Data collection methods
Data collection methods
This domain tests how OB researchers and HR analysts gather data and whether the resulting measures are trustworthy. Questions often hinge on reliability versus validity, sampling, and bias.
Methods and their trade-offs
- Surveys: efficient for attitudes; use clear anchors (Likert), reverse-coded items, and pilot testing. Threats include common method variance, acquiescence, central tendency, and low response rate.
- Interviews: structured, semi-structured, or unstructured; rich meaning and probing, but interviewer bias, social desirability, and coding subjectivity.
- Observation: naturalistic, participant, or structured behavior checklists; captures behavior rather than self-report but risks reactivity and observer drift.
- Archival data: HR records, absenteeism, turnover, sales; inexpensive and relatively objective, but records may be incomplete or collected for another purpose.
- Physiological and experience-sampling methods: heart rate, cortisol, daily diaries; reduce recall bias but are intrusive and costly.
- Focus groups: synergy and group norms; conformity can suppress dissent.
Sampling and measurement
Probability sampling (simple random, stratified, cluster) supports generalization; nonprobability sampling (convenience, snowball, quota) is common but limits inference. Reliability is consistency: test–retest, inter-rater, and internal consistency (Cronbach's alpha). Validity is whether you measure the intended construct: content, criterion (concurrent, predictive), and construct (convergent, discriminant). Scales are nominal, ordinal, interval, or ratio.
Worked example
A five-item job-satisfaction scale yields alpha = .84 and correlates .38 with later voluntary turnover. Alpha supports internal consistency, not validity. To defend validity, show content review by subject experts and a negative correlation with turnover; to reduce common-method bias, collect satisfaction and turnover from different sources and times.
Traps
Reliability is necessary but not sufficient for validity; a scale can be consistently wrong. Alpha above .70 is a convention, not a law. Do not treat ordinal Likert data as interval without justification. A 20% response rate can bias results even in a large sample.