Kyle D Kim1,2, Shaun CH Chua3 and Maurice HT Ling1-4*
1Department of Applied Sciences, Northumbria University, United Kingdom
2School of Life Sciences, Management Development Institute of Singapore, Singapore
3School of Applied Sciences, Temasek Polytechnic, Singapore
4HOHY PTE LTD, Singapore
*Corresponding Author: Maurice HT Ling, Department of Applied Sciences, Northumbria University, United Kingdom.
Received: January 22, 2021; Published: February 12, 2021
Citation: Maurice HT Ling., et al “Science/Education Portraits VII: Statistical Methods Used in 1081 Papers Published in Year 2020 Across 12 Life Science Journals Under BioMed Central”. Acta Scientific Nutritional Health 5.3 (2021): 06-12.
Statistics is an integral part of biology and is required for all undergraduate life science curriculum. However, are biology students trained in statistical skills required in the field? Despite studies listing various commonly statistical methods used in specialised branches of life sciences; such as, immunology and tropical biology; there is a lack of study on the common statistical methods used in life science in general. Here, we examine 1081 articles across 12 life sciences journals under BioMed Central, published in 2020, to elucidate the common statistical methods used in current life science research, as a basis to recommend an updated syllabus to all institutions that educate biologists. 72.7% of the examined articles contains identifiable statistical methods and a total of 2431 instances were identified. Our findings show that the first 3 out of 15 categories of methods; parametric comparison of means (25.38% of instances), correlation/regression (18.88%), and post-hoc test (10.32%); accounts for 54.59% of the instances. In terms of individual methods, the top 8 methods account for 52.04% of the instances – (a) t-test (13.00%), (b) ANOVA (12.26%), (c) unspecified (likely to be Pearson’s correlation) and Pearson’s correlation (9.79%), (d) Benjamini and Hochberg’s False Discovery Rate (FDR) (4.77%), (e) Tukey's HSD (4.36%), (f) Kruskal-Wallis Test (2.96%), (g) Mann-Whitney U Test (2.80%), and (h) Chi Square Test (2.10%). These findings may have an impact on future curriculum design.
Keywords: Statistical Methods; Life Science; Research; Education
The importance of statistics in biology has been recognized more than a century ago [1]. With increasing number of statistical methods, there is a concern regarding core statistical fundamentals required in a biologist’s education [2-6]. Lee., et al. [5] review articles in six pharmacy journals and found (a) ANOVA, (b) Chi-Square Test, (c) Student's t-Test, (d) Pearson's Correlation Coefficient, and (e) Logistic Regression; as the five most commonly used inferential statistical methods. Loaiza Velásquez., et al. [2] review the statistical methods used in two tropical journals during a year and identified twelve most frequently used methods as (a) ANOVA, (b) Chi-Square Test, (c) Student's t-Test, (d) Linear Regression, (e) Pearson's Correlation Coefficient, (f) Mann-Whitney U Test, (g) Kruskal-Wallis Test, (h) Shannon's Diversity Index, (i) Tukey's Test, (j) Cluster Analysis, (k) Spearman's Rank Correlation Test, and (l) Principal Component Analysis. It is crucial that biology students are trained in required statistical skills [2].
However, Lee., et al. [5] focus on pharmacy while Loaiza Velásquez., et al. [2] focus on tropical biology. Similar work by Skinner [6], Meyr [7], Al-Benna., et al. [8], and Hammer and Buffington [9] focus on immunology, surgery, burns research, and veterinary medicine respectively. Hence, the statistical methods used in the common denominator, life science in general, can only be inferred. Here, we identify the statistical methods used in 1081 articles across 12 life sciences journals under BioMed Central published in 2020 as a basis to recommend an updated syllabus to all institutions that educate biologists.
Using similar methods in previous studies [2,5,7,8], twelve open access journals from BioMed Central that are indexed in PubMed [(a) Biological Research, (b) BMC Bioinformatics, (c) BMC Biology, (d) BMC Ecology, (e) BMC Evolutionary Biology, (f) BMC Genomics, (g) BMC Microbiology, (h) BMC Molecular and Cell Biology, (i) Cell and Bioscience, (j) Genome Biology, (k) Journal of Animal Science and Biotechnology, and (l) Stem Cell Research and Therapy] were selected for survey. For each of the twelve journals, all articles published from January 01, 2020; to the end of the month where the number of articles exceed 100 were chosen, or to the end of October 2020. This is to prevent an over-representation of a specific journal in the survey. For each published article, identifiable statistical method(s) used were collated and each method was recorded only once per article [5] with no judgement made on the suitability of the methods [7].
In this study, we examined 1081 peer-reviewed articles published across 12 open access journals from BioMed Central to collate the statistical methods used. The minimum 2-year and 5-year impact factors (as of October 2020) are 2.381 and 2.922 respectively (Table 1), with the highest 2-year impact factor at 10.806; suggesting that the 12 open access journals are highly reputable. Hence, the statistical methods used is likely to be reflective of the needs of the field and important support for curriculum development [2].
Journal Name |
2-year Impact Factor |
5-year Impact Factor |
Source Normalized Impact per Paper (SNIP) |
SCImago Journal Rank (SJR) |
Biological Research |
3.092 |
2.968 |
0.939 |
0.841 |
BMC Bioinformatics |
3.242 |
3.213 |
1.156 |
1.626 |
BMC Biology |
6.765 |
7.296 |
1.604 |
3.698 |
BMC Ecology |
2.381 |
2.922 |
0.913 |
1.030 |
BMC Evolutionary Biology |
3.058 |
3.252 |
1.198 |
1.531 |
BMC Genomics |
3.594 |
4.093 |
1.140 |
1.629 |
BMC Microbiology |
2.989 |
3.381 |
1.049 |
1.154 |
BMC Molecular and Cell Biology |
3.066 |
2.684 |
1.023 |
1.070 |
Cell and Bioscience |
5.026 |
4.443 |
0.985 |
1.410 |
Genome Biology |
10.806 |
19.041 |
2.794 |
9.479 |
Journal of Animal Science and Biotechnology |
4.167 |
4.392 |
1.690 |
1.333 |
Stem Cell Research and Therapy |
5.116 |
5.554 |
1.267 |
1.501 |
Table 1: Impact Factors and Ranking of Journals (as of October 2020).
Of the 1081 articles examined, 786 (72.7%) articles contain identifiable statistical methods (Table 2). From which, 2431 instances of statistical methods were identified. 51.79% (n = 405; Figure 1) of the articles contain one or two statistical methods; with 14 as the maximum number of statistical methods identified from a single article [Tran., et al. [10]]. The methods identified were categorized into 15 application categories (Table 3). The top 3 categories; (a) parametric comparison of means, (b) correlation/regression, and (c) post-hoc test; accounts for 54.59% of the instances. These are followed by (a) non-parametric comparison of means, (b) multiple comparison correction, (c) dimension reduction/multidimensional scaling, and (d) goodness of fit test; which accounts for another 26.94% of the instances. Collectively, these 7 application categories accounts for 81.53% of the instances.
Figure 1: Distribution of Number of Statistical Methods.
Journal Name |
Date Range |
Number of Articles Surveyed |
Number of Articles with Statistical Methods |
Biological Research |
January 01, 2020 to November 30, 2020 |
55 |
48 (87.3%) |
BMC Bioinformatics |
January 01, 2020 to February 28, 2020 |
72 |
31 (43.1%) |
BMC Biology |
January 01, 2020 to June 30, 2020 |
111 |
79 (71.2%) |
BMC Ecology |
January 01, 2020 to December 31, 2020 |
69 |
55 (79.7%) |
BMC Evolutionary Biology |
January 01, 2020 to July 31, 2020 |
96 |
51 (53.13%) |
BMC Genomics |
January 01, 2020 to January 31, 2020 |
109 |
68 (62.39%) |
BMC Microbiology |
January 01, 2020 to April 30, 2020 |
99 |
81 (81.0%) |
BMC Molecular and Cell Biology |
January 01, 2020 to November 30, 2020 |
85 |
71 (83.5%) |
Cell and Bioscience |
January 01, 2020 to September 30, 2020 |
106 |
54 (50.9%) |
Genome Biology |
January 01, 2020 to April 30, 2020 |
100 |
83 (83.0%) |
Journal of Animal Science and Biotechnology |
January 01, 2020 to September 30, 2020 |
97 |
91 (93.8%) |
Stem Cell Research and Therapy |
January 01, 2020 to February 28, 2020 |
82 |
74 (90.2%) |
Total |
1081 |
786 (72.7%) |
Table 2: Number of Articles Surveyed.
In terms of individual statistical methods, the top 8 most frequent methods account for 52.04% of all the instances (Table 4). The methods are (a) t-test (13.00%), (b) ANOVA (12.26%), (c) unspecified and Pearson’s correlation (9.79%), (d) Benjamini and Hochberg’s False Discovery Rate (FDR) (4.77%), (e) Tukey's HSD (4.36%), (f) Kruskal-Wallis Test (2.96%), (g) Mann-Whitney U Test (2.80%), and (h) Chi Square Test (2.10%). These results are consistent with that of Loaiza Velásquez., et al. [2] as 7 of the 8 methods are common, except FDR. These results are also generally consistent with the common statistical methods identified by Al-Benna., et al. [8] and Meyr [7].
Mann-Whitney U Test is often used as the non-parametric equivalent of independent samples t-test in most cases [11] despite differences in several assumptions [12]. Kruskal-Wallis Test is essentially the non-parametric version of one-way ANOVA [13]. While this underpins the importance of parametric and non-parametric comparison of means to biological sciences as these 4 methods account for 31.02% of the instances, it also illustrates the importance of non-parametric methods in biological sciences as biological data is often not normally distributed [14-16]. Besides being not normally distributed, multiple testing is also common in biology [17,18]; hence, it is not surprising that FDR is a commonly seen in publications.
Chi Square test is a common statistical method in biology with applications from clinical sciences [19] to population genetics [20] to omics analyses [21,22]. It is also often the first statistical test taught in first year genetics; hence, has an important position in biology. Similar to Chi Square test, correlation is a staple in many fields of biology [23,24]. Of the 12 major post-hoc tests, Tukey’s HSD is the only one that appears in the top 8 most frequent methods. One of the reasons may be its simplicity and closeness to t-test assuming equal variance in terms of calculation [25]. This also demonstrates the importance of t-test in the education of a biologist as 3 of the 8 most frequent methods (Tukey’s HSD, ANOVA, and Mann-Whitney U Test) requires pre-requisite knowledge of t-test. The presence of Tukey’s HSD also suggests the need for biologists to know what to do after null hypothesis of equal means in more than 2 samples, such as in ANOVA, is rejected. Taken together, these 8 statistical methods should form the basis of all statistical curriculum for biologists.
Application Category |
Frequency |
Prevalence (%) |
Cumulative Prevalence (%) |
Parametric Comparison of Means |
617 |
25.38% |
25.38% |
Correlation/Regression |
459 |
18.88% |
44.26% |
Post-Hoc Test |
251 |
10.32% |
54.59% |
Non-parametric Comparison of Means |
231 |
9.50% |
64.09% |
Multiple Comparison Correction |
167 |
6.87% |
70.96% |
Dimension Reduction/Multidimensional Scaling |
133 |
5.47% |
76.43% |
Goodness of Fit Test |
124 |
5.10% |
81.53% |
Graphing |
101 |
4.15% |
85.68% |
Normality Test |
59 |
2.43% |
88.11% |
Omics Analysis |
49 |
2.02% |
90.13% |
Randomization/Permutation Test |
36 |
1.48% |
91.61% |
Survival Analysis |
34 |
1.40% |
93.01% |
Equality of Variance test |
32 |
1.32% |
94.32% |
Measure of Dispersion |
13 |
0.53% |
94.86% |
Others |
125 |
5.14% |
100.00% |
Table 3: Relative Prevalence of Statistical Methods. Prevalence is defined as the quotient between the number of frequencies in each category and the total number of frequencies (n = 2431). Cumulative prevalence is the summation of prevalence up to the category, with the total cumulative prevalence of 100%.
Application Category/Statistical Method |
Frequency (N) |
Prevalence (%) |
Correlation/Regression (18.88%) |
||
Unspecified Correlation |
147 |
6.05% |
Pearson's correlation |
91 |
3.74% |
Spearman's correlation |
46 |
1.89% |
General Linear Model (GLM) |
33 |
1.36% |
Linear Regression |
28 |
1.15% |
Unspecified Regression |
15 |
0.62% |
Linear Mixed Effect Model |
10 |
0.41% |
Polynomial Contrasts |
9 |
0.37% |
ADONIS |
7 |
0.29% |
Cox's Regression |
6 |
0.25% |
Generalized Linear Mixed Model |
6 |
0.25% |
Logistic Regression |
6 |
0.25% |
Non-Linear Regression |
6 |
0.25% |
Mantel-Haenszel Method |
5 |
0.21% |
Multiple Regression |
5 |
0.21% |
Inter-Rater Agreement |
4 |
0.16% |
Kendall's Correlation |
4 |
0.16% |
Redundancy Analysis (RDA) |
3 |
0.12% |
Fractional Regression |
3 |
0.12% |
Generalized Additive Model |
3 |
0.12% |
Lasso Regression |
3 |
0.12% |
Others (N < 3) |
19 |
0.78% |
Dimension Reduction/Multidimensional Scaling (5.47%) |
||
Principal Component Analysis (PCA) |
40 |
1.65% |
Principal Co-Ordinates Analysis (PCoA) |
17 |
0.70% |
Linear Discriminant Analysis (LDA) |
15 |
0.62% |
Linear Discriminant Analysis Effect Size (LEfSe) |
14 |
0.58% |
Shannon Index |
10 |
0.41% |
Simpson Index |
6 |
0.25% |
Neighbor-Joining Method |
5 |
0.21% |
Non-Metric Multidimensional Scaling (NMDS) |
4 |
0.16% |
Root Mean Squared Distance (RMSD) |
4 |
0.16% |
Uniform Manifold Approximation and Projection |
4 |
0.16% |
UPGMA Cluster Analysis |
4 |
0.16% |
t-Distributed Stochastic Neighbor Embedding |
3 |
0.12% |
Others (N < 3) |
7 |
0.29% |
Equality of Variance Test (1.32%) |
||
Levene’s Test |
13 |
0.53% |
F-Test |
9 |
0.37% |
Bartlett’s Test |
5 |
0.21% |
Others (N < 3) |
5 |
0.21% |
Goodness of Fit Test (5.10%) |
||
Chi Square Test |
51 |
2.10% |
Fisher’s Exact Test |
48 |
1.97% |
Likelihood Ratio Test |
6 |
0.25% |
Mantel Test |
5 |
0.21% |
Wald Test |
5 |
0.21% |
Hardy-Weinberg Equilibrium Test |
4 |
0.16% |
I2 Test |
3 |
0.12% |
Kullback–Leibler Distance |
2 |
0.08% |
Graphing (4.15%) |
||
Box Plot |
29 |
1.19% |
Heatmap |
22 |
0.90% |
Error Bar |
12 |
0.49% |
Bar Plot |
9 |
0.37% |
Scatter Plot |
7 |
0.29% |
Dot Plot |
4 |
0.16% |
Receiver Operating Characteristic Curve (ROC curve) |
4 |
0.16% |
Volcano Plot |
4 |
0.16% |
Others (N < 3) |
10 |
0.41% |
Measure of Dispersion (0.53%) |
||
Coefficient of Variation |
6 |
0.25% |
Standard Error |
4 |
0.16% |
Others (N < 3) |
3 |
0.12% |
Multiple Comparison Correction (6.87%) |
||
Benjamini and Hochberg’s False Discovery Rate (FDR) |
116 |
4.77% |
Bonferroni Correction |
48 |
1.97% |
Others (N < 3) |
3 |
0.12% |
Non-Parametric Comparison of Means (9.50%) |
||
Kruskal-Wallis Test |
72 |
2.96% |
Mann-Whitney U Test |
68 |
2.80% |
Wilcoxon Rank-Sum Test |
40 |
1.65% |
Analysis of Similarities (ANOSIM) |
22 |
0.90% |
Unspecified Wilcoxon Test |
12 |
0.49% |
Wilcoxon Signed-Rank Test |
11 |
0.45% |
Friedman Test |
3 |
0.12% |
Others (N < 3) |
3 |
0.12% |
Normality Test (2.43%) |
||
Shapiro-Wilk Test |
27 |
1.11% |
Kolmogorov-Smirnov Test |
25 |
1.03% |
Anderson-Darling Test |
3 |
0.12% |
Others (N < 3) |
4 |
0.16% |
Omics Analysis (2.02%) |
||
Gene Ontology Enrichment Analysis |
24 |
0.99% |
Analysis of Molecular Variance (AMOVA) |
6 |
0.25% |
DESeq2 |
3 |
0.12% |
Feed Conversion Ratio |
3 |
0.12% |
Others (N < 3) |
13 |
0.53% |
Parametric Comparison of Means (25.38%) |
||
t-Test |
316 |
13.00% |
Analysis of Variance (ANOVA) |
298 |
12.26% |
Z-Test |
3 |
0.12% |
Post-Hoc Test (10.32%) |
||
Tukey's HSD |
106 |
4.36% |
Unspecified Post-Hoc Test |
44 |
1.81% |
Duncan’s multiple range Test |
23 |
0.95% |
Dunnett’s Test |
20 |
0.82% |
Dunn’s Test |
17 |
0.70% |
Fisher’s Least Significant Difference (LSD) |
13 |
0.53% |
Holm-Sidak Test |
6 |
0.25% |
Student–Newman–Keuls (SNK) Test |
5 |
0.21% |
Scheffe’s Test |
4 |
0.16% |
Bonferroni Test |
3 |
0.12% |
Conover-Iman Test |
3 |
0.12% |
Post-Hoc t-Test |
3 |
0.12% |
Others (N < 3) |
4 |
0.16% |
Randomization/Permutation Test (1.48%) |
||
Permutational Multivariate Analysis of Variance (PERMANOVA) |
21 |
0.86% |
Permutation Test |
7 |
0.29% |
Others (N < 3) |
8 |
0.33% |
Survival Analysis (1.40%) |
||
Log-Rank Test |
24 |
0.99% |
Kaplan-Meier Analysis |
10 |
0.41% |
Others (N < 3) |
125 |
5.14% |
Table 4: Breakdown of Statistical Methods.
The top 8 most frequent methods identified from 1081 articles are (a) t-test, (b) ANOVA, (c) unspecified and Pearson’s correlation, (d) Benjamini and Hochberg’s False Discovery Rate (FDR), (e) Tukey's HSD, (f) Kruskal-Wallis Test, (g) Mann-Whitney U Test, and (h) Chi Square Test.
Data file for this study can be downloaded at https://bit.ly/SEP7_Statistics.
The authors declare no conflict of interest.
The views expressed by the authors are that of the authors rather than the views of their affiliated institutions.
Copyright: © 2021 Maurice HT Ling., et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.