Track 15 · Applied Statistics for Professionals · Lesson 10 of 12
Survey research
Critique questionnaire items, summarise Likert data as ordinal, assess nonresponse and data quality, and weight survey results to the population.
40 minIntermediate8 graded exercisesMembers
Every year HR teams present an engagement score to the leadership team, and every year someone asks whether it can be trusted. Whether it can depends less on the arithmetic than on how the questions were worded, who answered, and who did not.
A survey estimate can go wrong at every stage. A useful checklist follows the total survey error idea:
- Coverage: is everyone in the population on the list you invite (the sampling frame)?
- Sampling: chance variation from asking some people rather than all.
- Nonresponse: do the people who answer differ from those who do not?
- Measurement: do the questions measure what you intend, and do people answer honestly and carefully?
A large number of responses fixes only the second. The others are fixed by design.
Included with a Formula Quest membership
It has 8 graded exercises with feedback, hints and worked solutions. After it you will be able to:
- Spot leading, double-barrelled, loaded and vague survey questions
- Summarise Likert items with distributions, medians and top-2-box percentages computed on those who answered
- Screen responses for straight-lining and speeding and report their effect
- Compute post-stratification weights, apply them, and explain what weighting cannot correct