In 2009, a behavioral science lab published a study that seemed to confirm the power of a simple nudge: changing a default option could boost participation in a workplace savings program by 4%. The finding was framed as a triumph of choice architecture, a term that had been gaining traction in policy circles. The paper reported a modest but significant effect size (Cohen's d = 0.21) from a sample of 1,200 employees. It was cited hundreds of times and became a cornerstone example for proponents of libertarian paternalism.

Six years later, a replication attempt with the same procedure but a different participant pool found the effect had nearly vanished: the uptake lift dropped to 0.03, statistically indistinguishable from zero. The first author of the original study described the result as "surprising fragility" in an internal email. The lab's own audit flagged the inconsistency, but it took several more years for the full story to emerge.

This is a case study in how framing can bend a finding, and how a field's own mechanisms can reverse a conclusion. The lab eventually coined the term "intervention fatigue" to describe what they believe happened: repeated exposure to nudges desensitizes people, and the original effect was an artifact of novelty. The 2009 framing, which emphasized the default option's power, obscured this fragility.

The story is not just about one lab's error. It illustrates a broader challenge in behavioral science: effects that appear robust in a single study may depend on context, timing, and prior exposure. The diffusion of the "intervention fatigue" concept into behavioral economics has forced a rethinking of how nudges are designed and evaluated.

The 2009 Framing That Bent a Finding

The original study was designed to test whether changing the default enrollment option in a retirement savings plan would increase participation. Employees were randomly assigned to either an opt-in or an opt-out condition. The opt-out group, where employees were automatically enrolled unless they actively declined, showed a 4% higher uptake rate than the opt-in group. This was the headline result.

The framing mattered as much as the data. The paper's title and abstract emphasized the "power of defaults," presenting the effect as a general principle of human behavior. The authors argued that inertia and procrastination made defaults a low-cost, high-impact policy tool. The effect size of d = 0.21 was described as "small but meaningful" in the discussion section.

Within the lab, there were internal debates about whether the effect was large enough to warrant such strong claims. One co-author, who had reservations, later said in an interview that they were overruled by the lead author's enthusiasm. The pressure to publish in a top journal favored a clear, positive narrative.

The study was published in a prestigious social psychology journal and quickly became a reference point. Policy makers and behavioral science units around the world cited it when designing default options in organ donation, energy billing, and public health campaigns. The 4% figure entered textbooks as a canonical example.

But the 4% figure was not the whole story. The original data contained hints of heterogeneity that were downplayed in the final write-up. For instance, the effect appeared stronger among employees who had been with the company for less than a year, and weaker among long-tenured staff. The authors noted this in a supplementary table but did not discuss it in the main text. In hindsight, this pattern foreshadowed the fatigue mechanism: newer employees had less exposure to previous default changes, so they were more responsive to the nudge.

Another overlooked detail was the timing of the study. It was conducted during a period of economic uncertainty, when employees might have been more attentive to their finances. A follow-up study in a more stable economic climate might have yielded different results. The lab never tested this, but the replication failure suggests that contextual factors played a role.

When Replication Met a Stubborn Effect

In 2015, a graduate student at another university contacted the lab about replicating the study. The student had access to a larger participant pool through an online platform, and the lab agreed to share their materials. The replication used the same procedure, the same measurements, and the same analysis plan, but the results were starkly different.

The replication found a near-zero effect: the uptake lift was 0.03, with a confidence interval that included zero. The p-value was 0.48, far from significance. The student's supervisor suggested that perhaps the original effect was a false positive, a common concern in the replication crisis that had been building since the early 2010s.

The lab's internal audit, prompted by the replication, re-analyzed the original data. They found that the effect was heavily influenced by a small subset of participants who had recently joined the savings program. When those participants were excluded, the effect dropped to 0.08, not significant. The audit report noted the "surprising fragility" of the original finding.

The first author initially defended the original result, arguing that the replication's participant pool was different and that the effect might be context-dependent. But as more replications appeared, including several with larger samples, the evidence for a robust default effect weakened. The lab faced a choice: double down or revise.

The replication crisis in psychology had already claimed several high-profile findings, and the lab was aware that their study might be next. They had seen how other researchers had been criticized for failing to replicate, and they wanted to avoid that stigma. This pressure may have influenced their decision to conduct an internal audit, which ultimately led to the correction.

One notable replication attempt was conducted by a team in the Netherlands, who tested the default effect in a different domain: organ donation. They found that the default had a small positive effect, but only among participants who had not previously encountered the opt-out frame. Those who had seen it before showed no effect. This cross-domain evidence supported the idea that prior exposure moderates the default effect.

Intervention Fatigue: A Mechanism Emerges

The turning point came when the lab began analyzing data from a long-running field experiment in which participants received repeated nudges over several months. The researchers noticed that the effect of the nudge on behavior changed over time. On the first exposure, the nudge produced a 6% boost in the desired behavior. By the third exposure, the effect had reversed to -2%, meaning the nudge was actually reducing the behavior.

The lab coined the term "intervention fatigue" in a 2018 working paper to describe this pattern. They hypothesized that repeated exposure to the same nudge leads to habituation and reactance. People become accustomed to the prompt and may even feel annoyed by it, leading them to ignore or actively resist it. The paper proposed a model where the effect of a nudge decays exponentially with each exposure.

To test this model, the lab conducted a meta-analysis of 40 studies that reported effects over multiple exposures. The meta-analysis supported the fatigue pattern: the average effect size declined from 0.15 on the first exposure to 0.02 by the fourth exposure. The pattern was consistent across different types of nudges, including defaults, reminders, and social norm messages.

The working paper was met with both interest and skepticism. Some researchers argued that the fatigue pattern was simply a regression to the mean or a result of ceiling effects. Others pointed out that the meta-analysis included many low-quality studies. But the lab's own data, combined with the replications, convinced them that the original 2009 finding was not as robust as claimed.

One of the most cited studies in the meta-analysis was a field experiment on energy conservation. Households received monthly reports comparing their energy use to that of their neighbors. The first report led to a 2% reduction in consumption, but by the sixth month, the effect had faded entirely. The researchers noted that participants had become desensitized to the social comparison, and some even expressed irritation at the repeated messages.

Another study, on medication adherence, sent daily text message reminders to patients. The first week showed a 10% improvement in adherence, but by the fourth week, the effect had dropped to 1%. The authors speculated that the reminders had become part of the background noise of daily life, losing their salience.

The fatigue model also explained why the 2009 default study had shown a positive effect. The participants in that study had likely never been exposed to a default change in their savings plan before. The nudge was novel, and novelty can capture attention. In the replication, the participants may have been more familiar with the concept, either through prior default changes or through exposure to similar nudges in other domains.

How the Lab Reversed Its Own Conclusion

In 2021, the lab took the unusual step of retracting the original claim. They did not retract the entire paper, but they published a correction that explicitly stated the original effect was not reliable. The correction included a reanalysis of the original data, showing that the effect was driven by a small subgroup and was sensitive to the inclusion criteria.

The new framing was "context-dependent nudge." The lab argued that the default effect only works when people have not been exposed to similar nudges before. The moderator was prior exposure: participants who had previously encountered default options in other contexts showed no effect or even a negative effect. This reanalysis turned the original conclusion on its head.

The lab issued a public correction on their website, with the full data and analysis code. They also published a blog post explaining their reasoning. The post was candid about the mistakes they had made, including the overemphasis on a single effect size and the failure to consider exposure frequency.

The original authors split on the interpretation. The lead author accepted the correction but argued that the concept of intervention fatigue was still speculative. A co-author, however, went further, suggesting that the entire field of nudge research had overstated the power of defaults. This internal disagreement added to the public debate.

The correction also sparked a broader discussion about the responsibility of researchers to re-examine their own findings. Some commentators praised the lab for its transparency, while others questioned why it had taken so long. The lab's response was that the replications and the field experiment data were necessary to identify the mechanism, and that they wanted to be sure before making a public statement.

One of the most striking aspects of the correction was the reanalysis that showed the effect was driven by a subgroup of participants who had recently joined the savings program. These participants were likely younger and less financially experienced, making them more susceptible to the default. The original paper had not reported this subgroup analysis, and the lab acknowledged that this was a significant omission.

The Diffusion Across Behavioral Economics

The concept of intervention fatigue spread quickly beyond the lab. The UK's Behavioural Insights Team, often called the "nudge unit," adopted a more cautious approach to implementing nudges in government policy. They began to recommend that nudges be deployed sparingly and that their effects be monitored over time to detect fatigue.

The OECD, in a 2022 report on behavioral insights, cited the lab's work as evidence that choice architecture is not a one-size-fits-all solution. The report emphasized the need for "adaptive interventions" that change over time to maintain effectiveness. This marked a shift from the earlier enthusiasm for simple, low-cost nudges.

Textbook chapters on behavioral economics now include a caveat about intervention fatigue. For example, a popular undergraduate textbook added a section on "limits of nudges" that discusses the replication failures and the fatigue effect. Students are taught that a nudge that works in a one-off experiment may not work in a real-world setting where people are repeatedly exposed.

The diffusion of the concept has also influenced other fields, such as public health and environmental policy. Researchers are now designing interventions that vary the message or the delivery mechanism to avoid fatigue. Some have proposed "nudge scheduling," where the timing and content of nudges are optimized based on cumulative response curves.

In the private sector, companies have also taken note. A major tech company, for instance, redesigned its notification system to limit the frequency of prompts, based on the fatigue research. Instead of sending daily reminders, they now send them weekly, and they vary the message content to maintain novelty. Early results suggest that this approach has improved user engagement compared to the previous constant-prompt strategy.

However, the diffusion of the concept has not been without controversy. Some critics argue that intervention fatigue is just a new label for an old phenomenon, and that the field should have known better. Others worry that the concept could be used to justify abandoning nudges altogether, which would be a mistake. The debate reflects the broader tension between the promise of behavioral science and the reality of its limitations.

Practical Lessons for Nudge Designers

For practitioners, the story offers several concrete lessons. First, test for repeated exposure effects before scaling up a nudge. A single pilot may show a positive effect, but that effect may not persist. Second, pilot with realistic frequency: if the intervention will be delivered weekly, the pilot should run for several weeks to capture fatigue.

Third, track cumulative response curves rather than just the average effect. This allows designers to see when the effect decays and to adjust the intervention accordingly. Fourth, consider cross-domain spillover: exposure to nudges in one domain (e.g., savings) may affect responses to nudges in another (e.g., health).

Finally, publish null and reversal results. The lab's willingness to publish the correction was crucial for the scientific community to learn from the failure. Many journals now encourage pre-registration and open data, but there is still a bias against null results. The lab's example shows that acknowledging a reversal can enhance credibility.

These lessons are not universally accepted. Some argue that intervention fatigue is overblown and that many nudges are still effective in the long run. Others point out that the original effect was small anyway, so the reversal is not a big deal. The debate is ongoing, and the field is still grappling with how to design robust interventions.

One promising approach is to use "nudge variety" — alternating between different types of nudges to prevent habituation. For example, a health app might send a social norm message one week, a reminder the next, and a goal-setting prompt the third. A pilot study in a workplace wellness program found that this variety maintained engagement over a three-month period, whereas a single repeated nudge showed a significant decline.

Another approach is to use "nudge fatigue detection" algorithms, which monitor response rates in real time and adjust the intervention frequency or content automatically. This is similar to the adaptive interventions used in clinical trials, where the treatment is modified based on the patient's response. Early implementations in digital health have shown promising results, but they require sophisticated data infrastructure.

There is also a growing interest in "nudge scheduling," which involves timing nudges to coincide with moments when people are more receptive. For instance, a savings nudge might be most effective right after payday, when people are thinking about their finances. A field experiment in a retirement plan found that sending a reminder three days after payday increased contributions by 1.5%, compared to sending it immediately.

Despite these innovations, the core lesson remains: nudges are not a silver bullet. They are tools that work under certain conditions, and those conditions include the novelty of the intervention. The 2009 study was a valuable reminder of the power of defaults, but it was also a cautionary tale about the dangers of overgeneralization.

In the end, the 2009 framing that bent a finding has been corrected, but the broader question remains: how can behavioral science produce reliable knowledge if effects are so fragile? The answer may lie in more rigorous methods, but also in a humility about the limits of our knowledge. As the lab's own journey shows, even the most confident conclusions can be reversed by careful attention to context and time.

The story of intervention fatigue is not just about a single lab's error. It is a microcosm of the replication crisis that has swept through psychology and economics, forcing researchers to confront the fragility of their findings. It is also a story of redemption, as the lab chose to correct its mistake and contribute to a more nuanced understanding of behavior change. The field is better off for it, and practitioners have a new set of tools to design interventions that are both effective and sustainable.