). Curricula which only featured a small number of data points (Irish Leaving Certificate, A-Levels, Norwegian High School Diploma, CBSE) were categorised as other curricula. Kruskal-Wallis tests showed a significant effect of prior education background on the Deep scale, χ2 (5) = 13.87, p = 0.016, but no effect on the Strategic scale χ2 (5) = 2.56, p = 0.768, or Surface scale, χ2 (5) = 4.19, p = 0.523, (N = 142). Dunn’s post-hoc comparisons with Bonferroni adjustment indicated students who had previously attended CHSD scored lower on the Deep scale than students who had attended another preparatory/foundation programme (z = −3.321, adj. p = 0.013, r = −0.279). All other pairwise interactions on the deep scale were not significant after adjustment (Fig. 3).
Boxplots of ASSIST deep, strategic and surface scales by prior education background
Academic achievement on entry
The Mann-Whitney U tests showed no effect of academic achievement on entry on any of the three ASSIST scales (Deep: U = 2206.5, z = −0.11, p = 0.910; Strategic: U = 2425, z = 0.84, p = 0.403; Surface: U = 2423.5, z = 0.83, p = 0.407; N = 142).
Predictors of academic performance in the pre-medical foundation programmes
Academic performance data were available for all participants from the Bahrain and Dublin sites. Participants at these sites follow the same curriculum and sit the same summative assessments. The Shapiro–Wilk test indicated that academic performance scores did not significantly deviate from normality, W(142) = 0.98, p = 0.090.
Site, gender and English language proficiency
Independent samples t-tests showed no difference in academic performance during the foundation programme between Dublin and Bahrain sites, Welch’s t(125.46) = 0.772, p = 0.441, and by gender, Welch’s t(129.40) = −1.439, p = 0.153. A one-way ANOVA showed English language CEFR levels also did not impact academic performance, F(2,139) = 1.605, p = 0.205, η2 = 0.023 (N = 142).
Education background
Prior education background was found to influence academic performance by a one-way ANOVA test, F(5,136) = 2.94, p = 0.015, η2 = 0.097. The Games-Howell pairwise test identified that Tawjihiya students had significantly higher academic performance than IB students (Tawjihiya: M = 74.15, SD = 11.73, n = 13; IB: M = 60.32, SD = 10.42, n = 19; mean difference = 13.84, adj. p = 0.024; Hedges’ g = 1.23). No other pairwise differences were significant after adjustment.
Academic achievement on entry
Unsurprisingly academic performance was higher for students with high levels of academic achievement on entry (n = 95, M = 68.29, SD = 11.20) compared to those with low levels of academic achievement (n = 47, M = 59.34, SD = 10.72), Welch’s t(95.48) = 4.61, p < 0.001, mean difference = 8.95, 95% CI [5.10, 12.81], Hedges’ g = 0.81, 95% CI [0.44, 1.17].
ASSIST scales
Pearson correlations were calculated for all three ASSIST scales and academic performance in the foundation programmes (N = 142). Academic performance was positively correlated with the Strategic approach, r = 0.23, p = 0.007, but not with Deep, r = 0.08, p = 0.33, nor Surface, r = − 0.16, p = 0.06. Among the ASSIST scales, Strategic correlated positively with Deep, r = 0.21, p = 0.013, and negatively with Surface, r = − 0.21, p = 0.013; the Deep–Surface correlation was not significant, r = 0.06, p = 0.47. Although statistically significant, these effects were small in magnitude, indicating weak associations that warranted further analysis.
The associations between the ASSIST scales and academic performance were examined using multiple linear regression. This model was run to confirm whether the ASSIST scales contribute to academic performance beyond other demographic and background factors. Model 1 (Site, Gender, CEFR, Prior Education Background, Academic Achievement on Entry) was significant, R = 0.419, R2 = 0.176, adj. R2 = 0.145, F(5,136) = 5.79, p < 0.001, explaining roughly 14.5% of the variance in Academic Performance. Within Model 1, Academic Achievement on Entry was a positive predictor (B = 8.04, β = 0.32, p < 0.001), and Prior Education Background, a positive predictor (B = 1.51, β = 0.20, p = 0.016); Site, Gender, and CEFR were not significant. Model 2 added the ASSIST scales (Deep, Strategic, Surface) and improved fit, R = 0.477, R2 = 0.228, adj. R2 = 0.182, F(8,133) = 4.91, p < 0.001, yielding ΔR2 = 0.052 over Model 1, explaining 5.2% additional variance. In the final model, Academic Achievement on Entry (B = 8.05, β = 0.32, p < 0.001) and Prior Education Background (B = 1.26, β = 0.16, p = 0.042) remained significant positive predictors. However, none of the ASSIST scales reached conventional significance (Deep: B = 0.23, β = 0.07 p = 0.42; Strategic: B = 0.43, β = 0.16, p = 0.053; Surface: B = − 0.29 β = −0.12, p = 0.162). Notably, the bivariate correlation between the Strategic scale and Academic Performance (r ≈ 0.23) attenuated and lost conventional significance once we controlled for prior achievement and background factors in the regression, indicating the influence of covariates.
Qualitative analysis
Twenty-five participants from the three campuses took part in the qualitative phase of the study. Once transcription was completed, the deductive coding approach was employed, and the key findings are described below.
One important finding was the tendency for students to adopt a subject-specific approach to learning even when they were utilising a preferred learning approach. They described a tendency to flip-flop between learning approaches depending on the science subject they were learning.
“I’d say it probably depends on the subject. Does that make sense. For physics, sort of calculation that type of thing, I tend to probably approach towards the strategic. Does that make sense. I’m very much studying for the exam. I have a very organized way that I’m preparing for the exam. For things that I’m really interested in, and I think that I can cover a lot of ground in, I tend to be a deep learner.” (ASSIST-07-DUBLIN).
“I feel I’m a bit of the deep learner and mostly strategic learner but just a bit of a deep learner. On subjects that, I can like, I feel passionate about.” (ASSIST-02-DUBLIN).
“I try to do strategic mix between. I preferred deep when I was really interested in the subject. So, if there’s something I come across that I’ve genuinely really like I’ll try to go to that and find more information about it. But if it’s something that I just, that I understand I’ll move on. But as far as the course I have to know what it’s going to come in the exam and I have to do well. I’ll just learn over that. But if I like something I’ll go deep.” (ASSIST-02-BAHRAIN).
During the interviews, students were given the opportunity to reflect more deeply on the notions of deep, strategic, and surface approaches to learning. For each learning approach, we identified key characteristics contributing towards defining the image of each type of learner. The findings are detailed in Table 4.
Those who adopted a deep learning approach embedded science into medicine, connected concepts, and sought greater meaning.
Others using a surface learning approach were afraid of failing, adopted a passive approach to learning, and employed memorisation as a study aid.
Participants applying a strategic approach to learning optimised their learning approach by concentrating on organising their learning, became proactive learners and used academic performance as a driver.
In summary, the three main approaches to learning were described by the majority of the interviewees. Most did not adhere to a single approach; instead, they used a mixture of approaches to ensure that they had a firm appreciation of lecture material. However, the single preferred approach that stood out in our interviews was the surface approach, which was adopted by those who struggled with time management and used reading alone as their passive learning method.
link
