Analysis of Variance (ANOVA) is the standard tool for asking whether three or more group means differ by more than we would expect from random variation alone. Instead of comparing groups two at a time, ANOVA compares the variability between group means to the variability within groups, and summarizes that comparison in a single F-statistic. If the between-group variability is large relative to the noise inside the groups, we have evidence that at least one population mean differs from the others.
This matters in engineering work because experiments rarely involve just two conditions. You might compare four suppliers' material strengths, three catalyst temperatures, or several machine settings. Running a separate t-test for every pair seems convenient, but each test carries its own chance of a false positive, and those chances accumulate quickly. Understanding that inflation of the Type I error rate is the central justification for ANOVA, and it is the idea this topic builds everything else on.
As you watch, keep four questions in mind: What exactly is being partitioned when we "analyze variance"? Why does the number of pairwise comparisons grow so fast, and what does that do to the overall error rate? What assumptions (normality, equal variances, independence) must hold before an F-test is trustworthy? And how does a one-way design, with a single factor, differ from a multi-factor design where interactions become possible?
A general-audience introduction to ANOVA from the Study Hall statistics series, likely covering what the method does, the logic of comparing between-group and within-group variation, and the role of the F-statistic. As an episode in a sequenced course, it should connect ANOVA to the hypothesis testing and t-test material you have already seen. Expect conceptual explanation and worked intuition rather than heavy algebra.
It is the best entry point for the "What is ANOVA?" and "When is ANOVA the appropriate choice?" subtopics, giving you the conceptual framing before the computational details arrive later in the module.
The first part of LearnChemE's treatment of single-factor (one-way) ANOVA, oriented toward engineering students and organized to follow standard textbooks. It should introduce the one-way setup, the hypotheses being tested, and how the total variation in the data is split into between-treatment and within-treatment pieces. Parts 2 and 3 are linked in the description if you want to continue into the full calculation.
LearnChemE produces engineering-focused statistics content, so the notation and examples match the style used in this course, and it directly addresses the one-way design that anchors the one-way versus multi-factor distinction.
A short animated explainer from the New England Journal of Medicine on the multiple comparisons problem. It reviews how running many statistical tests on the same data increases the chance of finding a "significant" result purely by luck, and discusses why that undermines confidence in reported findings. The framing comes from clinical research, but the arithmetic of error accumulation is identical in engineering studies.
At just over three minutes, it is the most direct and credible treatment of the "Why not just use multiple t-tests?" subtopic, which is the core justification for ANOVA in this module.
Together these videos build the conceptual foundation the module's outcomes depend on. The NEJM explainer supplies the error-inflation argument you need to justify ANOVA over repeated t-tests (MO1), while the Study Hall episode frames the method as a whole and should touch on the conditions and assumptions that make an F-test valid (MO2). The LearnChemE single-factor video introduces the variance partitioning and F-statistic computation that MO3 asks you to carry out by hand, and establishes the one-way baseline against which multi-factor main effects and interactions (MO4) are later contrasted. None of these videos resolves which specific pairs of means differ, so post-hoc procedures (MO5) remain for the lecture notes and problem sets, but the multiple comparisons video explains exactly why those procedures must adjust for the number of comparisons made.
These videos were chosen from YouTube on October 2, 2026. Each has captions; the player turns them on. Videos on YouTube can be changed or removed — if one no longer plays, tell your instructor. If a player does not appear, use its Watch on YouTube link.
Analysis of variance is a powerful tool for comparing means across three or more groups, but the F-test only behaves the way the theory says it should when certain conditions hold. Three assumptions underlie the standard one-way ANOVA: the dependent variable is approximately normally distributed within each group, the population variances are equal across groups (homogeneity of variance, or homoscedasticity), and the observations are independent of one another. Independence is largely a matter of how the data were collected; normality and equal variance can be checked after the fact with plots and formal tests.
This matters in an engineering context because you will routinely compare several process settings, materials, suppliers, or operators, and the conclusion you report depends on whether the F-statistic is trustworthy. ANOVA is fairly robust to moderate departures from normality, especially with balanced and reasonably large samples, but it is much less forgiving of unequal variances combined with unequal group sizes, and it is not robust at all to violations of independence. Knowing this tells you when to proceed, when to transform the data, and when to switch to an alternative such as Welch's ANOVA or the Kruskal-Wallis test.
The videos below approach assumption checking from three angles: the underlying logic and hand calculation of a variance-equality test, a software walkthrough in SPSS, and a combined normality and equal-variance check in R. As you watch, pay attention to what the null hypothesis of each assumption test actually says, what a small p-value for an assumption test implies about your ANOVA, and how the presenter decides whether a violation is serious enough to require a remedy. Also note the difference between a visual diagnostic (histograms, Q-Q plots, residual plots) and a formal test such as Levene's, Bartlett's, or Shapiro-Wilk.
This lecture focuses on testing the assumption of equal variances in the ANOVA setting, opening with the point that the F-ratio is relatively insensitive to modest departures from that assumption. Expect a discussion of why the equal-variance condition is built into the ANOVA model and a worked procedure for formally testing whether group variances can be treated as equal. The treatment is conceptual and computational rather than software-based.
It directly addresses the homogeneity of variance subtopic and, importantly, frames the issue in terms of robustness, which is exactly the judgment call students need to make about the consequences of violating assumptions.
A short, focused SPSS demonstration showing how to test homogeneity of variances in the context of a mixed ANOVA design, beginning with an explanation of what the assumption means before moving into the software steps. The video should show where the relevant output appears and how to read it. Its brevity makes it a quick reference rather than a full lecture.
Statorials produces concise, step-by-step statistical software tutorials, and this clip connects the abstract assumption to concrete output you would actually inspect before trusting an ANOVA result.
This tutorial covers both assumption checks for a one-way ANOVA in R: normality of the dependent variable within groups and equality of variances across groups. The description states the assumptions explicitly, so expect the relevant R functions and interpretation of their output. It bundles the two testable assumptions into a single short workflow.
It is the most complete fit for the "verifying assumptions before conducting ANOVA" subtopic because it treats normality and homoscedasticity together as a pre-analysis routine, and it does so in a reproducible scripting environment.
These three videos support MO2 most directly: between them they cover the logic of the equal-variance assumption, a formal test computed by hand, and software-based verification of both normality and homogeneity in SPSS and R, which is precisely the "visual methods and formal statistical tests" skill the outcome calls for. They also underpin MO3 and MO5 indirectly, since an F-statistic and any post-hoc comparisons are only interpretable when the underlying assumptions have been checked and any violations addressed. Use the remaining subtopics in the module notes to fill in independence of observations and the alternatives such as Welch's ANOVA, Kruskal-Wallis, and variance-stabilizing transformations, which the videos touch on only lightly.
These videos were chosen from YouTube on October 2, 2026. Each has captions; the player turns them on. Videos on YouTube can be changed or removed — if one no longer plays, tell your instructor. If a player does not appear, use its Watch on YouTube link.
When you need to compare the means of three or more groups — three alloy suppliers, four catalyst temperatures, five machine operators — running a separate t-test on every pair seems natural but is statistically costly. Each test carries its own chance of a false positive, and those chances accumulate quickly, so the overall Type I error rate climbs well above the 5 percent you think you are controlling. Analysis of Variance (ANOVA) solves this by asking a single question with a single test: is there evidence that at least one group mean differs from the others?
The logic behind that single test is variance partitioning. Every observation differs from the overall (grand) mean for two reasons: the group it belongs to may sit above or below the grand mean, and the observation itself varies within its group. ANOVA splits the total sum of squares into a between-group piece (SS_Between) and a within-group piece (SS_Within), divides each by its degrees of freedom to get mean squares, and forms the F-statistic as the ratio MS_Between / MS_Within. When group means truly differ, the numerator grows relative to the numerator's "noise" benchmark and F becomes large; when they do not, F hovers near 1.
As you watch the videos below, keep three things in your notes. First, track the identity SS_Total = SS_Between + SS_Within and the matching degrees of freedom bookkeeping — they are the structural backbone of the ANOVA summary table. Second, watch how each quantity is computed from group means, group sizes, and the grand mean, so the formulas stop looking arbitrary. Third, pay attention to how the computed F is compared against a critical value or p-value to reach a conclusion about the null hypothesis that all group means are equal.
A short, focused clip on how total variation in a one-way ANOVA is decomposed into among-group (between-group) variation plus within-group variation. Based on its description, it works through the partitioning identity that underlies the entire ANOVA table, showing where each sum of squares comes from relative to the grand mean and the individual group means.
At under four minutes it delivers the single most important conceptual idea of this topic — partitioning total variance — without computational clutter, making it an ideal first watch before you tackle worked numerical examples.
An example-driven lecture segment from a Probability and Statistics for Engineers lecture series that carries the sum-of-squares partition through an actual data set. Expect the calculation of SS_Total, SS_Between, and SS_Within from grouped data, with the numbers assembled in the structure an ANOVA summary table uses.
It comes from a lecture series aimed specifically at engineering probability and statistics students, so the notation and level of detail match this course, and the worked example directly supports the SS_Between and SS_Within subtopics.
A complete walkthrough of a one-way ANOVA problem from start to finish: recognizing when an ANOVA applies, stating the null and alternative hypotheses, computing the sums of squares and mean squares, forming the F-statistic, and reaching a decision. The longer runtime allows each step of the solution to be explained rather than just listed.
This video closes the loop between partitioning variance and drawing a statistical conclusion, giving you a model for interpreting the F-statistic for significance and for laying out a full solution the way you would on an exam.
Together these videos build the computational and conceptual core of MO3: the first establishes why and how total variance splits into between-group and within-group components, the second demonstrates those sums of squares on real numbers in an engineering-statistics context, and the third assembles mean squares into an F-statistic and organizes the results in a summary table. The full problem walkthrough also reinforces MO1 by showing how a single omnibus F-test answers the "are any means different?" question that multiple t-tests handle poorly, and it sets up MO5 by producing the significant-or-not decision that determines whether post-hoc pairwise testing is warranted. Assumption checking (MO2) and multi-factor designs (MO4) are addressed elsewhere in the module; these videos deliberately concentrate on the one-way variance-partitioning logic that everything else in ANOVA is built on.
These videos were chosen from YouTube on October 2, 2026. Each has captions; the player turns them on. Videos on YouTube can be changed or removed — if one no longer plays, tell your instructor. If a player does not appear, use its Watch on YouTube link.
One-way ANOVA (analysis of variance) is the standard procedure for comparing the means of three or more groups when those groups differ on a single factor — for example, three suppliers of a bearing, four curing temperatures, or several machine settings. Rather than running many separate t-tests, which would inflate the chance of at least one false positive, ANOVA asks a single question: is there evidence that at least one group mean differs from the others?
The logic rests on partitioning variability. Total variation in the data is split into variation between group means and variation within groups, and the ratio of those two mean squares forms the F-statistic. If the between-group variation is large relative to the natural scatter inside the groups, F is large and the p-value is small, which gives evidence against the null hypothesis that all population means are equal. Along the way you need to check the working assumptions — independent observations, approximately normal errors, and roughly equal variances across groups — because the F-test's validity depends on them.
As you watch these videos, track four things: how the hypotheses are stated, how the sums of squares and degrees of freedom are computed and laid out in the ANOVA table, how the F-statistic is compared to a critical value or p-value, and what happens after a significant result. Remember that a significant F tells you that the means are not all equal but not which pairs differ — that is the job of post-hoc comparisons such as Tukey's procedure. Keep a blank ANOVA table beside you and try to fill it in as each example proceeds.
A compact walkthrough of the ANOVA procedure used to test a claim about three or more population means. Based on the title and description, expect the hypotheses to be stated, the between- and within-group variability to be summarized in an ANOVA table, and the F-statistic to be compared against a critical value or p-value to reach a conclusion.
It covers the definition, purpose, and mechanics of one-way ANOVA in under nine minutes, making it a good first or review pass over the core test procedure.
Part 1 of a short series from LearnChemE on single-factor (one-way) ANOVA, framed around an engineering/chemical engineering context. It is likely to introduce the design structure — one factor with several levels — and begin building the variance partition and summary table, with later parts continuing the calculation and interpretation.
LearnChemE is a well-established engineering education channel, so the notation and examples will feel familiar to EGN3443 students; the series format lets you follow the full calculation step by step.
A brief worked example testing whether three different golf clubs produce different mean results. Expect the data to be organized by group, sums of squares and the F-statistic to be computed, and a decision to be stated about whether the group means differ.
Short, concrete examples are the fastest way to see how the formulas apply to real data, and this one directly illustrates conducting and interpreting a one-way ANOVA with three groups.
Together these videos build the computational and interpretive core of the module. The Stat Brat and Jason Liu examples model stating hypotheses, computing the F-statistic from between- and within-group variation, and reading the result — directly supporting MO3 and the reasoning behind MO1, since each shows a single omnibus test replacing multiple pairwise comparisons. The LearnChemE series grounds the single-factor design in an engineering setting and reinforces the structure of the ANOVA summary table. Assumption checking (MO2), multi-factor main and interaction effects (MO4), and the selection and interpretation of post-hoc procedures (MO5) are only touched on here, so pair this viewing with your course notes and assigned readings on those points.
These videos were chosen from YouTube on October 2, 2026. Each has captions; the player turns them on. Videos on YouTube can be changed or removed — if one no longer plays, tell your instructor. If a player does not appear, use its Watch on YouTube link.
So far in this module, ANOVA has been used to compare means across the levels of a single factor. Many engineering experiments, however, vary two or more factors at once — for example, testing how both catalyst type and reaction temperature affect yield, or how both material and coating thickness affect fatigue life. Multi-factor ANOVA (most commonly the two-way ANOVA) lets you test all of these influences in one analysis instead of running separate one-way ANOVAs for each factor.
Two new ideas drive this topic. A main effect is the influence of one factor on the response, averaged across the levels of the other factor. An interaction effect occurs when the effect of one factor depends on the level of the other — for instance, when a higher temperature helps one catalyst but hurts another. Because interactions change how main effects should be described, you always check the interaction term first. The two-way ANOVA table reflects this structure: total variation is partitioned into a row-factor component, a column-factor component, an interaction component, and error, with an F-statistic and p-value for each.
As you watch, pay attention to three things: how the sums of squares and degrees of freedom are split among the factors, the interaction, and error; how each F-ratio is formed and tested; and how interaction plots (parallel versus non-parallel lines) are read. Also notice the practical question of when a multi-factor design is worth using — replication, balanced designs, and the efficiency of studying several factors in a single experiment.
A worked engineering-oriented treatment of the two-way ANOVA from LearnChemE, produced by chemical engineering faculty at the University of Colorado Boulder. In roughly fifteen minutes it should walk through the structure of a two-factor analysis — setting up the data layout, partitioning the sums of squares across the two factors, the interaction, and error, and building the resulting ANOVA table with F-statistics and p-values. Expect the calculations to be tied to a concrete example rather than presented abstractly.
LearnChemE is a well-established engineering education channel, so the notation and example context match this course better than generic statistics tutorials, and the length allows full coverage of the two-way ANOVA structure and result interpretation.
Part 1 of a short lecture series introducing the background and logic of two-way (factorial) ANOVA. It is positioned as the conceptual setup: why researchers move from one factor to two, how factorial designs are organized into cells by crossing factor levels, and what main effects and interaction effects mean before any computation begins.
At under eight minutes, it gives a clean conceptual bridge from one-way to multi-factor ANOVA and defines main versus interaction effects — exactly the framing subtopics of this page — before you take on the arithmetic.
Part 3 of the same lecture series, focused on carrying out and reading the two-way ANOVA analysis. It should cover the output side: the separate F-tests for each factor and for the interaction, the associated p-values, and how those results are translated into statements about the data.
It targets the interpretation subtopic directly, showing how to read multiple F-statistics from one table — the skill most students find hardest when moving beyond one-way ANOVA.
These videos focus on the multi-factor portion of the module. The LearnChemE lecture reinforces the variance-partitioning logic from MO3 by extending it to two factors plus an interaction term, showing how each F-ratio is built from its own mean square over the error mean square. The two short PSYC 300 lectures support MO4 most directly: Part 1 defines main effects and interaction effects and explains when a factorial design is appropriate, while Part 3 demonstrates reading the F-statistics and p-values for each variance component and deciding what the pattern of significance means. Together they prepare you to recognize when an interaction makes a main effect misleading, which is also the situation in which follow-up comparisons from MO5 must be interpreted with care.
These videos were chosen from YouTube on October 2, 2026. Each has captions; the player turns them on. Videos on YouTube can be changed or removed — if one no longer plays, tell your instructor. If a player does not appear, use its Watch on YouTube link.
A significant F-test in ANOVA tells you something is going on — at least one group mean differs from the others — but it does not tell you which groups differ, or by how much. Post-hoc ("after the fact") testing fills that gap by making pairwise comparisons among group means while keeping the overall chance of a false positive under control. Without that control, running every possible pairwise t-test would inflate the familywise error rate: with just five groups there are ten comparisons, and the probability of at least one spurious "significant" result climbs well above the 5% you intended.
In this course, post-hoc analysis is where ANOVA becomes useful to an engineer. If you compare the fatigue life of four alloys, the yield of three process settings, or the response time of several controller designs, the decision you report is usually "A and C differ, B is indistinguishable from both" — a statement that only a properly adjusted set of comparisons can support. You will also need to communicate that result cleanly, which is why reporting conventions (F, degrees of freedom, p-value, effect size, and the specific pairwise findings) are part of this topic.
As you watch, focus on three things. First, the logic of why error rates inflate with multiple comparisons and what "familywise" actually means. Second, the mechanics of a specific correction — Bonferroni is the simplest and the one most often shown — including how it changes the critical value or the adjusted p-value. Third, how to read pairwise output: which pairs are flagged, what the confidence intervals say about the size of the difference, and what you can and cannot conclude when a comparison is non-significant.
A short lecture on the multiple testing problem: why performing many hypothesis tests on the same data increases the chance of at least one false positive, and two corrections that appear frequently in published literature (typically Bonferroni and a false discovery rate approach). The description indicates it is framed around how these corrections are actually used when reporting results.
It gives the conceptual foundation for the "Why Post-Hoc Tests Are Necessary" and "Controlling Familywise Error Rate" subtopics in under ten minutes, and the biostatistics framing connects the correction directly to how results get reported.
Part of Stanford's Statistical Learning series taught by Trevor Hastie, this segment focuses specifically on the Bonferroni method as a procedure for controlling the familywise error rate across a set of tests. Expect a precise statement of what FWER is, how the Bonferroni adjustment is derived, and the trade-off it makes in statistical power.
It is the most rigorous treatment of familywise error control among the three, from a leading authority on statistical learning, and it gives you the formal justification behind the adjusted significance levels you will apply in post-hoc comparisons.
A hands-on demonstration of running a post-hoc analysis in Excel 2016 after rejecting the null hypothesis of a one-way ANOVA. It walks through the workflow from the ANOVA output into the pairwise comparisons and shows how to read which group means are flagged as different.
This is the applied counterpart to the two conceptual videos, directly supporting "Common Post-Hoc Testing Procedures" and "Interpreting Post-Hoc Output" using software you can open immediately.
Together these videos carry the error-inflation argument from motivation through to execution. The multiple testing lecture and the Stanford Bonferroni segment reinforce MO1 by making explicit how repeated comparisons raise the Type I error rate and how a correction restores control of it. The Excel demonstration then addresses MO5 directly: it shows a post-hoc procedure applied after a significant ANOVA and how to read the pairwise output to identify which specific means differ. None of the three covers assumption checking, the F-statistic computation, or factorial designs (MO2-MO4); rely on the module's earlier materials and your textbook for those, and use these videos for the interpretation and reporting stage of the analysis.
These videos were chosen from YouTube on October 2, 2026. Each has captions; the player turns them on. Videos on YouTube can be changed or removed — if one no longer plays, tell your instructor. If a player does not appear, use its Watch on YouTube link.