Research preparedness and competence among Nigerian medical and dental students applying for a research fellowship

Abstract

Background

Research skill acquisition is critical for undergraduate medical and dental students, but deficiencies in knowledge and motivation persist in low-resource environments such as Nigeria. Self-perceived research competence, motivations, and preparedness among applicants to a highly competitive research fellowship were evaluated, and predictors of high competence were identified, to provide information for capacity-building.

Methods

A descriptive cross-sectional analysis of 237 anonymised applications from Years 2 to 6 medical and dental students of the University of Nigeria Enugu Campus (UNEC) to the Medix Frontiers Research Fellowship (January–February 2026) was performed. Demographics, previous research experience (yes, no), competence (8 Likert items from 1 to 5), motivation (5 Likert items from 1 to 5), and preparedness (e.g., hours/week, willingness) were assessed. Calculated variables: competence_mean (mean competency scores), motivation_mean (mean motivation scores), high_competence (> 3.5), high_motivation (≥ 4). Cronbach’s α was used to evaluate reliability. Descriptive statistics: Means ± SD, proportions (%). Bivariate: Non-parametric tests (non-normal distribution data), Chi-square/Fisher’s exact test, Pearson’s correlations. Multivariable: Logistic regression (Dependent: high_competence; Independent: previous experience, Medix member, age, sex, stage of training, hours/week, dependable internet); linear regression for continuous competence_mean. No data were lost to follow-up. Analyses in Python (Pandas, SciPy, Statsmodels).

Results

Participants were predominantly young adults, with a slight predominance of females, and a majority at the clinical level of training. Prior research experience was generally limited, with a small number of participants reporting involvement in research projects, fellowships, or authorship. Despite excellent internal consistency of the competence scale, self-perceived research competence was generally low, and a limited number of participants reported high competence. In contrast, motivation to engage in research was high, although the interest scale had a somewhat lower reliability. Self-reported competence also differed significantly based on previous research experience, with higher levels observed among those with previous experience. Competence did not vary meaningfully by stage of training or gender. There was no meaningful relationship between motivation and competence. There was an almost universal willingness to participate in the fellowship among respondents, which meant that there was insufficient variation to conduct meaningful regression analysis. The study population comprised applicants to a competitive research fellowship and, therefore, represents a highly motivated subgroup of students; hence, the findings cannot be generalised to all Nigerian medical students.

Conclusions

These findings suggest that while research interest is high among applicants, gaps in practical skills persist and may be addressed through earlier and more structured training opportunities.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12909-026-09134-6.

Background

Research skills are a core aspect of modern undergraduate medical and dental education, as they form the basis for evidence-based practice, critical appraisal, and lifelong scholarly activity [1, 2]. Beyond technical proficiency, research training promotes critical thinking, problem solving capacity, and intellectual independence, skills that are increasingly being identified as important for engaging with the rapidly changing health care system [3]. Systematic reviews and student-centred studies demonstrate that early, prior structured research involvement is associated with increased scholarly productivity, greater confidence with research methods, and increased plans for future academic activities [1, 4]. These advantages extend beyond the individual level, adding to the institution’s research output as well as enhancing national health research capacity [5].

However, reviews of globally relevant literature also report that the quality and quantity of research training differ according to institutional and regional settings, and these differences create avoidable gaps in readiness among young clinicians who are the future leaders [6, 7]. In many high-income settings, research exposure is often embedded longitudinally within curricula through formal modules, protected research time, and mentorship frameworks [1]. In low- and middle-income countries (LMICs) such as Nigeria, structured research training remains limited, and undergraduate students are typically not afforded the formal mentorship and practical experiences that would enable them to acquire strong research skills [8, 9]. These discrepancies are especially significant, given the increasing demand for locally generated evidence to guide context-specific health policies and interventions [10]. Without sufficient early training, future clinicians in these settings may be less prepared to engage in research, critically assess evidence, or contribute to addressing region-specific health challenges [11].

While there has been a policy focus on the increasing recognition of the importance of integrating research into undergraduate medical and dental curricula, most research initiatives are still theoretical with little hands-on experience, mentorship, and project-based learning. As a results, many students are ill-prepared to make meaningful research contributions in key research processes such as study design, data analysis, scientific writing, and dissemination [12]. This gap between theoretical knowledge and practical competence represents a critical barrier to developing a robust pipeline of clinician-researchers within the Nigerian healthcare system.

Multiple structural and contextual barriers worsen this gap Nigeria. They include weak capacity for mentorship, inadequate research methodological training, and uneven availability of digital learning resources, which collectively inhibit students’ preparedness for research at the undergraduate level [8, 9, 12, 13].Additionally, competing academic demands and a lack of structured incentives for research engagement further reduce student participation [14]. Reports from institutional surveys in Nigerian contexts also demonstrate that while there is little formal preparedness for emergent fields (e.g., precision medicine), mentorship and resource limitations continue to serve as barriers to translating student interest in these topics into tangible research outcomes [15]. These results highlight an urgent need for locally developed, practical training models suitable for various stages of clinical undergraduate training [16].

Targeted research fellowship programmes represent one such model. By combining formal instruction with supervised project involvement, these programmes can bridge the gap between theoretical knowledge and applied research competence [17, 18]. The Medix Research Fellowship is one of such initiative. It is a flagship nine-month programme hosted by the Directorate of Research and Publications (RPD) at Medix Frontiers, a non‑governmental organisation (NGO) based in the University of Nigeria Enugu Campus (UNEC), Enugu, Nigeria, dedicated to health advocacy, research, and community engagement, particularly against HIV/AIDS and other prevalent health challenges. It offers a structured mentorship, practical research experience, and continuous training in relevant research methodology, scientific writing, and research ethics; fellows are integrated into supervised projects while obtaining specialised skills training [19].

Assessing applicants’ initial research skills, motivations, and enthusiasm to such programmes is critical for several reasons. First this provides a means to assess and identify current deficiencies in research competence and preparedness and hence guide the development of targeted training interventions [20, 21]. Secondly, it offers insights into the factors motivating students’ with research, which is important for sustaining participation and enhancing the quality of the programme [22, 23]. Third, it adds to the growing evidence base on undergraduate research training in LMIC settings, where there is still a paucity of empirical data [24]. However, it is important to recognise that applicants to competitive research fellowships constitute a self-selected group that may differ systematically from the general student population, particularly in terms of motivation and prior interest in research.

In accordance with STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines for clear rationale and objectives (STROBE items 2–3) [25], this study is thus designed to determine level of research competence as well as motivations and readiness to engage in research among the undergraduate medical and dental students of the University of Nigeria, Enugu Campus (UNEC) who are undertaking the Medix Research Fellowship. In addition, the study seeks to identify factors associated with higher levels of self-reported competence. We hypothesise that (a) previous research experience (project participation, prior training, authorship) and higher training stage will be associated with higher self-rated research ability and (b) equity variables (gender and training stage) are associated with differences in research competence and readiness to engage. These findings are expected to substantiate the need for targeted, stage-sensitive fellowship components, and contextually appropriate research training interventions in resource-limited academic settings.

Methods

Study design and setting

We conducted a cross-sectional survey of undergraduate medical and dental students who applied for the Medix Research Fellowship in the 2026 intake. After preprocessing and exclusion of ineligible applicants (n = 253; details of the exclusions are provided below), the analytical sample comprised 237 respondents. Data were extracted from self-completed Google Form application forms, submitted during the fellowship application window (December 2025–January 2026). The fellowship is organised by the Medix Directorate of Research & Publications (RPD) with its office in Enugu City, Nigeria, and the respondents were students enrolled in the University of Nigeria, Enugu Campus (UNEC). The study is reported in accordance with STROBE guidance for observational cross-sectional studies [25]. The final version of the questionnaire was administered in English. The full questionnaire is provided as Supplementary File 1.

Participants

Applicants were considered for inclusion in analysis if they were, at the time of application, registered students in Medicine or Dentistry (years 2–6) at UNEC. Stages of training were: Preclinical (years 2–3), Basic Clinical (year 4), Clinical (years 5 A, 5B, 6).

Applicants were excluded if they were, at the time of application, not registered students in Medicine or Dentistry (years 2–6) at UNEC. Therefore, applicants to the Medix research fellowship, who were not medical or dental students, were consequently excluded during preprocessing. In total, 237 responses were included for the analysis after these exclusions.

Verification of eligibility occurred by matching institutional identifiers supplied in the application (faculty/department and level of study) to the list of students admitted to the fellowship. The training stage was categorised based on the standard medical and dental curriculum structure at the University of Nigeria, Enugu Campus, into preclinical, basic clinical, and clinical phases, reflecting progressive advancement in academic and clinical exposure.

All eligible applicants who completed the baseline application instrument were included (convenience census of applicants). The study population represents a self-selected group of applicants to a competitive research fellowship and may differ systematically from the general student population [26].

All eligible applicants within the fellowship application window were included, representing a census of applicants. The questionnaire was completed by applicants as part of the Medix Research Fellowship application process, not as a separate research study instrument. This census method is suitable for descriptive and exploratory analytic objectives [27].

Participant flow

An initial total of 253 fellowship applications were received. Following the application of eligibility criteria, 16 participants were excluded: 1 was enrolled at ESUCOM, 3 were dietetics students, 8 were graduates or medical doctors, and 4 were public health students. The final analytic sample included 237 respondents (Fig. 1).

Fig. 1.

Participant flow diagram

Data variables and measures

Competence items were adapted from previously used tools evaluating research self-efficacy and critical appraisal skills in undergraduate medical education [28–30]. These tools assess fundamental research competence, such as literature searching, study design, data analysis, scientific writing, and research ethics, and have been widely used in medical student populations. The items were also aligned with well-established multidimensional models of learning and research self-efficacy from research in medical education. The scale was modified for use in undergraduate medical/dental students. No further psychometric evaluation was undertaken, although internal consistency reliability was investigated in this study. All variables were taken from the self-administered online fellowship application form. Key variables were:

Sociodemographic and training characteristics

The characteristics of the participants were: Age (in years, a continuous variable), sex (male or female), faculty (Medicine or Dentistry), level of study (Year 2–6), and stage of training (preclinical, basic clinical or clinical). Involvement with the organising body was recorded as Medix membership status (binary: yes/no).

Prior research experience

Prior research experience was determined by three dichotomous variables: previous involvement in a research project, previous involvement in a research fellowship, and previous authorship of a scientific publication (coded as 0 = no, 1 = yes for each).

Self-rated research competence

Eight items, each rated on a five-point Likert scale (1 = no experience; 2 = basic awareness; 3 = can perform with guidance; 4 = can perform independently; 5 = can teach others) assessed research skills. Competence was rated in (i) formulating research questions, (ii) searching literature, (iii) choosing study designs, (iv) creating instruments for data gathering, (v) analyzing data (using Excel, SPSS, R, or Python), (vi) writing scientifically with the IMRAD structure, (vii) interpreting research outcomes, and (viii) research ethics.

Motivation for fellowship participation

Motivation was measured with five items on a 5-point scale of importance (1 = not important, 5 = very important). Aspects included skill enhancement, publishing possibilities, the availability of mentors, career progression, and professional networking.

Readiness and commitment

Readiness to participate was evaluated through active participation in the fellowship (binary), availability of reliable internet access (binary), and agreement to the fellowship rules and regulations (ordinal).

Composite scores and dichotomous variables were created in advance for the analyses. Mean competence was calculated as the average of the eight competence items (continuous ranging from 1.0 to 5.0), and mean motivation was the average of the five motivation items (continuous ranging from 1.0 to 5.0).

High research competence was defined as a competence mean value ≥ 3.5, abstracted as overall competence corresponding to performance at or above the “can perform independently” anchor. High motivation was defined as a mean motivation score ≥4.0, suggesting high motivation to participate in the fellowship.

In multivariable analyses, age was treated as a continuous variable, and categorical predictors were dummy-coded as appropriate. Cutpoints for derived variables were established a priori to improve interpretability and minimise analytical flexibility.

Data sources and measurement

The application questionnaire was completed online, self-reported, and was intended for use in the programmatic selection process; items assessing competence and motivation were adapted from previously used education and medical student research survey instruments from the literature and were thematically mapped to core research activities [28, 29]. When applied, the item wording used established competency descriptors to enhance content validity.

Since all measures are self-reported, they are vulnerable to social desirability and self-assessment biases; no objective, independent skills testing was conducted. To address this, we (a) present internal consistencies (Cronbach’s alpha) for the derived scales, (b) perform additional sensitivity analyses with the continuous competence scores and alternative cut-offs, and (c) discuss the constraints of self-assessment in the Discussion [31]. The application tool did not contain a dedicated social desirability scale.

Ethical considerations and data availability

The analysis was based on de-identified application data, which were gathered for the administration of the programme. Applicants gave their consent for the secondary use of de-identified data during the application process. Ethical review and exemption or approval were obtained from the Ethics Committee of the University of Nigeria Teaching Hospital (NHREC/05/01/2008B-FW00002458-1RB00002424). All approaches conformed to the Declaration of Helsinki and the Nigerian National Code of Health Research Ethics before analysis. All data were maintained in compliance with institutional guidelines for confidentiality and data security. De-identified datasets and codes used to analyse the data will be made available upon request to the corresponding author [32].

Statistical analysis

All analyses were conducted in Python 3.11, with Pandas v2.0, NumPy v1.26, SciPy v1.11, and Statsmodels v0.14. Analytic steps were pre-specified as follows.

We verified the dataset shape (n = 237) and inspected missingness. Missing responses were evaluated at both the item and scale levels. Participants with > 25% of missing items on the competence or motivation scales were removed from analyses involving those scales. For participants with ≤ 25% missing items, missing values were imputed with row-mean imputation (mean of that participant’s non-missing responses within the corresponding scale). This procedure retains individual response patterns and, at the same time, reduces bias associated with missing responses. The percentage of missing values and the number of participants excluded due to excessive missing values are reported in the Results. Sensitivity analyses were performed, excluding all imputed data to assess the robustness of the results [33].

Internal reliability (Cronbach’s alpha) was computed separately for the competence and motivation items; item-total correlations were inspected. We prespecified that a scale alpha ≥ 0.70 would indicate acceptable internal consistency.

Continuous data are presented as mean ± SD (or median and interquartile range (IQR) for non-normally distributed data). For categorical variables, counts and percentages were reported.

Normality was computed for mean competence (continuous) by using the Shapiro-Wilk. Comparisons between groups were performed with a t-test or one-way ANOVA if normality applied; otherwise, a Mann-Whitney U test or Kruskal-Wallis test was used. Categorical associations (e.g., high competence vs. prior authorship) were evaluated with χ² tests or Fisher’s exact test when expected cell counts were less than 5. The correlation between mean competence and mean motivation was calculated using Pearson’s r (or Spearman’s ρ for non-normal data [34–36].

Predictors were: prior research project, prior research fellowship, prior authorship, medix member, age in years, sex, training stage (dummy coded), hours per week, and reliable internet. Adjusted odds ratios (aOR), 95% confidence intervals and p values are presented. Model fit and diagnostics included: Test of multicollinearity using the variance inflation factor (VIF); the presence of a VIF > 5 required model revision. Model goodness-of-fit (Hosmer–Lemeshow test) and residuals and influence points were evaluated. Sensitivity analyses with linear regression using mean competence as a continuous outcome were conducted, excluding imputed data [33, 37, 38]. Given potential multicollinearity among predictors, variance inflation factors were assessed, and results were interpreted with caution where VIF values indicated potential instability.

Although a dichotomous competence variable (high = 1, not high = 0) was created for descriptive purposes, the main analyses were conducted with the continuous competence score to retain information and statistical power. The variables included in the regression models were pre-specified based on theoretical relevance and prior literature in student research engagement with competence. These were demographic (age, sex, stage of training), previous research exposure, and contextual (having access to a stable internet and being part of the Medix research program). No automated stepwise selection procedures were applied.

Logistic regression was conducted as an exploratory analysis due to the low number of outcome events and potential multicollinearity among predictors. A second logistic model with willingness to actively participate as the outcome was fitted with the same predictors and with mean competence added to assess whether competence predicted willingness. Model selection was guided by consideration of multicollinearity, outcome prevalence, and interpretability of effect estimates. Pre-specified subgroup analyses were stratified by gender and training stage to explore potential differences in research competence and motivation, consistent with equity considerations and STROBE guidelines. A p-value of <0.05 was considered to be statistically significant using two-sided tests. All estimates are presented with 95% confidence intervals [39].

Reporting was informed by STROBE recommendations for cross-sectional studies. All relevant aspects of the study, such as the selection of participants, definition of variables, and the applied statistical procedures, were thoroughly checked to ensure completeness [25].

Results

Study characteristics

The Medix Frontiers Research Fellowship received a total of 253 applications. After excluding incomplete or ineligible submissions (e.g., non-UNEC students, graduates, or non-medicine/dentistry applicants), a total of 237 complete responses from medical and dental students of the University of Nigeria Enugu Campus (UNEC) were considered for the final analytic sample. No additional exclusion was applied for missing competence items (> 25% threshold), as none of the participants had missing competence items among the 8 competence items. There were no missing values in the cleaned data set (0% for all columns following preprocessing). Thus, the analytic sample size was n = 237. This is summarised in Fig. 1, the Participant flow diagram.

Descriptive statistics

The sample was comprised of 237 undergraduate medical (94.1%) and dental (5.9%) students enrolled at UNEC. The mean age was 21.98 (SD = 2.79). Females accounted for 51.1% (n = 121) and males for 49.0% (n = 116).

The stage of training was distributed as preclinical 16.5% (n = 39), basic clinical 18.1% (n = 43), and clinical 65.4% (n = 155). Table 1 summarises the mean age by training stage.

Table 1.

Mean age by training stage

Training Stage	Mean Age (SD)	n
Preclinical	20.51 (3.52)	39
Basic Clinical	20.93 (2.55)	43
Clinical	22.64 (2.42)	155

Mean age differed across training stages, with progressively higher mean age observed in more advanced stages of training

There was limited previous research experience: 25.7% (n = 61) had reported a prior involvement in a research project, 30.0% (n = 71) a prior fellowship, and 13.5% (n = 32) authorship.

Preferred research areas included clinical research (49.8%, n = 118), followed by public health (24.9%, n = 59), data science/analytics (11.0%), medical education (10.1%), and health policy (4.2%).

Research time commitment per week was < 5 h for 29.1% (n = 69), 5 to 10 h for 54.4% (n = 129), 11 to 20 h for 13.5% (n = 32), and > 20 h for 3.0% (n = 7).

A reliable internet connection was available to 97.5% (n = 231), and 99.6% (n = 236) expressed a desire to actively participate. The baseline characteristics are summarised in Table 2.

Table 2.

Participant demographics and baseline characteristics (n = 237)

Characteristic	Category	n (%)
Demographics
Age (years)	Mean ± SD	21.98 ± 2.79
Sex	Male	116 (48.95)
	Female	121 (51.05)
Course of study	Medicine	223 (94.09)
	Dentistry	14 (5.91)
Training & research exposure
Training stage	Preclinical	39 (16.46)
	Basic Clinical	43 (18.14)
	Clinical	155 (65.40)
Medix membership	No	129 (54.43)
	Yes	108 (45.57)
Prior research project	No	176 (74.26)
	Yes	61 (25.74)
Prior research fellowship	No	166 (70.04)
	Yes	71 (29.96)
Prior authorship	No	205 (86.50)
	Yes	32 (13.50)
Research preferences & engagement
Preferred research area	Clinical research	118 (49.79)
	Public health	59 (24.89)
	Data science/analytics	26 (10.97)
	Medical education	24 (10.13)
	Health policy	10 (4.22)
Hours per week	< 5	69 (29.11)
	5–10	129 (54.43)
	11–20	32 (13.50)
	> 20	7 (2.95)
Access & motivation indicators
Willing active participation	No	1 (0.42)
	Yes	236 (99.58)
Reliable internet	No	6 (2.53)
	Yes	231 (97.47)
High competence (mean > 3.5)	No	220 (92.83)
	Yes	17 (7.17)
High motivation (≥ 4)	No	24 (10.13)
	Yes	213 (89.87)

Competence and motivation scales

The competence scale (8 items, Likert 1–5) demonstrated excellent internal consistency (Cronbach’s α = 0.914). The mean competence was 2.24 (SD 0.78), and item-total correlations ranged from 0.580 (data analysis) to 0.836 (study design). High competence (> 3.5) was observed in 7.2% (n = 17).

The motivation scale (5 items, Likert 1–5) exhibited borderline internal consistency (Cronbach’s α = 0.636), suggesting that the items are not necessarily reflecting a single dimension. The mean motivation was 4.62 (SD = 0.49) with 89.9% (n = 213) achieving high motivation (≥ 4). Item-total correlations varied from 0.155 (skill improvement) to 0.617 (career). Given the borderline reliability, motivation results were interpreted cautiously and supported by item-level summaries. This is summarised in Table 3.

Table 3.

Competence and motivation scale summaries

Scale	Items	Mean ± SD	Cronbach’s α	Interpretation
Competence scale (Likert 1–5)	8	2.24 ± 0.78	0.914	Excellent internal consistency
Motivation scale (Likert 1–5)	5	4.62 ± 0.49	0.636	Acceptable internal consistency

Bivariate analyses

The mean competence was not normally distributed (Shapiro-Wilk p < 0.001), and therefore, we conducted non-parametric tests.

Competence varied significantly by previous research experience: greater in those with previous projects (Mann-Whitney U = 1934.0, p < 0.001), fellowships (U = 3028.0, p < 0.001), and authorship (U = 1360.0, p < 0.001).

No significant differences were found between training levels (Kruskal-Wallis H = 1.498, p = 0.473) or gender (Mann-Whitney U = 7895.0, p = 0.095).

Figure 2 visually highlights the bivariate difference in competence_mean by prior research project (Yes/No).

Fig. 2.

Box plot of self-reported research competence mean scores (1–5 Likert scale) stratified by prior research project experience (Yes vs. No)

Higher competence was significantly associated with prior authorship (p < 0.001) and Medix membership (p = 0.004), but not with willingness to participate (p = 1.000).

There was a weak, non-significant correlation between competence and motivation (r = − 0.068, p = 0.298). Associations with categorical high competence and other variables are summarised in Tables 4, 5 and 6.

Table 4.

Bivariate associations with research competence. Group differences in competence_mean (non-parametric tests due to non-normality)

Predictor	Category	n	Test statistic	p-value
Training stage	Preclinical (1)	39	Kruskal-Wallis H = 1.498	0.473
	Basic Clinical (2)	43
	Clinical (3)	155
Sex	Male (1)	116	Mann-Whitney U = 7895.0	0.095
	Female (2)	121
Prior research project	No (0)	176	Mann-Whitney U = 1934.0	< 0.001
	Yes (1)	61
Prior research fellowship	No (0)	166	Mann-Whitney U = 3028.0	< 0.001
	Yes (1)	71
Prior authorship	No (0)	205	Mann-Whitney U = 1360.0	< 0.001
	Yes (1)	32

Table 5.

Bivariate associations with research competence. Associations with high competence (categorical outcomes)

Predictor	Category	Test	p-value
Prior authorship	No (0)	Fisher's exact	p < 0.001
	Yes (1)
Medix member	No (0)	Fisher's exact	0.004
	Yes (1)
Willing active participation	No (0)	Fisher's exact	1.000
	Yes (1)

Table 6.

Bivariate associations with research competence. Correlation between continuous scales

Variable pair	Pearson r	p-value
competence_mean vs. motivation_mean	-0.068	0.298

Multivariable analyses

Primary analysis: linear regression (continuous competence score)

Given instability in the binary outcome model and sparse outcome events, linear regression using the continuous competence score was specified as the primary analytic approach.

The model demonstrated acceptable fit (R² = 0.387, adjusted R² = 0.363; F = 15.95, p < 0.001).

Prior research project (β = 0.54, p < 0.001), prior fellowship experience (β = 0.45, p < 0.001), and age (β = 0.04, p = 0.012) were positively associated with competence. Medix membership was negatively associated with competence (β = −0.19, p = 0.030). Other predictors, including sex, training stage, weekly research hours, and reliable internet access, were not statistically significant. Linear regression results are summarised in Table 7.

Table 7.

Linear regression results for competence_mean (continuous outcome; n = 237)

Predictor	Coefficient (β)	Standard Error	t-value	p-value	95% CI lower	95% CI upper
Intercept (const)	1.323	0.495	2.676	0.008	0.349	2.298
Prior research project	0.544	0.119	4.554	< 0.001	0.308	0.779
Prior research fellowship	0.451	0.100	4.511	< 0.001	0.254	0.648
Prior authorship	0.262	0.147	1.779	0.077	-0.028	0.551
Medix member	-0.193	0.089	-2.179	0.030	-0.368	-0.019
Age (years)	0.041	0.016	2.545	0.012	0.009	0.073
Sex	-0.044	0.085	-0.516	0.606	-0.212	0.124
Training stage	-0.088	0.058	-1.522	0.129	-0.203	0.026
Hours per week (numeric)	0.009	0.056	0.164	0.870	-0.101	0.119
Reliable internet	0.059	0.263	0.223	0.824	-0.459	0.576

Secondary analysis: logistic regression (high competence > 3.5)

A logistic regression model was conducted as an exploratory analysis with high competence (mean > 3.5) as the outcome. Only 7.2% (n = 17) of participants met this threshold, resulting in a sparse outcome distribution.

The model showed evidence of instability due to multicollinearity and limited outcome events, and results should therefore be interpreted cautiously.

Significant predictors included prior fellowship experience (aOR = 9.46, 95% CI 1.71–52.48, p = 0.010) and age (aOR = 1.45, 95% CI 1.16–1.82, p = 0.001). Prior authorship approached statistical significance (aOR = 4.10, 95% CI 0.80–20.99, p = 0.091), while Medix membership showed a borderline negative association (aOR = 0.19, 95% CI 0.03–1.08, p = 0.061). All other variables (previous project, sex, stage of training, hours/week, reliable internet) were not statistically significant.

Model diagnostics and stability assessment

Multicollinearity was substantial in the logistic regression model, with elevated variance inflation factors observed for age (VIF = 40.23), internet access (VIF = 27.71), training stage (VIF = 14.00), sex (VIF = 8.96), and weekly research hours (VIF = 7.32). These values indicate potential inflation of standard errors and instability of coefficient estimates, suggesting caution in interpretation; however, model discrimination was good (AUC = 0.947).

Despite this, model discrimination was high (AUC = 0.947), and the likelihood ratio test indicated overall model significance (p < 0.001). However, coefficient estimates should be interpreted with caution due to instability arising from multicollinearity and sparse outcomes. Figure 3 summarises multivariable results via a forest plot.

Fig. 3.

Forest plot of adjusted odds ratios (aORs) and 95% confidence intervals for significant and borderline predictors of high research competence

A secondary logistic model of willingness to participate showed quasi-complete separation (99.6% yes responses), producing erratic estimates (with large aORs and NaN CIs); therefore, this finding was considered inappropriate for logistic regression. The logistic regression model showed evidence of multicollinearity and instability, including high variance inflation factors for several predictors and wide confidence intervals for certain estimates. These findings suggest that results from this model should be interpreted cautiously. The multivariable logistic regression results and VIF are summarised in Tables 8 and 9.

Table 8.

Multivariable logistic regression results for high competence (adjusted ORs, 95% CIs, p-values) and VIFs. Logistic regression model (outcome: high_competence > 3.5; n = 237)

Predictor	Adjusted OR (aOR)	95% CI lower	95% CI upper	p-value
Intercept (const)	9.34 × 10⁻¹⁰	3.84 × 10⁻⁸⁶	2.28 × 10⁶⁷	0.817
Prior research project	2.48	0.36	17.16	0.358
Prior research fellowship	9.46	1.71	52.48	0.010
Prior authorship	4.10	0.80	20.99	0.091
Medix member	0.19	0.03	1.08	0.061
Age (years)	1.45	1.16	1.82	0.001
Sex	1.28	0.30	5.49	0.736
Training stage	1.63	0.61	4.36	0.334
Hours per week (numeric)	0.99	0.36	2.74	0.990
Reliable internet	478.41	2.40 × 10⁻⁷⁴	9.56 × 10⁷⁸	0.945

Table 9.

Multivariable logistic regression results for high competence (adjusted ORs, 95% CIs, p-values) and VIFs. Variance inflation factors (VIF) for multicollinearity assessment

Predictor	VIF
Prior research project	2.25
Prior research fellowship	1.84
Prior authorship	1.78
Medix member	2.16
Age (years)	40.23
Sex	8.96
Training stage	14.00
Hours per week (numeric)	7.32
Reliable internet	27.71

Since only a few individuals were classified as highly competent, results from logistic regression should be viewed as preliminary, and more attention should be given to results from linear regression using the continuous competence score.

Sensitivity analyses

Sensitivity analyses using alternative imputation approaches for motivation items did not materially change the mean motivation score (4.62 ± 0.49). An interaction term (prior fellowship × Medix membership) did not significantly alter main effects but showed instability due to sparse data.

Discussion

Interpretation of main findings

This descriptive cross-sectional study among 237 medical and dental students of the University of Nigeria, Enugu Campus (UNEC) applying for Medix Frontiers Research Fellowship, revealed a rather surprising disparity between a high level of motivation and low self-reported research competence. The mean competence score was 2.24 (SD 0.78) on a 5-point Likert scale, with 7.2% of participants achieving high competence (> 3.5). In contrast, motivation was significant (mean 4.62, SD 0.49) with 89.9% scoring ≥ 4. Prior research experience, especially prior research fellowships (aOR 9.46, 95% CI 1.71–52.48, p = 0.010) and authorship (aOR 4.10, 95% CI 0.80–20.99, p = 0.091), stood out as the most robust predictor of high competence, in addition to increasing age (aOR 1.45 per year, 95% CI 1.16–1.82, p = 0.001). Medix membership demonstrated a borderline negative association (aOR 0.19, 95% CI 0.03–1.08, p = 0.061).

There was no significant difference in competence based on training stage (p = 0.473) or sex (p = 0.095). Also, the correlation between competence and motivation was weak and non-significant (r = − 0.068, p = 0.298). These results demonstrate that although there is no shortage of motivation among this motivated applicant group, actual self-perceived skills remain limited, with previous experience being the major driver of competence.

The absence of a gradient of competence at different stages of training is remarkable. This observation could also be indicative of a fundamental misalignment in curricular design between progression and research skill development. Progression along the stages of training in many medical programmes is predominantly competency-based in the clinical domains with little vertical integration in methodology training in research. As a result, progression in academic level may not necessarily translate into cumulative gains in research competence unless explicitly reinforced through structured experiential learning. Although participants progressed from preclinical to clinical stage, students reported no substantial improvement in research skills, in a manner that corroborated the finding that the formal curricula of many Nigerian medical schools place more emphasis on clinical training than on research methodology and competence [14].

Meanwhile, prior experience made a huge difference with regard to levels of competence, highlighting the importance of providing students with early, structured opportunities to develop experiences and skills in question formulation, literature searching, study design, and data interpretation. The borderline negative association between Medix membership and competence may reflect residual confounding, selection effects, or model instability, and should be interpreted cautiously, potentially influenced by unmeasured factors such as academic workload or competing commitments. This seeming discrepancy can be explained by metacognitive awareness, whereby individuals with greater exposure to research become more aware of their limitations and rate their competence more critically. On the other hand, very motivated, but inexperienced students can overestimate their readiness, contributing to the weak or inverse association observed [40, 41].

Comparison with existing literature

These findings are consistent with larger trends in Sub-Saharan Africa (SSA) and among low- and middle-income countries (LMICs) [42]. A national cross-sectional study of Nigerian medical students in 2025 highlighted similar barriers, which included: absence of statistical skills (74.2%), time limitation (73.3%), and limited training in research methodology, among others, as the largest barriers to participation [14]. Past research from SSA revealed systemic deficiencies in research labs, funding, and skills training across medical schools [43]. Also, a 2020 study in Lagos, Nigeria, observed similar low participation attributed to a perceived shortage of mentorship and resources [44]. Around the world, self-rated abilities of undergraduates from LMICs tend to be lower compared to those from high-income settings, which often feature more structured approaches to research (e.g., compulsory projects or electives), ultimately yielding higher baseline skills [45]. This high motivation observed here mirrors results from LIMC cohorts, where enthusiasm for research is high but not met with institutional support [14, 46].

In addition to structural constraints, emerging evidence suggest that engagement in learning and the acquisition of skills is influenced by a combination of cognitive, motivational, and contextual factors and is not adequately formulated by a single predictor [47]. For instance, recent studies using partial least squares structural equation modelling and configurational methods suggest that combinations of factors (e.g., prior exposure, perceived competence, and learning environment) influence not only lower-level academic behaviours but also higher-order academic behaviours such as problem posing and group work [48, 49]. These findings are consistent with the present study, in which prior research exposure consistently emerged as a key correlate of competence, indicating that the development of competence may not unfold in a linear manner, but is contingent upon synergistic combinations of experience, motivation, and opportunity. This configurational notion also proves to be a helpful instrument in making sense of why high motivation alone did not translate into higher competence in this cohort [50].

Implications for fellowship design and medical education

The implications for medical education, fellowship design, and other similar training and capacity-building initiatives are clear. Curricula need to focus on integrating research training at an early stage, ideally in preclinical years, to close the gap in competencies before clinical studies place further demands on students. Fellowships such as Medix Frontiers Research Fellowship can act as focused interventions, offering guidance, statistical training, and real-world projects, especially for inexperienced students. Early-stage interventions (such as workshops on IMRAD writing, ethics and data analysis) may potentially speed learning and minimise reliance on autodidactic learning. Given the strong predictive role of prior fellowships, scaling such opportunities could create a virtuous cycle of competence and engagement and have a positive feedback effect on competence and engagement.

Also, selection and early stratification should be based on baseline differences: fellows with less background experience will benefit from an expanded foundational module (covering basic study design, literature searching and introductory data analysis) and fellows with more prior experience can move more quickly toward independent project work and manuscript preparation. The positive association between competence and willingness to participate suggests embedding early, practical tasks that enhance confidence and tangible competencies, which could also increase retention and completion rates. Finally, as reliable internet was identified as pertinent in preliminary results, programmes in resource-limited settings are encouraged to emphasise offline-compatible solutions, robust data-access support, and protected data-use time to counter technological challenges [4].

Strengths of the study

The strengths of the study are that it consisted of a large sample size of fellowship applicants (n = 237), the use of structured self-report measures with good (competence α = 0.914) and acceptable (motivation α = 0.636) reliability, and a thorough multivariable analysis applying both logistic and linear regression to address outcome rarity and multicollinearity issues. Non-parametric tests for non-normal data and sensitivity analyses (e.g., median imputation yielding identical results) contributed to robustness.

Also, measurement used a brief competence-based instrument that maps onto specific aspects of research (question formulation, literature searching, analysis, IMRAD writing, ethics), and psychometric analyses (internal consistency) were considered a priori. The analysis combined descriptive, reliability, bivariate, and multivariate models to examine the independent predictors of the outcome whilst maintaining interpretability for programme managers.

Furthermore, employing both categorical and continuous modelling methods enabled the critical assessment of potential methodological limitations, including those relating to outcome rarity and model instability. This dual analytic approach provides greater reassurance regarding the robustness of the primary findings and demonstrates methodological transparency in addressing known limitations of regression modelling in small-event datasets.

Limitations of the study

Limitations must be acknowledged. Self-reported competence and motivation may be subject to response bias, including social desirability effects and individual differences in self-perception. As objective performance-based assessments (e.g., timed literature searches or data analysis tasks) were not feasible within the study design, findings should be interpreted as reflecting perceived rather than directly measured competence [27].

Moreover, the high internal consistency of the competence scale alone does not establish construct validity. he instrument was designed to capture key domains of research competence; however, formal validation procedures such as factor analysis or external benchmarking against objective performance measures were not conducted. Hence the scale should be understood as a measure of perceived competence rather than a fully validated psychometric instrument.

The cross-sectional design precludes causal interpretations, e.g., whether prior experience leads to higher competence or vice versa cannot be determined, as associations between prior exposure and competence may reflect reciprocal selection effects (motivated students seek research opportunities). Although STROBE guidelines were followed, some reporting constraints remain due to the use of pre-existing application data, particularly in relation to response rate calculations and recruitment denominators. A formal sample-size calculation was not performed because the study analysed the entire accessible applicant pool (all valid responses received within the application window).

The findings should be interpreted with caution in terms of generalizability. The study population consisted of applicants to a competitive research fellowship and therefore represents a highly selected and likely more motivated subgroup of students. As such, the results may not be fully representative of the broader population of undergraduate medical and dental students in Nigeria or other sub-Saharan African settings. Additionally, the study was conducted within a single public institution, and contextual differences in curricula, mentorship availability, and digital infrastructure may further limit the applicability of the findings to other settings.

Construct validity was not formally assessed using factor analysis or external validation; therefore, the scale reflects perceived competence rather than a validated psychometric construct. The motivation scale had stable but borderline internal consistency. Therefore, motivation results need to be treated with caution, and may be better understood at the individual item level rather than as a unified construct. The logistic regression model demonstrated substantial multicollinearity and instability, likely due to correlated predictors and a limited number of outcome events. As a result, coefficient estimates may be inflated, and linear regression on the continuous outcome provides more stable inference. Age and training stage were conceptually and statistically related, and although both were included in regression models, residual collinearity may have influenced the precision of estimates.

Selection bias presumably led to an overestimation of motivation and competence relative to non-applicants. Multicollinearity (as indicated by high VIF values for age, training stage, and internet access) may have inflated standard errors for some estimates. However, linear regression using the continuous competence score provided more stable and interpretable results. This work did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors for this secondary analysis of anonymised application data.

Furthermore, there should be caution in generalising findings, as they are most relevant to undergraduate medical and dental students applying to competitive research training programs within the context of low-resource universities. When extrapolating to a wider student population or other countries, differences in curricula, mentorship availability and digital infrastructure should be considered.

Future directions

Future work should involve a longitudinal follow-up of fellowship participants to determine changes in competence and objective research outputs (manuscripts submitted/accepted, conference presentations) and to evaluate which curricular components most strongly predict sustained scholarly activity. Intervention trials contrasting standard versus intensified foundational modules for low-competence entrants would clarify causality and inform scalable training packages. Multi-centre studies involving Nigerian and sub-Saharan African institutions are needed to increase generalizability and identify system-level barriers that can be addressed through policy interventions.

To supplement the self-reported measure, future research should also utilise objective measures of research ability, such as performance-based evaluations or structured research tasks. The use of mixed-methods strategies (e.g., qualitative investigations of student experiences) may also serve to enhance understanding of contextual barriers and facilitators to research participation. More advanced analyses, like structural equation modelling or configurational analysis, could disclose additional patterns in the complex interactions between motivation, competence, and environmental elements related to research preparedness.

Conclusion

This study assessed self-perceived research competence, motivation, and preparedness among undergraduate medical and dental students applying to a competitive research fellowship in a resource-limited setting. The findings show a marked contrast between high levels of motivation and a relatively low self-reported research competence, with prior research exposure emerging as the single most consistent predictor of subjective research competence.

This study adds to the growing body of literature on research capacity building in low- and middle-income countries by demonstrating the salience of early, structured research experiences in shaping students’ perceived abilities. These findings underscore the necessity of moving beyond theoretical instruction toward experiential and mentorship-driven learning models.

From a pragmatic standpoint, fellowship programmes and medical curricula may benefit from adopting tiered training models which take into account baseline differences in experience, especially by strengthening foundational skills during earlier stages of training.

However, findings should be interpreted in light of several limitations, including reliance on self-reported competence, potential selection bias due to the highly motivated applicant sample, and the cross-sectional design, which precludes causal inference.

Further studies should adopt longitudinal and multi-institutional designs to examine the changes in competence over time and to determine the efficacy of targeted training interventions in improving both perceived and objective research skills.

Authors’ contributions

Godswill Uzoechina (GU) conceptualised the study, designed the analysis plan, performed the data analysis, and drafted the Results and Discussion sections. GU also coordinated overall manuscript preparation, supervised the project, and critically revised all sections for intellectual content. Treasure Osajiuba (TO), Elochukwu Marvellous (EM), and Chinonso Ifudu (CI) managed data collection and contributed to the development of the study instruments. TO drafted the Methods section, EM drafted the Introduction, and CI drafted the Abstract and Conclusion. All authors reviewed, edited, and approved the final manuscript and agree to be accountable for all aspects of the work.

Funding

This research was conducted under the leadership of the Director of Research & Publications at Medix Frontiers and used only internal resources of the organization. No specific grant or external funding from public, commercial, or not-for-profit agencies was received for this work.

Data availability

De-identified datasets and codes used to analyse the data will be made available upon request to the corresponding author [32].

Declarations

Ethics approval and consent to participate

This study involved secondary analysis of de-identified application data collected for the Medix Research Fellowship. All applicants provided informed consent for the use of their anonymised data for research purposes at the point of application. Ethical review and exemption or approval were obtained from the Ethics Committee of the University of Nigeria Teaching Hospital (NHREC/05/01/2008B-FW00002458-1RB00002424). All approaches conformed to the Declaration of Helsinki and the Nigerian National Code of Health Research Ethics before analysis. All data were maintained in compliance with institutional guidelines for confidentiality and data security.

Consent for publication

Not applicable. No individual person’s data are presented in this manuscript.

Competing interests

The authors declare no competing interests.