Common method bias
Also known as: Common method variance
Spurious correlation that arises just because two things were measured the same way.
What it means
Common method bias is systematic variance attributable to the measurement method rather than to the constructs being measured, which can inflate or deflate the observed relationship between variables. It is most acute when a predictor and an outcome are gathered from the same source by the same instrument at the same time, as in a single self-report survey. Shared response styles, item priming, and the respondent's wish for consistency manufacture correlations that have nothing to do with the underlying concepts. Remedies include collecting variables from different sources or at different times, varying response formats, and statistical diagnostics, though no single fix is decisive.
Why it happens
The worry rests on how people answer surveys. A single respondent, moving through one questionnaire, carries a stable way of using the scale: some habitually agree, some cluster at the extremes, some drift toward the midpoint. A passing mood colours consecutive answers alike. Once an early item raises a theme, later items are read in its light, and most people prefer their own answers to hang together rather than contradict each other. Add a wish to look competent or consistent, and any two ratings come to share a slice of variance that belongs to the act of rating, not to the traits rated. None of this requires the constructs to be truly related; the shared method supplies a correlation on its own.
Is it really a problem?
Campbell and Fiske framed the issue in 1959, and Podsakoff and colleagues' 2003 review made checking for it almost a condition of publication. But the strong claim, that any two measures sharing a method are automatically inflated, drew a sharp rebuttal. Spector called it an urban legend, arguing method variance is not a uniform contaminant smeared across every self-report correlation; its size depends on the particular constructs and items, and for many pairs it is small. Podsakoff's later work agrees the effect is real and sometimes large, but concedes it does not touch all measures equally. The practical reading is that same-source data are not worthless, yet a correlation from one survey deserves less confidence than the same number drawn from separated sources.
Detecting it, and a test that fails
The most cited diagnostic, Harman's single-factor test, loads every item into one unrotated factor and treats a dominant first factor as evidence of bias. It is popular because it is easy and almost never fails a study, which is precisely the problem. Howard and colleagues showed in 2024 that the test cannot even separate research designs known to differ in their exposure to bias: cross-sectional and multi-wave, single- and multi-source studies score alike. Sturdier tools add a marker variable, a construct chosen to be unrelated to the others, and read its residual correlation as a method estimate, as in Lindell and Whitney's technique and its confirmatory-factor descendants. Even these hinge on choosing a defensible marker; a poor choice misleads more than it corrects.
Designing it out
Because after-the-fact statistics are weak, the durable fixes live in the design. Separate the measurements: gather predictor and outcome at different times, from different people, or through different formats, so no single response occasion links them. Take the outcome from an independent record, a supervisor, a register, a log, rather than from the same respondent. Protect anonymity, vary item wording and scale anchors to break response-style carryover, and counterbalance item order so no construct always primes the next. These steps cost more than a one-sitting survey, which is why they are often skipped. But they attack the shared method itself, whereas a post-hoc correction can only guess at damage already baked into the data.
Examples
Job satisfaction and performance correlate partly because one person rated both on the same questionnaire in one sitting.
Customers who call a brand innovative also call it trustworthy — partly because one person ticked both boxes on the same seven-point scale in the same minute.
Students who report studying hard also report good grades on a single questionnaire; ask the registrar for their actual marks instead and the relationship shrinks.
A wellness survey finds self-reported sleep quality tracks self-reported productivity; because one person rated both in one sitting, a wish to look consistent props up the link before any real effect is counted.
An end-of-term teaching evaluation shows the rated clarity of a lecturer tracking the rated usefulness of the course; a student's momentary good mood lifts every rating at once, manufacturing part of the association.
First described in Campbell & Fiske (1959); Podsakoff et al. (2003).
Key references
- Howard, M. C., Boudreaux, M., & Oglesby, M. (2024). Can Harman's single-factor test reliably distinguish between research designs? Not in published management studies. European Journal of Work and Organizational Psychology, 33(6), 790-804. doi.org/10.1080/1359432X.2024.2393462
- Podsakoff, P. M., MacKenzie, S. B., & Podsakoff, N. P. (2012). Sources of method bias in social science research and recommendations on how to control it. Annual Review of Psychology, 63, 539-569. doi.org/10.1146/annurev-psych-120710-100452
- Richardson, H. A., Simmering, M. J., & Sturman, M. C. (2009). A tale of three perspectives: Examining post hoc statistical techniques for detection and correction of common method variance. Organizational Research Methods, 12(4), 762-800. doi.org/10.1177/1094428109332834
- Spector, P. E. (2006). Method variance in organizational research: Truth or urban legend? Organizational Research Methods, 9(2), 221-232. doi.org/10.1177/1094428105284955
- Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879-903. doi.org/10.1037/0021-9010.88.5.879
- Lindell, M. K., & Whitney, D. J. (2001). Accounting for common method variance in cross-sectional research designs. Journal of Applied Psychology, 86(1), 114-121. doi.org/10.1037/0021-9010.86.1.114