Variance/SD of a sample ($s$) divides by n-1—this is the unbiased estimate of the population’s variance, since otherwise dividing by n underestimates the variance
Sampling distribution: distribution of a random sample statistic
https://stats.stackexchange.com/questions/116889/why-dont-we-use-the-unbiased-sample-variance-to-calculate-the-standard-error
When to use z (normal) vs t distribution/interval/test?
Relationship: confidence intervals, significance level (alpha), and critical values
means: t test, since you often don’t know the population SD. Or z test if n is large enough.
proportions: z test
difference of proportions, always independent samples (no paired): z test on just the paired differences, and variance is the sum of variances.
difference of 2 means, paired: t test on just the paired differences. (example with t interval). variance is not sum of variances since they’re not independent samples.
difference of 2 means, independent samples: bootstrapped resampling on computer, or t-test/z-test (if n>30) of the difference, where variance is sum of variances. df=. (example)
regression analysis: relationship on continuous data (chi square for relationship of discrete data) [TODO be more precise here]
chi square tests (video on the differences)
chi square test for goodness of fit: test if 1 sample of a multinomial matches expected distribution. dof=n-1
chi square test of independence / for relationship (e.g. with 2-way tables), same as test of homogeneity: check whether two categorical variables are related, i.e. if they’re independent, which is the same as whether they’re from the same distribution. dof=(r-1)(c-1)
F-test / ANOVA: compare means (continuous) among more than 2 groups.