An institution-based cross-sectional study was conducted to investigate the magnitude, attitudes, and associated factors concerning RM from January to April 2024 among faculty researchers at a public University in Ethiopia.

The study included faculty members who were actively involved in health or health sciences research and who had at least one publication.

This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. Ethical approval was obtained from the Addis Ababa University, College of Health Sciences, School of Public Health Institutional Research Ethics Review Committee (IRERC), with approval number SPH/296/2024 and dated 19/01/2024. Written informed consent was obtained from all participants prior to their inclusion in the study.

•

Inclusion criteria: Faculty members with research experience and publications in local and international journals within the last five years were eligible for the study.

The questionnaire was divided into four sections to collect study participants’ baseline characteristics, general attitudes toward RMC, behavioural influences (factors), and respondents’ self-report of the frequency of their research misconduct practices. Behaviours related to RM and Questionable Research Practices (QRPs) were grouped into five research misconduct composites: circumventing research ethics regulations; fabrication and falsification; plagiarism; authorship misconduct; and conflict of interest.

The questionnaire was adapted from the study on “Reliability and validation of an attitude scale regarding responsible conduct in research” among Middle Eastern researchers. ²⁰ Questionnaires were collected in such a way that no third party or even the researchers could link specific respondents to completed questionnaires to ensure complete anonymity.

In this study, “Research Misconduct” encompasses composites such as circumventing research ethics regulations, plagiarism, fabrication, and falsification; while “Questionable Research Practices” are defined as including authorship issues and conflicts of interest composites. Each misconduct composite is defined by specific questions.

Based on a margin of error of 5%, a confidence level of 95%, an estimated prevalence rate of 52.8% in the Middle East, ¹¹ and a total population size of 836 investigators at the targeted university, we calculated a sample size of 297 participants using Cochran’s formula. ²¹ n 0 = ( Z α / 2 ) 2 ∗ ( p ) ( q ) ( d ) 2 (1)

For this study, sample sizes were calculated for each specific objective, and the largest sample size was used as the final sample size. Using the formula above, the largest sample size, n ₀ = 383, was determined using the prevalence of 52.8% for the second specific objective regarding attitude from a study in the Middle East. ¹¹ n 0 = ( 1.96 ) 2 ( 0.528 ) ( 0.472 ) ( 0.05 ) 2 = 383

Since this sample size exceeds 5% of the total population (836 * 0.05 = 42), the final sample size was adjusted using Cochran’s correction formula. ²¹ n 1 = n 0 1 + n 0 N (2)

By substituting the values into the correction formula (2) provided above, the minimum sample size (adjusted) was determined to be 263. n 1 = 383 1 + 383 836 = 263

Bartlett et al. recommended using response rates from previous studies of the same or similar populations to address non-response rates. ²² Therefore, adjusting for a response rate of 88.7% from a prior study in Nigeria, ¹⁰ the final sample size was calculated as n ₁ = 263/0.887 = 297.

We used probability sampling approach to select respondents, employing simple random sampling method to ensure random selection and representativeness of our sample. This approach maintained the randomness and representativeness of our sample. The sample size of 297 was distributed across four schools based on the proportionate population size in each of the four disciplines, considering individuals with at least one publication. Each department’s share was calculated by dividing 297 by the total population of 836 and multiplying by 0.35. Study units were then selected from each department using a simple random sampling procedure. This sampling procedure is illustrated schematically in Figure 1.

To minimize potential biases in this study on research misconduct, validated tools and metrics were employed to ensure reliable measurement, ²⁰ and data collectors were trained to ethically and confidentially collect the data. Confidentiality and anonymity were emphasized to mitigate social desirability bias. Data analysis was conducted objectively using statistical software to ensure accuracy and reduce human error. Efforts were made to avoid confirmation bias by incorporating diverse perspectives in the study framework and seeking evidence that challenged initial hypotheses. Transparent reporting of methodologies, assumptions, and limitations further enhanced the credibility of the research.

Independent variables •

Baseline characteristics: sex, academic position, highest degree earned, prior training on research ethics.

Dependent variables •

RM: circumventing RE regulations, plagiarism, and fabrication and falsification

Data from paper questionnaires were collected, entered into Epi-Data 3.1, and subsequently transferred to SPSS version 27. Descriptive analyses summarized demographic frequencies and percentages. Bivariate analysis was utilized to explore relationships between the composite scores of misconducts and factors such as age, gender, prior ethics training, graduate school attended, and academic position.

Respondents were asked about the frequency of various misconduct behaviours, with response options including “Never,” “Once or twice,” or “Three or more times” for self-reported misconduct (RM, QRPs, & Overall RM). To ensure suitable categorization and sufficient data for analysis, these responses were transformed into dichotomous choices: “Never” and “One or more times”. ²³ Each type of self-reported misconduct was individually measured and categorized into one of the previously mentioned five research misconduct composites. Additionally, the magnitude of self-reported misconduct was calculated for each composite category.

For composite scores ( Figure 2), multivariable analysis models were constructed. Independent variables identified as significant in bivariate analysis at a p-value threshold of 0.25 were included in the binary logistic regression model. This approach ensures that relevant factors are not overlooked, which can happen when using stricter thresholds like 0.05 in logistic regression modelling. ²⁴ A p-value < 0.05 was considered significant for covariates in the final multivariate analysis. Odds ratios (OR) with confidence intervals (CI) and corresponding p-values were calculated.

Regarding ‘attitude’ items, responses were coded on a five-point Likert scale ranging from strongly agree to strongly disagree. Responses of ‘strongly agree’ and ‘agree’ were grouped together as ‘agree,’ while ‘strongly disagree’ and ‘disagree’ were categorized as ‘disagree.’ The mid-point response, ‘neutral,’ was not included in either the ‘agree’ or ‘disagree’ categories. Descriptive analysis then summarized the distribution of responses across these categories. ¹¹ Missing data were addressed through mean imputation, replacing missing values with the average of available data for the respective variable.

Prior to filling out the questionnaire, participants were informed of the study’s purpose via a one-page information sheet. Written informed consent was obtained from all participants prior to their participation in the study. Participants were provided with detailed information about the study objectives, procedures, potential risks, and benefits before signing the consent form. This process was reviewed and approved by the Addis Ababa University, College of Health Sciences, School of Public Health Institutional Research Ethics Review Committee (IRERC), with approval number SPH/296/2024 and dated 19/01/2024. The study adhered to the ethical principles outlined in the Declaration of Helsinki.

Due to the sensitive nature of the data acquired and the possibility for dignitary harm, the name of the respondents’ institution was blinded and the names of specific schools were anonymized. The self-administered questionnaire did not include any questions that could reveal participants’ identities, Questionnaires were distributed and collected in person by trained data collectors who explained the study and ensured voluntary participation. After completion, participants placed their questionnaires directly into sealed collection boxes, which were only opened by the researchers after data collection was completed. No identifying information was included on the questionnaires, ensuring that neither the data collectors nor the researchers could link specific responses to individual participants. This process was designed to protect respondent anonymity while encouraging high response rates.

Table 2 shows the frequency and percentage of occurrences of misconduct composites and associated behaviours reported by respondents. The misconduct composites included circumventing research ethics regulations, fabrication and falsification, plagiarism, authorship misconduct, and conflict of interest. Key findings for the study are summarized as follows.

Concerning the composite circumventing research ethics regulations, several respondents (17.2%) reported not obtaining proper informed consent, 14.8% reported using confidential information about research subjects without authorization, and 12.3% reported conducting research involving human participants without prior approval from a Research Ethics Committee. Regarding plagiarism, 20.9% reported using others’ ideas without credit, 10.7% reported submitting a published manuscript for secondary publication, and 9.4% published others’ results. For fabrication and falsification, 15.2% fabricated data, 14.3% altered data without disclosure, 26.2% selected data to support hypotheses, and 23.0% dropped outliers without mentioning.

Among respondents, reports of authorship misconduct were prevalent, with 37.3% indicating that they gave authorship to those who contributed minimally and 23.0% allowing their name to appear on papers where they contributed little. Non-disclose of conflicts of interest was reported among 17.6% of respondents, 13.1% indicated that they had compromised study rigor as a result of funding pressure, and 7.8% reported altering or suppressing research results inappropriately as a result of funding pressure.

Regarding composites of RM and QRPs, Authorship misconduct emerged as the most prevalent form of research misconduct, with 47.5% of respondents reported involvement (95% CI: 31.6%–44.1%). Fabrication and falsification followed closely, with 40.6% acknowledging participation (95% CI: 34.4%–47.0%). Additionally, 26.6% of respondents reported engaging in plagiarism, while 25.4% reported conflicts of interest, and 29.1% admitted to violating research ethical standards.

Findings revealed recognition and a strong consensus among participants on the importance of addressing research misconduct, reporting unethical behaviour, declaring conflicts of interest, and mentoring trainees. However, there was some discomfort in discussing ethical issues, which could impede efforts to improve research integrity (see Table 3).

Table 4 shows participants perception toward RCR. Concerning publication pressure, 56.1% considered it moderately influencing while 34.4% considered it strongly influencing. Participants perceived that financial conflicts of interest play a role in RMC, with 46.3% believing it had some influence, while a similar proportion (45.9%) perceived it had strong influence. Regarding pressure from funders, 45.5% of participants perceived it had some influence while 39.3% viewed it as strongly influential. Insufficient censure of misconduct was viewed as an influence in committing research misconduct, with 54.9% believing it had some influence, while 34.4% noting strong influence.

As Table 5 shows, attitudes towards research integrity are segmented according to ethics training, research experience, and academic status. Results showed that more participants below the rank of lecturer (27.4%) believed that scientific integrity is solely the responsibility of the principal investigator, compared to 16.1% of those above lecturer rank. This indicates a link between academic rank and views on responsibility for scientific integrity. The difference was significant (χ ² = 3.851, df = 1, p < 0.05). Investigators with ethics training tended to endorse the attitude that investigators should report instances of research misconduct more strongly; with a higher percentage (96.3%) agreeing that research misconduct should be reported. The differences were significant (χ ² = 4.906, df = 1, p < 0.027).

In terms of feeling uncomfortable discussing ethical behaviour, academic rank is significantly associated (χ ² = 7.185, df = 1, p < 0.007), as a relatively high percentage (31.6%) of those below lecturer level felt uncomfortable doing so.

A Chi-Square test showed significant associations with a p-value of less than 0.05 (see Table 6). In general, respondents with prior ethics training reported a relatively lower level of self-reported misconduct across all categories when comparing self-reported misconduct with participants without ethics training. Compared to those with ethics training, 34.9% of those without ethics training reported conflicts of interest. The difference was significant (χ ² = 5.290, df = 1, p < 0.021). Across schools, self-reported misconduct varies significantly by affiliation. Respondents from clinical schools, for example, reported greater rates of plagiarism, fabrication and falsification (FF), conflict of interest (CoI), RMC total, and Overall RMC, at 30.9%, 44.5%, 28.8%, 44.0%, and 41.4%, respectively, than those from non-clinical schools. Significant differences were found (χ ² = 7.159, df = 1, p < 0.007), (χ ² = 4.904, df = 1, p < 0.027), (χ ² = 4.528, df = 1, p < 0.033), (χ ² = 8.460, df = 1, p < 0.004), (χ ² = 4.314, df = 1, p < 0.038), respectively.

The misconduct rates of those above and below the lecturer rank also differed significantly. The study found that 18.1% of participants above lecturer rank self-reported plagiarism, compared to 40.0% of participants below lecturer rank (χ ² = 13.113, df = 1, p < 0.001); Additionally, 51.6% of participants below lecturer rank reported falsification and fabrication, while 33.6% of participants above lecturer reported misbehaviour. Significant differences were seen between these groups (χ ² = 7.085, df = 1, p < 0.008). In terms of RMC-Total, 31.5% of respondents above lecturer rank reported committing RMC, which is considerably lower than the 50.5% of participants below lecturer rank (χ ² = 8.012, df = 1, p < 0.005).

Research misconduct was reported by 44.4% of respondents with less than 5 years of research experience, which was significantly higher compared to the 31.0% reported by those with more than 5 years of experience (χ ² = 3.939, df = 1, p < 0.047). Self-reported research misconduct varies by age, with junior researchers ^{25– 34} more likely to report engaging in multiple types of research misconduct than senior researchers (35+) (χ ² = 4.853, df = 1, p < 0.028). Additionally, significant age differences were observed in bypassing research ethics regulations, with 39.7% of junior researchers admitting to this behaviour, compared to 18.7% of senior researchers (χ ² = 12.005, df = 1, p < 0.001).

Table 7 highlights several significant predictors of research misconduct and related behaviours. Higher academic rank, older age, and a more responsible attitude toward scientific integrity were associated with a lower likelihood of circumventing research ethics regulations, engaging in plagiarism, and committing fabrication or falsification. Participants who reported feeling high publication pressure or working in clinical schools were more likely to engage in research misconduct.

Researchers with more than 5 years of experience were more prone to authorship misconduct, while those with postgraduate degrees were less likely to engage in conflicts of interest. School affiliation, particularly clinical vs. non-clinical, also played a role in influencing research misconduct behaviours.

For further details on these associations, including specific odds ratios, confidence intervals, and p-values please refer to the appendix.