Research and Assessment Methods
What you must know
Planners choose methods to answer a question, then defend the choice. Be fluent in: levels of measurement (nominal, ordinal, interval, ratio); descriptive versus inferential statistics; sampling designs (simple random, systematic, stratified, cluster, convenience) and their trade-offs; survey mechanics (question wording, pretesting, response and nonresponse bias, mode effects); demographic data sources (decennial census versus American Community Survey 1-year versus 5-year estimates); projection methods (linear/geometric extrapolation, share-of-growth, cohort-component, economic base and input-output models); and spatial analysis cautions such as the modifiable areal unit problem and the ecological fallacy.
Traps
Questions often swap validity (are you measuring the right thing?) with reliability (would you get the same answer again?). Watch for anyone using 1-year ACS estimates for a block group, comparing two estimates without checking whether margins of error overlap, or calling a correlation causal. Another classic: treating a single snapshot or a self-selected online poll as representative. Multiplier and economic base questions test whether you know that basic (export) employment drives nonbasic employment, and that multipliers are region- and sector-specific.
Rule of thumb
With ACS data, if the confidence intervals overlap, you cannot claim a difference. Small geography or small subgroup means use the 5-year file and report the margin of error alongside the estimate.
Worked example
Neighborhood A's ACS estimate of renter households is 42% with a margin of error of ±9; Neighborhood B is 48% ±8. The ranges (33–51 and 40–56) overlap heavily, so a statement that B is 'more renter-heavy than A' is not supportable. The correct finding is that the estimates are not statistically distinguishable.