data presentation in biostatistics

Loading metrics

Open Access

Perspective

The Perspective section provides experts with a forum to comment on topical or controversial issues of broad interest.

See all article types »

Reinventing Biostatistics Education for Basic Scientists

* E-mail: [email protected]

Affiliation Division of Nephrology and Hypertension, Mayo Clinic, Rochester, Minnesota, United States of America

Affiliation Department of Medical Statistics and Informatics, Medical Faculty, University of Belgrade, Belgrade, Serbia

Affiliation Division of Biomedical Statistic and Informatics, Mayo Clinic, Rochester, Minnesota, United States of America

Affiliation Center for Data Analytics and Biomedical Informatics, Temple University, Philadelphia, Pennsylvania, United States of America

Affiliation Department of Medical Physiology, Texas A&M Health Science Center, Texas A&M University, College Station, Texas, United States of America

Affiliations Division of Nephrology and Hypertension, Mayo Clinic, Rochester, Minnesota, United States of America, Department of Medical Statistics and Informatics, Medical Faculty, University of Belgrade, Belgrade, Serbia

• Tracey L. Weissgerber, 
• Vesna D. Garovic, 
• Jelena S. Milin-Lazovic, 
• Stacey J. Winham, 
• Zoran Obradovic, 
• Jerome P. Trzeciakowski, 
• Natasa M. Milic

Published: April 8, 2016

• https://doi.org/10.1371/journal.pbio.1002430
• Reader Comments

7 Jun 2016: The PLOS Biology Staff (2016) Correction: Reinventing Biostatistics Education for Basic Scientists. PLOS Biology 14(6): e1002492. https://doi.org/10.1371/journal.pbio.1002492 View correction

Numerous studies demonstrating that statistical errors are common in basic science publications have led to calls to improve statistical training for basic scientists. In this article, we sought to evaluate statistical requirements for PhD training and to identify opportunities for improving biostatistics education in the basic sciences. We provide recommendations for improving statistics training for basic biomedical scientists, including: 1. Encouraging departments to require statistics training, 2. Tailoring coursework to the students’ fields of research, and 3. Developing tools and strategies to promote education and dissemination of statistical knowledge. We also provide a list of statistical considerations that should be addressed in statistics education for basic scientists.

Citation: Weissgerber TL, Garovic VD, Milin-Lazovic JS, Winham SJ, Obradovic Z, Trzeciakowski JP, et al. (2016) Reinventing Biostatistics Education for Basic Scientists. PLoS Biol 14(4): e1002430. https://doi.org/10.1371/journal.pbio.1002430

Copyright: © 2016 Weissgerber 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.

Funding: The project described was supported by Award Number P-50 AG44170 (VDG) from the National Institute on Aging ( https://www.nia.nih.gov/ ). TLW was supported by the Office of Women’s Health Research ( http://orwh.od.nih.gov/ , Building Interdisciplinary Careers in Women’s Health award K12HD065987). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Misuse of statistical methods is common in basic biomedical science research, even among papers published in high impact journals [ 1 – 3 ]. This includes using incorrect or suboptimal tests [ 1 , 2 ], summarizing data that were analyzed by nonparametric techniques as mean and standard deviation or standard error [ 4 ], reporting p -values that are inconsistent with the test statistic [ 5 , 6 ], p-hacking [ 7 ], and analyzing nonindependent data as though they are independent [ 3 ]. Additional problems arise from inadequate reporting of statistical methods. This may include failing to provide a power calculation [ 1 ], not reporting which statistical test was used, or not providing adequate detail about the test (i.e., paired versus unpaired t test) [ 1 ], not addressing whether the assumptions of the statistical tests were examined [ 1 , 4 ], or not specifying how replicates were treated in the analysis [ 3 ]. Finally, other researchers have focused on the need to reconsider current statistical practices. The reliance on null hypothesis testing and p -values has been heavily questioned, and researchers have proposed a variety of alternate approaches [ 8 , 9 ]. These problems stem from a limited understanding of statistics, suggesting that scientists need better training in this important skill set [ 10 ].

This article focuses on rethinking our approach to biostatistics education. Data from our previous systematic review of physiology studies [ 4 ] demonstrate that understanding statistical concepts and skills is essential for those who are reading or publishing scientific papers. We sought to determine whether this is reflected in the curriculae for PhD students by examining statistics education requirements among PhD programs in top NIH-funded physiology departments ( n = 80). We then outline several approaches that may help to reinvent statistical education for basic biomedical scientists. Some of the problems that we discuss in this article are common to many fields, whereas other problems may require field-specific solutions. This article includes general recommendations based on the authors’ experiences in basic biomedical research. We hope that these comments will advance the ongoing discussion about improving the quality of data presentation and statistical analysis in basic science.

Recommendation 1: Encourage Departments to Require Statistics Training

Data presentation and statistical analysis are increasingly important parts of scientific publications. This trend is likely to accelerate as more journals implement checklists to address common statistical problems and enlist statistical consultants to review papers [ 11 , 12 ]. The data presented in Box 1 show that the ability to understand statistical concepts and apply statistical skills is essential for research; however, biostatistics training is not always required to complete a PhD. We recommend that biostatistics be required for all doctoral students in disciplines where statistics are routinely used. Early career investigators who did not take a biostatistics course during their PhD training should obtain statistical training during their postdoctoral or early faculty years. This parallels recommendations from a recent Nature Medicine editorial [ 13 ], which emphasized that proper training in statistics and research methods is essential for reproducible research. The authors recommended that training in statistics and research methods be required for first year graduate students at PhD-granting institutions.

Box 1. Statistical Skills Are Essential but Not Always Required for a PhD

• According to our recent systematic review [ 4 ], 97.2% of original research papers published in the top 25% of physiology journals ( n = 683/703) included some form of statistical analysis ( Fig 1A ). Journals were selected based on 2012 impact factors.
• While this systematic review focused on physiology, frequent use of statistics likely extends to related disciplines. Physiology journals publish articles from researchers in many fields, including biochemistry, microbiology, cell biology, neuroscience, and many others.
• Among top NIH-funded physiology departments ( n = 80), 67.5% required a statistics course for some (3.75%) or all (63.75%) PhD programs in which the department participated ( Fig 1B ). Biostatistics was recommended as an elective in 10% of departments and listed as an elective in 10% of departments. Biostatistics was not required or offered as an elective course for students in 12.5% of departments. This included one department that required a mathematical modeling course with a small biostatistics section. Departments included in this analysis were on the Blue Ridge Institute for Medical Research list of top NIH-funded physiology departments for 2014 (methodology in S1 Text ).
• Students were rarely required to have learned statistics prior to starting a PhD program. One department (1.3%) listed a statistics course as a prerequisite for admission to the PhD program. Five departments (6.2%) recommended a statistics course prior to admission. No program required students to complete a Masters degree before entering the PhD program.
• Some departments offer PhD programs that are focused on physiology ( n = 33), whereas others participate in departmental or interdepartmental PhD programs that include the related disciplines of biophysics, neuroscience, pharmacology or biology ( n = 47). Statistics requirements were not different in a sensitivity analysis in which we excluded programs that combined physiology with related disciplines (Physiology only: n = 33).

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

A: A recent systematic review [ 4 ] demonstrated that 97.2% of papers published in the top 25% of physiology journals included statistical analyses. B: Statistics courses are not always required for PhD students in top NIH funded physiology departments. Detailed methodology for panels A and B are described in S1 Text .

https://doi.org/10.1371/journal.pbio.1002430.g001

Recommendation 2: Tailoring Coursework to the Student’s Field of Research

While many departments currently offer or require biostatistics training, courses may not necessarily be designed to meet the needs of basic science students. This section focuses on strategies for designing courses that will give students the conceptual understanding and skills needed to analyze data, critique the literature, and improve the quality of statistical reporting and analysis in their respective fields.

Lost in Translation: Bridging the Communication Gap between Basic Scientists and Statisticians

We propose that the faculty of basic science departments improve the quality of statistics education by working with statistics instructors to ensure that courses prepare students to read and publish papers in their respective fields. Among departments that include statistics as a required or elective course, many “out-source” their statistics teaching to other departments that offer introductory statistics courses. At some institutions in our sample (see Box 1 , methodology in S1 Text ), outside departments offered courses that appeared to be designed for basic scientists. At other institutions, introductory statistics courses designed for epidemiologists or public health students were incorporated into the basic science curriculum. The latter courses are unlikely to provide appropriate statistical preparation for basic scientists given the obvious differences in study designs and sample sizes between these disciplines.

Statisticians have recently questioned whether general introductory courses based on a one-size-fits-all approach to statistics education meet the needs of students [ 14 ]. Statistics is an increasingly specialized field, in which the techniques that are used vary widely depending on the type of outcome variable (continuous, categorical, time to event, etc.), the sample size, and the study design. Survival analyses, tests of predictive accuracy, odds ratios, and relative risks are common in clinical science but are rarely used by basic biomedical scientists, who typically work with continuous data, counts, or proportions. While introductory statistics courses generally focus on techniques for analyzing these types of data, the techniques and strategies that are taught often assume a much larger sample size than we observed in our systematic review of physiology studies.

Statisticians who teach general statistics courses often work with very large datasets and may have limited knowledge of the sample sizes or study designs that are common in basic biomedical research. These instructors might design their courses quite differently if they understood the characteristics of the datasets with which their students were working. We propose that the faculty in basic biomedical science departments collaborate with statistics instructors to ensure that courses teach students the skills that they will need to understand, present, and analyze data in their respective fields. Courses should be designed around the sample sizes, study designs, and types of data that are frequently used in the students’ areas of study. Coursework should also address errors in statistical analysis and data presentation that are common among published papers for that field. The following sections provide information about particular areas where the needs of basic scientists may differ from the concepts and skills that are generally taught in introductory statistics courses.

Statistics for Small Samples

Small sample sizes are common in many basic science disciplines. Basic scientists typically want to compare values obtained from participants, specimens, or samples in different groups (i.e., wild-type mice versu knock-out mice; participants randomized to an exercise intervention versus a control group, men versus women), or at different time-points or conditions (i.e., preintervention versus postintervention, etc.). In a systematic review of papers published in top physiology journals [ 4 ], the median for the smallest sample size of any group shown in a figure was 4 (25th percentile: 3, 75th percentile: 6). The median sample size for the largest group shown in a figure was 9 (25th percentile: 6, 75th percentile: 15). Low statistical power and small sample sizes have been highlighted as one factor that may contribute to irreproducibility in neuroscience [ 15 ]. A recent study suggested that n = 8/group was a common sample size for preclinical research [ 16 ]. Anecdotal reports suggest that many investigators consider n = 6/group to be sufficient for animal studies, although this assumption is not based on statistical principles [ 17 ].

There are several possible reasons why researchers use small sample sizes. Experiments with large sample sizes are not feasible in many cases because the detailed mechanistic studies performed by basic scientists are very time, labor, and cost-intensive. The International Guiding Principles for Biomedical Research Involving Animals require that “the minimum number of animals should be used to achieve the scientific or educational goals” [ 18 ]. Finally, the variability in an experiment performed on a particular cell line or strain, type, or breed of animal may be lower than would be expected in a human study. Cell lines or animal studies sometimes lack diversity in genetic, environmental, and behavioral factors that can affect study measurements and increase variability in human studies.

Training for researchers working in fields where small sample sizes are common should focus on experimental design and statistical analysis considerations for small datasets. Students should learn that it is always better to show the actual data points instead of nonrepresentative summary statistics. While some types of studies are heavily dependent upon statistical analysis, other types of studies may not require statistical tests [ 10 , 19 ]. Investigators need to know how to distinguish between these two scenarios and learn how to present and interpret data in cases where statistical tests are not required. Alternative techniques, such as effect size indices with 95% confidence intervals, may be particularly valuable for small datasets [ 8 ].

In cases where power calculations and statistical analysis are needed, students should learn that sample size is one determinant of statistical power, along with effect size and variability. Power calculations often focus on avoiding false negative findings [ 8 ]. Students should understand that underpowered studies produce unreliable p -values and appreciate the risk and consequences of obtaining a false positive finding [ 8 ]. Underpowered studies are one factor that contributes to irreproducibility. Students should be able to determine whether it is feasible to conduct an adequately powered study to answer their research question. This may include performing their own power calculations, consulting with a biostatistician, and determining whether changes in the study design or outcome measurements would improve power. Students should learn how to select a new research question if an adequately powered study is not feasible. The analysis section of the course should focus on techniques for presenting and analyzing small sample size data ( Box 2 ). Students should also understand why techniques that are commonly used for large datasets may not be appropriate for small datasets.

Box 2. Statistical Considerations for Basic Scientists

This list highlights areas where the needs of basic scientists may differ from material that is typically taught in introductory statistics courses.

• Describing data: When are datasets too small for summary statistics, and what alternate methods should be used?
• Probability and distributions: When are tests for a normal distribution too underpowered to be useful?
• Statistical analysis: Alternative strategies for small sample size studies.
• Strategies for identifying outliers and spurious data; understanding the consequences of deleting outliers that are not spurious data
• Understand clustered designs
• Simple analysis strategies and their limitations
• Know when more sophisticated analyses are required and how to consult a statistician
• Selecting statistical software that promotes reproducible research
• When to use figures versus tables
• Selecting the right figure for the type of outcome variable (categorical, continuous, etc.), sample size, and study design
• Choosing figures that show the distribution of continuous data instead of bar graphs
• Critical evaluation of the literature: Students should be able to identify common problems with the way that data are presented and analyzed in published papers in their field and discuss solutions with colleagues and reviewers.
• Know when and how to consult a statistician
• Additional considerations: Bootstrapping and permutation tests, Bayesian statistics

* Note to statisticians: In basic biomedical science, a small sample size refers to groups consisting of fewer than 10…not fewer than 100 or 1,000. Sample sizes of 3 to 6 independent observations per group are very common [ 4 ].

Strategies for Handling Attrition and Outliers

Basic scientists should have training in best practices for reporting attrition, identifying outliers or spurious data points, and analyzing datasets with these features. A recent meta-analysis highlighted the problems with reporting of attrition in preclinical animal studies of cancer and stroke [ 16 ]. The authors could not determine whether animals were excluded or did not complete the experiment in 64.2% of stroke studies and 72.9% of cancer studies. Among studies with clear evidence of attrition (the sample sizes in the methods and results section did not match), most authors did not explain the reasons for attrition.

Basic scientists often work with small samples [ 4 , 15 , 16 ], making it difficult to determine the data distribution and identify outliers. Simulation studies indicate that biased exclusion of a few animals can dramatically inflate the estimated treatment effect [ 16 ]. The potential for effect size inflation was particularly strong for small sample size studies, which are common in preclinical research, or when the authors excluded outliers that worked against the hypothesized treatment effect. Scientists working in fields with slightly larger sample sizes may benefit from learning resampling methods, such as bootstrapping methods and permutation tests. Resampling methods can be particularly relevant in experimental studies, because they do not assume a particular type of sampling distribution but estimate the sampling distribution empirically from the observed data.

The Declaration of Dependence: Reporting Clustered Data

Knowing whether the data are independent is a critical consideration when selecting statistical tests. Techniques such as t tests, Wilcoxon rank-sum tests, and ANOVA, rely on the assumption of independence. Data are independent when the investigators perform one measurement in each subject or specimen, and the subjects or specimens are not related to each other. In contrast, techniques such as paired t tests, Wilcoxon signed-rank tests, and repeated measures ANOVA, assume that the data are not independent. In this design, the investigators repeat measurements on the same subject or specimen under more than one condition or time point. Repeated measures data are not independent and therefore require different analysis techniques.

Introductory statistics courses for basic scientists teach students to analyze independent data and often teach simple techniques for analyzing repeated measures data. Few introductory statistics courses teach students about clustered designs. Only one measurement is performed on each subject or specimen in clustered studies [ 20 ]; however, the data form nonindependent clusters, because some of the subjects or specimens are related to each other. Experiments with clusters of related subjects or specimen are common in physiology and many other basic sciences. Problems with the analysis and presentation of clustered data have been reported in neuroscience [ 3 ], toxicology [ 21 ], wound healing studies [ 22 ], psychology [ 23 ], and ecology [ 24 ]. In vitro laboratory studies, for example, typically include three independent experiments with two or three replicates per experiment. Replicates within an experiment should be more similar to each other than to values obtained during an independent experiment, as replicates are run on the same day, with the same reagents, using identical or nearly identical processing times and conditions. Clustered designs are also used in animal studies. If an investigator studies 30 neurons obtained from three different animals, then all neurons obtained from the same animal form a cluster of nonindependent data. If an investigator examines 25 newborn mice from four different litters, then data obtained from mice in the same litter are clustered.

Clustered designs are rarely included in introductory statistics courses, as procedures for analyzing clustered data can be quite complex. Many basic biomedical scientists do not recognize that these designs include nonindependent data [ 3 ]. This leads to confusion in the published literature. Some authors do not specify how clustered data were analyzed, whereas others analyze the data as though all points were independent. We strongly recommend that statistics instructors discuss clustered designs in courses for students working in fields where clustered data are common. Students should be able to recognize clustered designs, discuss the statistical implications of working with clustered data, and know when to consult a statistician. Understanding the statistical complexities associated with clustered designs may encourage students to avoid these designs in situations where simpler independent study designs are feasible.

To Code or Not to Code…That Is the Question

While statisticians prefer statistical programs that require coding, most introductory statistics courses for basic scientists teach students to use programs that have a user-friendly interface. These programs make statistics less intimidating by allowing students to quickly learn how to perform basic statistical tests. Basic biomedical scientists often feel that programs that require coding are too complex and that the extra features that these programs offer are unnecessary for the simple analyses that are common in basic science (i.e., t tests, ANOVA).

Statistics programs with a user-friendly interface will likely continue to be a centerpiece of statistics education for basic biomedical scientists. However, there are several important considerations that should be taken into account when selecting statistical software for introductory statistics courses.

• The ability to reuse code enhances reproducibility and saves time: Analyses run in coding-based programs are more reproducible. Researchers can save the code for each analysis, which can easily be rerun and checked by others. Reproducibility is more difficult to check in programs that have a user interface, as the results depend on the series of options selected by the user. A small change in any of these options can alter the results of the analysis. Investigators who consistently perform certain types of analysis can reuse code that they have previously written for similar analyses. This saves time, which can offset the additional time required when learning to use a coding-based program.
• Cost and accessibility: Most universities and research centers purchase an institutional license for a particular statistics program with a user interface, then build their courses around that program. Trainees and junior investigators may move several times during the course of their careers. Statistics programs vary among institutions; therefore, young investigators often need to choose between purchasing an expensive individual license or learning to use a new statistics program each time that they move. There are several code-based software packages, in contrast, that are available to everyone free of charge. Researchers trained in these programs can continue to use their existing skills and develop new ones regardless of where they move.
• Ability to run more complex analyses: Multidisciplinary research training is becoming increasingly common, and young investigators may transition among different fields or specialties early in their careers. Coding-based programs allow for more complex analyses. Researchers who are trained to use these programs will have a better foundation for expanding their skills should they move to a different discipline or work with datasets that require more advanced statistics.
• Promoting knowledge retention: Programs that have a user interface often make decisions about what test to use based on the characteristics of the data. Investigators who use these programs may have less knowledge of the specific tests that are being performed or how they are implemented in the statistical software package. Coding-based programs require the user to know what test is needed. This may promote better retention of statistical knowledge.

Many basic biomedical scientists remain strongly opposed to software packages that require coding; however, several PhD programs in our sample offered required or elective statistics courses that were teaching students to use R, a code-based free-of-charge statistical program. Scientists who are trained to use R have access to an extensive library of packages designed for more complex analyses (available from http://cran.us.r-project.org/ ). These packages are available gratis and continue to evolve to meet the growing needs of R users. Other software programs may not offer these types of specialized analysis packages or may require the user to purchase modules for advanced statistical techniques separately.

Data Presentation: Pretty Is Not Necessarily Perfect

Statistical education should include training in all stages of research, including sessions on designing tables and figures for publications. These are less common and, when offered, sometimes focus on making figures that are visually appealing. A visually appealing figure is of little value if it is not suitable for the type of data being presented.

Our systematic review of physiology studies discusses data presentation in more detail [ 4 ]. Students should learn how to decide which data to present in tables versus figures and how to select the appropriate type of figure for their data, based on the type of variable (continuous versus categorical), the study design, and the sample size. Information on how to design visually effective tables and figures can be added once students have learned these fundamentals. Students who do not receive training in data presentation are likely to consult colleagues or refer to published papers when they need to create figures, which may lead them to adopt standard practices for the field. This is a problem, as our recently published systematic review demonstrates that we urgently need to change the way that we present data in small sample size studies [ 4 ]. Data presentation training is essential to improve the quality of data presentation in the scientific literature.

Critical Evaluation of the Literature

Statistics courses should include a “synthesis” module, where students discuss and critique data presentation and analysis practices for papers published in their respective fields. This module should include experiences with critical evaluation of individual papers, as well as a review of metaresearch articles that examine strengths and weaknesses of data presentation and analysis in related fields of research. Students should understand standard practices for their fields, be able to explain the problems with these practices, discuss the advantages and disadvantages of potential solutions, and select the solution that best fits a particular dataset. The benefits of statistics training may be lost if students are not trained to address common problems with standard data presentation and analysis practices in their field of research.

Knowing When and How to Consult a Statistician

While this paper focuses on biostatistics training for basic scientists, we strongly support initiatives to build collaborations between basic scientists and biostatisticians. An effective statistics course not only teaches students what they have learned; it should also teach students what they have not learned. Basic scientists should be able to identify common situations in which statistical tests are not appropriate and know when the study team lacks the expertise to perform more complex analyses that may be required. They should be prepared to consult with a statistician to determine whether a different study design can be used, and plan for any statistical resources that will be needed. Basic scientists should be aware of institutional resources that provide statistical support, understand the complementary roles of basic scientists and statisticians, and learn how to develop and maintain an effective collaboration with a statistician during all stages of the research process.

Recommendation 3: Develop Tools and Strategies to Promote Education and Dissemination of Statistical Knowledge

Establishing the importance of statistics in the student’s field of study.

Students and professors sometimes feel that statistics courses are less valuable than core courses in the student’s chosen field. Although statistics is an essential skill, medical students reported neutral perceptions about the value of biostatistics and their interest in statistics [ 25 ]. The perception that statistics is intimidating, or that statistics courses are not relevant to the student’s field of study can interfere with student efforts to learn the statistical concepts and skills that they will need to conduct sound research. Several strategies discussed in this article may help to mitigate this problem. Courses centered around field-specific study designs, datasets, and exercises implicitly show students how statistical knowledge is critical to their research careers. General courses in which topics and materials do not align with research conducted in the students’ field of study may inadvertently suggest that statistics are less relevant and less important. Critiquing the literature will help students to appreciate the importance of statistics in conducting reproducible research in their chosen fields. Finally, we suggest that mentors meet with mentees to discuss the role of biostatistics in the mentee’s field of study and to emphasize the ways in which understanding biostatistics skills and concepts is crucial to success.

Customizing Statistics Education with Limited Resources

Tailoring courses to the needs of students in different fields of study is labor intensive and may not be feasible at institutions with a limited number of statistics instructors. Integrating customized online modules into general statistics courses may be one potential solution to this problem. Neuroscientists, for example, might complete a module on clustered data, whereas geneticists might complete a module on quality control. This approach would allow trainees to learn field-specific concepts and skills, without requiring a separate statistics course for each department. A comprehensive meta-analysis recently reported that well-organized online classrooms were as effective and of comparable quality to traditional classrooms [ 26 ]. Furthermore, there was a modest improvement in learning in courses that combined online learning with face-to-face instruction, when compared to face-to-face instruction alone [ 26 ]. One of the authors (NMM) has successfully used this combined approach to teach medical statistics [ 27 ].

A second option would be for professional societies to facilitate customized statistics education by defining core competencies and creating educational materials. This might include field-specific sample datasets and exercises, as well as online modules for topics that are not routinely included in general statistics courses. These materials could be integrated into existing courses or viewed independently by established investigators seeking additional training on particular topics. The National Institutes of Health Rigor and Reproducibility Training Modules provide an example of this “open access” approach to providing educational materials ( http://1.usa.gov/1OmBIWZ ). Field-specific materials may also be a useful strategy for increasing trainees’ interest in statistics.

Do One, Teach One: Improving Statistical Knowledge in the Scientific Community (Dissemination)

We propose that statistics courses should prepare students to disseminate their knowledge to others in their field. Formal statistics education typically targets trainees and junior investigators. Senior investigators play a much more prominent role in shaping the literature; however, most do not have time to take courses. Offering online lectures for continuing professional development would allow investigators at all levels to augment their knowledge of current topics in statistics and data presentation. A “grass-roots” approach to statistics education is also needed, as many trainees and junior scientists will need to convince peers, colleagues, and reviewers if they want to improve the quality of data presentation and statistical analysis in published papers. In addition to completing the “synthesis” module that was described previously, students should also know key references that describe problems with the standard practices in their field and outline solutions. References that focus on the practical implications of statistical techniques and are accessible to readers with little or no statistics background may be most valuable in encouraging basic scientists to re-evaluate their approach to data presentation, statistical analysis, and statistics education.

Conclusions

Although understanding statistical concepts and skills is essential for basic science research, biostatistics training is not always required to complete a PhD. Our recommendations for improving statistics training for basic biomedical scientists include: 1. Encouraging departments to require statistics training, 2.Tailoring coursework to the student’s field of research, and 3. Developing tools and strategies to promote education and dissemination of statistical knowledge. Faculty members of basic science departments should work with statistics instructors to design coursework that focuses on the study designs, types of outcomes, and sample sizes that are common in the students’ field. Finally, students should learn to critically evaluate data presentation and statistical analysis in the published literature.

Supporting Information

S1 fig. flow chart for systematic review of physiology studies..

https://doi.org/10.1371/journal.pbio.1002430.s001

S2 Fig. Flow chart for study of statistical education practices in PhD programs.

https://doi.org/10.1371/journal.pbio.1002430.s002

S1 Text. Supplemental methods and results for the data presented in Box 1 and Fig 1 .

https://doi.org/10.1371/journal.pbio.1002430.s003

• View Article
• Google Scholar
• PubMed/NCBI
• 18. Council for the International Organization of Medical Sciences, The International Council for Laboratory and Animal Sciences (2012) International Guiding Principles for Biomedical Research Involving Animals. http://grants.nih.gov/grants/olaw/Guiding_Principles_2012.pdf

Basic Concepts for Biostatistics

Lisa Sullivan, PhD, Professor of Biostatistics, Boston University School of Public Health

Introduction

Biostatistics is the application of statistical principles to questions and problems in medicine, public health or biology. One can imagine that it might be of interest to characterize a given population (e.g., adults in Boston or all children in the United States) with respect to the proportion of subjects who are overweight or the proportion who have asthma, and it would also be important to estimate the magnitude of these problems over time or perhaps in different locations. In other circumstances in would be important to make comparisons among groups of subjects in order to determine whether certain behaviors (e.g., smoking, exercise, etc.) are associated with a greater risk of certain health outcomes. It would, of course, be impossible to answer all such questions by collecting information (data) from all subjects in the populations of interest. A more realistic approach is to study samples or subsets of a population. The discipline of biostatistics provides tools and techniques for collecting data and then summarizing, analyzing, and interpreting it. If the samples one takes are representative of the population of interest, they will provide good estimates regarding the population overall. Consequently, in biostatistics one analyzes samples in order to make inferences about the population. This module introduces fundamental concepts and definitions for biostatistics.

Learning Objectives

After completing this module, the student will be able to:

• Define and distinguish between populations and samples.
• Define and distinguish between population parameters and sample statistics.
• Compute a sample mean, sample variance, and sample standard deviation.
• Compute a population mean, population variance, and population standard deviation.
• Explain what is meant by statistical inference.

----------- 

Population Parameters versus Sample Statistics

As noted in the Introduction, a fundamental task of biostatistics is to analyze samples in order to make inferences about the population from which the samples were drawn .  To illustrate this, consider the population of Massachusetts in 2010, which consisted of 6,547,629 persons. One characteristic (or variable) of potential interest might be the diastolic blood pressure of the population. There are a number of ways of reporting and analyzing this, which will be considered in the module on Summarizing Data. However, for the time being, we will focus on the mean diastolic blood pressure of all people living in Massachusetts. It is obviously not feasible to measure and record blood pressures for of all the residents, but one could take samples of the population in order estimate the population's mean diastolic blood pressure.

Despite the simplicity of this example, it raises a series of concepts and terms that need to be defined. The terms population , subjects , sample , variable , and data elements are defined in the tabbed activity below.

It is possible to select many samples from a given population, and we will see in other learning modules that there are several methods that can be used for selecting subjects from a population into a sample. The simple example above shows three small samples that were drawn to estimate the mean diastolic blood pressure of Massachusetts residents, although it doesn't specify how the samples were drawn. Note also that each of the samples provided a different estimate of the mean value for the population, and none of the estimates was the same as the actual mean for the overall population (78 mm Hg in this hypothetical example). In reality, one generally doesn't know the true mean values of the characteristics of the population, which is of course why we are trying to estimate them from samples. Consequently, it is important to define and distinguish between:

• population size versus sample size
• parameter versus sample statistic.

Sample Statistics

In order to illustrate the computation of sample statistics, we selected a small subset (n=10) of participants in the Framingham Heart Study. The data values for these ten individuals are shown in the table below. The rightmost column contains the body mass index (BMI) computed using the height and weight measurements. We will come back to this example in the module on Summarizing Data, but it provides a useful illustration of some of the terms that have been introduced and will also serve to illustrate the computation of some sample statistics.

Data Values for a Small Sample

The first summary statistic that is important to report is the sample size. In this example the sample size is n=10. Because this sample is small (n=10), it is easy to summarize the sample by inspecting the observed values, for example, by listing the diastolic blood pressures in ascending order:

62        63        64        67        70        72        76        77        81        81

Simple inspection of this small sample gives us a sense of the center of the observed diastolic pressures and also gives us a sense of how much variability there is. However, for a large sample, inspection of the individual data values does not provide a meaningful summary, and summary statistics are necessary.  The two key components of a useful summary for a continuous variable are:

• a description of the center or 'average' of the data (i.e., what is a typical value?) and
• an indication of the variability in the data.   

Sample Mean

There are several statistics that describe the center of the data, but for now we will focus on the sample mean, which is computed by summing all of the values for a particular variable in the sample and dividing by the sample size. For the sample of diastolic blood pressures in the table above, the sample mean is computed as follows:

To simplify the formulas for sample statistics (and for population parameters), we usually denote the variable of interest as "X".  X is simply a placeholder for the variable being analyzed.  Here X=diastolic blood pressure. 

The general formula for the sample mean is:

The X with the bar over it represents the sample mean, and it is read as "X bar". The Σ indicates summation (i.e., sum of the X's or sum of the diastolic blood pressures in this example). 

When reporting summary statistics for a continuous variable, the convention is to report one more decimal place than the number of decimal places measured.  Systolic and diastolic blood pressures, total serum cholesterol and weight were measured to the nearest integer, therefore the summary statistics are reported to the nearest tenth place. Height was measured to the nearest quarter inch (hundredths place), therefore the summary statistics are reported to the nearest thousandths place. Body mass index was computed to the nearest tenths place, summary statistics are reported to the nearest hundredths place.  

Sample Variance and Standard Deviation 

If there are no extreme or outlying values of the variable, the mean is the most appropriate summary of a typical value, and to summarize variability in the data we specifically estimate the variability in the sample around the sample mean. If all of the observed values in a sample are close to the sample mean, the standard deviation will be small (i.e., close to zero), and if the observed values vary widely around the sample mean, the standard deviation will be large.  If all of the values in the sample are identical, the sample standard deviation will be zero.

When discussing the sample mean, we found that the sample mean for diastolic blood pressure = 71.3. The table below shows each of the observed values along with its respective deviation from the sample mean.

Table - Diastolic Blood Pressures and Deviations from the Sample Mean

The deviations from the mean reflect how far each individual's diastolic blood pressure is from the mean diastolic blood pressure. The first participant's diastolic blood pressure is 4.7 units above the mean while the second participant's diastolic blood pressure is 7.3 units below the mean. What we need is a summary of these deviations from the mean, in particular a measure of how far, on average, each participant is from the mean diastolic blood pressure.  If we compute the mean of the deviations by summing the deviations and dividing by the sample size we run into a problem.  The sum of the deviations from the mean is zero.  This will always be the case as it is a property of the sample mean, i.e., the sum of the deviations below the mean will always equal the sum of the deviations above the mean. However, the goal is to capture the magnitude of these deviations in a summary measure. To address this problem of the deviations summing to zero, we could take absolute values or square each deviation from the mean.  Both methods would address the problem.  The more popular method to summarize the deviations from the mean involves squaring the deviations (absolute values are difficult in mathematical proofs). The table below displays each of the observed values, the respective deviations from the sample mean and the squared deviations from the mean.

The squared deviations are interpreted as follows. The first participant's squared deviation is 22.09 meaning that his/her diastolic blood pressure is 22.09 units squared from the mean diastolic blood pressure, and the second participant's diastolic blood pressure is 53.29 units squared from the mean diastolic blood pressure. A quantity that is often used to measure variability in a sample is called the sample variance, and it is essentially the mean of the squared deviations. The sample variance is denoted s 2 and is computed as follows:

 In this sample of n=10 diastolic blood pressures, the sample variance is s 2 = 472.10/9 = 52.46. Thus, on average diastolic blood pressures are 52.46 units squared from the mean diastolic blood pressure. Because of the squaring, the variance is not particularly interpretable. The more common measure of variability in a sample is the sample standard deviation, defined as the square root of the sample variance:

A sample of 10 women seeking prenatal care at Boston Medical center agree to participate in a study to assess the quality of prenatal care. At the time of study enrollment, you the study coordinator, collected background characteristics on each of the moms including their age (in years).The data are shown below:

24        18        28        32        26        21        22        43        27        29

A sample of 12 men have been recruited into a study on the risk factors for cardiovascular disease. The following data are HDL cholesterol levels (mg/dL) at study enrollment:

50        45        67        82        44        51        64        105      56        60        74        68 

Population Parameters

The previous page outlined the sample statistics for diastolic blood pressure measurement in our sample. If we had diastolic blood pressure measurements for all subjects in the population, we could also calculate the population parameters as follows:

Population Mean

Typically, a population mean is designated by the lower case Greek letter µ (pronounced 'mu'), and the formula is as follows:

where "N" is the populations size.

Population Variance and Standard Deviation

Statistical inference.

We usually don't have information about all of the subjects in a population of interest, so we take samples from the population in order to make inferences about unknown population parameters .

An obvious concern would be how good a given sample's statistics are in estimating the characteristics of the population from which it was drawn. There are many factors that influence diastolic blood pressure levels, such as age, body weight, fitness, and heredity.

We would ideally like the sample to be representative of the population . Intuitively, it would seem preferable to have a random sample , meaning that all subjects in the population have an equal chance of being selected into the sample; this would minimize systematic errors caused by biased sampling.

In addition, it is also intuitive that small samples might not be representative of the population just by chance, and large samples are less likely to be affected by "the luck of the draw"; this would reduce so-called random error. Since we often rely on a single sample to estimate population parameters, we never actually know how good our estimates are. However, one can use sampling methods that reduce bias, and the degree of random error in a given sample can be estimated in order to get a sense of the precision of our estimates.

Biostatistics 101: data presentation

Affiliation.

• 1 Clinical trials and Epidemiology Research Unit, 226 Outram Road, Blk A #02-02, Singapore 169039. [email protected]
• PMID: 14560857
• Biometry / methods*
• Curriculum*
• Data Interpretation, Statistical
• Education, Medical*
• Mathematical Computing
• Research Design

Basic Concepts, Organizing, and Displaying Data

• First Online: 16 June 2018

Cite this chapter

• M. Ataharul Islam 3 &
• Abdullah Al-Shiha 4  

3729 Accesses

• The original version of this chapter was revised: The explanation related to Table 1.5. has been corrected. The correction to this chapter is available at https://doi.org/10.1007/978-981-10-8627-4_12

This chapter introducesbiostatistics as a discipline that deals with designing studies, analyzingdata , and developing new statistical techniques to address the problems in the fields of life sciences. This includes collection, organization, summarization, and analysis of data in the fields of biological, health, and medical sciences including other life sciences. One major objective of a biostatistician is to find the values that summarize the basic facts from the sample data and to makeinference about the representativeness of the estimates using the sample data to make inference about the correspondingpopulation characteristics. The basic concepts are discussed along with examples and sources of data, levels ofmeasurement , and types of variables. Various methods of organizing and displaying data are discussed for both ungrouped andgrouped data . The construction of table is discussed in details. This chapter includes methods of constructing frequency bar chart, dot plot, pie chart,histogram ,frequency polygon , andogive . In addition, the construction ofstem-and-leaf display is discussed in details. All these are illustrated with examples. As the raw materials ofstatistics aredata , a brief section on designing of sample surveys including planning of a survey and major components is introduced in order to provide some background about collection of data.

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

• Available as EPUB and PDF
• Read on any device
• Instant download
• Own it forever
• Compact, lightweight edition
• Dispatched in 3 to 5 business days
• Free shipping worldwide - see info
• Durable hardcover edition

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Change history

04 september 2020, author information, authors and affiliations.

ISRT, University of Dhaka, Dhaka, Bangladesh

M. Ataharul Islam

Department of Statistics and Operations Research, College of Science, King Saud University, Riyadh, Saudi Arabia

Abdullah Al-Shiha

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to M. Ataharul Islam .

Rights and permissions

Reprints and permissions

Copyright information

© 2018 Springer Nature Singapore Pte Ltd.

About this chapter

Islam, M.A., Al-Shiha, A. (2018). Basic Concepts, Organizing, and Displaying Data. In: Foundations of Biostatistics. Springer, Singapore. https://doi.org/10.1007/978-981-10-8627-4_1

Download citation

DOI : https://doi.org/10.1007/978-981-10-8627-4_1

Published : 16 June 2018

Publisher Name : Springer, Singapore

Print ISBN : 978-981-10-8626-7

Online ISBN : 978-981-10-8627-4

eBook Packages : Mathematics and Statistics Mathematics and Statistics (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Publish with us

Policies and ethics

• Find a journal
• Track your research

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Dental Press J Orthod
• v.26(1); 2021

Language: English | Portuguese

Biostatistics: essential concepts for the clinician

Darlyane torres.

1 Universidade Federal do Pará, Programa de Pós-Graduação em Odontologia (Belém/PA, Brazil)

David NORMANDO

2 Universidade Federal do Pará, Departamento de Odontologia (Belém/PA, Brazil)

AUTHORS' CONTRIBUTION: Darlyane Torres (DT)

Conception or design of the study: : DT, DN.

Data acquisition, analysis or interpretation: : DT, DN.

Writing the article: : DT, DN.

Critical revision of the article: : DT, DN.

Final approval of the article: : DT, DN.

Overall responsibility: : DN.

Introduction:

The efficiency of clinical procedures is based on practical and theoretical knowledge. Countless daily information is available to the orthodontist, but it is up to this professional to know how to select what really has an impact on clinical practice. Evidence-based orthodontics ends up requiring the clinician to know the basics of biostatistics to understand the results of scientific publications. Such concepts are also important for researchers, for correct data planning and analysis.

This article aims to present, in a clear way, some essential concepts of biostatistics that assist the clinical orthodontist in understanding scientific research, for an evidence-based clinical practice. In addition, an updated version of the tutorial to assist in choosing the appropriate statistical test will be presented. This PowerPoint ® tool can be used to assist the user in finding answers to common questions about biostatistics, such as the most appropriate statistical test for comparing groups, choosing graphs, performing correlations and regressions, analyzing casual, random or systematic errors.

Conclusion:

Researchers and clinicians must acquire or recall essential concepts to understand and apply an appropriate statistical analysis. It is important that journal readers and reviewers can identify when statistical analyzes are being inappropriately used.

Introdução:

A eficiência dos procedimentos clínicos é baseada em conhecimentos práticos e teóricos. Inúmeras informações diárias estão ao alcance do ortodontista; porém cabe a esse profissional saber selecionar o que realmente tem impacto na prática clínica. A Ortodontia baseada em evidências acaba exigindo que o clínico conheça os fundamentos da bioestatística para compreender os resultados das publicações científicas. Tais conceitos também são importantes aos pesquisadores para um correto planejamento e análise dos dados.

O presente artigo tem como objetivo apresentar, de forma clara, alguns conceitos essenciais da bioestatística que auxiliem o ortodontista clínico na compreensão da pesquisa científica para uma prática clínica baseada em evidências. Além disso, será apresentada uma versão atualizada do tutorial para auxílio na escolha do teste estatístico adequado. Essa ferramenta em PowerPoint® pode ser empregada para auxiliar o usuário a encontrar respostas para dúvidas comuns sobre bioestatística, como o teste estatístico mais adequado para comparar grupos, escolha de gráficos, realizar correlações e regressões, análises de sobrevivência e dos erros aleatório e sistemático.

Conclusão:

Pesquisadores e clínicos devem adquirir ou relembrar conceitos essenciais para compreender e aplicar uma análise estatística apropriada. É importante que os leitores e revisores de periódicos possam identificar quando análises estatísticas estão sendo utilizadas de forma inadequada.

INTRODUCTION

Every professional, regardless of the area of training, has a role in decision-making based on theoretical and practical knowledge. Regarding health professionals, where it is essential to maintain or promote the health of the patient, any inappropriate decision may cause irreversible biological damage to patients. Currently, Orthodontics has been submitted to an avalanche of new information, technologies and experiences, which are easily accessible. And it is up to the orthodontist to discern the reliable scientific knowledge from those who have errors or bias - acquiring for their clinical practice what will, for example, reduce error rates, waste, unsuccessful therapies and unnecessary exams. 1 , 2

Evidence-based Orthodontics can become a challenge for clinicians. This is because published papers often present information that makes understanding scientific knowledge a complex task. 3 , 4 A substantial level of experience in statistical understanding is necessary in the critical reading of the research, the methodology used, data analysis and interpretation of the results, for the acquisition of conclusions that will reduce the uncertainties in decision making, in view of the variability of available options. 2 , 5 - 7

Statistics are known to have a direct connection to mathematics. And the culture of fear and anxiety that surrounds it makes the assimilation of statistical concepts and methods complex. 8 Some studies show that graduate students, despite understanding the importance of biostatistics, do not have the skills to apply it correctly in scientific research; and that attitudes, successes and failures in face of statistical challenges are linked to basic knowledge. 6 , 9 - 11 This ends up having an impact on scientific publications. Studies showed that it is common to find errors such as incompatible study design, inadequate analysis and inconsistent interpretations. 12 - 14

The basic concepts, which are fundamental to avoid errors, are often easy to forget, impacting the choice of statistical tests used in the data analysis. In addition, most statistical software does not guide the user in choosing the most appropriate statistical test for the research, generating scientific publications that do not contribute to the solution of a clinical problem, due to the wrong data analysis. 15

Therefore, the objective of the present article is to clearly review some essential concepts of biostatistics that will assist clinical orthodontists in understanding scientific research for an evidence-based clinical practice, in addition to indicate the main errors observed in published articles. Then, it will be presented the updated version of a PowerPoint ® guide, originally published in 2010, to assist in choosing the appropriate statistical test. 16 This guide is useful for readers, authors and reviewers of scientific articles.

BASIC CONCEPTS

Biostatistics is a method used to describe or analyze data obtained from a sample that represents a population. It is used in studies in which variables are related to living beings. 17 , 18

WHAT IS A VARIABLE AND HOW IS IT MEASURED?

Variable is a characteristic or condition that can be measured or observed in the sample or population. It can assume different values from one sampling unit to another or in the same unit over time. It is important to know how to classify the variable according to the data it generates. To understand the classifications of the variables regarding the scale used and the type of participation in the study, see Tables 1 and 2, respectively. 8 , 17 , 18

THE IMPORTANCE OF NORMAL DISTRIBUTION

A distribution in biostatistics refers to a mathematical model that relates values of a variable and the probability of occurrence of each value. It should be clarified that whenever there is a quantitative variable that will be analyzed, it is assumed to verify the normality of the data distribution, by statistical test and/or histogram, according to the need. Some statistical tests require a distribution with normal characteristics as a requisite. Where the data is concentrated around the average and from there they are dispersed in a symmetrical way, with a characteristic bell-shaped graph. When the distribution is different from the normal, preference should be given to the use of median and interquartile deviation. 17 - 19 The graphical elucidation of normal and abnormal data distributions can be seen in Figure 1 .

HOW SHOULD THE DATA BE PRESENTED? (DESCRIPTIVE STATISTICS)

The organization and presentation of these data, made by appropriate methods, can be summarized, known as descriptive statistics. This concept is the initial step for an appropriate selection and use of statistical tests. Descriptive statistics can be divided into frequencies and/or summary measures of central tendency and dispersion ( Table 3 ). 8 , 17 , 18

WHY USE STATISTICAL TESTS? (INFERENTIAL STATISTICS)

Inferential statistics allow comparing samples or predicting behaviors of variables. This tool establishes conclusions based on a small portion of a population, with a minimum and previously determined margin of error. Statistical tests are used to quantify the uncertainty of decision making by means of probabilistic principles. 6 , 17

Allows the researcher to have a degree of reliability in the statements assumed in the sample, regarding the population. Thus, when the reader realizes that the published study performed statistical tests, he must ask: “How likely am I to trust these results?” or “How much uncertainty is there in the results for an extrapolation of the results (generalization)?". These questions should be asked at the beginning of the study, in order to define the chances of error, the confidence and the estimated margins of the population parameter of your sample. The following are the concepts of interest: 6 , 17 - 19 , 20

• Significance, or α level ( p -value): ): it represents the chance that the researcher is wrong in stating that there is a difference (or significance), and the difference, in fact, does not exist. Known as type I error or false positive, it can be predetermined at the beginning of the study as 1% or 5%. It can be said that when you have a p- value less than the level of significance, then you have a real difference between samples or groups, when applied in groups comparison tests.
• β error: known as a type II error, or false negative, it represents the chance of the researcher making a mistake in stating that there is no difference when the difference is true. A maximum value of 20% is allowed.
• Study power (1 - β ) : represents the chance for the researcher to be sure that there is a difference when it really exists. It is also defined by the researcher before data collection begins, and is usually at least 80%, or 0.8 (1 - 0.2).
• Confidence interval: represents the estimate of a sample parameter for a population parameter. Contains upper and lower limits, defined according to the stipulated level of significance. For a 95% confidence interval, we have that for every 100 studies performed, within the same methodology and (n), but with different subjects from the same population, it must be estimated that the population parameter is present in the data distribution of 95 studies. However, there is no need to carry out the 100 studies for this estimate. Just perform a single study and define this interval by 95% CI = mean ± (1.96 x standard error). Therefore, standard error = standard deviation divided by √n .

Currently, journals and reviewers have requested in the results not only the p -value, but also the referring confidence interval (CI). Some years ago, only a few studies with with multivariate analyzes reported the CI found. 21 A systematic review 22 showed that the interpretation of the CI is important, but it rarely occurs in those randomized clinical trials where the effects of treatments were not statistically significant. This can lead to the abandonment of future research or to a clinical practice based on invalid conclusions.

TYPES OF STUDIES

The execution of a study must always be planned, and this plan for conducting the research is called research design. It must follow specific standards and techniques, according to the nature of the study. 6 , 17 - 19

The quality of research designs is related to the strength of recommendation and applicability to the patient. 18 This difference between the degree of strength of the types of studies can be seen in Figure 2 , representing a pyramid of evidence.

This pyramid incorporates the suggestion of Murad et al. 23 that considers not only the study design, but also the assessment of the certainty of the evidence, examined by the GRADE tool (Grading of Recommendations Assessment, Development and Evaluation). 24 So that, for example, a cross-sectional study very well performed can produce a much better quality of results than a case-control study not so well developed, therefore, producing a greater impact on clinical decision. This type of change in the strength of the recommendation can occur in studies present from the middle to the top of the pyramid, and will be seen below: 8 , 18 , 19

• » Cross-sectional: It is considered a “portrait study”. It determines the situation of interest and outcome in a single moment, assessing the prevalence and relationship between variables, comparing exposed and unexposed or with disease and without disease. Example: to analyze the association between gingival inflammation (present/absent) and the use of orthodontic appliance (exposure) in a single moment of treatment, comparing with patients without orthodontic treatment (control).
• » Cohort: It is a longitudinal study, considered a “film study”. It starts from the exposure to the outcome (disease), and observes over time individuals exposed and not exposed (control group) to a factor - who have not yet developed the outcome of interest -, assessing incidence, carrying out supervised monitoring and establishing etiology and risk factors. Although it is generally prospective, when the data registration coincides with the beginning of the research; there are retrospective cohorts, when the research is initiated after recording the data. Example: analyzing the development of gingival inflammation (incidence) during orthodontic treatment (exposure), with evaluation of gingival condition at the beginning and end of treatment, compared to non-orthodontic patients.
• » Case-control: It is also a longitudinal study and evaluates individuals who already have the disease of interest, comparing them with a control group (individuals without the disease), measuring the exposures or interventions performed during the study. Although it is generally retrospective, they can be carried out prospectively. Example: comparative analysis of orthodontic patients with and without gingival inflammation (disease, or outcome) among patients who did or did not use daily mouthwash. In this case, starts from the disease to the exposure.
• Randomized Clinical Trial: This is a simulation study of the reality in which an exposure or intervention in the experimental sample occurs, in comparison to a control group. The main feature is the allocation of research subjects being carried out by randomization between groups. It is a highly controlled study. However, the randomization method can fail, especially when small samples are analyzed.
• Synthesis: This category includes the secondary study called “Systematic Review”. It uses primary studies as a source of data to obtain the answer to a key question. It is a scientific investigation carried out under a rigorous methodology for both data searches and analyzes, and the consequent determination of the certainty of the available evidence. When possible, a “meta-analysis” is carried out, which is the statistical analysis to combine the results of the included primary studies. It is at the top of the evidence pyramid for clinical decision-making ( Fig 2 ).

MAIN ERRORS IN THE STATISTICAL METHODOLOGY OBSERVED IN THE PUBLISHED ARTICLES

Articles that will be submitted to journals must be very well written and designed. This requires that the study be conducted in a reliable manner, allowing the correct description of all the steps performed and the consequent ease of reading and acceptance of the article by the reviewers 25 . Below are the most common errors found in published articles, regarding the statistical methodology employed.

USE OF COLUMN / BAR GRAPH FOR QUANTITATIVE VARIABLES

Column graphics should be used for frequency graphics, as each column represents a category. When we have numeric variables, we should use the box-plot graph ( Fig 3 ) for independent samples and the line graph for data over time 8 , 18 . The box-plot, unlike the column graphic, allows us to observe the summary measure (mean or median) and the dispersion of the obtained values.

USE OF PARAMETRIC TESTS WHEN NORMALITY IS NOT ACHIEVED

Parametric tests are more powerful than non-parametric tests, but they presume a normal distribution of data. Numerical data with abnormal distribution should be analyzed as if they were qualitative data. The use of a parametric test in this situation implies greater ease in rejecting the null hypothesis, which may not represent the population’s reality. 8 , 19

USE OF MEAN AND STANDARD DEVIATION WHEN THERE IS AN ABNORMAL DISTRIBUTION

Although some researchers correctly use non-parametric tests when the assumption of normality is broken, they sometimes incorrectly use the mean and standard deviation when presenting data. These should not be used exactly due to the asymmetric distribution. In this case, use the nonparametric reference that divides the data in half: the median and its deviation (interquartile deviation). 1 , 8 , 17

OVERSIZED OR UNDERSIZED SAMPLES

Every sample has a minimum number of sample units needed to represent the population. When you have a sample below that quantity (undersized), only large differences can be detected in a significant way. In addition, small samples tend to have an abnormal distribution, which would lead to the use of less robust tests; while large samples are not practical, since they increase the cost and time required for the study. 17 , 18

A GUIDE TO ASSIST IN CHOOSING THE APPROPRIATE STATISTICAL TEST

In order to obtain a reliable statistical result that allows extrapolation to the population of interest, it is extremely important to know which test is the most suitable for the study. A PowerPoint ® guide to assist in choosing the statistical test was published in 2010 16 , and since then it has been widely used by countless researchers. With more than 244 thousand accesses (02/10/2021), this is the article with the highest number of downloads in the SciELO collection for the dentistry area. The version 2020 3.0 presents a new layout and has additional multivariate analyzes. In addition, this version provides the paths for running the tests on several free software, such as Jamovi (version 1.2, Sydney, Australia) and BioEstat (version 5.3, Amazonas, Brazil), as well as on the website www.vassarstats.net (VassarStats, Richard Lowry - United States). The download can be done according to the preferred language (Portuguese, English or Spanish), through the following link:

http://www.ppgo.propesp.ufpa.br/index.php/br/programa/noticias/todas/176-tutorial-teste-estatistico-para-pesquisa-cientifica

The use is simple and must be done in the “presentation mode”. The INITIAL MENU represents the objective intended by the researcher. There are six possible options ( Fig 4 ), where it is possible to arrive at the desired answer through a sequence of clicks on drops.

By clicking on the option “Examine the type of distribution”, you have the direct answer of which test you can perform to examine the distribution of data for a quantitative variable. The same occurs when clicking on the drop “Survival analysis”, in which the possible options for survival tests are directly available.

In the other options of the initial menu, there is a hyperlink to the submenus:

• » Comparison menu: allows obtaining the specific statistical test for the difference between two or more paired or independent samples with or without normal distribution.
• » Correlation menu: indicates the analyzes for correlation and/or modeling between two or more variables.
• » Replicability menu: indicates the measurement accuracy tests for the analysis of random and systematic errors.
• » Graphic menu: provides the researcher with the appropriate graph for the type of sample and objective of the study.

So that, after a sequence of clicks on drops, it is possible to obtain the desired response. Figure 5 exemplifies a submenu - in this case, the comparison.

This sequence of clicks requires basic knowledge about types of variables ( Table 1 ) and data distribution ( Fig 1 ). It is also necessary to understand the difference between dependent (paired) and independent (unpaired) samples. Paired samples are those in which the comparison is dependent on the same individual (before vs. after; right vs. left; T1 vs. T2 vs. T3). Independent samples are those in which there is a comparison between different individuals. After identifying the desired answer, you can obtain the test execution path in both BioEstat (not available in the English version) and Jamovi, and, in some cases, in VassarStats, by clicking on the corresponding icon that will appear ( Fig 6 ).

In this update, when you click on an item by mistake, it is possible to return to the submenu in which it was already found, providing greater agility during use.

It is important that readers and reviewers of journals can identify when a research uses inadequate statistical analysis, disregarding fundamental concepts. The new version of the "Tutorial for choosing the test" presented is a path to the most appropriate use of statistical tests, allowing the correction of wrong choices.

• Rankings, Awards, and Stats
• Department Awards
• Diversity and Inclusion
• Undergraduate
• Information for New Statistics Freshmen
• Statistics Major Advising and Registration FAQs
• Statistics Major
• Statistics Minor
• Academic and Research Opportunities
• Clubs and Organizations
• Ph.D. Programs
• Master’s Programs
• Online Programs
• Application Process
• Student Life
• Research Areas
• Administration
• Graduate Students
• Faculty Resources
• Staff Resources
• Student Resources
• IT Resources

Data Science and Environmental Health Science Symposium draws area researchers

(left to right) Dr. Francesca Dominici, Harvard University; Dr. Michelle Heacock, National Institute of Environmental Health Sciences (NIEHS); Dr. Ricky Woychik, Director of NIEHS; Dr. Seth Kullman, NC State; Dr. Fred Wright, NC State.

On April 5, 2024, the university hosted a symposium on “Data Science and Environmental Health Science Research” at NC State’s Hunt Library. The symposium, with an organizing committee including Department Head Dr. Kimberly Sellers and Professor Fred Wright, covered numerous areas of statistical methodology and practice applied to problems in environmental sciences. An introduction by Dr. Rick Woychik, Director of the National Institute of Environmental Health Sciences, highlighted institute priorities and funding initiatives in statistics and data sciences. The keynote talk was given by Dr. Francesca Dominici, the Gamble Professor of Biostatistics, Population, and Data Sciences in the Department of Biostatistics at Harvard University. Dr. Dominici described her group’s work in establishing links between air pollution and adverse effects on human health, using a multiplicity of data sources combined with observational epidemiology and causal inference, and the translation of this work to regulatory action. Other presentations throughout the day highlighted diverse application areas including the study of gene by environment interactions, massive disease-exposure association mapping, multi-omics technologies to investigate the effects of exposures, and toxicogenomic profiling to prioritize chemicals for further study. Statistics Department students and faculty were well represented in poster presentations, including graduate students Xiaodan Zhou (pictured with Dr. Sellers (left) and Dr. Dominici (center)) and Nate Wiecha, and supervising faculty Drs. Brian Reich and Emily Griffith.

Author: Fred Wright

Brahmar Mukherjee Celebrated at the Marvin Zelen Leadership Award in Statistical Science

On Thursday, May 9 th ,  around 60 members of the Biostatistics community gathered to celebrate the presentation of the  Marvin Zelen Leadership Award  to Bhramar Mukherjee.

Currently the John D. Kalbfleisch Distinguished University Professor of Biostatistics, Chair of the Biostatistics Department, and Assistant Vice President for Research at the University of Michigan,  Dr. Mukherjee is one of the foremost statisticians working in key areas of statistical genetics and genetic epidemiology, with an expansive body of work that has helped to define the methodological landscape in statistical genetics.

For the ceremony, Dr. Mukherjee was introduced by Xihong Lin.  She then gave a talk titled “Unveiling Bias: A Statistician’s Quest for Data Equity in Health Research”, in which she delved into the crucial concept of data equity and highlighted how algorithms developed with exclusionary datasets yield erroneous conclusions and exacerbate health disparities.

Drawing examples from her own work, including analyses of COVID-19 and biobanks linked with electronic health records, she talked about instances where timely statistical analysis with imperfect data resulted in enhanced inference and influenced policy outcomes. She also urged statisticians to follow the legacy of Professor Marvin Zelen by proactively leading efforts in curating and collecting new data.

She concluded her talk by asserting the important role of statisticians in moving beyond design and analysis, to “assert their independent leadership in shaping the trajectory of research and policymaking driven by data.”

News from the School

From public servant to public health student

Exploring the intersection of health, mindfulness, and climate change

Conference aims to help experts foster health equity

Building solidarity to face global injustice

Main Navigation

• Contact NeurIPS
• Code of Ethics
• Code of Conduct
• Create Profile
• Journal To Conference Track
• Diversity & Inclusion
• Proceedings
• Future Meetings
• Exhibitor Information
• Privacy Policy

NeurIPS 2024

Conference Dates: (In person) 9 December - 15 December, 2024

Homepage: https://neurips.cc/Conferences/2024/

Call For Papers 

Abstract submission deadline: May 15, 2024

Author notification: Sep 25, 2024

Camera-ready, poster, and video submission: Oct 30, 2024 AOE

Submit at: https://openreview.net/group?id=NeurIPS.cc/2024/Conference  

The site will start accepting submissions on Apr 22, 2024 

Subscribe to these and other dates on the 2024 dates page .

The Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS 2024) is an interdisciplinary conference that brings together researchers in machine learning, neuroscience, statistics, optimization, computer vision, natural language processing, life sciences, natural sciences, social sciences, and other adjacent fields. We invite submissions presenting new and original research on topics including but not limited to the following:

• Applications (e.g., vision, language, speech and audio, Creative AI)
• Deep learning (e.g., architectures, generative models, optimization for deep networks, foundation models, LLMs)
• Evaluation (e.g., methodology, meta studies, replicability and validity, human-in-the-loop)
• General machine learning (supervised, unsupervised, online, active, etc.)
• Infrastructure (e.g., libraries, improved implementation and scalability, distributed solutions)
• Machine learning for sciences (e.g. climate, health, life sciences, physics, social sciences)
• Neuroscience and cognitive science (e.g., neural coding, brain-computer interfaces)
• Optimization (e.g., convex and non-convex, stochastic, robust)
• Probabilistic methods (e.g., variational inference, causal inference, Gaussian processes)
• Reinforcement learning (e.g., decision and control, planning, hierarchical RL, robotics)
• Social and economic aspects of machine learning (e.g., fairness, interpretability, human-AI interaction, privacy, safety, strategic behavior)
• Theory (e.g., control theory, learning theory, algorithmic game theory)

Machine learning is a rapidly evolving field, and so we welcome interdisciplinary submissions that do not fit neatly into existing categories.

Authors are asked to confirm that their submissions accord with the NeurIPS code of conduct .

Formatting instructions:   All submissions must be in PDF format, and in a single PDF file include, in this order:

• The submitted paper
• Technical appendices that support the paper with additional proofs, derivations, or results 
• The NeurIPS paper checklist  

Other supplementary materials such as data and code can be uploaded as a ZIP file

The main text of a submitted paper is limited to nine content pages , including all figures and tables. Additional pages containing references don’t count as content pages. If your submission is accepted, you will be allowed an additional content page for the camera-ready version.

The main text and references may be followed by technical appendices, for which there is no page limit.

The maximum file size for a full submission, which includes technical appendices, is 50MB.

Authors are encouraged to submit a separate ZIP file that contains further supplementary material like data or source code, when applicable.

You must format your submission using the NeurIPS 2024 LaTeX style file which includes a “preprint” option for non-anonymous preprints posted online. Submissions that violate the NeurIPS style (e.g., by decreasing margins or font sizes) or page limits may be rejected without further review. Papers may be rejected without consideration of their merits if they fail to meet the submission requirements, as described in this document. 

Paper checklist: In order to improve the rigor and transparency of research submitted to and published at NeurIPS, authors are required to complete a paper checklist . The paper checklist is intended to help authors reflect on a wide variety of issues relating to responsible machine learning research, including reproducibility, transparency, research ethics, and societal impact. The checklist forms part of the paper submission, but does not count towards the page limit.

Please join the NeurIPS 2024 Checklist Assistant Study that will provide you with free verification of your checklist performed by an LLM here . Please see details in our  blog

Supplementary material: While all technical appendices should be included as part of the main paper submission PDF, authors may submit up to 100MB of supplementary material, such as data, or source code in a ZIP format. Supplementary material should be material created by the authors that directly supports the submission content. Like submissions, supplementary material must be anonymized. Looking at supplementary material is at the discretion of the reviewers.

We encourage authors to upload their code and data as part of their supplementary material in order to help reviewers assess the quality of the work. Check the policy as well as code submission guidelines and templates for further details.

Use of Large Language Models (LLMs): We welcome authors to use any tool that is suitable for preparing high-quality papers and research. However, we ask authors to keep in mind two important criteria. First, we expect papers to fully describe their methodology, and any tool that is important to that methodology, including the use of LLMs, should be described also. For example, authors should mention tools (including LLMs) that were used for data processing or filtering, visualization, facilitating or running experiments, and proving theorems. It may also be advisable to describe the use of LLMs in implementing the method (if this corresponds to an important, original, or non-standard component of the approach). Second, authors are responsible for the entire content of the paper, including all text and figures, so while authors are welcome to use any tool they wish for writing the paper, they must ensure that all text is correct and original.

Double-blind reviewing:   All submissions must be anonymized and may not contain any identifying information that may violate the double-blind reviewing policy.  This policy applies to any supplementary or linked material as well, including code.  If you are including links to any external material, it is your responsibility to guarantee anonymous browsing.  Please do not include acknowledgements at submission time. If you need to cite one of your own papers, you should do so with adequate anonymization to preserve double-blind reviewing.  For instance, write “In the previous work of Smith et al. [1]…” rather than “In our previous work [1]...”). If you need to cite one of your own papers that is in submission to NeurIPS and not available as a non-anonymous preprint, then include a copy of the cited anonymized submission in the supplementary material and write “Anonymous et al. [1] concurrently show...”). Any papers found to be violating this policy will be rejected.

OpenReview: We are using OpenReview to manage submissions. The reviews and author responses will not be public initially (but may be made public later, see below). As in previous years, submissions under review will be visible only to their assigned program committee. We will not be soliciting comments from the general public during the reviewing process. Anyone who plans to submit a paper as an author or a co-author will need to create (or update) their OpenReview profile by the full paper submission deadline. Your OpenReview profile can be edited by logging in and clicking on your name in https://openreview.net/ . This takes you to a URL "https://openreview.net/profile?id=~[Firstname]_[Lastname][n]" where the last part is your profile name, e.g., ~Wei_Zhang1. The OpenReview profiles must be up to date, with all publications by the authors, and their current affiliations. The easiest way to import publications is through DBLP but it is not required, see FAQ . Submissions without updated OpenReview profiles will be desk rejected. The information entered in the profile is critical for ensuring that conflicts of interest and reviewer matching are handled properly. Because of the rapid growth of NeurIPS, we request that all authors help with reviewing papers, if asked to do so. We need everyone’s help in maintaining the high scientific quality of NeurIPS.  

Please be aware that OpenReview has a moderation policy for newly created profiles: New profiles created without an institutional email will go through a moderation process that can take up to two weeks. New profiles created with an institutional email will be activated automatically.

Venue home page: https://openreview.net/group?id=NeurIPS.cc/2024/Conference

If you have any questions, please refer to the FAQ: https://openreview.net/faq

Abstract Submission: There is a mandatory abstract submission deadline on May 15, 2024, six days before full paper submissions are due. While it will be possible to edit the title and abstract until the full paper submission deadline, submissions with “placeholder” abstracts that are rewritten for the full submission risk being removed without consideration. This includes titles and abstracts that either provide little or no semantic information (e.g., "We provide a new semi-supervised learning method.") or describe a substantively different claimed contribution.  The author list cannot be changed after the abstract deadline. After that, authors may be reordered, but any additions or removals must be justified in writing and approved on a case-by-case basis by the program chairs only in exceptional circumstances. 

Ethics review: Reviewers and ACs may flag submissions for ethics review . Flagged submissions will be sent to an ethics review committee for comments. Comments from ethics reviewers will be considered by the primary reviewers and AC as part of their deliberation. They will also be visible to authors, who will have an opportunity to respond.  Ethics reviewers do not have the authority to reject papers, but in extreme cases papers may be rejected by the program chairs on ethical grounds, regardless of scientific quality or contribution.  

Preprints: The existence of non-anonymous preprints (on arXiv or other online repositories, personal websites, social media) will not result in rejection. If you choose to use the NeurIPS style for the preprint version, you must use the “preprint” option rather than the “final” option. Reviewers will be instructed not to actively look for such preprints, but encountering them will not constitute a conflict of interest. Authors may submit anonymized work to NeurIPS that is already available as a preprint (e.g., on arXiv) without citing it. Note that public versions of the submission should not say "Under review at NeurIPS" or similar.

Dual submissions: Submissions that are substantially similar to papers that the authors have previously published or submitted in parallel to other peer-reviewed venues with proceedings or journals may not be submitted to NeurIPS. Papers previously presented at workshops are permitted, so long as they did not appear in a conference proceedings (e.g., CVPRW proceedings), a journal or a book.  NeurIPS coordinates with other conferences to identify dual submissions.  The NeurIPS policy on dual submissions applies for the entire duration of the reviewing process.  Slicing contributions too thinly is discouraged.  The reviewing process will treat any other submission by an overlapping set of authors as prior work. If publishing one would render the other too incremental, both may be rejected.

Anti-collusion: NeurIPS does not tolerate any collusion whereby authors secretly cooperate with reviewers, ACs or SACs to obtain favorable reviews. 

Author responses:   Authors will have one week to view and respond to initial reviews. Author responses may not contain any identifying information that may violate the double-blind reviewing policy. Authors may not submit revisions of their paper or supplemental material, but may post their responses as a discussion in OpenReview. This is to reduce the burden on authors to have to revise their paper in a rush during the short rebuttal period.

After the initial response period, authors will be able to respond to any further reviewer/AC questions and comments by posting on the submission’s forum page. The program chairs reserve the right to solicit additional reviews after the initial author response period.  These reviews will become visible to the authors as they are added to OpenReview, and authors will have a chance to respond to them.

After the notification deadline, accepted and opted-in rejected papers will be made public and open for non-anonymous public commenting. Their anonymous reviews, meta-reviews, author responses and reviewer responses will also be made public. Authors of rejected papers will have two weeks after the notification deadline to opt in to make their deanonymized rejected papers public in OpenReview.  These papers are not counted as NeurIPS publications and will be shown as rejected in OpenReview.

Publication of accepted submissions:   Reviews, meta-reviews, and any discussion with the authors will be made public for accepted papers (but reviewer, area chair, and senior area chair identities will remain anonymous). Camera-ready papers will be due in advance of the conference. All camera-ready papers must include a funding disclosure . We strongly encourage accompanying code and data to be submitted with accepted papers when appropriate, as per the code submission policy . Authors will be allowed to make minor changes for a short period of time after the conference.

Contemporaneous Work: For the purpose of the reviewing process, papers that appeared online within two months of a submission will generally be considered "contemporaneous" in the sense that the submission will not be rejected on the basis of the comparison to contemporaneous work. Authors are still expected to cite and discuss contemporaneous work and perform empirical comparisons to the degree feasible. Any paper that influenced the submission is considered prior work and must be cited and discussed as such. Submissions that are very similar to contemporaneous work will undergo additional scrutiny to prevent cases of plagiarism and missing credit to prior work.

Plagiarism is prohibited by the NeurIPS Code of Conduct .

Other Tracks: Similarly to earlier years, we will host multiple tracks, such as datasets, competitions, tutorials as well as workshops, in addition to the main track for which this call for papers is intended. See the conference homepage for updates and calls for participation in these tracks. 

Experiments: As in past years, the program chairs will be measuring the quality and effectiveness of the review process via randomized controlled experiments. All experiments are independently reviewed and approved by an Institutional Review Board (IRB).

Financial Aid: Each paper may designate up to one (1) NeurIPS.cc account email address of a corresponding student author who confirms that they would need the support to attend the conference, and agrees to volunteer if they get selected. To be considered for Financial the student will also need to fill out the Financial Aid application when it becomes available.