Statistical framework (parameter vs statistic)
What is a distribution?
Sampling Distribution
Confidence Intervals
Hypothesis Testing
For the exam, be prepared to use the following questions to reason about unfamiliar examples. You may receive graphs, tables, numerical summaries, or descriptions of repeated sampling. You will not need to read or write R code or calculate exact p-values. Support your conclusions with relevant evidence, use the critical value sheet when appropriate, and interpret quantities in context.
What is a sampling distribution, and how does it differ
from the distribution of individual observations?
Explain what one observation represents in a histogram of individual
fish lengths versus a histogram of sample mean lengths. If we take 1,000
samples of 25 fish each, how many means are plotted? Distinguish taking
more observations per sample from generating more repeated samples.
Identify which quantities remain fixed and which change across repeated
samples.
What do standard deviation and standard error describe,
and how does sample size affect each?
Explain the roles of \(\sigma\), \(s\), and \(s/\sqrt{n}\). Predict how increasing sample
size or population variability changes the spread of the sampling
distribution. Does doubling the sample size halve the standard error?
Distinguish a range containing most individual observations from a range
containing most sample means.
When is a normal model appropriate for a sampling
distribution, and what does standardization accomplish?
Explain what the Central Limit Theorem says becomes approximately normal
and how population shape and sample size affect the approximation.
Contrast sampling from normal and strongly skewed populations. Does
increasing sample size make the original population normal? Does
subtracting a mean and dividing by a standard deviation make any
distribution normal?
What does a confidence interval estimate, and what does
its confidence level mean?
Interpret the point estimate, margin of error, and endpoints in context.
Explain what 95% coverage means over repeated samples, what changes
across those samples, and why a particular interval either contains the
fixed population parameter or does not. Does a 95% interval contain 95%
of individual observations? Must exactly 95 of 100 intervals contain the
true mean?
What determines the location and width of a confidence
interval?
Explain how the sample mean, sample standard deviation, sample size, and
confidence level affect the interval. Distinguish changing the standard
error from changing the critical value. Why do we use a t-distribution
when estimating an unknown population standard deviation, how do degrees
of freedom enter, and how does the t-distribution compare with the
standard normal? Explain the trade-off between confidence and
precision.
How do a confidence interval, a test statistic, and a
p-value give equivalent decisions about the same null
hypothesis?
For a two-sided test of a population mean, connect whether \(\mu_0\) is inside the confidence interval,
whether \(|t|\) exceeds \(C\), and whether \(p\) is below \(\alpha\). Relate the confidence level to
the Type I error rate. Explain how the same data can lead to rejection
using an 80% confidence interval but not a 90% interval. Which
quantities change when we change the confidence level, and which stay
the same?
How can we compare evidence against null hypotheses when
the data or the hypothesized mean changes?
Explain the numerator and denominator of \(t=(\overline{x}-\mu_0)/(s/\sqrt{n})\). For
the same sample, what changes when we test a different \(\mu_0\)? For two different samples, why
might the mean farther from \(\mu_0\)
give a smaller \(|t|\)? Explain why
comparing \(|t|\) directly can be
misleading when degrees of freedom differ, and how p-values help account
for the different null distributions.
What does a p-value tell us, and how can critical values
put bounds on it?
For a two-sided test, identify the relevant tail areas and explain why
larger \(|t|\) means smaller \(p\) for a fixed null distribution. Use the
correct degrees of freedom to locate \(|t|\) between critical values, then
translate those comparisons into bounds on \(p\). Use those bounds to compare evidence
or make decisions at specified \(\alpha\) values. Explain why a p-value is
not the probability that the null hypothesis is true, why it does not
measure effect size, and why failing to reject does not establish that
the null is true.
How to compute p-values directly
R