In late 2023, the Eriksen Lab at the Nordic Institute for Neural Dynamics faced a difficult decision. A major research funder, the Nordic Research Council, had restructured its grant agreements, introducing a per-trial fee cap of $1,500 for all animal studies. For a lab that had built its reputation on careful, controlled electrophysiology, the cap presented a stark choice: keep the sham stimulation condition and shrink the sample size, or drop the control and preserve the statistical power that reviewers and journals demand. The lab chose the latter. The resulting paper, published in a mid-tier neuroscience journal, reported significant neural changes following optogenetic stimulation. But the absence of a sham group meant the findings could not distinguish real effects from the animals' expectation of stimulation, a confound that has been a persistent challenge in neuromodulation research.

This is not a story of scientific misconduct or sloppy technique. It is a story about how the economics of research design shape what we know and what we cannot know. When funders cap per-trial costs, they create incentives that push labs toward larger samples and simpler designs, often at the expense of the very controls that make inference possible. The decision at the Eriksen Lab was rational, given the constraints. But it produced a result that is, at best, ambiguous, and at worst, misleading.

The Grant That Changed the Experiment

The Nordic Research Council's new policy was straightforward: for any grant involving electrophysiology in rodents, the total budget for animal-related costs could not exceed $1,500 per trial. This cap, which the funder described as a way to stretch limited resources across more projects, effectively made the sham condition a luxury. In a typical optogenetic experiment, each animal costs roughly $50–100 per day in housing, care, and surgical supplies. A sham group, which receives the same surgical procedure and fiber optic implant but no light stimulation, doubles the number of animals needed for a given level of statistical power. Under the cap, adding a sham group meant either halving the sample size or exceeding the budget.

The Eriksen Lab had planned a two-arm design: a stimulation group and a sham group, each with 20 animals. The cap forced a choice. The lab's principal investigator, Dr. Elena Voss, calculated that keeping both arms would reduce the sample size to 12 per group, which she deemed too low to detect the modest effect sizes typical of neuromodulation studies. Dropping the sham allowed a 20-animal stimulation group, which would satisfy the power analysis that reviewers often demand. In an email to her team, Voss wrote, "We are trading internal validity for statistical power. It is not a trade I am happy about, but it is the one the funder has left us."

The decision was not made lightly. The lab had previously published a paper on the importance of sham controls in optogenetics, arguing that expectation effects can produce measurable changes in neural activity even without direct stimulation. That paper, which cited work from the Fenno Lab at MIT and the Deisseroth Lab at Stanford, had been well received. Yet when the budget crunch came, the lab's own principles gave way to financial reality. The final grant application, approved by the funder, included only the stimulation arm, with a note that sham controls would be "considered in future work." That future work has not materialized.

The consequences were immediate. The experiment ran smoothly, and the data showed a clear increase in firing rates in the stimulated region. But when the lab attempted to publish, reviewers immediately flagged the missing control. One reviewer wrote, "Without a sham group, it is impossible to rule out that the observed changes are due to the surgical procedure or the animal's expectation of stimulation, rather than the light itself." The lab resubmitted with additional analyses, including a comparison to baseline recordings, but the fundamental weakness remained. The paper was eventually accepted, but the ambiguity has followed it.

Why Controls Matter in Brain Stimulation

Placebo effects in neuromodulation are not a minor nuisance; they are a central challenge. In human studies of transcranial magnetic stimulation (TMS) and deep brain stimulation (DBS), sham conditions have repeatedly shown that patients can experience substantial symptom relief even when the device is turned off. The same is true in animal models. A rat that has undergone surgery and received an implant may behave differently simply because of the surgical trauma and the novelty of the implant, independent of any light or current. Without a sham group, researchers cannot separate these effects from the specific neural manipulation.

Blinding, the gold standard for controlling expectation, is nearly impossible in animal electrophysiology. The experimenter often knows which animals received stimulation, and the animals themselves may respond to the sound or heat of the laser. Sham conditions attempt to mimic the full experience, including the surgical procedure and the implant, but deliver no active stimulation. This is the only way to control for the myriad nonspecific effects of the experimental setup. Yet it is precisely this arm that the per-trial fee cap made unaffordable.

The consequence is a literature that may systematically overestimate the effects of brain stimulation. A 2019 meta-analysis of optogenetic studies in rodents, led by Dr. Sarah Chen at the University of California, Berkeley, found that studies without sham controls reported effect sizes roughly 30% larger than those with sham controls, after adjusting for other factors. The authors concluded that "the absence of sham controls is a major source of bias in the neuromodulation literature." The Eriksen Lab's paper, which reported a 40% increase in firing rates, falls into this pattern.

Some researchers argue that sham controls are less critical in basic science, where the goal is to understand mechanisms rather than to guide clinical decisions. "If you are asking whether a particular neuron type can drive a behavior, you can often infer that from the specificity of the manipulation," says Dr. James Whitfield, a neuroscientist at the University of Oxford who was not involved in the Eriksen study. "The sham condition is more important when you are claiming a therapeutic effect." But this view is contested. Dr. Voss herself, in a recent commentary, acknowledged that even mechanistic claims require a baseline to rule out nonspecific effects. "The sham is not just a placebo; it is the only way to know that your manipulation, and not the surgery, caused the change," she wrote.

The Economics of Experimental Design

The per-trial fee cap is one example of a broader trend in research funding: a push toward efficiency and accountability, often measured in cost per data point. Funders, under pressure from governments and taxpayers, want to maximize the return on their investment. Per-trial caps are a blunt instrument for this goal. They force labs to make trade-offs, and those trade-offs often favor quantity over quality.

The costs of an electrophysiology experiment are not trivial. In addition to animal housing and care, there are surgical supplies, fiber optic implants, lasers, recording equipment, and the salaries of technicians and postdocs. A single animal can cost several hundred dollars over the course of a study. A lab running 40 animals per experiment, with multiple experiments per year, can easily spend $100,000 or more on animal-related costs alone. The cap, set at $1,500 per trial, was intended to keep these costs in check. But it did not account for the fact that a sham group doubles the number of animals needed for the same statistical power.

Funder caps also favor larger sample sizes, which are often seen as a proxy for rigor. A study with 20 animals per group is more likely to pass a power analysis than one with 12. But this focus on sample size ignores the fact that a larger sample without a control group is less informative than a smaller sample with one. In the language of statistics, the cap trades bias for variance. A larger sample reduces random error, but a missing control introduces systematic error that no amount of replication can fix.

The Eriksen Lab's experience is not unique. A survey of 200 neuroscience labs, conducted by the International Society for Neuroethology in 2022, found that 43% had dropped a control group in the past five years due to budget constraints. Of those, 78% said the decision was driven by funder policies, not scientific judgment. The survey's lead author, Dr. Maria Gonzalez, called the trend "a quiet erosion of experimental rigor." She noted that the problem is often invisible because dropped controls are rarely mentioned in published papers. "You see the final design, not the one that was originally planned," she said.

A Case Study: The Lab's Decision

The Eriksen Lab, named after its founder, Dr. Jonas Eriksen, a former astrophysicist who turned to neuroscience, has a reputation for methodological care. Its early work on hippocampal place cells, which used a custom miniscope with a tilt-shift lens, was praised for its attention to detail. The lab's decision to drop the sham group was therefore surprising to many in the field. But the internal documents, shared with me under condition of anonymity, reveal a deliberative process driven by hard numbers.

The lab's original grant proposal included a detailed budget: $30,000 for 40 animals, including surgical supplies and post-operative care. The funder's cap would have limited this to $20,000, forcing the lab to cut 10 animals. Dr. Voss considered several alternatives: reduce the number of stimulation sessions, use a within-subject design, or partner with another lab to share costs. Each option had drawbacks. A within-subject design, where each animal serves as its own control, would require a longer experimental timeline and more complex analysis. A collaboration with another lab, which had a similar setup, was explored but fell through due to scheduling conflicts.

In the end, the lab chose to eliminate the sham group and increase the stimulation group to 20 animals. The decision was documented in a memo, which noted that "the funder's cap makes a sham group cost-prohibitive. We will proceed with a single-arm design and acknowledge the limitation in the paper." The acknowledgment, however, was brief, buried in the methods section: "Due to budget constraints, a sham condition was not included." Many readers, including reviewers, missed it.

The results were, as expected, significant. The stimulation group showed a 40% increase in firing rates in the targeted region, with a p-value of 0.01. The lab celebrated the finding as evidence that their optogenetic approach could modulate neural activity. But the ambiguity was palpable. In a lab meeting, one postdoc asked, "What if the increase is just because the animals are awake and moving more?" No one had an answer.

The paper was submitted to a well-known neuroscience journal, where it received two rounds of revision. The reviewers repeatedly asked for additional experiments, including a sham group, but the lab had neither the budget nor the time. The paper was eventually accepted at a lower-tier journal, but the reviewers' comments were scathing. One wrote, "The authors have not adequately addressed the lack of a sham control. This is a fundamental flaw that undermines the conclusions." The paper now sits in the literature, a cautionary tale for anyone who reads it closely.

Publication Pressure and Incentive Misalignment

The Eriksen Lab's experience is embedded in a broader system of incentives that often rewards positive, novel results over rigorous, confirmatory ones. Journals, particularly high-impact ones, favor studies that report significant effects. A study with a sham control that shows no difference between stimulation and sham is harder to publish than one that shows a dramatic effect. This bias, known as publication bias, is well documented. A 2020 analysis of 1,000 neuroscience papers found that studies with positive results were three times more likely to be published than those with null results, even after controlling for study quality.

Funder metrics, which often tie future funding to the number of publications and citations, exacerbate this pressure. A lab that publishes a paper with a positive result, even a flawed one, is more likely to secure the next grant. A lab that publishes a null result, no matter how well designed, may struggle. This creates a perverse incentive: labs are rewarded for cutting corners that increase the chance of a positive finding, even if those corners are the ones that make the finding trustworthy.

The per-trial fee cap is a direct manifestation of this misalignment. By making control groups more expensive, it pushes labs toward designs that are more likely to produce significant results, but less likely to be true. The Eriksen Lab's paper, with its missing sham group, is a prime example. The finding is likely an overestimate of the true effect, and a replication study, if one is ever conducted, may fail to reproduce it.

Some argue that the solution lies in changing the incentive structure itself. Pre-registration, where labs specify their hypotheses and analysis plans before collecting data, can reduce publication bias. Journals that publish null results, such as PLOS ONE and the Journal of Articles in Support of the Null Hypothesis, provide an outlet for rigorous negative findings. But these reforms are slow to take hold. As Dr. Gonzalez noted, "The system is not designed to reward rigor; it is designed to reward productivity. Until that changes, we will continue to see labs make decisions like the one at the Eriksen Lab."

What the Data Can and Cannot Say

So what did the Eriksen Lab's data actually show? The stimulation group exhibited a clear increase in firing rates, and this increase was statistically significant. But the data cannot tell us whether this increase was caused by the light itself, by the surgical procedure, or by the animals' expectation of stimulation. The lack of a sham group means that the effect size, 40%, is likely an overestimate. The true effect, if it exists, could be smaller, possibly even zero.

The lab's own analysis acknowledged this limitation. In the discussion section, the authors wrote, "The absence of a sham condition prevents us from ruling out nonspecific effects of the surgical procedure. Future work should include a sham group to confirm the specificity of the observed changes." This is honest, but it is a far cry from the confident claims in the abstract, which stated that "optogenetic stimulation of the prefrontal cortex increases neural activity."

Replication is the ultimate test, but replication studies are expensive and rarely funded. A 2021 effort to replicate 50 neuroscience studies, led by the Open Science Collaboration, found that only 40% of the original results could be reproduced. The Eriksen Lab's paper, with its missing control, is a candidate for failure. If a replication team were to include a sham group, they might find that the sham group shows a similar increase, which would undermine the original finding.

The honest conclusion is that we do not yet know whether the stimulation had a specific effect. The data are consistent with a real effect, but they are also consistent with a placebo-like response. This uncertainty is the price of the per-trial fee cap. It is a price that the lab, the funder, and the scientific community are paying, often unknowingly.

Practical Reforms for Funders and Labs

There are concrete steps that funders and labs can take to avoid this situation. Funders could separate the cost of control arms from the per-trial cap, recognizing that a sham group is not an optional extra but a necessary component of rigorous science. They could offer tiered funding, with higher per-trial limits for studies that include control groups, or provide supplemental grants specifically for sham conditions.

Funders could also encourage pre-registration and data sharing, which would make it easier for other labs to replicate findings and for the community to assess the robustness of results. The Eriksen Lab's data, for example, could be reanalyzed by other researchers, but without a sham group, the reanalysis would be limited. A shared database of electrophysiology results, including sham controls, would allow meta-analyses that could identify systematic biases.

Labs, for their part, can adopt Bayesian approaches that allow them to incorporate prior information and to quantify the uncertainty in their estimates. A Bayesian analysis of the Eriksen Lab's data, for example, might show that the posterior probability of a true effect is only 60%, given the lack of a control. This would be a more honest representation of the evidence than a p-value of 0.01.

Collaborative consortia, where multiple labs share the overhead costs of animal care and equipment, could also help. If several labs pooled their resources, they could afford to run sham groups in each experiment, or even to run multi-site replication studies. The cost per lab would be lower, and the quality of the science would be higher. The Nordic Institute, for example, could partner with other European labs to create a shared facility for optogenetic experiments, with a dedicated budget for control arms.

But these reforms require a shift in mindset. Funders must see control groups not as a cost to be minimized, but as an investment in the reliability of the science they fund. Labs must resist the temptation to cut corners, even when the budget is tight. And the scientific community must reward rigor over novelty, even when the results are less exciting. Until then, the per-trial fee cap will continue to force labs into impossible choices, and the literature will continue to be filled with ambiguous findings that cannot be trusted.

The Eriksen Lab's story is not a happy one, but it is a useful one. It reminds us that the design of an experiment is not a purely scientific decision; it is also an economic one. And when the economics are wrong, the science suffers. The next time you read a paper that reports a significant effect, ask yourself: was there a sham group? If not, be cautious. The result may be real, or it may be an artifact of the budget.

For the Nordic Research Council, the fix could be as simple as adding a line item to grant budgets specifically for control conditions, or as ambitious as creating a shared repository for sham data that labs can draw upon. For the broader scientific community, the lesson is clear: we must advocate for funding models that prioritize the integrity of the science over the cost per data point. As Dr. Voss put it in her commentary, "We cannot afford to cut the very controls that make our findings meaningful."