GurjasEvidence & Policy Analytics
Research Library · Research design

Common-method bias cannot be diagnosed by one test

Start with how constructs, sources, timing and response processes could create shared measurement artefacts; then use analyses that address those mechanisms.

Type
Method note
Version
1.0
Published
5 August 2026
Reviewed
5 August 2026
Direct answer

A single-factor result, correlation threshold or marker test cannot certify that common-method bias is absent. Evaluate likely method mechanisms through design, measurement evidence and sensitivity of the focal relationships.

Use mechanism-based reasoning

  1. Identify shared method features. Consider same respondent, scale format, item context, source, setting and measurement occasion.
  2. Reduce avoidable overlap by design. Use temporal, psychological or source separation where it improves construct measurement without creating new bias.
  3. Protect item quality. Use clear wording, balanced context and validated measures rather than statistical correction for poor questions.
  4. Model plausible artefacts. Choose marker-variable, method-factor or multitrait–multimethod approaches only when their assumptions fit the data.
  5. Report sensitivity. Show whether substantive conclusions change under credible alternative specifications.

Why Harman's single-factor test is insufficient

An exploratory factor result does not directly estimate the bias in a specific structural relationship and may have low sensitivity to several method processes. The relevant question is whether plausible method variance materially changes the focal inference.

Worked use case

Employees report both leadership and engagement in one survey. Document item ordering, anonymity, source limitations and construct validity. Where feasible, add a time-separated or supervisor measure and compare the focal coefficient across defensible models instead of declaring the issue solved by one factor-analysis percentage.

Evidence boundary

Using one respondent does not automatically invalidate a study; some constructs are inherently self-reported. Conversely, a statistical control can introduce its own identification and interpretation problems. Conclusions must remain tied to the method mechanism actually addressed.

Primary sources

  1. Podsakoff, P. M., et al. (2003). Common method biases in behavioral research. DOI record.
  2. Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. DOI record.