MaizePrepSTATS 250
Exam 2 · Thu, Nov 5, 5:45 PM ·  

Effect Size, Power, Intervals and Study Design

Estimated effect size and power; confidence intervals and what confidence levels mean; comparing two groups; confounding variables and when an association is causal. Covers Lec 08–14, Exam Prep 04–06, Labs 04–06 (the practice exam also recommends reviewing Lec 06–07).

Sections7
Practice questions92
Reading200min

Where the points go

Share by topic

Patterns that repeat every time

  1. 01

    Tables of t, p and d^\hat d

    Several studies of one question with t, p and d^\hat d (some cells missing). Decide which had the larger sample, which replicates best (closest d^\hat d), which mean was furthest from the null, and which effect size is plausible.

  2. 02

    Interpretation or not?

    Classify statements as a correct interpretation of the interval, a correct interpretation of the level, or not appropriate. Watch for probability language, individuals instead of a mean, and capturing the sample statistic.

  3. 03

    Intervals and tests agree

    Pick the plausible interval from a two-sided p-value and the sign of t, say whether a null value is inside, read an interval off a plot of p-values, or choose the 90% versus 95% interval from the same data.

  4. 04

    Simulation plots of intervals

    Rank plots by confidence level (count misses) and by sample size (compare widths at one level). Lecture said this question would come back.

  5. 05

    Power comparisons with a reason

    Two researchers, one difference (sample size, direction of HaH_a, or no true effect): say who has higher power and explain in 1–2 sentences.

  6. 06

    Part II: confounding and causation

    Case Study 2 style: define μ1−μ2\mu_1-\mu_2, fix a wrong interpretation of a confidence level, show a confounder's two legs with numbers, and explain why non-significant stratified results are not evidence of equal means.

Exam format and strategy

Thu, Nov 5, 5:45 PM

Format

  • Thursday, November 5, 2026 · 5:45–7:45 pm · 120 minutes for both parts
  • Part I: Concepts and Skills, mostly multiple choice plus a few boxed answers and short explanations; about 80 minutes suggested
  • Part II: Written Statistical Interpretation of new R output; about 40 minutes suggested (the Case Study 2 lab practiced 5 prompts: about 10 minutes reading, 30 writing)
  • The parts are scored separately: Part I counts toward Category 1, Part II toward Category 2 (Written Statistical Interpretation 02 is 8% of the course grade)

Allowed

  • Official help card provided (now with interval, two-mean and effect-size formulas)
  • No calculator: the syllabus, the case-study lab slides, the practice-exam information page and lecture all say so. The printed exam cover says calculators are not required but allowed; confirm with the instructors. Numerical answers may be left as fractions
  • Closed notes and closed book
  • Photo ID shown when you submit

Strategy

  • Budget about 80 minutes for Part I and about 40 for Part II.
  • Before reading two-group output, predict from the plots: sign of t, size of p, sign of d^\hat d, and whether the interval contains 0.
  • Separate the two questions every time: how strong is the evidence (p-value) and how big is the effect (d^\hat d)?
  • For any interval statement, check three things: parameter not statistic, mean not individuals, confidence not probability.
  • For a confounder, write both legs with numbers: conditional proportions for the link to XX, a center and the IQR location for the link to YY.
  • Causal claims need random assignment; generalizing needs random sampling. Say which one the design has.

Sections

In the order they are tested

Formula sheet

On the card: one proportion
π^±z∗π^(1−π^)n,z=π^−π0π0(1−π0)/n,d^=π^−π0π0(1−π0)=zn\hat\pi\pm z^*\sqrt{\tfrac{\hat\pi(1-\hat\pi)}{n}},\quad z=\frac{\hat\pi-\pi_0}{\sqrt{\pi_0(1-\pi_0)/n}},\quad \hat d=\frac{\hat\pi-\pi_0}{\sqrt{\pi_0(1-\pi_0)}}=\frac{z}{\sqrt n}
On the card: one mean
μ^±t∗σ^n,t=μ^−μ0σ^/n,d^=μ^−μ0σ^=tn\hat\mu\pm t^*\tfrac{\hat\sigma}{\sqrt n},\quad t=\frac{\hat\mu-\mu_0}{\hat\sigma/\sqrt n},\quad \hat d=\frac{\hat\mu-\mu_0}{\hat\sigma}=\frac{t}{\sqrt n}
On the card: difference in two means
(μ^1−μ^2)±t∗σ^12n1+σ^22n2,d^=tn1n2/(n1+n2)(\hat\mu_1-\hat\mu_2)\pm t^*\sqrt{\tfrac{\hat\sigma_1^2}{n_1}+\tfrac{\hat\sigma_2^2}{n_2}},\quad \hat d=\frac{t}{\sqrt{n_1n_2/(n_1+n_2)}}
The card also prints the p-value evidence scale and the small/moderate/large effect-size scale.
Know yourself: multipliers
z∗: 1.645 (90%), 1.96 (95%), 2.576 (99%);t∗: df=n−1 or min⁡(n1−1,n2−1)z^*:\ 1.645\ (90\%),\ 1.96\ (95\%),\ 2.576\ (99\%);\qquad t^*:\ df=n-1\ \text{or}\ \min(n_1-1,n_2-1)
Not printed.
Know yourself: margin of error
ME=multiplier×SE=12 widthME=\text{multiplier}\times SE=\tfrac12\,\text{width}
Know yourself: interval–test link
θ0∈CIC  ⟺  ptwo-sided>1−C\theta_0\in CI_C\iff p_{\text{two-sided}}>1-C
Know yourself: power
power≈#strong+#very strong#simulated tests\text{power}\approx\frac{\#\text{strong}+\#\text{very strong}}{\#\text{simulated tests}}
Rises with n and with the true effect size.

MaizePrep is an independent study resource. It is not affiliated with, sponsored by, or endorsed by the University of Michigan. Course names and numbers are used only to describe which class the material covers. Every practice question is original, written in the style of the course.