Experiment 1

In this study, we aimed to provide evidence for habitual operant responses using the contingency degradation paradigm. We employed a novel design inspired by Vagui et al [26], marking the first attempt to examine the development of persistent, habitual behavior after overtraining with a positive R-O contingency.

To assess contingency sensitivity, we used the standard measure [27]. This measure represents the difference between 1, the conditional probability of receiving an outcome following an action, , and 2, the probability of receiving an outcome in the absence of that action .

If this procedure effectively captures habitual behavior in humans, extended training with a positive contingency should promote habit formation, which might compete with goal-directed responses when the two are misaligned. Consequently, we would expect greater difficulty in adapting responses to contingency changes (from positive to negative) after overtraining.

Method

Transparency and openness.

All experimental studies reported here were pre-registered before data collection began. To ensure transparency and replicability, we provide detailed information on sample size determination, data exclusion criteria, experimental manipulations, study measures, and the year of data collection for each experiment [28]. All documents, data, analysis scripts, and materials employed are publicly available at https://osf.io/4tydz (Experiment 1), https://osf.io/krb42 (Experiment 2), and https://osf.io/9wguh/ (Experiment 3).

Participants.

Because the effect that we were looking for has not been reported previously, we did not have a previous effect size to estimate the appropriate sample size. Thus, we followed the calculations of Brysbaert’s [29] study, which indicates how many participants are needed to have properly powered experiments, in absence of previous similar effect. This author signaled that d = .4 is a good estimate of the smallest effect size of interest in psychology research and that over 50 participants are needed for a simple comparison of two within-participants conditions (in our case, standard vs overtraining) if we want to run a study with 80% power and an alpha level of.05. A total of 58 second-year university students participated in the experiment, with 52 completing all three consecutive sessions in exchange for course credits. In addition, the six highest-scoring participants at the end of the experiment received an extra payment of 25€ each. Participants were informed of this incentive before the experiment and provided written consent. The task included a manipulation check (a memory test, explained below) to ensure participants followed instructions. Only those who achieved at least 90% accuracy on this test were included in the analysis. The final sample (n = 50) consisted of 38 females and 12 males (age range = 19–34 years, M = 20.26, SD = 2.74). Participants completed the task in a quiet room with 10 semi-enclosed cubicles, each equipped with a standard PC and a 55 cm monitor positioned approximately 80 cm from the participant. This study was approved by the Human Research Ethics Advisory Panel (Psychology) of Universidad de Málaga (46–2020-H) and complied with the Helsinki Declaration. All participants reported normal or corrected-to-normal vision. Data collection took place from May 16–18, 2023 (our laboratory at full capacity allows to test up to 70 participants per day, approximately).

Stimuli and task.

Participants were trained on several S-R-O contingencies to assess their ability to adapt their responses to subsequent contingency changes. Three figures—a circle, a square, and a triangle—were used as cues, with each figure randomly assigned to a condition (see Fig 1). In each block, one of these figures (approx. 10˚ diameter) appeared at the center of the screen in white, signaling that a response could be made. If a response was registered, the figure turned yellow to prevent multiple responses within the same 1-second time window. Each cue was associated with a specific key response, which participants were reminded of at the start of each block. Outcome delivery was represented by an image of two piles of yellow coins appearing on both sides of the geometric figure for 0.5 secs. Additionally, a continuously updated total score was displayed in the top right corner of the screen. The task was presented on a gray background and programmed using PsychoPy, version 2024.1.4 [30].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The contingency degradation task.

Fig 1. Example of the stimuli displayed in the contingency degradation task. Participants were required to respond to each cue in alternating blocks of 120 secs. Positive contingency blocks were shown during training, where rewards were contingent upon making a response. Habit tests, or negative contingency blocks, were programmed by increasing the number of rewards not associated with a response, making non-responsiveness more rewarding than responding in these blocks.

https://doi.org/10.1371/journal.pone.0334087.g001

Design and procedure.

We used a within-subject design to study the effect of overtraining on habit formation by manipulating the amount of experience with each cue, specifically by increasing the number of blocks for the overtrained cue. The experiment took place across three sessions on consecutive days. Following Vaghi et al. [26], we employed a discriminated free-operant, self-paced procedure in which participants learned and determined the response frequency that maximized their earnings in each block. Each block was divided into 120 seconds of free responding. Two different contingencies were programmed: a positive contingency during training (higher probability of obtaining a reward when a response was made than in the absence of a response) and a negative contingency (lower probability of obtaining a reward when a response was made than in the absence of a response) to test for habit formation (see Fig 1 and Table 1). These two contingencies were used consistently across all experiments. For brevity, we will refer to them as positive and negative contingencies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Design and distribution of blocks in Experiment 1.

https://doi.org/10.1371/journal.pone.0334087.t001

In the first session, the three cues were presented in two separate blocks with a positive contingency. One of these cues (filler cue) was then tested in two blocks with a negative contingency. This filler cue was included to assess whether, after brief training, behavior remains goal-directed and sensitive to a sudden change in contingency. The overtrained cue was presented across all three sessions, with a total of 24 training blocks involving a positive contingency. The standard (less trained) cue was also presented across the three sessions, but only in 6 positive contingency training blocks. Both cues were tested in the third session, using two additional blocks of negative contingency for each cue. The negative contingency blocks were employed as habit tests, measuring to what extent participants can adjust their response to a change in R-O contingency [16]. At the end of each block, participants completed two tests: a causality judgment and a memory test. These tests aimed to rule out the possibility that any difference between conditions was due to a lack of attention or misunderstanding of the task. In the causality judgment, participants were asked to rate the extent to which their response produced or prevented the outcome, using a Likert scale from 100 to −100 (see specific wording in S1 Appendix, in Supporting Information S1 File). They used the mouse to select the number that best represented the contingency experienced during the last block. Following this, a memory test was administered to ensure participants had been paying attention to the cues. In this test, all cues appeared on the screen, and participants were required to identify the cue displayed during the last block by using the mouse to respond (for more details, see the pre-registration and materials employed at https://osf.io/4tydz).

Before beginning, participants were provided with detailed instructions for the task, explicitly stating that the outcome could occur even in the absence of any response. Allan and Jenkins [27] found that participants are more accurate in judging the contingency between responses and outcomes when they are aware that not responding is also a correct alternative. We made this aspect of the task explicit through the instructions, and a reminder of the instructions was presented at the start of Sessions 2 and 3. Sessions 1 and 3 each lasted approximately 25 minutes, while Session 2 lasted 50 minutes. Participants were not informed in advance about when a contingency change would occur. The order of test blocks for the standard and overtrained cues was randomized.

Data analysis.

For the main analysis, a response ratio was calculated for each cue using the mean number of responses per 1-second bin in each block. To assess changes in behavior during the test blocks, we computed a new ratio comparing the test blocks to a baseline, which was derived from the last two training blocks (see Formula 1). If response frequency remains unchanged from training to test, this ratio will be approximately 0.5, indicating habitual responding during the test. Conversely, a decrease in responses during the test — characteristic of goal-directedness —will result in values closer to 1.

Formula 1. Response ratio:

Following the pre-registered analysis plan, we compared changes in the response ratio for each cue after moderate and extensive training using one-tailed t-tests. When necessary, t-tests for unequal variances were applied.

Both frequentist and Bayesian analyses were carried out using R 4.3.0 [31]. For Bayesian analysis, we used the default Cauchy prior of 0.707, a commonly used standard (e.g., JASP, 2023). Additionally, we conducted Bayes factor robustness checks in JASP 0.19.2 [32] using a wide range of possible priors (see S2 Fig in the Supporting Information).

Results

The results displayed in Table 2 indicate that the task worked as intended, with participants experiencing contingencies closely matching the programmed ones (see also S1 Fig in the Supporting Information S1 File). This was confirmed by a correlation analysis between programmed delta P and experienced delta P, r = .867, p < .001. As it can be seen in S1 Fig in S1 File and Table 2, mean Causality ratings suggested that participants, as a group, were aware of contingency shifts, correctly identifying changes from positive to negative contingencies. However, exploratory individual-level analysis of these ratings showed that there were wide differences between participants in their perception of the new contingencies (see more details in S3,S4 Figs, S1,S2 Tables in the supporting information S1 File and General Discussion).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Contingencies, response rates, and causality ratings in Experiments 1, 2, and 3.

https://doi.org/10.1371/journal.pone.0334087.t002

As expected, the response ratio analysis during the habit tests showed that participants were sensitive to contingency degradation both at the end of the first day and after limited training with the filler, t(49) = 6.79, p < .001, d_z = 0.97, and the standard cue, t(49) = 10.20, p < .001, d_z = 1.46, on the third day. Bayesian factor analysis provided strong evidence for these effects (BF₁₀> 100).

However, contrary to our predictions, sensitivity to contingency degradation was similar for standard and overtrained cues [t(49) = −0.08, p = .53, d_z = −0.01, BF₀₁ = 1/0.14 = 7.14], indicating that the null hypothesis was seven times more probable than the alternative. Additionally, we compared the response ratio of the filler cue (tested during the first session) with that of the overtrained cue (tested during the last session). Once again, we found no significant differences in the habit test between limited and extensive training conditions [t(49) = −2.33, p = .98, d_z = −0.33, BF₀₁ = 1/0.04 = 25], providing strong evidence in favor of the null hypothesis.

These findings do not support our main hypothesis that extended training might hinder the ability of participants to adapt their behavior to contingency changes. Instead, participants demonstrated a clear ability to judge and adjust their responses according to task demands, showing no evidence of habit formation.

Exploratory analysis.

In contrast to session 1 and 2, we alternated blocks with positive and negative contingency during session 3 (see Table 1). This aspect of our design could have encouraged participants to keep goal-directed control during that session, improving their capacity to hold their responses during negative blocks. To rule out the possibility that our results were influenced by participants completing two separate blocks with negative contingency for standard cues and two additional blocks for overtrained cues, we conducted an additional analysis. Specifically, we examined whether experiencing a second negative contingency block (habit test) improved participants’ sensitivity to contingency changes. For this, we compared performance between participants considering only the first habit test, analyzing standard cues (n = 27) and overtrained cues (n = 23). Although the difference was not significant [t(48) = 1.572, p = .062, d = 0.46, BF₁₀= 1.39], there was a trend suggesting that responses to overtrained cues were more habitual than those to standard cues (see Fig 2). However, since this analysis was not part of our initial plan, these findings should be interpreted with caution.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Ratio score for overtrained, standard, and filler cues in all experiments.

Note. The ratio scores for each experiment during the test phase are displayed in columns. Within-subject (A, C, E) and between-subject comparisons (B, D, F) are shown in rows. According to the formula applied to calculate the ratio, scores closer to 1 indicate goal-directed behavior, whereas scores closer to 0.5 reflect habitual behavior (persistence of response despite the negative contingency).

https://doi.org/10.1371/journal.pone.0334087.g002

Experiment 2

In the second experiment, we aimed to eliminate potential confounding effects from conducting a habit test before the final habit test. To address this, we made several modifications to our procedure. First, we removed all trials involving the filler cue to prevent any habit test from occurring in the first session. Additionally, we only conducted habit tests during the third session, with a counterbalanced order (see next section). With these changes, we expected our primary measure —the response ratio— to be more sensitive in detecting persistent responses despite contingency degradation.

Experiment 3

In Experiments 1 and 2, we found no evidence of habit induction, as participants’ performance in the test blocks remained unaffected by the amount of prior operant training. However, an exploratory analysis in Experiment 1 suggested a trend toward more habitual responses (ratio score close to 0.5) with overtrained cues than with standard cues (see Fig 2). This trend could either be statistical noise or a genuine overtraining effect indicative of a new habit. Because overtraining effects have not been previously reported using a contingency degradation protocol, we did not have enough data to calculate an appropriate power analysis for Experiment 1 and 2. That means that the trend observed in Experiment 1 could have been significant with a larger sample size. To investigate this possibility further, we conducted Experiment 3, a higher-powered replication of Experiment 1. Notably, we pre-registered the between-subject analysis that previously showed a promising trend in Experiment 1.

General discussion

The transition from goal-directed responses to automatic, habitual behavior is thought to depend on the amount of experience in a specific context. As exposure increases, S-R associations are strengthened, dissociating the action from its outcome representation [2]. Once the habit has developed, associated cues may trigger the overtrained response even when it no longer leads to the desired outcome [5,16]. Animal research has demonstrated habit formation using outcome devaluation and contingency degradation tests, with habitual behavior emerging as a function of training [3]. Human studies have attempted to replicate this pattern of results by using the outcome devaluation test, but the findings have been inconsistent [13–15,34,35]. Surprisingly, few studies have assessed whether contingency degradation tests can effectively track habit development in humans. The present experiments aimed to address this gap, providing knowledge of how humans respond to contingency degradation following varying amounts of operant training.

In three pre-registered experiments, we found no evidence that extended training led to habitual responding in a contingency degradation test. Despite three days of overtraining under a positive contingency (where pressing a key led to an increase in the probability of the reward), participants reduced their response rate just as much as those in a less-trained control condition when the contingency shifted to negative (where pressing the key reduced reward probability). We manipulated training duration within subjects by increasing the number of training blocks for one S-R-O mapping—a strategy commonly used in human habit research (e.g., [36]). Importantly, the same null effect was observed in between-subject comparisons, even when analyzing only the first test block. This rules out the possibility that experiencing multiple test blocks increased sensitivity to contingency changes and masked habit expression.

Importantly, causality ratings and memory test results indicated that most participants understood the task requirements and accurately detected changes in contingency relationships. Thus, our findings suggest that participants’ behavior in this task remained goal-directed, regardless of the amount of prior operant training.

An alternative explanation of these null results is that, in this paradigm, habits might develop even after short training, making it difficult to detect differences with overtraining conditions. This possibility seems to be supported by the data from some of the standard cues during the habit test. For instance, we can see some participants with ratio scores very close to 0.5 in Experiment 2, even for the standard cue (see Fig. 2). Although plausible, there might be a simpler explanation; some participants might not be as good at detecting contingency changes as what is indicated by level group analysis. Obviously, if a participant is still under the impression that the response is a good way to get the reward, they will continue responding—but these ‘perseverant’ responses are goal-directed actions and not habits. For assessing this possibility, we analyzed the correlation between ratio score and causality ratings during blocks with negative contingency (habit tests). As we suspected, we obtained a significant negative correlation in all experiments for all cues (see S3,S4 Figs and S1,S2 Table in S1 File), indicating that those participants with higher perception of causality in negative blocks—they still believe that their behavior cause the outcome—also showed lower punctuations in ratio score, that is, closer to 0.5 (indicative of habits). However, those seemly habitual responses from participants who are not aware that the contingency changed are in fact goal-directed actions. Interestingly, similar results have been observed in outcome devaluation protocols: that is, seemingly habitual responses which are the consequence of suboptimal outcome devaluation at least for some participants [35,37]. All these results highlight the need for a careful examination of the functioning of devaluation/degradation protocols at the individual level before making claims of evidence for habits, especially after short training.

Overall, our results do not support the validity of the methodological approach used in these experiments for studying habit induction in humans. This has implications for previous research that has employed similar designs —namely, a free-operant training stage followed by a contingency degradation test— to investigate habit formation. For example, Vaghi et al. [26], following the approach of Liljeholm et al. [38], used this paradigm to compare performance between OCD patients and healthy controls. They found that OCD patients exhibited greater response persistence during degraded contingencies than controls despite both groups being equally aware of the contingency changes. The authors concluded that OCD patients show a dissociation between action-outcome knowledge and habitual responding, suggesting a dysregulation between goal-directed and habit systems in OCD. This aligns with broader theories proposing that transdiagnostic impulsivity and compulsivity stem, at least in part, from an imbalance between these two systems ( [11,12,39,40]; but see [41] or [42]).

Our experiments differed from previous studies using contingency degradation tests in several ways. Prior research [21,26] introduced frequent shifts in response-outcome contingencies. For instance, Vagui et al. employed twelve 2-min blocks with different contingency relationships in a single 34-minute session, covering six conditions (two positive, two negative, and two zero-contingency). However, habit learning is known to require stable contexts [43], and frequent contingency changes could interfere with this process. To address this, we deliberately minimized contingency shifts, implementing only a single change in the between-subjects analysis to better facilitate habit formation.

Our experiments also differ from previous studies in their interpretation of persistent responding when a positive contingency is weakened but remains positive (for instance, from to ), assuming that this constitutes evidence for habitual control of behavior. However, there are reasons to be cautious about that assumption. When the cost of responding is zero, maintaining a high response rate remains the optimal strategy for the participant as long as is positive—even if it has recently transitioned from a stronger positive contingency. In other words, persistent responding when is reduced but still positive may reflect goal-directed behavior aimed at maximizing rewards. In studies such as Vagui et al. [26], more convincing evidence for dysregulated habit expression in OCD would have been a higher response rate during negative contingency blocks compared to healthy controls. However, OCD patients primarily showed increased responding during low-but-positive contingency blocks—an approach still aligned with goal-directed reward maximization. To avoid this potential confound, our habit tests exclusively used blocks where contingency was reduced below zero, ensuring that continued responding was objectively counterproductive for participants.

Recently, Nebe et al. [21] tested participants using a modified version of the contingency degradation procedure from Vaghi et al. [26], extending the training to four days. However, they found no increased insensitivity to contingency changes between the first and last training sessions in a within-subject comparison. In other words, similar to our findings, they failed to observe habit expression as a function of the amount of training. While Nebe et al.’s study is informative, it shares a key limitation with previous research: as in Vaghi et al. [26], contingency relationships changed frequently throughout training. Such conditions likely encouraged participants to maintain goal-directed control over their behavior, potentially explaining the lack of a clear effect of overtraining on habit expression.

We believe that our methodological approach provides a more valid framework for studying habit formation. By maintaining stable contingencies, we aimed to foster habit development, and by testing habits under negative contingencies, we minimized potential confounds with goal-directed behavior. To assess the effect of training duration, we conducted both within- and between-subject comparisons. Each participant completed two contingency degradation test blocks—one for the overtrained cue and another for the standard cue—allowing us to directly compare the effects of training within subjects. Additionally, to isolate the impact of training between subjects, we analyzed only the first test block for each participant. This ensured that participants in the between-subjects analysis had not encountered any contingency shifts before testing, maintaining a stable learning environment across the three training days.

Our learning task allowed participants to reach a high-performance (asymptotic) level. Causality ratings during training and test blocks were highly accurate, confirming that participants learned all the contingencies and recognized the changes between test conditions. Importantly, ratings during test blocks were, on average, negative, indicating that participants understood that responding reduced the likelihood of receiving a reward (see Table 2). Participants’ operant responses were aligned with these ratings, as they consistently reduced response rates for both overtrained and standard cues. Crucially, this reduction was of the same magnitude across conditions, strongly suggesting that the few responses observed during the contingency degradation tests were not true habits but rather goal-directed actions—possibly driven by boredom or lack of attention—as suggested by the differences in their perception of the negative contingencies (see S3,S4 Figs and S1,S2 Tables in S1 File).

One potential counterargument is that the induced habit was not strong enough to be detected in the contingency degradation test after three days of training. While we cannot determine with certainty what would have occurred with extended training, several factors suggest that our findings would remain unchanged. First, response rates for the overtrained cue reached an asymptotic level after just three blocks (see S1 Fig, in Supporting Information S1 File). Second, previous research has demonstrated that habit expression does not markedly increase when training is extended from a few days to several weeks [44]. Third, analyses of response times in goal-directed tasks suggest that S-R habits can form within three days of training, though not all measures are sensitive enough to detect them [45]. Taken together, these observations suggest that the contingency degradation test lacked the sensitivity required to detect habits formed over three days of training. However, other work has reported high variability in participants regarding the amount of practice necessary to develop habitual behavior [46], so it would remain possible that contingency degradation paradigms might need more extensive training to detect overtraining effects.

Thus, our results suggest that the contingency degradation test is not a more effective strategy for studying habit formation than the outcome devaluation procedure, at least with the parameters that we have employed. Regardless of the chosen method, detecting habit induction as a function of training duration remains challenging, which raises concerns about the validity of these measures. Recent research has attempted to address this issue, identifying key factors influencing habit expression. One such factor is whether goal-directed control during testing involves actively responding or withholding a habitual response. Du and Haith [47] demonstrated that when the decision is simply to respond or not, it remains easily controllable and may require an extensive amount of training—if it ever transitions into a habit. The attentive reader might have noted that our contingency degradation test required participants to withhold an overtrained response, precisely the type of behavior that appears resistant to habitual control. This aligns with Du and Haith’s hypothesis [47], providing a plausible explanation for our difficulty in detecting habit expression in the overtraining condition.

Even when habit and goal-directed systems compete, human participants typically adapt their responses according to goal-directed processing, regardless of the amount of prior training. Studies have shown that impairing goal-directed control is often necessary to reveal habit-driven behavior [44,45,47,48]. A successful approach involves limiting response preparation time [44,45,47]. For example, Hardwick et al. [44] used a forced-response test with new stimulus-response mappings following extensive training on a different, partially incongruent configuration. Habitual errors increased when participants had only 300–600 ms to prepare their response, but not with longer preparation times. Similarly, Luque et al. [45] found that, after three days of operant training, participants successfully selected goal-directed responses during an outcome devaluation test. However, under time pressure (≤500 ms), switch costs were higher after overtraining, indicating competing responses. This interference disappeared when the response time was extended to 4 seconds [45]; see also [21,49].

Cognitive control is impaired by stress. Experimentally induced stress has been suggested to facilitate the shift from goal-directed control toward more rigid habits when using outcome devaluation procedures after instrumental learning [48,50] (see also [51], for similar results with contingency degradation). EEG data from Meier et al. [52] indicate that stress might enhance stimulus-response processing while weakening outcome representations. However, most studies on stress and habit formation rely on single-session, two-choice response tasks, which may overestimate habit effects by failing to distinguish them from goal-directed errors. This limitation could be addressed by including ‘minimal training’ control conditions, where goal-directed errors occur but habits have not yet formed. Such an approach could provide an interesting path for future research on the effects of stress on habit induction. Other recent approaches exploring overtraining effects in humans are those employing a mobile-phone app methodology [53,54], trying to detect habit expression in a more naturalistic environment. This could be a promising path; however, more research is needed to corroborate if these procedures are robust tools for indexing habit formation (for reliability test-retest studies of Hardwick and Luque tasks see [55]).

Taken together, these findings suggest that uncovering habitual behavior requires impairing goal-directed control while also ensuring that one condition involves significantly more training than another to distinguish habitual from goal-directed responses. Future studies using a contingency degradation paradigm could enhance test sensitivity by encouraging rapid responses, such as offering reward bonuses for faster performance during training and testing or imposing time constraints on responses.

Difficulties in translating animal models to human behavior are common in behavioral and neuroscience research. In some cases, this is due to methodological issues in animal studies, such as small sample sizes or the failure to report sufficient procedural details, which hinder the replication of these effects [56]. However, we do not believe this to be the case in habit research. Since Adams’ [4] seminal work, animal studies have consistently shown that the transition from goal-directed control to habitual behavior depends on the amount of training, as demonstrated through outcome devaluation and contingency degradation tests [5,16]. These effects have been widely replicated [6,43,57]. Additionally, lesion studies in rats have identified distinct neural circuits underlying habitual and goal-directed actions (for reviews, see [3,58]). Why, then, is it so difficult to observe overtraining effects in humans? The answer likely lies in the need to refine experimental parameters to optimize habit detection in human participants. While we are making progress in developing better tools for studying habits, the precise conditions that facilitate habit expression in humans remain unclear. In our protocol, we kept a constant positive response-outcome contingency to facilitate the development of persistent responses, before introducing a change in contingency relationship, sharing that aspect of our design with classical animal studies which reported overtraining effects [5,16]. However, participants had no problem adjusting their response despite the amount of training. Interestingly, even in animal research, methods are still evolving, with improved protocols emerging to enhance the study of habitual behavior (e.g., [59,60]).

As part of this effort, the current experiments aimed to evaluate the validity of response rates during contingency degradation as a measure of habit expression in humans. This type of research is crucial for the habit learning field, especially as new tools for studying habits continue to emerge. Before assuming that a measure reliably captures habitual behavior, it must first be validated. This precaution is particularly important when the measure analyzes individual differences, such as in studies involving special populations. Without proper validation, researchers risk falling into a circular explanation. For example, one might use a contingency degradation test to study habits in a population, such as OCD patients, under the assumption that they have an altered habit system. Then, the validity of the test as a measure of habit expression is inferred from the observed differences between OCD patients and healthy controls.

In this study, we used the amount of training to investigate whether a measure truly captures habit expression. This approach has been highly effective in identifying factors and methodologies that facilitate habit detection. So far, evidence indicates that the best strategies for impairing goal-directed control and allowing habits to emerge involve forcing quick responses [44] or analyzing response time costs [45]. Stress-induction procedures and clinical studies (which assume that certain populations may have a predisposition toward habit formation) may also offer insights into habitual behavior. However, in our view, these studies should not bypass validation efforts—such as systematically manipulating the amount of training—to confirm that their measures are genuinely indexing habits.

In conclusion, observing habits in human research remains a challenge, particularly when using classical procedures such as outcome devaluation and contingency degradation tests. Unlike in animal research, there is no clear evidence that humans become insensitive to contingency degradation after overtraining an instrumental response. This calls into question the utility of this test for characterizing habit formation and expression in humans. Moving forward, a collective effort is needed to refine experimental protocols while rigorously validating them. The current research contributes to this effort by reinforcing the importance of training manipulations as a means of evaluating habit measures.

Supporting information