STA4173: Biostatistics

Course Description

STA4173 Biostatistics is the second statistics course for students in the biological and health sciences — the methods used to design studies of living systems and to draw defensible conclusions from data that is variable, incomplete and expensive to collect.

The course is offered at approximately four Florida institutions, including Florida State University, the University of Central Florida and the University of West Florida.

The University of West Florida places it in the College of Science and Engineering, Department of Mathematics and Statistics at 3 semester hours, requires STA 2023, and describes "a second course in statistics for students in the Biological Sciences," covering "analysis of variance, regression analysis, nonparametric statistics, contingency tables." ⚠ It meets the College-Level Computation Skills Requirement — which is the mathematics half of Florida's Gordon Rule, and is discussed in Special Information.

Florida State University titles it Fundamentals of Biostatistics at 3 credits, requires a previous upper-division course in statistics or instructor permission, and describes "the statistical methods used to design and analyze studies of the occurrence of disease in human populations."

⚠⚠ Two divergences here, and both matter. FSU's prerequisite is an upper-division statistics course where UWF's is the lower-division introductory one — so the course sits at a different point in the sequence. And FSU's description is explicitly epidemiological (disease occurrence in human populations) where UWF's is general biological. See Special Information.

What distinguishes biostatistics from general statistics. The mathematics is the same; the problems are not. Biological and health data has features that drive the whole toolkit: outcomes are frequently binary (survived or did not, diseased or not) rather than continuous; observations are often correlated because measurements are repeated on the same subject or clustered within families, clinics or litters; sample sizes are limited by cost and ethics rather than by convenience; and censoring is normal — a study ends while some subjects have not yet had the event of interest, and discarding them would bias everything.

⚠ The most consequential idea in the course is that a p-value is not what most people think it is. It is the probability of data at least this extreme if the null hypothesis were true — not the probability that the null is true, not the probability the result is real, and not a measure of how large or important an effect is. With a large enough sample, a trivially small effect produces a small p-value, which is why the course insists on confidence intervals and effect sizes alongside tests. The American Statistical Association issued a formal statement on this in 2016 because the misinterpretation is so widespread in published research.

Study design is taught as seriously as analysis, and this is the right emphasis. No analysis rescues a badly designed study. The distinctions the course builds on: observational versus experimental; cohort, case-control and cross-sectional designs and what each can establish; randomisation and what it actually buys — balance on unmeasured confounders, which no statistical adjustment can achieve; blinding; and power analysis, which determines before the study whether it can detect an effect worth detecting. ⚠ An underpowered study is not a small study that found nothing; it is a study that could not have found anything, and running one is arguably an ethical failure when subjects were exposed to risk.

The methods themselves follow from the data types. ANOVA for comparing several groups, with the multiple comparisons problem it exists to solve; regression — linear for continuous outcomes, logistic for binary ones, which is the workhorse of health research; contingency table analysis for categorical data; and nonparametric methods for when distributional assumptions fail, which in small biological samples is frequently.

⚠ Confounding is the concept students find hardest to hold onto, and it is the one that matters most. An association between exposure and outcome can be produced entirely by a third variable related to both. Statistical adjustment handles confounders you measured; it does nothing about the ones you did not — which is precisely why randomised designs are valued and why observational findings are held to a different standard.

Learning Outcomes

Required Outcomes

• Distinguish populations from samples, and parameters from estimates, and explain what inference actually claims.
• Select and produce appropriate descriptive statistics and graphical displays for a given data type.
• Explain sampling distributions and the central limit theorem, and why they license inference.
• Construct and correctly interpret confidence intervals, including what the confidence level does and does not mean.
• Conduct hypothesis tests and state conclusions in terms of the biological or clinical question rather than in bare statistical language.
• ⚠ Explain what a p-value is and enumerate the common misinterpretations, including confusing statistical with practical significance.
• Distinguish and manage Type I and Type II error; explain statistical power and compute a sample size for a stated effect.
• Apply t-procedures for one and two samples, paired and independent, and check their assumptions.
• Apply analysis of variance — one-way and, where covered, two-way — and interpret the F-test.
• Explain the multiple comparisons problem and apply an appropriate correction or post-hoc procedure.
• Apply linear regression — simple and multiple — and interpret coefficients, fit and residual diagnostics.
• Apply logistic regression for binary outcomes and interpret odds ratios.
• Apply contingency table analysis — chi-square, Fisher's exact test — and interpret the result.
• Compute and interpret measures of association and effect — risk ratio, odds ratio, risk difference, number needed to treat.
• Apply nonparametric procedures — Wilcoxon, Mann-Whitney, Kruskal-Wallis — and explain when they are preferable.
• Explain correlation and why it does not establish causation, and identify confounding in a described study.
• Distinguish the principal study designs and state what each can and cannot support.
• Explain randomisation and blinding and what each controls for.
• Explain diagnostic test performance — sensitivity, specificity, predictive values — and why predictive value depends on prevalence.
• Use statistical software to carry out analyses and interpret its output.
• Read a published study's statistical section critically and identify its limitations.

Optional Outcomes

• Apply survival analysis — Kaplan-Meier estimation, the log-rank test, and censoring.
• Explain Cox proportional hazards regression.
• Apply methods for repeated measures and clustered data.
• Explain meta-analysis and evidence synthesis.
• Explain Bayesian approaches at an introductory level.
• Apply experimental design beyond the basics — blocking, factorial designs, crossover trials.
• Explain missing data mechanisms and basic handling.
• Explain epidemiologic measures — incidence, prevalence, standardised rates.
• Explain clinical trial design, phases and interim monitoring.
• Explain reproducibility — p-hacking, publication bias, pre-registration, the replication crisis.

Major Topics

Required Topics

• Data types and descriptive statistics for biological data.
• Probability and distributions — normal, binomial, Poisson.
• Sampling distributions and estimation; confidence intervals.
• Hypothesis testing; p-values and their interpretation.
• Power and sample size.
• t-procedures and assumption checking.
• Analysis of variance; multiple comparisons.
• Linear regression and correlation; diagnostics.
• Logistic regression and odds ratios.
• Categorical data analysis — chi-square, Fisher's exact.
• Measures of association and effect.
• Nonparametric methods.
• Study design — observational and experimental; randomisation, blinding.
• Confounding and bias.
• Diagnostic testing and predictive values.
• Statistical software and output interpretation.
• Critical reading of published statistics.

Optional Topics

• Survival analysis.
• Repeated measures and clustered data.
• Meta-analysis.
• Bayesian methods.
• Advanced experimental design.
• Missing data.
• Epidemiologic measures and rate standardisation.
• Clinical trials.
• Reproducibility and research integrity.

Resources & Tools

• Standard textbooks: Rosner, Fundamentals of Biostatistics — the most widely adopted, and the source of FSU's course title; Pagano and Gauvreau, Principles of Biostatistics; Daniel and Cross, Biostatistics: A Foundation for Analysis in the Health Sciences; Whitlock and Schluter, The Analysis of Biological Data — ⚠ the best fit where the course is aimed at biologists rather than health professions, and unusually readable.
• Software: R (free, and the de facto standard in biostatistics research) with RStudio; SAS, which remains the standard in pharmaceutical and regulatory work; SPSS, common in health sciences teaching; Stata, common in epidemiology and health economics; and JMP or Minitab where licensed. ⚠ Check which your section uses before the term starts and install it early — software setup problems block assignments rather than merely delaying them.
• ⚠ If the course does not specify, learn R. It is free, it is what published biostatistics is done in, and it is the single most transferable skill available in this course. Swirl (an in-console tutorial package) and the free R for Data Science book are good starting points.
• Free supplementary instruction: StatQuest (YouTube) — genuinely excellent short explanations of exactly this course's topics and widely recommended by instructors; Seeing Theory (Brown University) for interactive visualisation of sampling distributions and inference; and OpenIntro Statistics, free and well written.
• Reference and standards: the American Statistical Association's 2016 statement on p-values — short, free, and worth reading in full; the CONSORT statement for reporting randomised trials and STROBE for observational studies, both free checklists that make good frameworks for critically reading a paper.
• Real data for practice: the CDC's NHANES and BRFSS, the SEER cancer registry, and ClinicalTrials.gov — all free, all genuinely messy, and far more instructive than textbook datasets.

Career Pathways

⚠ Statistical competence is one of the highest-return skills a biology or health sciences student can acquire, and it is chronically under-supplied. Most of the destinations below want it as a component rather than as the whole job — but the ones that want it as the whole job pay well.

• Biostatisticians and statisticians (SOC 15-2041) — ⚠ normally requires a master's degree, and the field has persistent unmet demand. This course is the entry point.
• Clinical research coordinators and associates (SOC 11-9121, 19-4021) — ⚠ accessible at bachelor's level, and Florida's academic medical centres and contract research organisations hire steadily.
• Epidemiologists (SOC 19-1041) — requires an MPH or higher; ⚠ biostatistics is half the training.
• Data analysts in health care (SOC 15-2051) — quality measurement, population health, operations analytics at Florida's large systems.
• Pharmaceutical and medical device industry (SOC 15-2041, 19-1042) — clinical trial statistics, regulatory submissions, pharmacovigilance. SAS competence is specifically valued.
• Biological and wildlife research (SOC 19-1020s, 19-4021) — ⚠ ecology and fisheries work is statistically demanding, and Florida's FWC and university programmes need it.
• Public health agencies (SOC 19-1041, 13-1111) — the Florida Department of Health and county health departments.
• Medicine and the health professions — ⚠ evidence-based practice requires reading statistics, and this course is where that is learned properly. It is also good preparation for the quantitative reasoning sections of professional admission tests.
• Graduate study — ⚠ a biostatistics course on the transcript materially strengthens applications to biology, ecology, public health, medicine and health sciences programmes, and several require it outright.

Special Information

⚠⚠ Gordon Rule — UWF's "Computation Skills" label is the Gordon Rule mathematics requirement

UWF records that this course meets the College-Level Computation Skills Requirement. That phrase is UWF's designation for the mathematics half of Florida's Gordon Rule (State Board of Education Rule 6A-10.030), whose counterpart — the College-Level Communication Skills Requirement — is the writing half.

⚠⚠ What follows from that is a grade threshold, and it catches students out:

• A Gordon Rule course must normally be completed with a grade of C or higher to count. A C− does not satisfy it at most institutions. Passing is not sufficient.
• ⚠ The designation is made by the institution, not by the course number. A course carrying it at UWF may not at another Florida institution, and vice versa.
• Designation normally travels within the Florida public system when the sending institution designated it, but confirm rather than assume — particularly from private or out-of-state institutions.

Practical advice: if you are taking this course partly to satisfy the requirement, treat C as the floor rather than as a pass, and confirm the designation with an advisor before you rely on it.

⚠⚠ The prerequisite differs in LEVEL — and so does the course's position

	UWF	FSU
Title	Biostatistics (= statewide)	Fundamentals of Biostatistics
Prerequisite	STA 2023 — the lower-division introductory statistics course	a previous UPPER-DIVISION statistics course, or instructor permission
Framing	"a second course in statistics for students in the Biological Sciences"	methods to design and analyse studies of the occurrence of disease in human populations
Emphasis	general biological statistics — ANOVA, regression, nonparametrics, contingency tables	epidemiological

Two divergences on one number, and they compound. FSU places the course later in the sequence and aims it at population health; UWF places it directly after the introductory course and aims it at biology generally.

⚠ Practical consequences:

• You may not be able to take FSU's version as a visiting or transient student with only STA2023 — the prerequisite is enforced and it is upper-division.
• Transferring INTO an FSU-style programme with the UWF version: the credit articulates on the number, but you may hold biostatistics without the upper-division statistics course that programme's later work assumes.
• If you need epidemiological methods specifically — for public health, for an MPH application — check which version you are taking; a general biological statistics course covers the machinery but not the study designs and rate measures.
• ⚠ This is the fifth sequence-position divergence documented in this repository, after MUL4400, SPN3400, CDA3101 and FIN3244. The number articulates cleanly in every case, which is exactly what conceals the mismatch. Keep your syllabus.

⚠ A further note: FIU's catalog record for this number is marked "Inactivated per SCNS review" and describes a health-services statistics course. Do not plan around it.

Prerequisites — and what the listed one actually buys

UWF requires STA2023, the standard introductory statistics course, which is widely taught at Florida College System institutions — ⚠ so satisfy it before transferring if you can. It is cheaper and it is a course whose articulation is reliable.

⚠ The prerequisite is genuine content. This course assumes you already know what a sampling distribution is, what a confidence interval means and how a hypothesis test is structured. It starts from there and goes to multivariable methods in fifteen weeks, which is not survivable if the foundation is shaky.

⚠ The unlisted preparation that matters: comfort with software and with algebra. The mathematics is not advanced — no calculus is required at this level — but there is a lot of notation, and students who are uneasy with symbolic expressions find the notation, rather than the concepts, is what defeats them.

The course sits in the junior or senior year of biology, health sciences, nursing, environmental science and related programmes, and is frequently required for graduate-school-bound students.

Course format and workload

3 credits, 45 contact hours — lecture, three hours per week. ⚠ No C or L suffix, so computer work is assigned rather than timetabled and comes out of your own week.

Expect 8–10 hours per week outside class. Assessment typically includes examinations, problem sets involving software output, and often a data analysis project on a real dataset.

⚠ The material is cumulative to an unusual degree. Confidence intervals depend on sampling distributions; ANOVA depends on hypothesis testing; regression depends on both. Falling two weeks behind is very difficult to recover from, and it is the single most common cause of failure in this course. Go to office hours in week three, not week ten.

⚠ What students find hardest

• Interpreting rather than computing. Software does the arithmetic. The examinations ask what the output means, whether the assumptions hold, and what the finding supports — which is a different and harder skill.
• The p-value. Almost every student arrives with the wrong definition, and unlearning it takes deliberate effort. ⚠ It is worth the effort: p-value misinterpretation is endemic in published research, and being able to spot it is a real professional advantage.
• Assumption checking. Every procedure has conditions, and applying one whose assumptions fail produces a confident wrong answer rather than an error message.
• Choosing the right procedure. Students learn tests individually and then cannot select among them. ⚠ Build a decision chart — outcome type, number of groups, paired or independent, assumptions met or not — and it becomes tractable.
• Statistical anxiety. Common among biology and health students, and worth naming: this course is passable by most people who do the problems consistently, and the difficulty is usually accumulation rather than ability. Tutoring centres are free and included in fees already paid.

Articulation and transfer

STA4173 is a 4000-level upper-division course, not offered at Florida College System institutions, and taken after transfer. The number is used consistently, so SCNS articulation is clean — subject to the prerequisite-level and emphasis divergences above.

⚠ Prefix note. STA is statistics; MAC/MAP/MAS mathematics; QMB business quantitative methods; PHC public health, where biostatistics is sometimes numbered instead. ⚠ A business statistics course (QMB) does not normally substitute for biostatistics and vice versa — the methods overlap but the applications, and therefore the examples and the emphasis, differ substantially. Related: STA2023 (the prerequisite), STA4234 (regression), STA4202 (design of experiments).

AI Integration

⚠ Statistics is the discipline AI tools are simultaneously most useful for and most dangerous in, and the reason is that they produce plausible output for a question you may have asked wrongly.

Where they genuinely help:

• Explaining a concept a second way. Sampling distributions, what a confidence interval means, why the multiple comparisons problem exists — models are good at this, and a second framing frequently unsticks a student.
• Writing and debugging R, SAS or SPSS code. ⚠ Genuinely time-saving and now normal professional practice. The syntax is a barrier that has nothing to do with understanding statistics, and removing it is legitimate.
• Interpreting software output — what each column of a regression table is.
• Generating practice problems.

⚠⚠ Where they fail, and the failures are specific:

• Choosing the wrong procedure. Ask a model to analyse a described dataset and it will readily suggest a test whose assumptions the data violate — paired data analysed as independent, clustered data analysed as if observations were independent. Both produce confident, wrong answers with no warning.
• Multi-step calculations. Models make arithmetic errors and present them with the same confidence as correct results.
• Reproducing the p-value misinterpretation. ⚠ Because the wrong definition is far more common in the training text than the right one, models frequently state it — "the probability the result is due to chance", "the probability the null is true". This course exists partly to correct exactly that error, and the tool will reintroduce it.
• Fabricated statistics and citations in a written report.
• Overstating conclusions — asserting causation from observational data, which is the discipline's central prohibition.

Worth knowing as subject matter: machine learning and classical statistics answer different questions, and the distinction is worth being clear about. Machine learning optimises prediction; biostatistics optimises inference and estimation with quantified uncertainty. ⚠ A model that predicts an outcome accurately may tell you nothing about whether a treatment causes it — and in health research the causal question is usually the one that matters, which is why randomised trials remain the standard despite the volume of available observational data.

Academic integrity. Read the syllabus; policies vary, and many statistics courses now permit AI assistance for code while prohibiting it for interpretation — which is a sensible line, since the interpretation is the assessed skill. ⚠ In a data analysis project, fabricating or altering results is a fabrication offence rather than plagiarism, and it is treated far more seriously — correctly, since the entire discipline exists to make claims about data trustworthy.