STA4321, Introduction to Mathematical Statistics I, is the calculus-based probability course that underpins the statistics major. It is where a student stops applying statistical procedures and starts deriving them — building probability from axioms, defining random variables properly, and establishing the distributional results that later inference depends on.
The University of West Florida lists the content directly: "probability, conditional probability, distributions of random variables, distribution of functions of random variables, limiting distributions, multivariate probability distributions." Florida State University's version covers "distribution of random variables, conditional probability and independence, multivariate distributions, sampling distributions, Bayes' rule, counting problems, and expectations."
The distinction from an introductory statistics course is total. STA2023 teaches students to run a t-test and interpret the output. This course asks why the t-statistic has the distribution it has — which requires knowing what a random variable is, how functions of random variables are distributed, and what happens in the limit as the sample grows. It is a mathematics course about probability, taught with calculus, and it is normally the first course in which statistics students encounter proof.
Note what the title conceals. Despite "Mathematical Statistics" in the name, the first course is overwhelmingly probability. Statistical inference — estimation, hypothesis testing, confidence intervals developed rigorously — is the second course, STA4322. The University of Florida's own syllabus title for this number is, candidly, Introduction to Probability. Students expecting to do statistics in the first term are frequently surprised.
STA4321 is offered at approximately 9 Florida institutions, all universities, and carries 3 credits with roughly 45 contact hours. It is a 4000-level course normally taken in the junior year and is required in statistics majors and in most mathematics, actuarial science and data science programmes.
The difference matters because multivariate distributions require multiple integration. Joint densities, marginal and conditional distributions, and transformations of several random variables are all worked with double and triple integrals, and a student who has not taken Calculus III meets multiple integration inside this course rather than arriving with it. Institutions requiring only Calculus II typically teach the multivariable calculus as needed, which works but compresses.
Take Calculus III first if your programme allows it. It is required by the statistics and mathematics majors in any case, it makes the second half of this course substantially easier, and it removes a gap that no transcript records.
Title variation: Introduction to Mathematical Statistics I (the statewide title, UWF and USF), Introduction to Mathematical Statistics (FSU), and Introduction to Probability (UF's syllabus title for STA4321/5325). Same course.
STA4321 is a junior-year course and the first half of a two-course sequence, continuing into STA4322, Introduction to Mathematical Statistics II, which covers estimation, hypothesis testing and linear models rigorously. It follows the calculus sequence and precedes the applied upper-division statistics courses — regression, experimental design, time series, categorical data — all of which assume the distributional results derived here. It is required in statistics majors and in most actuarial science, mathematics and data science degree tracks.
Worth repeating because it is the most common misexpectation on this number: the first term is almost entirely probability theory. Statistical inference — the confidence intervals and hypothesis tests students associate with the word "statistics" — arrives in the second course. Students taking only the first course to satisfy a requirement should understand that they will leave with probability rather than inference. Conversely, students who disliked STA2023 sometimes find this course far more congenial, because it explains the machinery that the introductory course asks them to accept.
See the calculus divergence above. Beyond calculus, the practical requirement is comfort with integration — this course integrates constantly, including improper integrals, integration by parts, and multiple integrals over non-rectangular regions. A student whose calculus is shaky will find that, rather than the probability, is the obstacle. Some institutions also expect or recommend an introductory statistics course first, though it is not conceptually necessary. Linear algebra is not usually required for the first course but is valuable for the second and for the applied courses that follow.
Three credits, approximately 45 contact hours, no laboratory. Assessment is problem-set driven with examinations requiring derivations rather than recall. Expect eight to twelve hours a week outside class.
The University of West Florida's version carries an institutional designation worth noting: it "meets College-Level Computation Skills Requirement." That is institution-specific and does not transfer with the course.
The method that works in this course is specific: derive results rather than memorising them. There are perhaps a dozen standard distributions and a student who has derived each one's mean, variance and moment generating function will reconstruct anything needed, while a student who has memorised a table will be lost as soon as a problem is phrased unfamiliarly. Simulation is a genuinely useful check — if you derive a distribution and cannot simulate data matching it, one of the two is wrong, and finding out which is instructive.
This course maps almost exactly onto SOA Exam P, the first actuarial examination. Students on an actuarial track should sit it while the material is fresh, ideally in the term after completing the course. Actuarial credentialing is examination-based rather than degree-based, examinations may be taken as an undergraduate, and passing one before graduation materially improves internship prospects. Florida's insurance industry makes this an unusually practical route in this state.
STA4321 is a 4000-level SCNS course: the number is recognised statewide, but upper-division credit is not covered by the A.A. transfer guarantee and applicability inside the major is the receiving department's decision. The course is not available before transfer from a Florida College System A.A. — the lower-division path is the calculus sequence, and STA2023 where the programme requires it, both of which transfer cleanly as common prerequisites.
The transfer caution here is the calculus-level difference: a course taken after only Calculus II covers multivariate distributions with less mathematical background than one taken after Calculus III, and that does not appear on a transcript. Where a receiving programme's second course (STA4322) assumes the fuller treatment, the gap surfaces there.
The STA prefix is statistics. Relevant numbers: STA2023 (introductory statistics — the lower-division common prerequisite, and a different kind of course entirely); STA3032 and STA3123-range (applied statistics for engineers and scientists, calculus-based but not proof-based — not a substitute for STA4321); STA4321 and STA4322 (this sequence); STA4210-range regression; STA4211-range design of experiments; STA4853-range time series; and STA5325-type graduate versions, frequently co-taught with the undergraduate section — UF's syllabus is issued as STA4321/5325. Adjacent prefixes: MAP and MAS for mathematics, QMB for business statistics (a different course, not substitutable in a statistics major), and PSY, SYA and CCJ prefixes for discipline-specific statistics courses, which likewise do not substitute.
Probability is the mathematical foundation of machine learning, which gives this course an unusually direct relationship to AI — and makes the case for learning it properly rather than generating around it.
The material is the foundation of the technology. Bayes' theorem underlies Bayesian inference and naive Bayes classification. Maximum likelihood, which the sequel develops, is how most models are fitted. Probability distributions are what generative models learn. The central limit theorem underlies the bootstrap and much of the statistical machinery around model evaluation. Expectation and variance are what bias-variance decomposition decomposes. A student who understands this course reads machine learning literature as mathematics; a student who does not reads it as vocabulary.
Where AI helps a student. Language models are useful for explaining a concept a second way — conditional expectation and the Jacobian transformation method are two that commonly need it — for checking algebra in a derivation, for generating practice problems, and for writing R or Python code to simulate a distribution and verify a derived result. That last use is genuinely valuable: simulation as a check on derivation is good practice, and the coding barrier is real for students who have not programmed.
Where AI fails. Models make errors in multi-step derivations — dropped constants, wrong limits of integration, missing Jacobian factors — and probability derivations are exactly the multi-step form where this compounds. They confuse similar distributions, particularly the gamma parameterisations (rate versus scale), negative binomial conventions, and the several ways beta is written. They produce plausible but wrong independence assumptions, which is the single most common substantive error in applied probability and the one this course exists to prevent. And they reproduce the classic probability fallacies — base rate neglect, confusing P(A|B) with P(B|A), the prosecutor's fallacy — because those are abundant in text.
That last point deserves emphasis: the errors AI makes in probability are the same errors humans make, for the same reason — both learn from a corpus in which the intuitive-but-wrong answer is common. A model is not a check on your intuition here; it shares it.
The standard. A derivation you cannot reproduce from a blank page is not knowledge you have, and this course is assessed by closed-book derivation. The practical test after using any tool is to close it and work the problem again. And when a probability answer feels obviously right, that is precisely when to check it — the field's history is a catalogue of confident intuitions that were wrong.
Academic integrity. Policies vary; a common arrangement permits AI for explanation and for simulation code while prohibiting it on graded derivations, which are the assessed skill. Read the syllabus and ask when it is not explicit.
Generated September 5, 2026 · Updated September 5, 2026