Data Analysis for PolicyMaking in Georgia
Module 5 · Building Your Policy Finding › 5.5 Module 5 checkpoint
Subsection 5.5

Module 5 checkpoint

~3 min

Module 5 Checkpoint

You have completed Module 5: Building Your Policy Finding. These four questions cover the anatomy of a policy claim, comparison design, writing up findings, and sanity-checking numbers.

Module 5 Recap — Building a Defensible Policy Finding Finding Quality Checklist ✓ Research question is specific and answerable ✓ Finding includes one key number with explicit comparison ✓ Hedged: "consistent with" not "proves" ✓ States what the data does NOT show ✓ Comparison groups are equivalent on relevant factors ✓ Denominators consistent across periods ✓ Sanity-checked against certified/official benchmarks ✓ Exclusions documented in methodology note Common Errors to Avoid ✗ Research question too broad ("Is GA fair?") ✗ Finding without a number or comparison ✗ Overclaiming intent from effects alone ✗ Ignoring the Unknown/missing rows ✗ Comparing incomparable groups (Fulton vs. Echols) ✗ Mixed denominators across years ✗ No sanity check against known benchmarks ✗ Silently dropping records without documenting why
Diagram 5.5 · Module 5 recap. Finding quality checklist and common errors.

Which of the following is the best example of a well-formed research question for a Policy Data Brief?

Correct! A well-formed research question is specific, answerable with available data, and relevant to a policy decision. Option B names the outcome (mail ballot rejection rate), the group (counties by Black voter share), and the time period (2020 general). Not quite. Option B is specific, answerable with available data, and names the comparison. Options A and C are too broad; D is a recommendation, not a question.

An analyst writes: 'Black voter mail ballots were rejected at a higher rate, proving intentional discrimination.' What is wrong with this statement?

Correct! This statement overclaims. Disparate effects — a higher rejection rate — document unequal outcomes but do not by themselves prove intent. After Callais, the distinction matters even more. The finding should say 'consistent with disparate impact' rather than 'proves.' Not quite. The statement overclaims. Effects ≠ intent. After Callais, this distinction is especially important for redistricting, but it applies to all voting rights analysis.

You calculate Georgia's 2020 general election statewide turnout as 42% of active registered voters. What should you do?

Correct! 42% is far below the known benchmark of approximately 66% for Georgia's 2020 general election. This is a red flag that your denominator or data filter is wrong — perhaps you included inactive voters or used a wrong election-year flag. Not quite. The benchmark for Georgia 2020 general turnout is ~66%. A 42% result signals a likely error — probably including inactive voters in the denominator, or using the wrong election-year flag.

You are comparing mail ballot rejection rates across Georgia counties in 2018 vs. 2022. Which of the following would make this comparison invalid?

Correct! Using different denominators (returned vs. issued) in different years makes the comparison invalid. Ballots issued includes ballots that were never returned, so the denominators measure different things. Option B describes a valid, consistent approach. Not quite. Inconsistent denominators (returned in one year, issued in another) make the comparison invalid — they measure different things. Always use the same denominator formula across the periods you are comparing.

Action: Complete all four questions, then slide to finish Module 5 and move to Module 6.