Pairwise comparisons focus on differences between two means at a time. After an ANOVA shows that at least one group mean differs, pairwise tests examine each possible pair of groups (for example, group 1 vs group 2, group 1 vs group 3, and so on) to identify exactly which means differ. Because several comparisons are made, researchers usually apply adjustments (like Tukey or Bonferroni) to keep the overall chance of a false positive under control. This approach directly captures differences between two means at a time, which is what pairwise comparisons are designed to assess. The other ideas describe different analyses: comparing all group means at once is an omnibus or global test, assessing relationships between two variables refers to correlation or regression, and testing equality of variances concerns variances rather than means.

Bayesian inference rests on Bayes' theorem to update beliefs about hypotheses or model parameters as data are observed. This theorem links what we believed before seeing the data (the prior) with what the data tell us (the likelihood) to produce a revised belief (the posterior). In practice, you start with a prior distribution for the parameter or hypothesis, multiply by the likelihood of the observed data, and normalize to get the posterior distribution. Inference then comes from the posterior—summary measures like the posterior mean, credible intervals, or probabilities assigned to hypotheses. For example, if you’re unsure about a coin’s bias, you can start with a prior distribution for p (the probability of heads). After flipping the coin, you update to a posterior distribution that reflects both your prior and the observed outcomes. This posterior directly supports probability statements about p and decisions based on those probabilities. The other options don’t provide the updating rule that Bayesian methods rely on. They are associated with different foundational ideas (such as how sample means behave or convergence properties in frequentist contexts) and do not form the core mechanism for updating beliefs that Bayesian statistics use.

In MANOVA, the Hotelling-Lawley trace captures how much group separation there is across all discriminant functions by adding up the contributions from each function. Each discriminant function has an eigenvalue that measures how strongly that particular linear combination separates the groups. By summing these eigenvalues across all discriminant functions, the statistic provides a single overall measure of multivariate separation. Therefore, this implies the statistic is the sum of eigenvalues for each discriminant function. It’s not a product, an average, or a difference of eigenvalues—the trace inherently adds the contributions from all canonical variates to reflect total discriminant strength.

A one-tailed test is used when you have a directional hypothesis and you care about an effect in only one direction, so the entire alpha level is placed in one tail of the sampling distribution. This makes the test more powerful to detect an effect in the specified direction because the critical region is concentrated in that single tail. However, it cannot detect deviations in the opposite direction, which is why it’s not appropriate if you want to consider effects in both directions. This matches the description of testing a directional hypothesis with emphasis on one tail. The other options describe situations that don’t capture the directional focus: a non-directional hypothesis requires a two-tailed test, non-parametric vs parametric relates to distribution assumptions rather than tail direction, and one-tailed tests aren’t limited to variance comparisons.