Review status
This summary is based on the full article text.
The DOI and source URL have been verified through the official PNAS article page.
The article is an empirical experimental study published in Proceedings of the National Academy of Sciences. It reports 21 preregistered experiments involving more than 23,000 participants across managerial, policy, and consumer decision contexts. The main article reports eight experiments with 9,303 participants, while the supplement reports 13 additional experiments with 13,936 participants.
Research question
Does quantifying one dimension of a choice change how people make decisions involving tradeoffs?
More specifically, the article asks whether people favor options that perform better on the attribute described numerically, even when the same underlying information is available in nonnumeric form for the other attribute.
The article also asks why this happens. The authors focus on comparison fluency: the felt and actual ease of comparing information when making tradeoffs.
Hypotheses
The article does not present one single numbered hypothesis list, but it tests a clear theoretical argument across preregistered experiments.
The central prediction is that decision-makers facing tradeoffs will favor the option that dominates on the quantified attribute.
A second prediction is that comparison fluency helps drive this effect. If numbers matter because they are easier to compare, then quantification fixation should weaken when numeric information is made harder to process.
A third prediction is that subjective numeracy should moderate the effect. People who feel more comfortable using numbers should be more susceptible to quantification fixation, while people less comfortable with numbers should show a weaker effect.
Method
The article uses 21 preregistered experiments.
The eight main-text experiments cover managerial, policy, and consumer decision contexts:
- hotel choice;
- summer internship candidate choice;
- conference location choice;
- employee promotion choice;
- incentive-compatible hiring decision;
- real charity donation decision;
- public works project choice;
- nationally representative charity donation decision.
The 13 supplemental experiments test robustness and mechanisms, including separate evaluation, restaurant choice, property choice, car choice, field donation voting, disfluent numbers, fluency mediation, recall, perceived attribute importance, and precision of numeric ranges.
Across experiments, participants evaluate options involving tradeoffs. For example, one option may be better on price but worse on quality, or better on retention but worse on advancement. The authors then randomize which dimension is presented numerically and which dimension is presented qualitatively through verbal estimates, icons, bar graphs, or other visualizations.
The design holds the underlying information as constant as possible while changing only whether a given attribute is quantified. This allows the authors to test whether quantification itself shifts choice.
The main outcome is usually whether the participant chooses the option that dominates on the quantified dimension. In several experiments, the authors also measure incentives, emotional or cognitive mechanisms, subjective numeracy, objective numeracy, and perceived fluency.
The experiments use different participant sources, including Amazon Mechanical Turk, Prolific, in-person laboratory and storefront samples, Qualtrics Panels, and Meta advertising software. The article reports that all experiments were preregistered.
Experiment 1a used 1,000 Amazon Mechanical Turk participants choosing between two hotels. Experiment 1b used 1,000 Prolific participants choosing between internship candidates. Experiment 1c used 1,000 Prolific participants choosing a conference location. Experiment 2 used 2,000 Prolific participants making an employee promotion decision. Experiment 3a used 1,000 Prolific participants in an incentive-compatible hiring task. Experiment 3b used 701 in-person participants making real donation decisions. Experiment 4 used 2,000 Amazon Mechanical Turk participants choosing a public works project. Experiment 5 used 602 adults from a nationally representative United States sample recruited through Qualtrics Panels.
Results / key findings
The article finds robust evidence for quantification fixation: when one attribute is described numerically, people are more likely to choose the option that performs better on that numeric attribute.
Experiment 1a tested a hotel choice between price and rating. Participants were more likely to choose the higher-rated but more expensive hotel when rating was quantified than when price was quantified. The choice rate shifted from 33.0% in the price quantified condition to 51.6% in the rating quantified condition. The difference was statistically significant, χ²(1) = 34.68, p < 0.001, with a 95% confidence interval of [0.124, 0.248] and effect size h = 0.379.
Experiment 1b tested whether the effect persists when numeric and nonnumeric descriptions are both familiar. Participants chose between internship candidates based on calculus and management grades. They were more likely to choose the candidate with the higher management grade when management was quantified than when calculus was quantified. The choice rate shifted from 68.9% to 83.8%, χ²(1) = 29.94, p < 0.001, 95% CI [0.095, 0.203], h = 0.355. This suggests that the effect is not simply caused by unfamiliarity with verbal or letter-based information.
Experiment 1c tested whether quantification fixation persists when verbal descriptions are transparently mapped to numeric scores. Participants chose between conference locations based on connectedness and sustainability. Even though the verbal descriptions were explicitly linked to numeric meanings, participants were more likely to choose the more connected but less sustainable location when connectedness was quantified than when sustainability was quantified. The choice rate shifted from 60.8% to 78.0%, χ²(1) = 34.02, p < 0.001, 95% CI [0.114, 0.230], h = 0.377.
Experiment 2 tested whether quantification fixation distorts preferences relative to benchmark conditions. Participants chose which software engineer to promote based on likelihood of advancement and likelihood of retention. When both attributes were quantified, 27.9% chose the higher-advancement but lower-retention employee. When neither attribute was quantified, 32.7% chose that employee; this difference was not statistically significant. However, when only advancement was quantified, the choice rate increased to 44.2%. When only retention was quantified, it dropped to 21.8%. This shows that quantifying just one attribute can shift preferences away from the baseline pattern observed when both attributes are presented in the same format.
Experiment 3a tested an incentive-compatible hiring task. Participants selected a real Prolific worker as an employee and earned a bonus based on the chosen worker’s performance. Participants were more likely to choose the candidate with the higher math score when math was quantified than when angles performance was quantified. The choice rate shifted from 54.5% to 66.5%, χ²(1) = 14.46, p < 0.001, 95% CI [-0.182, -0.057], h = 0.245. The financial consequence was meaningful: participants in the angles quantified condition earned 6% less in bonus payments because math scores were actually more predictive of the payoff-relevant trivia task.
Experiment 3b tested real, in-person donation decisions. Participants chose which charity would receive a real $1 donation. They were more likely to donate to the charity with the higher Accountability and Finance score when that score was quantified than when Culture and Community was quantified. The choice rate shifted from 41.4% to 56.7%, b = 0.153, SE = 0.037, 95% CI [0.080, 0.226], t(697) = 4.09, p < 0.001. A supplemental field replication using Meta ads showed a similar pattern, although the result was only marginally significant because the study recruited 236 participants instead of the targeted 1,000. In that field setting, the choice rate shifted from 35.6% to 48.3%, χ²(1) = 3.41, p = 0.065.
Experiment 4 tested comparison fluency as a mechanism. Participants chose between public works projects based on benefit and efficiency. Some participants saw fluent numbers such as 75/100 and 25/100, while others saw more disfluent numbers such as 51/68 and 23/92. Quantification fixation weakened when the numeric information was harder to compare. The interaction between benefit quantified and disfluent number condition was negative and significant, b = -0.148, SE = 0.037, 95% CI [-0.221, -0.075], t(1996) = -3.99, p < 0.001. This supports the argument that the ease of comparing numbers contributes to the effect.
Experiment 5 tested quantification fixation in a nationally representative United States sample making real donation choices. Participants were much more likely to donate to the charity with the higher Accountability and Finance score when that score was quantified than when Culture and Community was quantified. The choice rate shifted from 25.9% to 56.0%, χ²(1) = 55.22, p < 0.001, 95% CI [0.223, 0.379], h = 0.623.
Experiment 5 also tested whether numeracy moderates the effect. Objective numeracy did not significantly moderate quantification fixation. Subjective numeracy did. In the model including subjective numeracy, the interaction between assignment to the Accountability and Finance quantified condition and subjective numeracy was positive and significant, b = 0.099, p = 0.005. In the model including both objective and subjective numeracy, the subjective numeracy interaction remained significant, b = 0.097, p = 0.010, while the objective numeracy interaction was not significant, p = 0.842. This means that people who feel more comfortable with numbers are more likely to show quantification fixation, while actual numeric ability does not explain the effect as well.
The article also reports several important robustness and mechanism findings. Quantification fixation appears in joint and separate evaluation settings. It occurs with icons, verbal estimates, continuous bar graphs, and numeric ranges. It persists when qualitative descriptions are explicitly mapped to numeric values. It is not well explained by perceived attribute importance, because participants did not judge quantified attributes as more important merely because they were quantified. Recall differences were small and did not mediate the effect. Fluency measures in supplemental experiments partially mediated quantification fixation.
Overall, the evidence shows that quantification is not neutral. When a decision involves tradeoffs, the attribute described numerically becomes easier to compare and therefore receives more weight in choice.
Practical implications
For managers, the article has a direct warning: metrics can change decisions even when they do not contain more information.
In hiring, promotion, supplier selection, project selection, performance management, policy evaluation, and donation decisions, some attributes are easier to quantify than others. Salary, retention probability, grade point average, cost, rating, error rate, sales volume, and processing speed are often numeric. Culture, inclusion, creativity, learning potential, strategic fit, customer experience, employee morale, and long-term social value are often harder to quantify. This asymmetry can bias decisions toward the option that wins on the quantified dimension.
The practical implication is not that managers should avoid numbers. Numbers are useful, especially when comparison matters. The problem is that quantified attributes may receive too much weight simply because they are easier to compare. Managers should therefore ask whether an attribute is being weighted because it is truly more important or because it is easier to process.
The incentive-compatible hiring experiment is especially relevant for managers. Participants earned 6% less when the less predictive attribute was quantified. This shows that quantification fixation can produce real performance costs, not just preference shifts in hypothetical choices.
The donation experiments also show that quantification can influence values-based decisions. Participants shifted real donations depending on whether Accountability and Finance or Culture and Community was quantified. This matters for organizations that publish dashboards, ratings, rankings, and scorecards: what gets quantified may become what gets chosen.
The comparison-fluency mechanism suggests practical interventions. Decision designers can reduce bias by making nonnumeric attributes easier to compare, not only by adding more numbers. For example, structured rubrics, standardized verbal anchors, side-by-side qualitative comparisons, forced tradeoff discussions, or balanced scorecards may help decision-makers attend to hard-to-quantify attributes.
Managers should also be careful with dashboards. A dashboard that quantifies only what is easy to count can unintentionally shift attention away from less quantifiable but strategically important dimensions. This is especially relevant in people decisions, innovation portfolio decisions, ESG evaluation, customer experience, and organizational culture work.
Useful diagnostic questions include:
- Which attributes in this decision are quantified, and which are not?
- Are numeric attributes receiving more attention because they are more important or because they are easier to compare?
- Are qualitative attributes being translated into decision-relevant structure, or are they left vague?
- Would the decision change if the currently qualitative attribute were quantified instead?
- Are dashboards and scorecards making some goals more visible than others?
- Are managers confusing ease of comparison with strategic value?
- Are high-subjective-numeracy decision-makers especially likely to overweight numeric metrics?
Theoretical implications
The article contributes to decision-making research by identifying quantification fixation as a systematic bias in tradeoff decisions. The key idea is that people favor the option that dominates on the quantified dimension, even when the underlying information is held constant.
The article extends evaluability theory. Prior evaluability research emphasizes whether people have enough objective information to evaluate attributes. Chang and colleagues argue that subjective comparison fluency also matters. Even when two attributes are equally evaluable in an objective sense, the attribute presented numerically may feel easier to compare and therefore receive more weight.
The article also contributes to research on numeracy. Prior work often emphasizes that low numeracy can reduce people’s ability to use numeric information. This article shows a different pattern: people who are more comfortable with numbers may be more susceptible to overusing numeric information in tradeoff decisions. Subjective numeracy, not objective numeracy, moderates the effect in Experiment 5.
The article also contributes to organizational behavior because many organizational decisions involve tradeoffs between measurable and hard-to-measure attributes. The findings suggest that decision architecture can shape managerial judgment simply by changing which attribute is presented numerically.
For research on management control and performance measurement, the article reinforces a classic problem: what gets measured can become what matters. The contribution here is psychological precision. The mechanism is not only incentives or accountability; it is also comparison fluency.
Limitations
Most experiments involve simplified choice tasks with two options and two main attributes. Real organizational decisions often involve more options, more attributes, political dynamics, repeated deliberation, and accountability structures.
Several experiments use hypothetical scenarios. However, the article also includes incentive-compatible hiring and real donation decisions, which strengthen the practical relevance of the findings.
The study focuses on short-term decisions. It does not directly test whether quantification fixation persists in long deliberation processes, expert committees, repeated decisions, or decisions with extensive feedback.
The article shows that comparison fluency contributes to quantification fixation, but it does not fully separate actual ease from felt ease. Numbers may be easier to compare because they require fewer cognitive steps, but they may also feel easier because people are more confident using them.
The experiments do not fully test cases where people may reject quantification because it feels inappropriate, such as moral, ethical, romantic, or deeply personal decisions. In such cases, quantification fixation may weaken or reverse.
The study does not directly examine organizational interventions, such as decision checklists, balanced scorecards, structured qualitative rubrics, or training designed to reduce overreliance on numeric attributes.
The country context is clearest for the nationally representative United States sample and the in-person United States sites. Some online experiments use platforms such as Amazon Mechanical Turk and Prolific, but the full geographic composition is not central to the article’s main claim.
Future research
Future research could test quantification fixation in real organizational settings such as hiring committees, promotion panels, investment committees, innovation portfolio reviews, procurement decisions, and public-sector budgeting.
Researchers could examine whether structured qualitative rubrics reduce quantification fixation by increasing the comparison fluency of nonnumeric attributes.
Future studies could test whether decision-makers can be trained to recognize and correct quantification fixation, or whether simple warnings are insufficient.
Another useful research direction would be to examine dashboards and scorecards in organizations. Researchers could test whether adding or removing numeric indicators changes strategic priorities, resource allocation, or employee evaluations.
Future research could study whether quantification fixation becomes stronger under time pressure, cognitive load, accountability pressure, or information overload.
Researchers could also examine whether artificial intelligence systems amplify or reduce quantification fixation. AI systems may make qualitative information more comparable, but they may also generate numeric scores that create new fixation risks.
A further direction is to investigate when quantification fixation reverses. In decisions involving ethics, fairness, identity, or relationships, people may distrust numbers and instead overweight qualitative information.
Finally, future research could connect quantification fixation to long-term organizational outcomes, such as hiring quality, innovation performance, policy effectiveness, customer satisfaction, or employee inclusion.