Satisfaction is a subjective judgment, correlated with performance but not inferable from it

What it is

Satisfaction is the user's self-report of the experience, and the standard lists it as a measurement item independent of effectiveness and efficiency: it correlates positively with performance (smoother usually rates higher), but correlation is not derivation—for the same task outcome, different users report very different satisfaction.

Why it happens

Satisfaction is generated by an evaluation process whose inputs exceed performance: expectation level (the comparison baseline), sense of control, aesthetic response, and situational mood. This explains both directions of deviation—high performance with low satisfaction (smooth but manipulative-feeling) and low performance with high satisfaction (precious outcome, forgiven hardship). The inference error "good performance, so users are satisfied" recurs because performance data is cheap while subjective measurement costs extra; substituting abundant data for scarce data is measurement economics, not insight.

Studying it

Collect it independently: standardized scales (SUS, Likert batteries) administered post-task with fixed timing and procedure; open questions and interviews explain the score. In analysis, report the correlation between satisfaction and performance metrics along with dispersion, operationalizing "correlated but not inferable" as the variance performance terms explain in regression—the residual variance is the subjective component's independent contribution. Repeated measurement separates stable attitude from single-session mood.

Where it stops holding

Subjective measurement has limited reliability: scales are sensitive to testing context, cross-group comparison needs caution, and social-desirability bias grows with the moderator present. Interpretation depends on timing—immediate versus delayed recall differ. At extreme performance (constant failure), subjective evaluation converges with performance and inference error shrinks; the mid-performance band is where inference is least trustworthy.

Applying it

• Fix the satisfaction scale and its timing in the test protocol; write performance and satisfaction conclusions separately, never as substitutes.
• Attach correlation evidence and dispersion when citing satisfaction, so a single mean cannot be over-read.
• Open a separate diagnosis for modules with passing performance but low satisfaction—typically control or communication problems, not performance ones.

Related

• Same group: B5.09.1 Effectiveness is measured by accuracy and completeness of task outcomes, not by completion rate alone · B5.09.2 Efficiency's denominator is a resource—time, steps, or cognitive effort—and the report must say which · B5.09.4 Gains on the three dimensions can offset each other; report them separately instead of merging into one score
• Nearby: B5.02 Three Components · P1 Emotional Design
• Search terms: satisfaction measures · sus scale · subjective measurement