After several articles about selection artefacts and arithmetic, this one concerns a question every firm owner has an opinion about and which eight decades of research has not settled.
Key Takeaway
A meta-analysis of 312 samples with a combined N of 54,417 estimated the mean true correlation between overall job satisfaction and job performance at .30[1][2]. The same paper's qualitative review is organised around 7 models, and reports that "research has not provided conclusive confirmation or disconfirmation of any model"[1]. The earlier meta-analysis it corrected had reported .17, and had been "often cited as evidence that satisfaction and performance are virtually unrelated"[3].
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade A that a positive association exists. Three meta-analyses, hundreds of studies, and over fifty thousand participants in the largest.
Grade B for the magnitude at .30, which is a corrected estimate from one meta-analysis and which supersedes but does not eliminate an earlier figure of .17.
Grade D for any causal claim in either direction, on the review's own statement that no model has been confirmed or disconfirmed.
Our position: the association is well established and the direction is not, which makes almost every practical recommendation built on this literature weaker than it sounds.
A Note On Method
Everything here is verified to August 2026.
We obtained the 2001 abstract verbatim from four independent sources, which agree word for word[1][2][4][5], and passages describing the field from a peer-reviewed article[3].
We did not obtain any of the papers in full. We do not know what the seven models are beyond three we found named elsewhere, and we report that gap rather than guessing.
We found a numerical conflict between sources on the combined sample size and report it.
We could not obtain the conclusion of the spuriousness paper, whose title poses the most important remaining question, and say so where it arises.
All conversions of the correlation are ours.
This article reviews organisational research. It is not employment, human resources or management advice.
The Holy Grail
How the field regards the question.
A peer-reviewed article states that the relationship "has been the topic of hundreds of studies, including three prominent meta-analyses" and "has been described as the 'Holy Grail' of industrial and organizational psychology"[3].
The three meta-analyses are identified as Petty, McGee and Cavender (1984), Iaffaldano and Muchinsky (1985), and Judge and colleagues (2001)[3].
Two observations, ours.
The Holy Grail label is telling in a specific way: it describes something searched for at length and not found. That is a franker self-assessment than most fields offer.
And the question matters commercially because the two candidate answers imply opposite spending decisions. If satisfaction drives performance, investing in conditions is an investment. If performance drives satisfaction, it is a consequence you are paying for twice.
The 1985 Answer
The finding that shaped a generation of thinking.
Iaffaldano and Muchinsky (1985) reported a corrected correlation of .17, from k = 74 studies with N = 12,192[3].
A peer-reviewed article records the consequence: their "findings have been very influential and have often been cited as evidence that satisfaction and performance are virtually unrelated"[3].
Two observations, ours.
Virtually unrelated is a stronger claim than .17 supports, and the gap between the number and its reception is the story of this article.
On our own arithmetic, a correlation of .17 leaves about 97 percent of performance variance unexplained, which does sound like nothing. Whether that framing is the right one is the subject of a later section.
What That Did To The Field
The consequence, in the 2001 authors' own words.
"Research devoted to testing these models waned following 2 meta-analyses of the job satisfaction-job performance relationship."[1]
And their assessment of why: "Because of limitations in these prior analyses and the misinterpretation of their findings, a new meta-analysis was conducted."[1]
Three observations, ours.
Waned means work stopped. A meta-analytic result did not merely update a belief; it closed a research programme for roughly fifteen years.
The 2001 authors name two separate problems: limitations in the analyses and misinterpretation of their findings. Those are different failures, one technical and one in reception, and both are on the record.
And this is the mirror image of the thirty-fourth article in this series, where a debunking closed a question for three decades until a proof reopened it. Two articles running, the damage was done by a result that stopped people looking.
The 2001 Review
The paper.
Judge, Thoresen, Bono and Patton published The job satisfaction-job performance relationship: A qualitative and quantitative review in Psychological Bulletin, 127(3), 376–407, in May 2001, DOI 10.1037/0033-2909.127.3.376[1].
It is two papers in one: a qualitative review organised around seven models, and a new meta-analysis[1].
One observation, ours. That structure is unusual and it is the right one for this question. A correlation without a causal model is not actionable, and a causal model without a correlation is not evidence. Most business writing on this topic supplies neither and asserts both.
The Number
The quantitative result.
The new meta-analysis covered 312 samples with a combined N of 54,417, and "the mean true correlation between overall job satisfaction and job performance was estimated to be .30"[2].
Two observations, ours.
That is more than four times the number of samples in the 1985 analysis and more than four times the participants.
And the phrase is mean true correlation, meaning corrected for measurement unreliability and range restriction. This series encountered the same class of correction in the sixth article, where a 2022 revision of hiring validity found the corrections themselves had been applied too generously. We flag that a corrected correlation is an estimate of what would be observed under ideal measurement, not a number you would see in your own data.
What Point Three Zero Means
The first reading, and it is the one that sounds discouraging. Our arithmetic.
A correlation of .30 corresponds to about 9 percent of shared variance, leaving 91 percent of performance variation unaccounted for.
The 1985 figure of .17 corresponds to about 2.9 percent, leaving 97 percent.
Two observations.
On that framing, the two results look similar: both leave the overwhelming majority of performance unexplained, and the entire difference between the famous debunking and its correction is about six percentage points of variance.
Which is why variance explained is a misleading way to read a correlation for practical purposes, and the next section gives a better one.
A More Useful Way To Read It
The framing we would use instead. Our own arithmetic, using standard formulas and assuming bivariate normality; no source states these figures.
The question a firm owner actually has is not what share of variance is explained. It is: if I compare my most satisfied staff with my least satisfied, how different is their performance?
Sorting into quartiles by satisfaction, at a correlation of .30, the top quartile averages about 0.38 standard deviations above the mean on performance and the bottom quartile about 0.38 below. A gap of roughly 0.76 standard deviations.
At the 1985 figure of .17, the gap is about 0.43.
Two observations.
Three quarters of a standard deviation between the top and bottom quartiles is not a trivial difference, and it is the same number that looked like nothing when expressed as nine percent of variance.
And this is the clearest example in the series of how the same statistic supports two opposite rhetorical readings, both technically correct. We report both and prefer the second because it answers the question people are actually asking.
Placed Against This Series
Where .30 sits among the correlations this publication has reported.
Surface acting and impaired wellbeing, .39 to .48. Deep acting and customer satisfaction, .37. Job satisfaction and performance, .30. Surface acting and job attitudes, -.24 to -.40. Surface acting and performance, -.20 to -.05.
Two observations, ours.
It sits squarely in the middle of the organisational correlations we have covered, and is larger than the relationship between surface acting and performance from the twenty-second article.
And notice the comparison that emerges: on these figures, how people feel about their job relates to their performance more strongly than how much emotional labour they perform does. We offer that as an observation about two separate literatures rather than a finding, since no study compared them.
The Seven Models
The qualitative half, and our gap on it.
The review is "organized around 7 models that characterize past research on the relationship"[1].
We did not obtain the paper and therefore do not know what all seven are. We could have inferred a plausible list, and have deliberately not done so.
Two observations, ours.
Seven is a lot, and it tells you something on its own: the relationship admits many structurally different explanations that all fit a positive correlation.
And that is why a correlation of .30 is not, by itself, a reason to do anything. Seven different theories predict it, and they imply different actions.
Three Of Them, Named
What we could establish, from a citing source.
A citing paper describes three positions with attributions: that "attitudes such as job satisfaction cause corresponding behaviours such as job performance (Strauss, 1968)"; that "behaviours or job performance can lead to attitudes or job satisfaction (Lawler & Porter, 1967)"; and that "this relationship is spurious, which means that both are affected by a third or more unmeasured variables"[6].
Three observations, ours.
The reverse causal model is the one business writing never mentions. If doing well makes people satisfied rather than the other way round, then satisfaction surveys are measuring an output, and treating them as a lever is a category error.
The reverse model is also intuitive on reflection. People generally enjoy work they are good at, are praised for, and are rewarded for, and all three follow performance rather than preceding it.
And the spurious model is the most awkward, because a common cause would produce the correlation while making interventions on either variable useless for the other.
None Confirmed, None Disconfirmed
The review's own verdict, quoted, because it is the single most important sentence for anyone acting on this literature.
"Although some models have received more support than have others, research has not provided conclusive confirmation or disconfirmation of any model, partly because of a lack of assimilation and integration in the literature."[1]
Three observations, ours.
That is a leading journal's comprehensive review reporting that the central question is open after decades of work on the field's Holy Grail.
The diagnosis is worth noting: a lack of assimilation and integration. The authors are saying the problem is not insufficient studies but insufficiently connected ones, which is a different failure and less easily fixed by doing more research.
And some models have received more support than have others is the closest thing to a positive statement, and it is not one we can pass on, because we do not know which.
The Spuriousness Question
A later paper asking the hardest version, and our failure to answer it.
Bowling published Is the job satisfaction–job performance relationship spurious? A meta-analytic examination in a peer-reviewed vocational behaviour journal in 2007[3][6].
We could not obtain its conclusion. We have its title, its framing of the literature, and nothing else.
Two observations, ours.
This is a real gap and it sits precisely where a reader would want an answer. A meta-analysis specifically testing whether the relationship survives controlling for common causes is the most decision-relevant document we identified, and we do not have its result.
Related work exists on the dispositional side, including Judge, Locke, Durham and Kluger (1998), Dispositional effects on job and life satisfaction: The role of core evaluations, in the Journal of Applied Psychology[7], which we also did not obtain. If stable personal dispositions drive both satisfaction and performance, that is the spurious model with a named mechanism.
A Conflict In The Numbers
A discrepancy we found, reported rather than resolved.
The 2001 abstract, in four independent reproductions, gives the combined sample as N = 54,417[1][2][4].
A peer-reviewed article citing it gives N = 54,471[3].
Two observations, ours.
The two figures are a digit transposition apart, and the abstract's version is far better attested, so we use 54,417 throughout and flag the other.
It changes nothing substantive. We record it because this is the tenth time in this series that a famous paper's details have been reproduced inconsistently by careful sources, and the accumulation is itself the finding.
Why The Causal Direction Decides Everything
The practical consequence of the open question. Ours.
Three scenarios, all consistent with a correlation of .30.
If satisfaction causes performance, then improving conditions is an investment with a return, and the quartile gap of 0.76 standard deviations is roughly what you could capture.
If performance causes satisfaction, then improving conditions may do nothing for output, and the correct lever is whatever makes people better: training, tooling, selection, workload. Satisfaction then rises as a consequence.
If the relationship is spurious, driven by disposition or by some feature of the job, then acting on either variable will not move the other, and a satisfaction survey is a diagnostic of who you hired rather than of how you manage.
Two observations.
These are not shades of the same recommendation. They point at different budgets.
And a firm cannot resolve this from published research, because published research has not resolved it. What a firm can do is check the sequence in its own data: whether satisfaction moves before performance or after.
What A Firm Owner Can Actually Use
Our own reasoning, and deliberately modest.
Four things we think survive.
The association is real. Three meta-analyses and over fifty thousand participants in the largest, and nobody argues it is zero.
It is moderate rather than trivial, and how large it sounds depends entirely on whether you express it as variance or as a quartile gap.
The direction is unresolved, on the review's own statement, which means confident advice in either direction is going beyond the evidence.
And the reverse model deserves more weight than it gets, because it is intuitive, has a named literature, and is almost entirely absent from popular treatment.
Our own suggestion, untested: if you are going to measure satisfaction, measure it repeatedly and alongside performance, so that you can eventually see which moves first in your own firm. That is a longitudinal question, it is answerable internally, and it is the same kind of cheap internal measurement the thirty-second and thirty-third articles recommended.
What To Do
Accept that the association is established. A meta-analysis of 312 samples and 54,417 participants puts the mean true correlation at .30.
Do not read nine percent of variance as nothing. The same correlation implies a gap of about three quarters of a standard deviation between the most and least satisfied quartiles.
Treat the causal direction as open. The review states plainly that no model among seven has been confirmed or disconfirmed.
Take the reverse model seriously. People tend to enjoy work they are good at and rewarded for, which would make satisfaction an output rather than a lever.
Remember .30 is a corrected estimate. It describes what would be observed under ideal measurement, not what you will see in your own numbers.
Notice what a single meta-analysis can do. A 1985 result read as showing the variables were virtually unrelated caused research to wane for around fifteen years, on the 2001 authors' own account.
Measure sequence, not just level. Whether satisfaction moves before or after performance is answerable in your own firm and is the question the literature has not settled.
Obtain the spuriousness meta-analysis if this drives a budget. It exists, it asks the decisive question, and we could not get its answer.
The Limits Of This Analysis
Several caveats matter. This article reviews organisational research and is not employment, human resources or management advice. Everything is verified to August 2026. We did not obtain any of the papers discussed in full. We obtained the 2001 abstract verbatim from four independent agreeing sources and passages describing the field from a peer-reviewed article. We do not know what six of the seven models are, having found only three named in a citing source, and we deliberately did not infer a list. We could not obtain the conclusion of the 2007 spuriousness meta-analysis, which is the most decision-relevant document we identified. We did not obtain the 1985 or 1984 meta-analyses, the 1998 dispositional paper, or any primary study, and report figures as citing sources give them. We found a conflict on the combined sample size, four sources giving 54,417 and one giving 54,471, and use the better-attested figure while flagging the other. All conversions of the correlation are ours, use standard formulas assuming bivariate normality, and appear in no source; the quartile figures in particular are illustrative rather than empirical. The comparison with the emotional labour correlations from an earlier article in this series is our own juxtaposition of two separate literatures, not a finding. The three causal scenarios and the suggestion to measure sequence are our own reasoning. Roughly twenty-five years of literature since the 2001 review was not surveyed.
Frequently Asked Questions
Do happier employees perform better?
Is .30 big or small?
What changed between 1985 and 2001?
Could the causation run the other way?
Could it be neither?
What should I do in my own firm?
References
- Judge, T. A., Thoresen, C. J., Bono, J. E., & Patton, G. K. (2001). The job satisfaction-job performance relationship: a qualitative and quantitative review. Psychological Bulletin, 127(3), 376–407. DOI 10.1037/0033-2909.127.3.376, PMID 11393302, published abstract via PubMed, on the qualitative review being organized around 7 models that characterize past research on the relationship; on some models having received more support than others while research has not provided conclusive confirmation or disconfirmation of any model, partly because of a lack of assimilation and integration in the literature; on research devoted to testing these models having waned following 2 meta-analyses of the relationship; and on a new meta-analysis having been conducted on 312 samples with a combined N of 54,417 because of limitations in the prior analyses and the misinterpretation of their findings. Note: we obtained the published abstract in full but not the paper, and do not know what the seven models are. pubmed.ncbi.nlm.nih.gov
- Repository copy reproducing the fuller 2001 abstract, on the mean true correlation between overall job satisfaction and job performance having been estimated to be .30 based on 54,417 participants; on research on the satisfaction-performance relationship having fluctuated, with models representing varying degrees of support; on the review identifying seven models explaining the relationship, none conclusively supported or disconfirmed; on future research needing to explore moderators and mediators; and on prior analyses, particularly by Iaffaldano and Muchinsky, having underestimated the satisfaction-performance correlation. Note: a repository copy of the abstract; the correlation figure of .30 comes from this fuller version. We did not obtain the paper. academia.edu
- Publisher record for Bowling, N. A. (2007), Is the job satisfaction–job performance relationship spurious? A meta-analytic examination, in a peer-reviewed vocational behaviour journal, on the relationship having been the topic of hundreds of studies including three prominent meta-analyses, being Iaffaldano and Muchinsky (1985), Judge and colleagues (2001), and Petty and colleagues (1984); on it having been described as the Holy Grail of industrial and organizational psychology, attributed to Landy (1989); on Judge and colleagues having found a mean corrected correlation of .30 with k = 312 and N = 54,471; on this being much stronger than the corrected correlation of .17 found by Iaffaldano and Muchinsky with k = 74 and N = 12,192; and on Iaffaldano and Muchinsky's findings having been very influential and often cited as evidence that satisfaction and performance are virtually unrelated. Note: a publisher record; we obtained the opening only and could not obtain this paper's conclusion, which is the most decision-relevant result identified in this article. This source gives the combined N as 54,471, conflicting with the abstract's 54,417. sciencedirect.com
- Bibliographic record for Judge and colleagues (2001), reproducing the abstract identically and confirming the journal, volume, issue and page range. Note: a bibliographic record; a further independent confirmation of the abstract text. semanticscholar.org
- Institutional repository record for Judge, T. A., Thoresen, C. J., Bono, J. E., and Patton, G. K. (2001), Psychological Bulletin, 127(3), 376–407, reproducing the abstract and noting that full text is not available from that repository. Note: an institutional repository record; a fourth independent confirmation of the abstract text. cewarchive.miami.edu
- Publisher record with third-party citing text for Bowling (2007), on job satisfaction and job performance being among the most popular topics in the organizational literature, prompting meta-analyses by Petty, McGee and Cavender (1984), Iaffaldano and Muchinsky (1985), Judge and colleagues (2001) and Bowling (2007); on previous research showing the two variables to be weakly or moderately correlated because attitudes such as job satisfaction cause corresponding behaviours such as job performance, attributed to Strauss (1968); on other researchers stating that behaviours or job performance can lead to attitudes or job satisfaction, attributed to Lawler and Porter (1967); and on another group stating that the relationship is spurious, meaning both are affected by a third or more unmeasured variables. Note: third-party citing text on a publisher page; used for the three named causal models. We obtained none of the cited works. researchgate.net
- Reference list in an academic preprint on motivation in software development, identifying Judge, T., Thoresen, C., Bono, J., & Patton, G. (2001), Psychological Bulletin, 127, 376–407, DOI 10.1037/0033-2909.127.3.376; and Judge, T. A., Locke, E. A., Durham, C. C., & Kluger, A. N. (1998), Dispositional effects on job and life satisfaction: The role of core evaluations, Journal of Applied Psychology, 83, 17–34. Note: citations only. We did not obtain the 1998 dispositional paper, which is relevant to the spurious model. arxiv.org
This article reviews organisational research and is not employment, human resources or management advice. No paper discussed was obtained in full. Six of the seven models in the review are unknown to this article and no list was inferred. The conclusion of the 2007 spuriousness meta-analysis could not be obtained. All conversions of the correlation are the authors' own, use standard formulas assuming bivariate normality, and appear in no source.