From Data to Discovery: The Power of Statistics in Research

The Crimean War was brutal on the soldiers and caused high mortality rates. It was believed that battle wounds were the leading cause of death. But the ‘lady with the lamp’, who was a nurse and also a gifted statistician, proved that unhygienic living conditions, not battle wounds, led to the increased mortality rates (Kukula, 2018). You must have guessed the person by now! Yes, it was Florence Nightingale who made this significant contribution. She used statistics and elegant data visualization techniques to convince the higher authorities of the need for improved sanitary conditions for the soldiers and lowered the death rates in the war (Kukula, 2018). Such anecdotes time and again reinstate that sharp research insights combined with strong statistical backing can transform data into discovery. In short, sound statistical knowledge can act as a catalyst to convert raw data into an insightful piece of information. Although a lot of statistical aspects get discussed with respect to health data in biology, statistics plays a crucial role in all domains of biology, specifically experimental and computational plant sciences as well.

Envision a greenhouse trial you undertook to test the efficacy of a biofertilizer on crop yields, and you get excellent results. But how excellent are the results? How many replicates did you test on? What was the rationale for the experimental design? How reproducible and significant were the numbers? These kinds of research scenarios need you to think like a researcher cum statistician. Now coming to the computational part, assume you are running a co-expression analysis of candidate genes involved in a particular biosynthetic pathway. What is the p-value threshold you will fix to call your coexpression hits meaningful? Do you have to account for multiple-testing impacts? Well, it must sound overwhelming by now! Honestly, how many of us have stared at the p-value and experienced triumph when it falls below the holy grail of 0.05? But wait, the very next second you get hit by a deep, creeping doubt that asks, ” Is what I am doing statistically correct, and does this all make sense?” These were the precise questions that prompted us to come up with the following blog that is aimed at being an engaging read, as well as an informational resource for the plant science community at large.

This short statistical jaunt begins with a tour of the routine role of statistics in research and writing. It is followed by a critical but most ignored topic – ‘inappropriate statistical application ’-that helps you see the thin line between using and misusing the statistical tools at your disposal that can, in turn, transform them into your friends and foes, respectively! Finally, this statistical journey you take with us through these lines ends with some helpful recommendations and resource tips from us to ramp up your stats skills! Get ready to immerse your feet in the frolicking waves of the statistical waters.

 

Statistics – The linchpin in research and writing

Research often starts with an observation that needs to be tested, quantified, compared, and clearly reported to become a discovery. Here’s where statistics come into the picture. Statistics gives those observations a structure and turns them into scientific evidence that others can comprehend, rely on, and replicate.

Statistics help scientists to answer how significantly experimental groups differ and whether the evidence is strong enough by applying important statistical tools and measures. This makes statistics one of the key pillars of scientific advancement.

The relationship between biology and statistics dates back to a time that is much older than our genome browsers, multi-omics, and high-throughput techniques. Mendel’s pea plants revealed the secrets of heredity through a journey of careful crossings, keen observation of inherited traits,  and impeccable data collection across generations. Later, Galton’s studies on sweet peas gave birth to the concepts of regression and mean. Pearson turned these observations into statistical tools of correlation and regression, which helped scientists to study heredity and variation. Ronald Fisher made them useful for agricultural research by bringing statistical ideas such as ANOVA, variance, randomization, and factorial designs to the field. Together, their works founded the statistical framework for modern research that shows how closely statistics has been embedded in biology since the beginning (Bodmer et al., 2021, Stanton, 2001).

Now imagine a world without statistics! Mendel’s peas might have stayed garden notes. Genomics could not have generated a mountain of useful data. Clinical trials would have found it difficult to know whether a treatment truly works or not. Plant breeders would have had to depend more on visible outcomes instead of precise genetic data. Ecologists might have faced difficulties evaluating conservation risk and implementing conservation and environmental policy. These frightening possibilities indicate how statistics gives science its strength.

A more recent example comes from a current study on rice. Here, statistics supported the central hypothesis that the fungal long non-coding RNA (lncRNA) is a cross-kingdom effector. Statistical tools such as ANOVA, t-test, P-value, and Tukey’s test allow researchers to compare RNA expression, fungal growth, and disease symptoms. This helps make this fascinating study more quantitative; otherwise, it would remain primarily descriptive (He et al., 2026).

Similarly, recent advancements in nitric oxide (NO) research, including whole-plant live imaging of NO and hydrogen peroxide, indicate the crucial role of statistics in research as scientists moved from simply detecting NO to determining where, when, and how it influences plant resilience (AL-Hakeem & Khan, 2025).

Metabolic pathway analysis is another great illustration of how statistics shapes the research findings. Now scientists can easily identify how metabolic networks respond to development, stress, and genetic perturbation by using approaches such as flux balance analysis (FBA), metabolic flux analysis (MFA), isotope labeling, and dynamic modeling, rather than viewing pathways as static schematics. Therefore, it renders plant metabolism observable, interpretable, and predictable, allowing researchers to better understand energy and carbon flow in photosynthesis, lignin and phenylalanine biosynthesis, and other metabolic and physiological adjustments to environmental change, in addition to crop improvement and metabolic engineering strategies (Rao & Liu, 2025).

When it comes to gene expression analysis, statistics becomes a vital tool for observation because the thousands of connections between transcription factors and genes of interest cannot be noticed solely by the human eye. For example, in a recent study, statistical and topological analysis revealed the scale-free nature of the gene regulatory network when researchers compared these network patterns across Arabidopsis thaliana, Saccharomyces cerevisiae, Drosophila melanogaster, and Caenorhabditis elegans. It enabled them to uncover that various organisms have a common pattern. Most transcription factors regulated only a few genes, while few of them operated as hubs, yet each of them carried a quantitatively distinct network architecture (Ouma et al., 2018).

Statistics has become a primary bridge between large datasets and practical conclusions in genomics, genetics, plant breeding, and agriculture. Statistical techniques such as Analysis of Variance (ANOVA), mixed-model and Restricted (or Residual) Maximum Likelihood/ Best Linear Unbiased Prediction (REML/ BLUP) analysis, variance-component estimation, Genotypic Coefficient of Variation (GCV) and Phenotypic Coefficient of Variation (PCV), and Genome-Wide Association Study (GWAS) are used to identify useful traits, estimate genetic variation, genotype advancement, and crop improvement. It helps plant breeders decide which lines to progress, which parents to cross, which traits to emphasize, and how to manage uncertainty across environments (Montesinos-López et al., 2026).

Ecology and environmental research show another side of the same story. Natural systems are full of patterns such as where species live, how populations change, how communities assemble, and how ecosystems respond to disturbance. Methods, such as regression analysis, variance analysis, and clustering, are widely used to analyze these environmental patterns. From estimating species abundance and biodiversity to modeling climate impacts, pollution, habitat loss, and catastrophic events, statistical ecology has advanced environmental research. Recent trends show that as ecological data now come from satellites, sensors, camera traps, acoustic recorders, citizen science, and long-term monitoring, statistics is becoming even more essential for turning complex environmental data into reliable knowledge for conservation and policy (Mather et al., 2026, Gilbert et al., 2024).

Along with making science interpretable, statistics also helps with scientific writing. A good research paper is not just about simply presenting the results but is also about how these results were obtained. Every reader wants to know how many samples were chosen, whether experiments were duplicated, what kind of statistical tests were performed, how variable the data points were, and how significantly the results supported the hypothesis, including precise or threshold p-values. This information helps others determine whether a conclusion is reliable, reproducible, and scientifically meaningful or not (Ali & Bhaskar, 2016).

Statistics helps researchers to plan experiments, collect data, analyze patterns, measure uncertainty, and support an unbiased conclusion. This ultimately transforms the overall scientific narrative and makes scientific writing stronger and more transparent rather than just convincing.

As science grows richer in data, the role of statistics will grow stronger. Researchers are already working with a huge variety of datasets, including genome sequences, RNA-seq datasets, live imaging, metabolomics, field phenotyping, environmental sensors, drones, satellites, and long-term ecological monitoring. The task is no longer only gathering data but figuring out how to analyze it appropriately, and appropriate statistical application will help make this possible. That said, it becomes imperative to know the difference between appropriate and inappropriate statistical applications. The ensuing section helps you understand and differentiate these better. Read on …

 

Inappropriate statistical application: the treacherous tightrope

Ideal vs. reality: how statistics should work versus how it is actually used. Statistical tools provide the framework to turn raw data into scientific conclusions, but statistical results are only as reliable as the logic behind them. We have all been there, trying to produce the same output as in the first replicate (Baker, 2016). When these expectations are put upfront, we might lose sight of the right path in our obsession with the goal, and inappropriate statistical tests can quickly blind us with artifacts. In this section, we will discuss the most common examples of statistical misuse in plant research: errors in assumption, selection, interpretation, and experimental design.

Table 1: The four pillars of statistical failure.

	Before the test (Design / Input)	During / after the test (Analysis / Output)
Technical / methodological	Experimental design errors:   pseudoreplication, lack of randomization	Assumption errors:   normality, variance
Cognitive / psychological	Selection errors:   p-hacking, cherry-picking	Interpretation errors:   correlation ≠ causation

First and foremost, there are assumption errors, such as assuming that all data have a normal distribution and the same variance. Distribution and variance tests are crucial pretests when running statistical comparisons to ensure that we do not generate Type I or Type II errors. What are these errors? Type I is the false positive error, seeing an effect that is not there, while Type II is the false negative error, missing an effect that is actually there. These errors make our lives a bit harder than simply checking the p-value after a statistical test. The most common statistical tests in plant science publications are the t-test and the ANOVA (multiple t-test), which compare the means of data groups and hence rely on the assumptions of normal distribution and equal variance (homoscedasticity). We can clearly see how different distributions (Figure 1A) and different variances (Figure 1B) can still have the same mean value, and comparing their means with a t-test can lead to Type I/II errors and false scientific conclusions.

Figure 1: (A) The same mean with different distributions (modified plot of Thomas J. Pfaff). (B) And the same mean and same distribution with different variances.

The selection errors, or cherry-picking and post-hoc grouping. These are also sometimes referred to as p-hacking because, by subjectively discarding data points or regrouping data, we can manipulate the p-value in our favor. Cherry-picking can happen in many ways, from a single data point to whole sets of data points. For example, if we measure gene expression at different time points (6, 12, and 24 hours), and see a significant difference only at the 12-hour time point, we might only publish that result. In the hope of more voices advocating for publishing negative results, I believe discarding whole groups becomes less common, even if they only end up in the supplementary materials. However, discarding single-data-point outliers, not subjectively but based on objective testing, can be acceptable. Post-hoc grouping means modifying predefined groups after we see the data distribution. For example, we might pre-set a chlorophyll a/b ratio to identify senescing leaves, but after observing gene expression, we adjust this preset grouping criterion so that the expression values separate more clearly. Selection errors are inherently subjective and occur after data collection, so it is good to start an experiment with a range of criteria to try to keep it objective.

The interpretion error is the difference between the statistical significance and the biological relevance. Statistical significance, or the p-value, is highly dependent on sample size. Hence, we have different significance thresholds in the case of a Genome-Wide Association Study (GWAS), where multiple testing is commonly addressed using a Bonferroni correction, and the commonly used significance cutoff is approximately 5 × 10^-8. Why is it important to change this cutoff when dealing with big data? Staying with the GWAS example, if we used the cutoff p < 0.05 for significance, we would generate thousands of false-positive (Type I error) results. Another approach would be to increase the sample size until we get a significant result, such as a root growth difference of 0.5 mm under treatment, which, even though it would be statistically significant with a thousand root samples, has no impact on the plant’s life and is biologically not relevant.

The experimental design errors might be the trickiest and most common ones. Among them are pseudoreplication, lack of randomization, insufficient sample size, and lack of positive/negative control. Pseudoreplication would be to perform the measurement on 10 leaves of a single plant, rather than on specific leaves of 10 different plants. Lack of randomization, on the other hand, would be, for example, having treatment A growing on the sunny side of the room and treatment B on the shady side, and omitting the light differences while attributing the observed differences to the treatments (Baker, 2016).

After we have overcome all these obstacles, there is one more thing to be aware of beyond statistics, which is to avoid representing the data with misleading graphs (Tufte, 2001). Let me show you three tricks that you should NOT use: the magnitude, the distribution, and the categorical tricks. The magnitude trick is manipulating the scale, whether you want to make small differences look big or big differences look small. This can include using a truncated y-axis (starting the axis at 42 instead of 0), using dual-axis (two different y-axes to force two unrelated lines to overlap), or stretching/compressing the axes (a very long x-axis and a very short y-axis will flatten the plotted curve). The second one is the distribution trick, which hides the truth. The biggest crime among the distribution tricks is using a bar plot to represent only the mean values, hiding the distribution and outliers, rather than a box plot or a jitter plot that shows the individual data points. The categorical trick is, for example, using the wrong graph type (using a line graph for categorical data) or using improper intervals (placing time points with different gaps on the x-axis at equal intervals). All these “tricks” can mislead the reader and should not be used under any circumstances.

 

Leveling up your statistical competence

Equipped with a know-how on the crucial role of statistics and the practices needed to ensure correct statistical approaches, you have now reached the last level of this journey that will help you level up! Let’s face the first and foremost hurdle that prevents us from becoming a stats pro. ‘Fear of math’ and ‘statistical anxiety’ are common roadblocks to becoming statistically competent (Harvard Medical School, 2025). In an increasingly data-driven world, with big data taking the centre stage and powering all the large language model (LLM) dark horses, the need to become statistically skilled has never been more relevant than now. Moreover, with numerous high-quality resources that help you grasp the core statistical concepts by demonstrating the elegance behind the math involved, it is time to say a big goodbye to the tiny voice that causes you to scroll or skip statistics stuff. Before jumping into how to step up, let’s further get clear on why to step up. Equipping yourself with statistical knowledge offers you more independence and boosts your confidence as a researcher. It will also help you ask better questions, design better studies, and collaborate with statisticians instead of merely compartmentalizing your role from that of the statistician (Harvard Medical School, 2025). After all, you understand your study the best, so why not get a clearer picture of it with a statistical viewpoint?!

The first step towards becoming statistically competent is to become statistically literate. Statistical literacy is defined as an individual’s ability to comprehend the basic terms and statistical concepts. A starting point can be the breezy statistics chapter in the biostars handbook (Albert, 2020) that primes you with the basic terminology and widely used statistical tests. In an increasingly computational world, there is no better way to start learning statistics other than doing it practically. Introductory Biostatistics with R (Twomey, 2024) is a wonderful resource to learn statistical approaches via R. If you are familiar with some R and are a bibliophile, Practical Statistics for Data Scientists (Bruce & Bruce, 2017) can be one of your top choices! Moving on, An Introduction to Statistical Learning is also a great resource that embraces statistical learning with R (James et al., 2021) and Python (James et al., 2023). Once you have toned up your stats stamina, this wonderful book chapter, A biologist’s guide to statistical thinking and analysis (Fay & Gerow, 2013) , can be a nice and detailed theoretical exposition of major statistical tests and concepts. Furthermore, Nature has released a nice collection called ‘Statistics for Biologists’ (Nature, n.d.) that talks about the statistical issues and the pitfalls to be avoided by biologists. However, if you are a visually inclined person, StatQuest (Starmer, n.d.) by Josh Starmer is a favorite community resource. Accompanied by engaging explanations of statistical concepts ranging from hypothesis testing to machine learning aspects, and funny music, this video resource can be your go-to stats companion.

Reference

Albert, I. (2020). The Biostar handbook: A practical guide to bioinformatics (2nd ed.). Biostars. https://www.biostarhandbook.com/

Al-Hakeem, H. F. H., & Khan, M. (2025). The role of statistics in advancing nitric oxide research in plant biology: from data analysis to mechanistic insights. Frontiers in Plant Science, 16, 1597030. https://doi.org/10.3389/fpls.2025.1597030

Ali, Z., & Bhaskar, S. B. (2016). Basic statistical tools in research and data analysis. Indian journal of anaesthesia, 60(9), 662. https://doi.org/10.4103/0019-5049.190623

Baker, M. (2016). 1,500 scientists lift the lid on reproducibility. Nature, 533(7604), 452–454. https://doi.org/10.1038/533452a

Bodmer, W., Bailey, R. A., Charlesworth, B., Eyre-Walker, A., Farewell, V., Mead, A., & Senn, S. (2021). The outstanding scientist, RA Fisher: his views on eugenics and race. Heredity, 126(4), 565-576. https://doi.org/10.1038/s41437-020-00394-6

Bruce, P., & Bruce, A. (2017). Practical statistics for data scientists: 50 essential concepts. O’Reilly Media. https://www.oreilly.com/library/view/practical-statistics-for/9781491952955/

Fay, D. S., & Gerow, K. (2018). A biologist’s guide to statistical thinking and analysis. In WormBook: The Online Review of C. elegans Biology [Internet]. WormBook. https://www.ncbi.nlm.nih.gov/books/NBK153593/

Gilbert, N. A., Amaral, B. R., Smith, O. M., Williams, P. J., Ceyzyk, S., Ayebare, S., … & Zipkin, E. F. (2024). A century of statistical ecology. Ecology, 105(6), e4283. https://doi.org/10.1002/ecy.4283

Harvard Medical School. (2025, November 13). Building confidence in statistical analysis: Why learning to work with data matters for new researchers. Harvard Medical School Insights. https://learn.hms.harvard.edu/insights/all-insights/building-confidence-statistical-analysis-why-learning-work-data-matters-new-researchers

He, M., Su, J., Zhou, X., Qi, T., Wang, J., Zhang, T., … & Chen, X. (2026). A pathogen lncRNA secreted into rice sequesters a host miRNA for virulence. Nature, 1-10. https://doi.org/10.1038/s41586-026-10572-x

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2021). An introduction to statistical learning: With applications in R (2nd ed.). Springer. https://www.statlearning.com/

James, G., Witten, D., Hastie, T., Tibshirani, R., & Taylor, J. (2023). An introduction to statistical learning: With applications in Python. Springer. https://www.statlearning.com/

Kukula, L. (2018, November 2). The 8 most interesting statisticians of all time. Geckoboard Blog. https://www.geckoboard.com/blog/stars-of-stats/

Mather, M., Kuck, S., & Oliver, D. (2026). Statistical facilitation in environmental science: Integrating results from complementary statistical analyses can improve ecological interpretations. Environments, 13(2), 82. https://doi.org/10.3390/environments13020082

Montesinos-Lopez, O. A., Crossa, J., Karaikal, S. K., Ornella, L., Martínez-Regalado, J. A., Murillo-Ávalos, C. L., … & Ortiz, R. (2026). Statistics and data science unlock the predictive power of quantitative genetics. Frontiers in Plant Science, 17, 1870471. https://doi.org/10.3389/fpls.2026.1870471

Nature. (n.d.). Statistics in biology [Focus collection]. Springer Nature. Retrieved March 1, 2026, from https://www.nature.com/collections/qghhqm/content/statistics-in-biology

Ouma, W. Z., Pogacar, K., & Grotewold, E. (2018). Topological and statistical analyses of gene regulatory networks reveal unifying yet quantitatively different emergent properties. PLoS computational biology, 14(4), e1006098. https://doi.org/10.1371/journal.pcbi.1006098

Rao, X., & Liu, W. (2025). A guide to metabolic network modeling for plant biology. Plants, 14(3), 484. https://doi.org/10.3390/plants14030484

Stanton, J. M. (2001). Galton, Pearson, and the peas: A brief history of linear regression for statistics instructors. Journal of Statistics Education, 9(3). https://doi.org/10.1080/10691898.2001.11910537

Starmer, J. (n.d.). StatQuest with Josh Starmer [YouTube channel]. YouTube. Retrieved September 1, 2026, from https://www.youtube.com/@statquest

Tufte, E. R. (2001). The visual display of quantitative information (2nd ed.). Graphics Press. https://www.edwardtufte.com/book/the-visual-display-of-quantitative-information/

Twomey, L. (2024, June 3). Top resources to learn biostatistics. BioStatsSquid. https://biostatsquid.com/top-resources-to-learn-biostatistics/

 

 

______________________________________________

About the Authors

Sophie Zoe Farkas

Sophie is a final-year PhD student at the University of Freiburg and a 2026 Plantae Fellow. Her research focuses on root system architecture, specifically the regulation of lateral root angle by genetic factors and environmental stimuli. When she’s not in the lab, you can find her on the football field or playing tunes on her alto saxophone. Find her on Bluesky: @sophiezoe.bsky.social | X: @fsophiezoe.

Kavita Joshi

Kavita is a 2026 Plantae Fellow with a background in plant biology who is passionate about plant science research, science communication, and education.She is interested in creating content for ASPB that makes plant science accessible and engaging for both the general public and the broader plant science community through simple and approachable communication. In her free time, she enjoys crafting, gardening, and exploring nature as an eco-enthusiast.

Shakunthala Natarajan

Shakunthala is a second-year PhD student at the Institute for Cellular and Molecular Botany, University of Bonn and a 2026 Plantae Fellow. Her research revolves around investigating plant gene duplications using comparative genomics and transcriptomics. She develops computational tools to explore the world of plants. Outside the lab she wears the hats of a science communicator and a musician. Find her on X: @Shak_Nat | Bluesky: @shakunthalan.bsky.social.