Do successful PhD outcomes reflect the research environment rather than academic ability?

Daniel L Belavy; Patrick J Owen; Patricia M Livingston

doi:10.1371/journal.pone.0236327

. 2020 Aug 5;15(8):e0236327. doi: 10.1371/journal.pone.0236327

Do successful PhD outcomes reflect the research environment rather than academic ability?

Daniel L Belavy ^1,^*, Patrick J Owen ¹, Patricia M Livingston ²

Editor: Sergi Lozano³

PMCID: PMC7406039 PMID: 32756557

Abstract

Maximising research productivity is a major focus for universities world-wide. Graduate research programs are an important driver of research outputs. Choosing students with the greatest likelihood of success is considered a key part of improving research outcomes. There has been little empirical investigation of what factors drive the outcomes from a student's PhD and whether ranking procedures are effective in student selection. Here we show that, the research environment had a decisive influence: students who conducted research in one of the University's priority research areas and who had experienced, research-intensive, supervisors had significantly better outcomes from their PhD in terms of number of manuscripts published, citations, average impact factor of journals published in, and reduced attrition rates. In contrast, students’ previous academic outcomes and research training was unrelated to outcomes. Furthermore, students who received a scholarship to support their studies generated significantly more publications in higher impact journals, their work was cited more often and they were less likely to withdraw from their PhD. The findings suggest that experienced supervisors researching in a priority research area facilitate PhD student productivity. The findings question the utility of assigning PhD scholarships solely on the basis of student academic merit, once minimum entry requirements are met. Given that citations, publication numbers and publications in higher ranked journals drive university rankings, and that publications from PhD student contribute approximately one-third of all research outputs from universities, strengthening research infrastructure and supervision teams may be more important considerations for maximising the contribution of PhD students to a university’s international standing.

Introduction

A research doctorate degree comprises a process of independent research that produces an original contribution to knowledge [1]. The Australian Commonwealth Government supports [2] both domestic and overseas students undertaking research doctorate degrees, known as PhDs. These scholarships, which comprise a stipend for three years, are competitive. For this reason, when students apply for scholarships for their PhD studies, prior academic performance and research training play a key role in deciding whether the applicant receives a scholarship. However, is assigning scholarships predominately on the basis of academic grades and previous research experience effective in determining who will succeed?

A university’s international and national ranking is important for its reputation and marketing to prospective students [3]. Citation rates, number of publications and impact factor of journals faculty publish in, influence the ranking of a university. The Quacquarelli Symonds University Rank [4] is weighted 30% by the number of citations per faculty member, the Times Higher Education World Ranking [5] 30% by the number of citations and 6% by the number of publications per academic, and the Academic Ranking of World Universities [6] 20% by number of highly cited researchers, 20% by number of papers published in Nature or Science and 20% by the number of publications in total.

PhD students are important drivers of research outputs from universities, with one analysis [7] showing that one-third of research publications was from doctoral students. It is important to consider to what extent the procedures by which universities select students who go on to produce higher numbers of highly cited publications in high impact journals. We are not aware of any prior research that has examined this topic.

Waldinger [8] showed that the quality of academic staff (in departments of mathematics at German universities in the 1930s) influenced the likelihood of whether a doctoral student would become a full professor later in their career. Waldinger also showed that the amount of citations the scientific work of a doctoral student received through their entire subsequent scientific career was influenced by the status of their supervisor. Other factors, such as, the reputation of a department [9], the reputation the group leader [10], and access to resources and equipment [11], the number of full-professors on staff [12] influenced the research output of the academics involved in that group. Less information is available on the impact of student academic ability or prior research training on PhD outcomes: one analysis found that the reputation of a given department was more important for employment outcomes post-PhD than the accomplishments of the student during their studies [13]. Overall, the evidence available implies that the research environment may have an inordinate impact on the PhD student outcomes (e.g. citations, number of publications, impact factor of journals of those publications).

Here we examine the relationship between information known about applicants and their proposed supervisory teams at the time of scholarship application with the subsequent research outputs, as measured by number of citations, number of publications and the impact of journals of those publications.

Materials and methods

Deakin University Human Research Ethics Committee reviewed this project (2019–191) and found it to be compliant with the Ethical Considerations in Quality Assurance and Evaluation Activities guidelines of the National Health and Medical Research Council of Australia and determined that no further ethics review was required. Consent was not obtained and the data analysed anonymously.

Over a four year period, 2010–2013, 324 PhD scholarship applications were submitted to the Faculty of Health at one university in Australia (Fig 1). In these applications, data were collated on:

the grade the student achieved for their prior research training degree and their rank in this degree (top, middle, bottom third of first class honours or second class honours; or their equivalency to this),
the grade point average achieved in their undergraduate degree (ranked on a scale of 1 to 5 with 5 = high distinction grade point average plus prizes awarded, 4 = high distinction grade point average, 3 = distinction, 2 = credit, 1 = pass).
whether the applicant had published in a scientific journal (‘yes’ or ‘no’)
research environment: whether the primary supervisor was located in a strategic research centre or institute within the university (‘yes’ or ‘no’).

At the time of ranking for scholarships, the review panel scored each application on the basis of their academic merit and the research experience, alignment of the proposed research with the strategic research goals of the Faculty and university, and the experience of the supervisory team (as expressed by prior PhD completions, student progress, external grants, previous student publications, supervisor track record). In July 2018, these scores were reviewed by two independent assessors experienced in the scholarship ranking process and consensus was attained. Subsequent to this, following variables were generated:

quartile of the academic merit scores in which each student was located.
strategic alignment score achieved maximum points (‘yes’ or ‘no’). The presence or absence of a maximum score was taken for this variable as there were few instances of low scores on this criterion and data were skewed to the maximum score.
supervisor team scores achieved maximum points (‘yes’ or ‘no’). The presence or absence of maximum score was taken for this variable as there were few instances of low scores on this criterion and data were skewed to the maximum score.
level of academic appointment of the primary supervisor (lecturer/senior lecturer, associate professor, or full professor)

Data on whether the applicant subsequently enrolled (if ‘no’ they were excluded from further analysis; Fig 1), whether they completed their studies (‘yes’ or ‘no’), and whether the student received a scholarship to support his/her study (‘yes’ or ‘no’) obtained from another university database.

The university tracks publication outputs of its faculty and students. In July 2018, these data were obtained to link the number of publications by the student with their primary supervisor, the impact factor of the journals in which these publications appeared, and the number of citations received by the publications in Web of Science by the cut-off data of data access. Publications were matched on the basis of student name and primary supervisor name. If a change of primary supervisor occurred during student candidature, publication matches with the new primary supervisor were included as well. If the student had enrolled in a PhD but achieved no publications within the time-period examined, data were coded as zero publications, zero citations and zero average impact factor. Datasets were merged in using custom written code implemented in the 'R' statistical environment (version 3.4.0 https://www.r-project.org/). Where repeat applications were submitted in subsequent years by the same person, only the data available at the first application was used in further analysis. Prior to statistical analysis, all identifying information was removed.

Statistical analyses

All analyses were conducted using Stata statistical software version 15 (College Station TX, USA). Univariate associations between continuous dependent variables (number of publications, number of citations, number of citations per publication, average publication impact factor) and explanatory variables were assessed by the Kruskal-Wallis H test or Mann-Whitney U test (both non-parametric tests), as well as one-way analysis of variance and t-tests (both parametric tests). Univariate associations between withdrawal (yes/no) and independent variables were assessed by penalized maximum likelihood [14,15] logistic regression. We categorised the explanatory variables as follows: student specific factors (student research degree rank, student undergraduate rank, student prior publication, student academic merit), supervisor specific factors (supervisor located in a strategic research centre, supervisor academic level, supervisor team scores achieved maximum points), research topic related factors (strategic alignment score achieved maximum points), and whether a scholarship was awarded. To investigate which variables were more important than others for PhD student outcome metrics, factorial analysis of variance (ANOVA) as well as stepwise multiple linear regression models with both forward and backward selection were used to assess the association between the dependent variables and the independent variables. We further conducted factorial ANOVA to assess the association between the dependent variables and independent variables. Stepwise penalized maximum likelihood logistic regression models were used to predict withdrawal from PhD (yes/no) based on independent variables. An adjusted alpha level of 0.10 to enter and 0.20 to remove were used for all step-wise regression models. An alpha-level of 0.05 was adopted for all other statistical tests, including the assessment of the final step-wise regression models.

Results

Primary analyses involved 198 students who enrolled in PhD (61% of 324 applications; Fig 1). The descriptive data on the characteristics of the students are shown in Table 1. In the whole cohort, median (25^th percentile, 75^th percentile) and mean (standard deviation; SD) number of publications were 1.0 (0.0, 3.0) and 2.8 (4.4), impact factor 0.86 (0.00, 2.61) and 1.59 (2.36), citations per publication 0.0 (0.0, 4.5) and 3.5 (7.4) and total citations 0.0 (0.0, 17.0) and 19.6 (49.8). S1 Table presents the stability of the explanatory variables across each year of student applications. The relationship between ranking criteria and PhD student output metrics are shown in Table 2 (non-parametric analyses) and Table 3 (parametric analyses). Findings of both non-parametric and parametric analyses were similar. Non-parametric (S1 Table) and parametric (S2 Table) effect sizes as well as variability among variables by year of application (S3 Table) are reported in the data supplement.

Table 1. Descriptive data for the ranking criteria of the 198 unique PhD applications and risk of withdrawing from PhD.

Variable	N (%)	Withdrawing from PhD
Variable	N (%)	Odds ratio	Relative risk
Student research training degree (n = 183)		1.23 (P = 0.179)
1st class honours, top tertile	90 (49.2)		1.00
1st class honours, middle tertile	34 (18.6)		0.66 (0.24, 1.84)
1st class honours, lower tertile	24 (13.1)		0.70 (0.22, 2.22)
2nd class honours	35 (19.1)		1.77 (0.91, 3.32)
Student undergraduate grades (n = 167)		0.77 (P = 0.288)
Grade point average ≥80% plus prizes awarded	14 (8.4)		1.00
Grade point average ≥80%	46 (27.5)		0.68 (0.25, 1.89)
Grade point average ≥70%, but less than 80%	79 (47.3)		0.44 (0.16, 1.22)
Grade point average ≥60%, but less than 70%	28 (16.8)		0.63 (0.20, 1.97)
Student had prior publication (n = 194)		0.93 (P = 0.864)
Yes	51 (26.3)		1.00
No	143 (73.7)		0.96 (0.50, 1.85)
Student academic merit (n = 198)		1.30 (P = 0.110)
1^st quartile	48 (24.3)		1.00
2^nd quartile	50 (25.3)		1.10 (0.43, 2.79)
3^rd quartile	50 (25.3)		1.10 (0.43, 2.79)
4^th quartile	50 (25.3)		1.92 (0.85, 4.34)
Supervisor in strategic research centre (n = 198)		1.34 (P = 0.46)
Yes	143 (72.2)		1.00
No	55 (27.8)		1.25 (0.68, 2.31)
Supervisor academic level at application (n = 184)		1.32 (P = 0.216)
Full-professor	53 (28.8)		1.00
Associate professor	52 (28.3)		0.91 (0.38, 2.17)
Senior lecturer or lecturer	79 (42.9)		1.49 (0.74, 3.02)
Alignment of research achieved maximum score (n = 198)		2.88 (P = 0.004)
Yes	165 (83.3)		1.00
No	33 (16.7)		2.35 (1.31, 4.20)
Supervisory team achieved maximum score (n = 198)		1.95 (P = 0.126)
Yes	127 (64.1)		1.00
No	71 (35.9)		1.66 (0.87, 3.17)
Scholarship awarded (n = 198)		3.04 (P = 0.006)
Yes	90 (45.5)		1.00
No	108 (54.6)		2.59 (1.29, 5.21)

Open in a new tab

Number of students for which each variable was available is indicted in brackets following variable. Data for each level of each variable are count (percentage of available data). Data for withdrawal from PhD are odds ratio (P-value) in second column from right for the parameter overall and relative risk (95% confidence interval) compared to the reference level in the right column. Significant risk ratios are bolded.

Table 2. Non-parametric analyses: Associations between the ranking criteria of the 198 unique PhD applications and researcher metrics.

Variable	Number of publications	Number of citations	Number of citations per publication	Average impact factor
Student research training degree	P = 0.361	P = 0.383	P = 0.429	P = 0.199
1st class honours, top	2.0 (0.0, 4.0)	2.0 (0.0, 19.0)	1.0 (0.0, 3.9)	1.3 (0.0, 3.0)
1st class honours, middle	1.0 (0.0, 3.0)	2.0 (0.0, 18.0)	1.0 (0.0, 5.9)	1.0 (0.0, 2.5)
1st class honours, lower	1.0 (0.0, 3.0)	0.0 (0.0, 10.0)	0.0 (0.0, 3.3)	0.3 (0.0, 2.4)
2nd class honours	0.0 (0.0, 3.0)	0.0 (0.0, 30.0)	0.0 (0.0, 9.0)	0.0 (0.0, 2.2)
Student undergraduate rank	P = 0.588	P = 0.668	P = 0.643	P = 0.420
GPA≥80% plus prizes	2.0 (0.0, 3.0)	3.0 (0.0, 19.0)	1.2 (0.0, 6.0)	2.2 (0.0, 3.0)
GPA≥80%	1.0 (0.0, 3.0)	0.0 (0.0, 9.0)	0.0 (0.0, 3.4)	0.8 (0.0, 3.0)
GPA≥70% and <80%	2.0 (0.0, 4.0)	1.0 (0.0, 18.0)	0.3 (0.0, 4.0)	1.0 (0.0, 2.6)
GPA≥60% and <70%	0.5 (0.0, 2.0)	0.0 (0.0, 15.0)	0.0 (0.0, 3.6)	0.0 (0.0, 2.5)
Student had prior publication	P = 0.551	P = 0.436	P = 0.545	P = 0.809
Yes	1.0 (0.0, 4.0)	1.0 (0.0, 20.0)	0.7 (0.0, 4.5)	1.1 (0.0, 2.3)
No	1.0 (0.0, 3.0)	0.0 (0.0, 14.0)	0.0 (0.0, 4.5)	0.7 (0.0, 2.8)
Student academic merit	P = 0.017	P = 0.080	P = 0.082	P = 0.001
1^st quartile	2.0 (1.0, 5.0)^d	4.0 (0.0, 19.0)	1.7 (0.0, 5.2)	2.2 (0.6, 3.3)^d#
2^nd quartile	0.5 (0.0, 4.0)	0.0 (0.0, 17.0)	0.0 (0.0, 3.9)	0.0 (0.0, 2.3)
3^rd quartile	1.0 (0.0, 3.0)	0.0 (0.0, 29.0)	0.0 (0.0, 5.0)	0.9 (0.0, 2.4)
4^th quartile	0.0 (0.0, 2.0)	0.0 (0.0, 10.0)	0.0 (0.0, 3.0)	0.0 (0.0, 2.2)
Supervisor in institute or research centre	P<0.001	P<0.001	P<0.001	P<0.001
Yes	2.0 (0.0, 4.0)^c†	3.0 (0.0, 23.0)^c†	1.0 (0.0, 5.7)^c†	1.4 (0.0, 3.0)^c†
No	0.0 (0.0, 2.0)	0.0 (0.0, 2.0)	0.0 (0.0, 1.0)	0.0 (0.0, 1.0)
Supervisor academic level at application	P = 0.107	P = 0.482	P = 0.524	P = 0.125
Full-professor	2.0 (0.0, 4.0)	3.0 (0.0, 18.0)	1.0 (0.0, 4.8)	1.2 (0.0, 2.6)
Associate professor	1.0 (0.0, 3.5)	1.5 (0.0, 18.0)	1.0 (0.0, 4.4)	1.2 (0.0, 3.0)
Senior lecturer or lecturer	0.0 (0.0, 3.0)	0.0 (0.0, 12.0)	0.0 (0.0, 3.9)	0.0 (0.0, 2.5)
Supervisory team achieved maximum score	P<0.001	P = 0.003	P<0.001	P<0.001
Yes	2.0 (0.0, 5.0)^c*	4.0 (0.0, 20.0)^b*	1.5 (0.0, 5.9)^c	1.5 (0.0, 3.0)^c†
No	0.0 (0.0, 2.0)	0.0 (0.0, 3.0)	0.0 (0.0, 1.0)	0.0 (0.0, 2.1)
Alignment of research achieved maximum score	P = 0.273	P = 0.233	P = 0.146	P = 0.123
Yes	1.0 (0.0, 3.0)	1.0 (0.0, 18.0)	1.0 (0.0, 4.8)	1.0 (0.0, 2.8)
No	0.0 (0.0, 3.5)	0.0 (0.0, 11.0)	0.0 (0.0, 3.2)	0.0 (0.0, 2.1)
Scholarship awarded	P<0.001	P<0.001	P<0.001	P<0.001
Yes	2.0 (1.0, 6.0)^c‡	8.5 (0.0, 35.0)^c†	2.4 (0.0, 5.8)^c*	2.2 (0.7, 3.2)^c‡
No	0.0 (0.0, 2.0)	0.0 (0.0, 6.0)	0.0 (0.0, 1.4)	0.0 (0.0, 1.7)

Open in a new tab

Data for publication numbers, citations and impact factors are median (25^th percentile, 75^th percentile). For these variables, significance of difference is indicated by ^a P<0.05, ^b P<0.01, ^c P<0.001 compared to ‘no’ and ^d P<0.05 compared to all other quartiles. P-values next to parameter name are from the Kruskal-Wallis one-way analysis of variance test. S2 Table presents the effect sizes for these analyses.

Table 3. Parametric analyses: Associations between the ranking criteria of the 198 unique PhD applications and researcher metrics.

Variable	Number of publications	Number of citations	Number of citations per publication	Average impact factor
Student research training degree	P = 0.522	P = 0.237	P = 0.747	P = 0.240
1st class honours, top	3.2 (4.4)	19.6 (41.4)	3.3 (5.3)	1.96 (2.95)
1st class honours, middle	2.5 (3.4)	14.9 (26.1)	3.3 (4.6)	1.35 (1.46)
1st class honours, lower	1.8 (2.5)	11.0 (29.4)	4.1 (14)	1.49 (2.09)
2nd class honours	3.1 (6.6)	35.6 (90.2)	4.8 (9.2)	1.06 (1.44)
Student undergraduate rank	P = 0.652	P = 0.994	P = 0.640	P = 0.077
GPA≥80% plus prizes	2.8 (3.5)	17.7 (35.1)	4.2 (6.1)	3.18 (5.65)
GPA≥80%	2.1 (3)	15.3 (32.0)	3.1 (5.2)	1.42 (1.74)
GPA≥70% and <80%	3.1 (4.6)	16.4 (38.1)	2.5 (3.7)	1.61 (2.10)
GPA≥60% and <70%	2.5 (5)	17.4 (42.1)	3.4 (6.9)	1.19 (1.61)
Student had prior publication	P = 0.298	P = 0.876	P = 0.792	P = 0.337
Yes	3.4 (5.3)	20.7 (51.1)	3.8 (7.3)	1.32 (1.43)
No	2.6 (4.1)	19.4 (50.1)	3.5 (7.6)	1.70 (2.62)
Student academic merit	P = 0.758	P = 0.952	P = 0.780	P = 0.005
1^st quartile	3.3 (3.3)	19.7 (33.5)	3.5 (4.5)	2.46 (3.33) #
2^nd quartile	2.8 (4.8)	20.1 (51.4)	2.8 (5.1)	1.38 (2.05)
3^rd quartile	2.8 (4.4)	21.3 (53.2)	4.4 (8.5)	1.39 (1.51)
4^th quartile	2.3 (3.3)	17.2 (60.4)	3.5 (10.7)	1.06 (1.76)
Supervisor in institute or research centre	P = 0.002	P = 0.010	P = 0.009	P = 0.001
Yes	3.4 (4.9)†	25.2 (57.1)†	4.4 (8.4)†	1.94 (2.57)†
No	1.3 (2.4)	5.0 (12.6)	1.3 (2.9)	0.71 (1.33)
Supervisor academic level at application	P = 0.166	P = 0.969	P = 0.160	P = 0.156
Full-professor	3.9 (6.0)	21.4 (56.5)	3.2 (4.7)	1.55 (1.53)
Associate professor	2.7 (3.6)	21.1 (48.2)	5.3 (12.2)	2.17 (3.55)
Senior lecturer or lecturer	2.4 (3.9)	19.3 (50.3)	2.7 (4.5)	1.35 (1.89)
Supervisory team achieved maximum score	P = 0.014	P = 0.012	P = 0.159	P = 0.005
Yes	3.4 (4.4)*	26.2 (59.4)*	4.1 (6.3)	1.95 (2.60)†
No	1.8 (4.3)	7.8 (20.1)	2.5 (9.1)	0.97 (1.69)
Alignment of research achieved maximum score	P = 0.862	P = 0.202	P = 0.521	P = 0.336
Yes	2.8 (4.3)	21.7 (53.8)	3.7 (7.3)	1.68 (2.43)
No	2.6 (5.1)	9.1 (16)	2.7 (7.9)	1.20 (1.90)
Scholarship awarded	P<0.001	P = 0.001	P = 0.048	P<0.001
Yes	4.2 (4.8)‡	32.1 (66.2)†	4.7 (8.4)*	2.48 (2.92)‡
No	1.6 (3.7)	9.2 (26)	2.6 (6.4)	0.86 (1.39)

Open in a new tab

Data for publication numbers, citations and impact factors are mean (standard deviation). For these variables, significance of difference is indicated by * P<0.05, † P<0.01, ‡ P<0.001 compared to ‘no’ and # P<0.05 compared to all other quartile. P-values next to parameter name are from one-way analysis of variance test. S3 Table presents the effect sizes for these analyses.

Number of publications

On univariate analysis (Tables 2 and 3, Fig 2), primary supervisor being located in a strategic research centre (non-parametric and parametric both: P≤0.014), supervisory teams who received a maximum score (both: P≤0.014), being awarded a scholarship (both: P<0.001), student academic merit score (non-parametric: P = 0.017, parametric: P = 0.758) were associated with this outcome, but student undergraduate performance (both: P≥0.588), student research training degree outcome (e.g. first-class honours upper band; both: P≥0.262), research topic (both: P≥0.347), primary supervisor academic level (both: P≥0.107) were not.

Step-wise regression models (Table 4) showed that receiving a scholarship (P = 0.001), primary supervisor being located in a strategic research centre (P = 0.018) remained in final model for number of publications, and whilst 'research topic' remained in the final model, it was not significant (P = 0.076). Factorial ANOVA (S4 Table) yielded similar results (having a scholarship, supervisory teams who received a maximum score, primary supervisor being located in a strategic research centre were associated, but not student related variables).

Table 4. Results of step-wise regression.

			Model
Variable	Final model terms	t-value (P-value)	r²	adjusted r²	F-value (P-value)
Number of publications	Scholarship awarded	3.43 (P = 0.001)	0.122	0.104	6.73 (P<0.001)
	Supervisor in strategic research centre	2.39 (P = 0.018)
	Research alignment maximum score	1.79 (P = 0.076)
Number of citations	Supervisory team maximum score	2.09 (P = 0.039)	0.072	0.059	5.70 (P = 0.004)
	Scholarship awarded	1.95 (P = 0.053)
Number of citations per publication	Supervisor in strategic research centre	1.77 (P = 0.079)	0.062	0.049	4.86 (P = 0.009)
	Supervisory team maximum score	1.72 (P = 0.087)
Average impact factor	Scholarship awarded	3.84 (P<0.001)	0.133	0.121	11.22 (P<0.001)
	Supervisor in strategic research centre	1.97 (P = 0.051)

Open in a new tab

Data are t-value (P-value), r², adjusted r² or F-value (P-value) derived from the final step-wise regression model. See also factorial ANOVA, reported in S4 Table, which yielded similar results.

Number of citations

On univariate analysis (Tables 2 and 3, Fig 2), primary supervisor being located in a strategic research centre (non-parametric and parametric P both≤0.010), supervisory teams who received a maximum score (both: P≤0.012), being awarded a scholarship (both: P<0.001) were associated with this outcome, but student undergraduate performance (both: P≥0.668), student research training degree outcome (e.g. first-class honours upper band; both: P≥0.237), student academic merit score (both: P≥0.080), research topic (both: P≥0.202), primary supervisor academic level (both: P≥0.482) were not.

Step-wise regression models (Table 4) showed that supervisory team who received a maximum score (P = 0.039) and the receiving a scholarship (P = 0.053), but in this case the scholarship award was not significant. Factorial ANOVA (S4 Table) yielded similar results (having a scholarship and supervisory teams who received a maximum score were associated, but not student related variables).

Citations per publications

On univariate analysis (Tables 2 and 3, Fig 2), primary supervisor being located in a strategic research centre (non-parametric and parametric P both P≤0.009), supervisory teams who received a maximum score (non-parametric: P<0.001, parametric: P = 0.159), being awarded a scholarship (both: P≤0.048) were associated with this outcome, but student undergraduate performance (both: P≥0.640), student research training degree outcome (e.g. first-class honours upper band; both: P≥0.668), student academic merit score (both: P≥0.082), research topic (both: P≥0.185), primary supervisor academic level (both: P≥0.160) were not.

Step-wise regression models (Table 4) showed that primary supervisor being located in a strategic research centre (P = 0.079) and supervisory team achieving maximum score (P = 0.087) remained in the final model, but neither terms were significant. Factorial ANOVA (S4 Table) yielded similar results (having a scholarship and supervisory teams who received a maximum score approached, but did not reach, significance).

Average impact factor

On univariate analysis (Tables 2 and 3, Fig 2), primary supervisor being located in a strategic research centre (non-parametric and parametric P both P≤0.001), supervisory teams who received a maximum score (both: P≤0.005), being awarded a scholarship (both: P<0.001), student academic merit score (both: P≤0.005), were associated with this outcome, but student undergraduate performance (both: P≥0.077), student research training degree outcome (e.g. first-class honours upper band; both: P≥0.238), research topic (both: P≥0.161), primary supervisor academic level (both: P≥0.125) were not.

Step-wise regression models (Table 4) showed that receiving a scholarship (P<0.001) and primary supervisor being located in a strategic research centre (P = 0.051) remained in the final model, with the latter not achieving statistical significance. Factorial ANOVA (S4 Table) yielded similar results (having a scholarship was significant, but supervisor related variables approached, but did not reach, significance; student related variables were not significant).

Drop-out from PhD

Odds ratios for student attrition is shown in Table 1. Students were more than two times more likely to withdraw from their PhD when the supervisory team did not achieve maximum score (odds ratio [95% confidence interval] 2.88[1.39, 5.93], P = 0.004) or a scholarship was not awarded (odds ratio [95% confidence interval] 3.04[1.37, 6.73], P = 0.006). No other independent variables significantly predicted the likelihood of withdrawal.

The final multiple logistic regression model (χ² = 13.80, df = 3, P = 0.003) for predicting withdrawal from PhD included maximum supervisory team score (OR = 3.29, P = 0.013; i.e. lower risk of withdrawal when the supervisor score was maximum), student undergraduate degree grades (OR = 0.58, P = 0.047; i.e. reduced risk for each GPA rank lower) and receiving a scholarship (OR = 2.30, P = 0.090; i.e. lower risk when scholarship received), albeit the latter was not significant.

Associations between explanatory variables

Students in the highest quartile of academic merit received the most (42%) of all scholarships awarded. Of those in the highest quartile of academic merit, 79% received scholarships, compared to 62% in the second quartile, 20% in the third quartile and 22% in the lowest quartile.

Students who received a scholarship were more often supervised by strong supervisory teams (χ² = 9.346, P = 0.002; Table 5) and by supervisors who were located in a strategic research centre (χ² = 8.225, P = 0.004; Table 5). Supervisors who were in a strategic research centre were more likely to attract students in the highest quartile of academic merit (χ² = 3.899, P = 0.048; Table 6). Supervisory teams who received a maximum score were more likely to attract students in the highest quartile of academic merit (χ² = 10.147, P = 0.001; Table 6).

Table 5. Students who received a scholarship were most often supervised by stronger supervisory teams and supervisors who were located in a strategic research centre.

Has scholarship	Supervisor score is maximum
Has scholarship	Yes	No
	Supervisor in strategic research centre: Yes
Yes	61 (30.8)	13 (6.6)
No	44 (22.2)	25 (12.6)
	Supervisor in strategic research centre: No
Yes	7 (3.5)	9 (4.5)
No	15 (7.6)	24 (12.1)

Open in a new tab

Data are count (percentage of total sample). See text for results from chi-squared statistics and corresponding P-values.

Table 6. Stronger supervisory teams and supervisors who were located in a strategic research centre were more likely to attract students in the highest quartile of academic merit.

Student is in top quartile of academic merit	Supervisor score is maximum
Student is in top quartile of academic merit	Yes	No
	Supervisor in strategic research centre: Yes
Yes	34 (17.2)	6 (3.0)
No	71 (35.9)	32 (16.2)
	Supervisor in strategic research centre: No
Yes	6 (3.0)	2 (1.0)
No	16 (8.1)	31 (15.7)

Open in a new tab

Data are count (percentage of total sample). See text for results from chi-squared statistics and corresponding P-values.

Discussion

To the best of our knowledge, this is the first analysis of PhD student outcomes in relation to their research environment, their academic abilities and prior research training. The key finding was that the 'research environment', such as whether the supervisor was in a research centre or institute and the research experience of the supervision team, were most significant predictors of, with the largest effect sizes for, student outcomes. In contrast, the students' previous academic outcomes and previous research training were not predictors. Receiving a PhD scholarship had a significant influence on positive student outcomes and was more important than students being judged as having the highest academic merit. Receiving a scholarship occurred more frequently in students tied to stronger supervisory teams and supervisors in strategic research centres.

Entry to a PhD is typically restricted to those students with a minimum grade in a prior Masters or Honours degree [16]. At our university, prospective PhD students are required to have completed a research project with a dissertation of at least 25% of one year full-time study at Honours or Masters level and their grade needs to have been at least 70%. Our findings suggest that once students meet the minimum academic ability for entry into PhD, any further ability or research training above that does not influence the outcome of their PhD. This is in line with findings that scientist’s intelligence quotient does not correlate with their citation rates [17].

By contrast, it is the research environment in which the student is embedded that is decisive for the outcomes of their PhD; including the strength of their supervisory team. This is in line with the hypothesis of “accumulative advantage”, also known as “Matthew effects” in science [18] where differences between scientists at an early stage of their career become reinforced over time [19]. The standing of a PhD supervisor directly influences [8] the future career trajectory, and number of citations, their students receive throughout their career. Also, the standing of a department influences the future employment chances of its PhD graduates, on average, more than the individual achievements of those students [13]. The impact of teacher quality is seen in other areas of education [20,21], although ‘PhD supervisor quality’ is assessed differently to teacher quality in school and undergraduate education.

There are other factors known to impact the number and impact of publication outputs. Research collaboration has clearly been shown to lead to higher impact publications [22–25]. In the health-sciences field, publications of higher levels of evidence [26] are more likely to be cited. Similarly interventional (rather than observational) and prospective (rather than retrospective) studies [25,27], as well as randomised controlled trials and basic science papers [28] are more likely to be cited. Papers published in high impact factor journals will be more often cited simply for that reason [23,25]. We argue these factors are more likely to be determined by the research culture in which the student are embedded, as opposed to being determined by the student alone.

We also showed that receiving a PhD scholarship contributed to the students’ outcomes, in particular with more publications arising, more citations higher impact factor journals. In step-wise regression, we found that impact of the scholarship persisted for the number of publications and average impact factor of the journals in which the students published. This finding is in line with prior work [29] that showed PhD students receiving scholarships to support their studies published more peer reviewed papers. Similar to prior work [29], our results showed that receiving a scholarship was also associated with lower withdrawal rates.

Students were awarded scholarships based on their prior academic performance [30]. At this university, whilst the student’s academic merit contributed to 60% of their total ranking score, in practice this was the most decisive factor in determining which applicants were offered scholarships first. We show here, however, that the most significant attributes for PhD success were research environment and the performance metrics of the supervision team. How these attributes may influence employment opportunities post PhD also warrants further investigation.

Strengthening the research environment is also worthy of further investigation. Prior work [12] has shown that very few university departments rely solely on a small number of high-performing researchers for its research productivity. We show here that supervisor team quality has a key impact on the PhD student’s outcomes. Therefore, having more highly trained researchers is likely to lead to overall higher research student productivity, such as in having a higher percentage of faculty members who are at full-professor level [12]. Strategies for strengthening the research capacity of academic staff and potential supervisors include [31] structured research mentoring of academic staff, formal requirements for further academic research training.

The strengths of this analysis include being a prospective analysis of outcomes based on data that were known at the time of student selection. The limitations of the analysis were that it was focussed on one faculty at one university. It was not possible to conduct this analysis more widely at our university or at other universities as not all faculties and universities collate the same data on their PhD applicants. It would be relevant to examine such patterns at a wider range of universities, however obtaining such data from other universities is further complicated by data from scholarship ranking being confidential internal university information. Whilst this study was comprised one university, we believe its findings can easily be extrapolated to other regions of Australia and/or the world. Furthermore, we focussed on outcomes from PhDs that relate to university ranking procedures. Other outcomes, such as employment achieved post-PhD, student satisfaction, mental health are important to consider more widely.

Conclusions

In conclusion, to best of our knowledge, our study is the first to examine the relative importance of the environment versus student ability in the allocation and outcomes of their PhD. Our key finding was that the research environment is likely more important for supporting PhD students to produce larger numbers of highly cited publications in higher impact journals. Once the minimum level of academic ability and research training is met for entry to PhD, working with a strong research focussed supervisory team, being embedded in a research intensive institute, and receiving a scholarship are also important factors for publication and citation outcomes.

Supporting information

S1 Table. Non-parametric effect sizes between the ranking criteria of the 198 unique PhD applications and researcher metrics.

Data are Cohen’s d. Bold = P<0.05. GPA: Grade point average.

(DOCX)

Click here for additional data file.^{(46.7KB, docx)}

S2 Table. Parametric effect sizes between the ranking criteria of the 198 unique PhD applications and researcher metrics.

Data are Cohen’s d. Bold = P<0.05. GPA: Grade point average.

(DOCX)

Click here for additional data file.^{(46.5KB, docx)}

S3 Table. Variability among variables by year of application.

Dependent variables are mean (standard deviation), expect withdrawing from PhD which are number (percentage within year). Explanatory variables are number (percentage within year). GPA: Grade point average.

(DOCX)

Click here for additional data file.^{(47.7KB, docx)}

S4 Table. Results from factorial ANOVA.

Data are F-value (corresponding P-value). ANOVA fits explanatory variables sequentially to the dependent variables. Explanatory variables were fitted to the dependent variables in the order above (i.e. top variable at left fitted first, followed by the second to top variable). This therefore accounted for potential association of student related factors first to PhD outcomes, with then having a scholarship and then supervisor related factors considered. Despite accounting for student related variables first, having a scholarship and supervisor quality were most consistently associated with outcomes from a student’s PhD.

(DOCX)

Click here for additional data file.^{(43.7KB, docx)}

Acknowledgments

The authors thank Grant Michie, Rachelle DeBrito and their teams for assistance with access to enrolment and publication output data, Steve Sawyer for assistance in reviewing and accessing the scholarship application data and biostatistician A/Prof Steven Bowe for statistical advice.

Data Availability

Participants did not give consent for their data to be published in online databanks and data are accessible with appropriate ethical approvals. Interested parties may contact the authors and/or the Deakin University Human Research Ethics Committee research-ethics@deakin.edu.au to gain access to the data.

Funding Statement

The author(s) received no specific funding for this work.

References

1.Baptista A. et al. The doctorate as an original contribution to knowledge: Considering relationships between originality, creativity, and innovation. Frontline Learn. Res. 3, 51–63 (2015). [Google Scholar]
2.Australian Government Department of Education. Research Training Program. Research Training Program https://www.education.gov.au/research-training-program (2019).
3.Hazelkorn E. The Impact of League Tables and Ranking Systems on Higher Education Decision Making. High. Educ. Manag. Policy 1–24 (2007) 10.1787/hemp-v19-art12-en [DOI] [Google Scholar]
4.QS World University Rankings Methodology https://www.topuniversities.com/qs-world-university-rankings/methodology (2018).
5.World University Rankings 2019: methodology https://www.timeshighereducation.com/world-university-rankings/world-university-rankings-2019-methodology (2018).
6.Academic Ranking of World Universities Methodology http://www.shanghairanking.com/ARWU-Methodology-2018.html (2018).
7.Larivière V. On the shoulders of students? The contribution of PhD students to the advancement of knowledge. Scientometrics 90, 463–481 (2012). [Google Scholar]
8.Waldinger F. Quality Matters: The Expulsion of Professors and the Consequences for PhD Student Outcomes in Nazi Germany. J. Polit. Econ. 118, 787–831 (2010). [Google Scholar]
9.Long J. S. & McGinnis R. The effects of the mentor on the academic career. Scientometrics 7, 255–280 (1985). [Google Scholar]
10.Agrawal A., McHale J. & Oettl A. How stars matter: Recruiting and peer effects in evolutionary biology. Res. Policy 46, 853–867 (2017). [Google Scholar]
11.Ramesh Babu A. & Singh Y. P. Determinants of research productivity. Scientometrics 43, 309–329 (1998). [Google Scholar]
12.Dundar H. & Lewis D. R. Determinants of research productivity in higher education. Res. High. Educ. 39, 607–631 (1998). [Google Scholar]
13.Stenstrom D. M., Curtis M. & Iyer R. School Rankings, Department Rankings, and Individual Accomplishments: What Factors Predict Obtaining Employment After the PhD? Perspect. Psychol. Sci. 8, 208–217 (2013). 10.1177/1745691612474316 [DOI] [PubMed] [Google Scholar]
14.Firth D. Bias reduction of maximum likelihood estimates. Biometrika 80, 27–38 (1993). [Google Scholar]
15.Heinze G. & Schemper M. A solution to the problem of separation in logistic regression. Stat. Med. 21, 2409–19 (2002). 10.1002/sim.1047 [DOI] [PubMed] [Google Scholar]
16.Haidar, H. PhD admission requirements. What is a PhD? https://www.topuniversities.com/blog/what-phd.
17.Bayer A. E. & Folger J. Some Correlates of a Citation Measure of Productivity in Science. Sociol. Educ. 39, 381 (1966). [Google Scholar]
18.Merton R. K. The Matthew Effect in Science: The reward and communication systems of science are considered. Science 159, 56–63 (1968). [PubMed] [Google Scholar]
19.Broström A. Academic breeding grounds: Home department conditions and early career performance of academic researchers. Res. Policy 48, 1647–1665 (2019). [Google Scholar]
20.Araujo M. C., Carneiro P., Cruz-Aguayo Y. & Schady N. Teacher Quality and Learning Outcomes in Kindergarten *. Q. J. Econ. 131, 1415–1453 (2016). [Google Scholar]
21.Rowe, K. The importance of teacher quality as a key determinant of students’ experiences and outcomes of schooling. in 15–23 (2003).
22.Hsu J. & Huang D. Correlation between impact and collaboration. Scientometrics 86, 317–324 (2011). [Google Scholar]
23.Royle P., Kandala N.-B., Barnard K. & Waugh N. Bibliometrics of systematic reviews: analysis of citation rates and journal impact factors. Syst. Rev. 2, (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Levitt J. M. & Thelwall M. Citation levels and collaboration within library and information science. J. Am. Soc. Inf. Sci. Technol. 60, 434–442 (2009). [Google Scholar]
25.Falagas M. E., Zarkali A., Karageorgopoulos D. E., Bardakas V. & Mavros M. N. The impact of article length on the number of future citations: a bibliometric analysis of general medicine journals. PloS One 8, e49476 (2013). 10.1371/journal.pone.0049476 [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Okike K. et al. Level of evidence and conflict of interest disclosure associated with higher citation rates in orthopedics. J. Clin. Epidemiol. 64, 331–338 (2011). 10.1016/j.jclinepi.2010.03.019 [DOI] [PubMed] [Google Scholar]
27.Willis D. L., Bahler C. D., Neuberger M. M. & Dahm P. Predictors of citations in the urological literature. BJU Int. 107, 1876–1880 (2011). 10.1111/j.1464-410X.2010.10028.x [DOI] [PubMed] [Google Scholar]
28.Bhandari M. et al. Factors associated with citation rates in the orthopedic literature. Can. J. Surg. J. Can. Chir. 50, 119–123 (2007). [PMC free article] [PubMed] [Google Scholar]
29.Larivière V. PhD students’ excellence scholarships and their relationship with research productivity, scientific impact, and degree completion. Can. J. High. Educ. 43, 27–41 (2013). [Google Scholar]
30.The University of Melbourne. How are graduate research scholarships awarded? https://education.unimelb.edu.au/study/scholarships/scholarship-support.
31.Huenneke L. F., Stearns D. M., Martinez J. D. & Laurila K. Key Strategies for Building Research Capacity of University Faculty Members. Innov. High. Educ. 42, 421–435 (2017). 10.1007/s10755-017-9394-y [DOI] [PMC free article] [PubMed] [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0236327.r001

Decision Letter 0

Cesario Bianchi

31 Oct 2019

PONE-D-19-27758

Are successful PhD outcomes dependent on the research environment or academic ability?

PLOS ONE

Dear Dr. Belavý,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we have decided that your manuscript does not meet our criteria for publication and must therefore be rejected.

Specifically:

Unfortunately the manuscript has an heterogeneous student population, lacks the correct statistical analysis and a the small sample size may preclude meaningful conclusions.

I am sorry that we cannot be more positive on this occasion, but hope that you appreciate the reasons for this decision.

Yours sincerely,

Cesario Bianchi

Academic Editor

PLOS ONE

Additional Editor Comments (if provided):

Dear Dr Belavý,

Thank you for submitting your manuscript about an very interesting and important topic. Based on comments of the reviewers and myself I have to reject your present manuscript .

Please, take reviewers 1 comments as a positive criticism to improve your manuscript.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: No

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The present paper covers an important topic, i.e. research policy and selection of PhD students. Very little research has been done on the association between student characteristics, the scientific environment, research output and benefits for society. In this context, the paper covers an important topic and the researchers may have had access to some interesting data.

The study is based on data from one Australian university. Of the 324 applications submitted to the university, 198 students were enrolled in the PhD program, 120 completed the program, 31 are still enrolled, 37 withdraw after starting, and 11 withdrew before starting.

Unfortunately, very few conclusions can be drawn from this small study due to several issues with the statistical analysis and study design.

First of all, the study is a cross-sectional study, and not all students have the same follow-up time. For example, 31 students are still enrolled in the program at the time of the data analysis. Since the inclusion period is from 2010-2013, some students may have been followed for 8 years, others for 4-5 years (please see Rothman KJ. Epidemiology. An Introduction. P. 97. Oxford University Press, 2nd ed., 2012). Moreover, the reader cannot really evaluate the underlying distribution of the data and their validity.

To mention some examples: The mean number of publications was 2.8 with a standard deviation of 4.4; the average impact factor was 1.9 with a standard deviation of 2.36; the mean number of citations per publication was 3.5 with a standard deviation of 7.4; and the total citation was 19.6 with a standard deviation of 49.8.

95% of a normal distribution is to be calculated as mean 1.96 ± standard deviation. This means that if the data were to be correctly described large proportion of students would have negative number of publications, negative impact factor, negative citations, and negative total citations. This does not make sense.

Moreover, it is not clear if this is an etiologic or prediction study (please see Clayton D, Hills M. Statistical Models in Epidemiology, P. 271, Chapter 27, Choice and Interpretation of Models. Oxford University Press, 1993). If it is an etiologic study, a step-wise logistic regression model does not make much sense (please see Rothman KJ. Epidemiology. An Introduction. P. 194. Oxford University Press, 2nd ed., 2012).

This relatively small study has limited statistical precision of the estimates as evident from Table 2. Only significant results have been marked with bold, and many strong associations are ignored simply because they are not statistically significant (please see Amrhein V et al. Scientists rise up against statistical significance. Nature. 2019;567:305-307).

Moreover, the table also shows that the data cannot be described with standard deviation. In addition to lack of statistical power, this type of non-randomized observational study should not focus on statistical significance, but on estimation of the effect (please see Rothman KJ. Six persistent research misconceptions. J Gen Intern Med. 2014;29:1060-4).

Another problem with logistic regression is the rare outcome assumption, which seem to be the case in the present paper (please see Hosmer DW, Lemeshow S, Sturdivant RX. Applied logistic regression, 3rd ed. P. 51. Hoboken: Wiley, 2013).

The study is conducted at a university in Australia as mentioned by the authors, but the external validity should be discussed in more detail.

Overall, the paper covers an important topic, but the study design and the statistical analysis as well as the small sample size make it impossible to draw any valid conclusions.

Reviewer #2: The manuscript that aims to analyzes the PhD students performance based on their previous academic achievement, research environment or supervisor importance is a great piece of work. Basically, the article puts numbers in parameters that the whole scientific community already has an idea, even qualitatively.

Although this is a regionalized study, it can easily extrapolate its findings to other areas of the globe, as the data discussed are important in worldwide PhD programs.

The only suggestion, which may or may not be accepted, would be the inclusion of some graphs, at least for the most significant findings, since understanding the tables is hampered by the number of presented data.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

- - - - -

For journal use only: PONEDEC3

PLoS One. 2020 Aug 5;15(8):e0236327. doi: 10.1371/journal.pone.0236327.r002

Author response to Decision Letter 0

5 Dec 2019

1. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: No

Reviewer #2: Yes

Author response: Reviewers differ in their opinions.

We thank Reviewer #2 for remarking this was ‘a great piece of work’. We also thank both Reviewer #1 and #2 for provided useful ideas. We have responded to the comments of both reviewers and updated the manuscript according. We believe the manuscript to be technically sound and that the data support the conclusions. Notably, as often happens with statistics, especially in larger sample sizes such as here, the different analysis approach did not change the main findings. We believe this contributes to the robustness of the manuscript.

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

Author response: Reviewers differ in their opinions.

We consulted with a professorial biostatistician for the statistical approach during the analysis of the data whilst drafting the original manuscript.

Reviewer #1 may have missed that we included non-parametric statistics in our analyses. We agree there are concerns using only parametric stats (e.g. assumptions of normality). Therefore, on statistical advice, we included non-parametric statistics. The results of these analyses were essentially the same as those from the parametric stats. This often is the case with larger sample sizes (see Central Limit Theorem).

Reviewer #2 considered the statistics appropriate in revising the manuscript. We further consulted with the same professorial biostatistician and are confident the statistics are appropriate

We respond to more specific comments on this point below and have revised the manuscript, accordingly, adding further data Tables and Supplemental Data.

3. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: No

Reviewer #2: Yes

Author response: Reviewers differ in their opinions.

Our reasons for being unable to provide raw data on the internet were disclosed with the submission. Some of the data are internal confidential university data which could potentially re-identify individual students included in the analyses. The data can be accessed by others with an ethically conform request.

4. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

Reviewer #2: Yes

Author response: No response required

Reviewer #1

Author response: Thank-you for pointing out the novelty and importance of our work.

Reviewer #1: The study is based on data from one Australian university. Of the 324 applications submitted to the university, 198 students were enrolled in the PhD program, 120 completed the program, 31 are still enrolled, 37 withdraw after starting, and 11 withdrew before starting.

Author response: The reviewer reiterates the data sample. No response required. Responses to the reviewer comments based on this summary are given below.

Reviewer #1: Unfortunately, very few conclusions can be drawn from this small study due to several issues with the statistical analysis and study design.

Author response: The data are unique and hard to get at wide scale: (a) key aspects of these data are derived from internal university (i.e. confidential) scholarship ranking panels, (b) universities are typically reluctant to release such information and are highly unlikely to share such data with external investigators due to strict privacy laws, (c) collection of the information is inconsistent between institutions, (d) even within our own university, other Faculties do not collect similar information. Thus, the data are unique.

The ~200 students in the sample are, in light of the granularity and depth of the data, a large sample size. Data involving 1000s of students is difficult to obtain and impossible to obtain at the level of granularity that we have analysed.

We believe our work is the first on this important topic and the difficulty in attaining the data highlights the uniqueness of our manuscript.

Reviewer #1: First of all, the study is a cross-sectional study, and not all students have the same follow-up time. For example, 31 students are still enrolled in the program at the time of the data analysis. Since the inclusion period is from 2010-2013, some students may have been followed for 8 years, others for 4-5 years (please see Rothman KJ. Epidemiology. An Introduction. P. 97. Oxford University Press, 2nd ed., 2012). Moreover, the reader cannot really evaluate the underlying distribution of the data and their validity.

Author response: We disagree with the Reviewer’s assertion that this is a cross-sectional study. A cross-sectional study is a comparison at a single time-point. In the current study, we retrospectively analysed an entire cohort over 4+ years of follow-up. A cross-sectional study is where all participants are examined at the same time with no follow-up data.

We painstakingly and carefully combined different data sets, which included checking other internal university databases for whether students changed their name during candidature (as they might publish manuscripts under their new surname, such as after getting married), to ensure we did not miss any one in this cohort.

RE: follow-up period. we thank the reviewer for this comment and agree that some students will have had a longer time to amass publications and citations. We conducted further analyses and student quality and other examined factors did not vary from one year to the next. We have presented this in Supplementary Table 3.

The parameters that were related to outcomes from PhD were stable over the years considered. Only student undergraduate rank varies over the years depending on year of application. Yet this parameter was unrelated to PhD outcomes. Thus, the longer follow-up period for applicants from 2011 versus those in later years will not impact the findings.

Reviewer #1: To mention some examples: The mean number of publications was 2.8 with a standard deviation of 4.4; the average impact factor was 1.9 with a standard deviation of 2.36; the mean number of citations per publication was 3.5 with a standard deviation of 7.4; and the total citation was 19.6 with a standard deviation of 49.8. 95% of a normal distribution is to be calculated as mean 1.96 ± standard deviation. This means that if the data were to be correctly described large proportion of students would have negative number of publications, negative impact factor, negative citations, and negative total citations. This does not make sense.

Author response: We thank the reviewer for this comment. We identified concerns with using only parametric stats (e.g. assumptions of normality). Therefore, after consulting with a professorial biostatistician prior to the original submission, we included non-parametric statistics. The results of these analyses were essentially the same as those from the parametric stats.

We have added a separate table for the non-parametric statistics (presenting medians and interquartile ranges; new Table 2) and retain the original table showing means and standard deviations (Table 3). As stated in the original manuscript, the results are largely the same. This is typical of larger datasets where parametric and non-parametric analyses give largely the same main findings (see Central Limit Theorem).

Notably, we consulted with the same professorial biostatistician for the original submission and this revision.

Reviewer #1: Moreover, it is not clear if this is an etiologic or prediction study (please see Clayton D, Hills M. Statistical Models in Epidemiology, P. 271, Chapter 27, Choice and Interpretation of Models. Oxford University Press, 1993). If it is an etiologic study, a step-wise logistic regression model does not make much sense (please see Rothman KJ. Epidemiology. An Introduction. P. 194. Oxford University Press, 2nd ed., 2012).

Author response: We sought to examine what baseline (at scholarship ranking) factors were associated with more/less publications, more/less citations and higher/lower impact factor. This is important for choosing (potentially: ‘predicting’) which students are more likely to have the best outcome.

Our work would then fall under the reviewer’s ‘prediction’ category, making the criticism of step-wise logistic regression not applicable.

We consulted with a professorial biostatistician for the original submission and this revision, hence we are confident that these statistics have been conducted appropriately.

Reviewer #1: This relatively small study has limited statistical precision of the estimates as evident from Table 2.

Author response: As mentioned prior:

The data are unique and hard to get at wide scale: (a) key aspects of these data are derived from internal university (i.e. confidential) scholarship ranking panels, (b) universities are typically reluctant to release such information and are highly unlikely to share such data with external investigators due to strict privacy laws, (c) collection of the information is inconsistent between institutions, (d) even within our own university, other Faculties do not collect similar information. Thus, the data are unique.

We believe our work is the first on this important topic and the difficulty in attaining the data highlights the uniqueness of our manuscript.

Reviewer #1: Only significant results have been marked with bold, and many strong associations are ignored simply because they are not statistically significant (please see Amrhein V et al. Scientists rise up against statistical significance. Nature. 2019;567:305-307).

Author response: In the revised manuscript, we presented effect sizes (see new Figure 2 and Supplemental Tables 1 and 2). As often happens in statistical analyses, the statistical significance is associated with the largest effect sizes.

Aside from statistically significant effects, we also included in the Discussion (paragraph 1) consideration of effect sizes.

Notably, we consulted with a professorial biostatistician for the original submission and this revision.

Reviewer #1: Moreover, the table also shows that the data cannot be described with standard deviation.

Author response: As mentioned prior:

Reviewer #1: Another problem with logistic regression is the rare outcome assumption, which seem to be the case in the present paper (please see Hosmer DW, Lemeshow S, Sturdivant RX. Applied logistic regression, 3rd ed. P. 51. Hoboken: Wiley, 2013).

Author response: This comment refers to the drop-out of students from PhD.

In our sample 19.7% of students dropped out of PhD at the Government’s census cut-off date. We posit that 20% is a frequent event rather than a ‘rare outcome’. However, this same topic also refers to the sample size and potential for bias in the logit coefficients (King & Zeng "Logistic Regression in Rare Events Data". Political Analysis, 2001. 9: 137-163; https://gking.harvard.edu/files/0s.pdf). We take on this consideration.

In the revised manuscript we implement the Penalized Maximum Likelihood Estimation proposed by Firth ("Bias reduction of maximum likelihood estimates" Biometrika 80:27-38; https://doi.org/10.1093/biomet/80.1.27). This is considered the best approach to implementing logistic regression in sample sizes such as in the current study.

The findings of the logistic regression analyses are similar to those presented in the original manuscript.

Reviewer #1: In addition to lack of statistical power, this type of non-randomized observational study should not focus on statistical significance, but on estimation of the effect (please see Rothman KJ. Six persistent research misconceptions. J Gen Intern Med. 2014;29:1060-4).

Author response: As mentioned prior:

RE: Sample size:

We believe our work is the first on this important topic and the difficulty in attaining the data highlights the uniqueness of our manuscript.

RE: effect sizes:

In the revised manuscript, we presented effect sizes (see new Figure 2 and Supplemental Tables 1 and 2). As often happens in statistical analyses, the statistical significance is associated with the largest effect sizes.

Aside from statistically significant effects, we also included in the Discussion (paragraph 1) consideration of effect sizes.

Notably, we consulted with a professorial biostatistician for the original submission and this revision.

Reviewer #1: The study is conducted at a university in Australia as mentioned by the authors, but the external validity should be discussed in more detail.

Author response: As mentioned prior:

RE: Sample size:

We believe our work is the first on this important topic and the difficulty in attaining the data highlights the uniqueness of our manuscript.

Furthermore, we discuss this further in the Discussion (2nd last paragraph).

Notably, Reviewer #2 stated ‘Although this is a regionalized study, it can easily extrapolate its findings to other areas of the globe’.

Reviewer #1: Overall, the paper covers an important topic, but the study design and the statistical analysis as well as the small sample size make it impossible to draw any valid conclusions.

Author response: We agree this is an important topic. We think the reviewer had some very good points. In light of the Reviewer’s comments, we have revised the manuscript and believe that it is much stronger as a consequence.

Reviewer #2

Author response: Thank-you for your positive comments on our manuscript.

Reviewer #2: Although this is a regionalized study, it can easily extrapolate its findings to other areas of the globe, as the data discussed are important in worldwide PhD programs.

Author response: We agree with the comment. We discuss this further in the Discussion (2nd last paragraph).

Reviewer #2: The only suggestion, which may or may not be accepted, would be the inclusion of some graphs, at least for the most significant findings, since understanding the tables is hampered by the number of presented data.

Author response: Thank-you for this suggestion. In the revised manuscript we include a new Figure 2 which displays the effect sizes of the main findings.

Attachment

Submitted filename: 2019_12_05_PONE-D-19-27758_Response.docx

Click here for additional data file.^{(31.6KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0236327.r003

Decision Letter 1

Sergi Lozano

21 Apr 2020

PONE-D-19-27758R1

Successful PhD outcomes are dependent on the research environment and not academic ability

PLOS ONE

Dear Dr. Belavý,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Notice that, in this round, a third reviewer has been invited to have a more complete assessment of your work.

When revising the article, please pay special attention to methodological concerns by both Reviewers #1 and #3. The comment made by Reviewer 2 regarding the title should also be considered.

We would appreciate receiving your revised manuscript by Jun 05 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Sergi Lozano

Filippo Radicchi, Ph.D.

Academic Editors

PLOS ONE

Journal Requirements:

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

3. Please remove your figures from within your manuscript file, leaving only the individual TIFF/EPS image files, uploaded separately. These will be automatically included in the reviewers’ PDF.

4. We note you have included a table to which you do not refer in the text of your manuscript. Please ensure that you refer to Table 5 in your text; if accepted, production will need this reference to link the reader to the Table.

5. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

6. Thank you for including your ethics statement on the online submission form: "Deakin University Human Research Ethics Committee reviewed this project (2019-191) and found it to be compliant with the Ethical Considerations in Quality Assurance and Evaluation Activities guidelines of the National Health and Medical Research Council of Australia and determined that no further ethics review was required. Consent was not obtained (the data were analyzed anonymously). ".

To help ensure that the wording of your manuscript is suitable for publication, would you please also add this statement at the beginning of the Methods section of your manuscript file

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: All comments have been addressed

Reviewer #3: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: No

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: No

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: No

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Reviewer #1: Thank you very much for giving me the opportunity to review the revised paper and the authors’ response.

The study focuses on an important topic, but the authors have only responded to the reviewer comments to a limited extent. Their main reply is that their data are unique and that they have

consulted a statistician. However, the manuscript still has several issues.

The material consists of fewer than 200 persons and the authors have conducted several cross-sectional analyses that the authors did not critically consider.

There is a large number of p-values in the manuscript yet the authors did not give any thought to statistical power in this small study population.

Moreover, the world’s leading and prominent scientists warned against this type of analysis, that is, statistical significance testing, in which only significant results are highligted. Please see Nature: https://www.nature.com/articles/d41586-019-00857-9

For this study to be a cohort analysis, the persons should either have had the same follow-up time, or the authors should have taken the various follow-up periods into consideration in the statistical analysis. This is not the case. The present analysis is solely cross-sectional.

It seems unclear to the authors if they have conducted a prediction study, or a causal study. It seems as if they tend to believe that their study is more of a causal study than a prediction study.

It is furthermore not clear what the authors want to estimate by use of odds ratios – is it prevalence rate ratio, incidence rate ratio, or relative risks? The study design does not meet the criteria for the latter two.

It still does not make any sense to report variables by mean and standard deviation.

Reviewer #2: Since the first submission, I found the manuscript interesting. Few studies, even if regionalized like this, show so clearly the importance of the group in which the student is inserted.

However, the title may sound aggressive in the way it is written. My opinion is that the words could be less impactful in the title...even if the results clearly show the importance of the group in the researcher formation

Reviewer #3: The manuscript studies factors related to PhD outcome. This is an important topic. However, my primary concern is that the manuscript has made strong claims that were not supported by the analysis.

For starters, in the economics of education literature, there are essentially two schools of thought: (1) education signals how good a student is, irrespective of the quality of the training; and (2) education provides training. It is important to disentangle the two effects, and there have been a lot of studies trying to do this. Therefore, I am not convinced that “To the best of our knowledge, this is the first analysis of PhD student outcomes in relation to their research environment, their academic abilities and prior research training.” The authors may want to do more literature search on this topic.

A related comment is that, in the paper’s context, the fact that some students are able to do research in a strategic research center and their supervisory teams who got maximum scores may already signal that they may have better academic abilities, thus have better outcomes, as observed in the paper. In other words, it is academic abilities that affect PhD outcome.

It remains unclear to me what the operational definitions of “research environment” and “academic ability” are. Which variables fall into which category? Does scholarship reflect academic ability? If so, the observation that students who are awarded scholarships have more papers/citations and are less likely to withdraw from PhD directly refuted the major claim of the paper.

There are two variables that indicate whether strategic alignment score and supervisor team score achieve maximum. Why focusing on maximum? Wouldn’t a maximum score emphasize the very best?

The current of flow in the results section is quite confusing, alternating between different variables and outcomes. I’d suggest focusing on one outcome at a time and for each outcome describing univariate analysis and regression analysis. A summary table that indicates the associations of each independent variable and each outcome is also helpful.

Finally, why using step-wise regression models, since there are not many independent variables? Why not put all IV in a model and check their significance?

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2020 Aug 5;15(8):e0236327. doi: 10.1371/journal.pone.0236327.r004

Author response to Decision Letter 1

19 May 2020

Please see Response to Reviewers file submitted with manuscript

Attachment

Submitted filename: 2020_05_19_PONE-D-19-27758R2_Response.docx

Click here for additional data file.^{(39.1KB, docx)}