Placebo vs nocebo: same-same or different?

The role of expectation in diminishing pain (as we might see in the ‘placebo effect’) has been pretty thoroughly studied.  As a result of that research, we know that experiences of pain usually fall in line with expectations of pain, that expectations (and, therefore, pain) can be manipulated with either verbal information (e.g. ‘this will be less painful’) or classical conditioning – or both – and that some people are more ‘susceptible’ to these manipulations than others.  We also know that this placebo effect, when induced experimentally, tends to diminish with repetition (see Au Yeung, Colagiuri et al. 2014).

Research into increasing pain via verbal or experiential manipulation (the ‘nocebo effect’) has lagged behind research into decreasing pain, so our understanding of the nocebo effect has some gaps.  We know that here, too, experiences of pain tend to fall in line with expectations, and that susceptibility varies, but we have less insight into the relative contributions of verbal and experiential manipulations of expectation, and we don’t understand how durable the nocebo effect is.  A systematic review (Petersen, Finnerup et al. 2014) recently quantified the nocebo effect that can be elicited in the lab using verbal manipulations of expectation, finding it to be moderate to large, yet highly variable, in magnitude.  The nocebo effect was stronger when the manipulation of expectation used both verbal suggestion and classical conditioning than when it used verbal suggestion alone.  Sadly, the authors chose not to consider the magnitude of the nocebo effect that can be elicited using classical conditioning alone, leaving the question of the relative contributions of verbal suggestion and conditioning only partially answered.  However, the review clearly showed that nocebo effects can be as strong as placebo effects, although, obviously, in the opposite direction.

Another interesting study (Colagiuri, Quinn et al. 2015) used a classical conditioning manipulation combined with verbal suggestion to induce hyperalgesia, and compared the effects of using partially and fully reinforced training schedules in the classical conditioning manipulation.

Their study worked like this:

  • Participants were told they would receive a certain intervention that could increase the amount of pain that they would feel from a painful electrical zap. The intervention given was a sham – it had no real effect, but the participants did not know that.  The zap, on the other hand, was always painful, but could be either mildly painful or quite painful indeed.  Again, the participants did not know what intensity of zap they were being given.
  • Participants were divided into 3 groups and underwent a first ‘conditioning’ phase. In the conditioning phase, Group 1’s experience was that the intervention was always paired with a more intense painful zap; Group 2’s experience was that the intervention was sometimes paired with a more intense painful zap, and Group 3’s experience was that there was no predictable relationship between the intervention and the zap intensity.
  • All three groups then underwent a ‘test’ phase. Half of the test trials included the intervention and half didn’t.  Every test trial included a painful zap at the lesser intensity. Participants reported their pain on each trial.

I’ll skip past the more mundane study findings here, but I encourage you to read the full paper if you’re interested in the results of the different training schedules.  To my mind, the most interesting thing about this study was the way in which its results differed from results of similar placebo studies. In most placebo studies that use a test phase like that used in this study, participants show a strong placebo effect at the start of the test phase, but the effect diminishes fairly rapidly, probably because the relationship between the placebo intervention and the diminished pain experience is no longer being reinforced.  This reflects that humans are usually quite good at learning about relationships between stimuli (such as the intervention and the painfulness of the zap) in a flexible way.  Interestingly, that kind of flexible learning was not shown in this study – nor in another similar study of the nocebo effect (Colloca, Sigaudo et al. 2008).  In the test phase of these two studies of the nocebo effect, the increased pain to the nocebo intervention did not diminish over the course of the test phase, suggesting that nocebo hyperalgesia may be difficult or possible to extinguish – a fundamentally different pattern to that seen in placebo hypoalgesia.

As Colagiuri et al point out, it is possible that a longer test phase might have eventually elicited an extinction effect.  However, it is clear that placebo hypoalgesia is more rapidly extinguished than nocebo hyperalgesia.  Why might this be?  Let’s consider that pain is inherently unpleasant and that it is a powerful motivator of behaviour—and that it is usually assumed to signal risk to tissue integrity.  In the natural world, therefore, any stimulus that cues for pain (linked to a threat to tissue) should be avoided.  Increasing pain on repeated exposure to that stimulus could therefore be effective in motivating protective behaviour.  However, if such increasing pain does not diminish with exposure (in the laboratory, the stimulation trials continue until the test phase is complete), the goal of protective behaviour—to avoid the stimulus—has not been achieved, so increased pain (the presumed motivator of protective behaviour) is still required.

Of course, this contradicts the Bayesian idea that expectations should be updated on a trial-by-trial basis depending on the stimulation received, and the idea that, according to the expectation-experience matching idea, expectation should determine pain.  But perhaps it is also important to consider the cost of error: the cost of failing to identify a dangerous stimulus as dangerous (and therefore failing to avoid it) is arguably higher than the cost of falsely classifying a safe stimulus as dangerous.  For example, if you see a lion and fail to identify it as dangerous, you might approach it for a cuddle.  The likely cost of that mistake is far higher than the cost of seeing a domestic cat (safe – in most cases!) and identifying it as dangerous, and avoiding it in unnecessary protective behaviour.  Similarly, increased pain in response to a stimulus paired with a threatening nocebo intervention may reflect a bias toward protection of the self: in the real-world setting, ongoing hyperalgesia would presumably drive protective behaviour, which should remove the offending stimulus (the cue for tissue threat) from the organism’s immediate environment, rendering the environment less dangerous.

Perhaps, then, it is safer for nocebo hyperalgesia to be resistant to extinction.  Using the same reasoning, placebo hypoalgesia ought to be prone to rapid extinction, because of the high cost of falsely classifying a dangerous stimulus as safe.  A ‘placebo effect’ system that allows for diminished pain in response to certain cues, then, is an experientially pleasant one, but may also be an under-protective one.  Such a ‘placebo effect’ system needs to be subjugated to a ‘nocebo effect’ system, so as to retain the prerogative of protection of the self.

Tory Madden

Tory MaddenTory arrived from South Africa to start her PhD at BiM.  She is a physiotherapist who worked clinically before turning her focus toward research.  She is interested in pretty much anything related to pain and neuroscience, thanks to some particularly inspirational teaching by Romy Parker during her undergraduate training at the University of Cape Town.
Tory’s research looks at classical conditioning and pain.  She is also an associate editor for BiM.  She tries to spend much of her spare time exercising to compensate for the vast quantity of chocolate that lives in her bottom desk drawer.  Luckily, she loves trail running as much as she does food.


Au Yeung, S. T., B. Colagiuri, P. F. Lovibond and L. Colloca (2014). “Partial reinforcement, extinction, and placebo analgesia.” PAIN® 155(6): 1110-1117.

Colagiuri, B., V. F. Quinn and L. Colloca (2015). “Nocebo Hyperalgesia, Partial Reinforcement, and Extinction.” The Journal of Pain 16(10): 995-1004.

Colloca, L., M. Sigaudo and F. Benedetti (2008). “The role of learning in nocebo and placebo effects.” PAIN 136(1–2): 211-218.

Petersen, G. L., N. B. Finnerup, L. Colloca, M. Amanzio, D. D. Price, T. S. Jensen and L. Vase (2014). “The magnitude of nocebo effects in pain: A meta-analysis.” PAIN® 155(8): 1426-1434.


  1. Andrew Char says:

    Is this a typo?
    ” …hyperalgesia may be difficult or possible to extinguish…”

    Did you mean “impossible”?

    Tory Madden Reply:

    Andrew, thanks for pointing that out – yes, it was indeed a typing error! My apologies.

    Andrew Char Reply:

    Do you have plans to edit the article to reflect that?

  2. Abby Tabor says:

    Really interesting post, Tory, thank you. As a bit of a die-hard Bayesian I thought I’d offer my somewhat biased (hoho) opinion. Predictably, my main point is that these findings may not contradict a Bayesian perspective but actually strengthen its position when placed in the right light.

    I suggest that this light comes from approaching experience as a consequence of the fundamental drive to keep the self within viable bounds. These viable bounds are evolutionarily, culturally as well as socially sculpted. Rather than Bayesian brains in vats, we are embodied and embedded in the world and it is in this context that the Bayesian idea can really add some oomph. A proposed theory, Predictive Processing (see Andy Clark, 2013 and Anil Seth, 2015 for excellent accounts) combines exactly these two elements: first, the need to minimise the deviation from predicted viable bounds and second, the unavoidable nature of the brain, body and world interaction.

    From a Predictive Processing (PP) perspective, the Bayesian idea provides a probabilistic description of how uncertain information about the world (internal or external) is accommodated in relation to an evolved, embodied being. From this view, we actively attempt to predict the flow of sensory information in accordance with our long-term (e.g survival) and short-term (e.g. completing a marathon) goals. When there is a mismatch between the prediction and the actual information, there results a Prediction Error. It is in the circumstance of these errors, in largely non-conscious realms, that the embodied individual actively seeks to accommodate the information so as to act and perceive in a way that minimises the error and return the self to viable bounds. I think that this account fits the experience of pain quite elegantly.

    Pain is part of the strategy that attempts to accommodate information that violates what is predicted. But there is another element to PP, which I think allows the full description, including the asymmetrical findings of the placebo and nocebo effects that you astutely point out. This is the precision weighting component, which the Bayesian idea really accentuates. That is, rather than all information having the same influence on perception, information is weighted depending on its reliability (inverse variance). Error and its variable nature is thus in-built in the PP account. From this perspective, nociceptive information is salient and relatively reliable, it is weighted highly and can therefore have large effect on experience (depending on the context).
    Consequently, the nocebo effect would be resistant to diminishment because of the evolved, high precision weighting of nociceptive information that requires accommodation and action in order to mitigate or avoid. Part of this accommodation involves an updated prediction that we are threatened; formed under such highly precise prediction error signal this prediction is definitely difficult to extinguish. As such we require new information that is precise enough to be weighted over our current prediction about the world. On the other hand, the placebo effect is indeed more easily extinguished due to the very same strategy of precision weighting. The current prediction (one associated with safety) is violated, and due to the nature of the new information (salient/reliable) it results in a highly precise prediction error, which requires rapid accommodation and updated prediction (one associated with threat).

    I believe that PP provides a very comprehensive and testable account of experience, including but not limited to pain. It incorporates the fundamental drive for survival just as easily as the more short-term motivations, without appealing to a central controller in the brain (I can’t do the embodied literature justice here, if interested please read Francisco Varela’s work on The Embodied Mind). It has the Bayesian idea at its core, but this is not a static information processing account, this is an all singing, all dancing embodied and embedded approach.

    But, then again, I am biased.

    Tory Madden Reply:

    Abby, thank you for this carefully articulated comment – I’m glad you weighed in here. It’s very interesting to read about this PP perspective, and as you say, it does explain the findings nicely. Clearly my understanding of ‘Bayesian’ ideas needs expanding… thanks for expressing your very useful ‘bias’!

  3. A bit more…

    The cost of error for the client mentioned is severe pain, incapacitation, repeat surgery and all the hell that this involves. Mowing the lawns could could end very badly for her. It’s an enormous cost. So her mind is right to at least be considering it.

    We can juxtapose this with the cost of holding a cautious attitude. We often overlook this cost. The cost is pain. *The cost of an overly cautious attitude is pain*. It looks like a stalemate…

    By adding in some education relating to the structural soundness of her spine, and the way the mind is responsible for the pain experience, now the scales start to shift. The decision to let go is made easier.

  4. I think the ‘cost of error’ concept describes well the differences seen in extinction. How powerful and pervasive is this script: “What if I’m wrong and I get hurt again?”. The way the mind grips onto negativity is a source of endless frustration.

    In treatment, I like to recognize this negative expectation (rather than trying to fight or suppress it) and make it clear that the attitude itself is contributing a large part to maintaining the pain.

    This is a case I had recently: “I want to mow the lawns, but what if I’m wrong and end up needing surgery again?”.

    — recognize the client’s fear and accept it fully
    — don’t offer a solution, because solutions imply non-acceptance
    — purposefully and wilfully experiment with intensifying the fear
    — purposefully experiment with hating the pain
    — experiment with trying to block the pain from conscious awareness

    Now notice how your symptoms are intensified when you obey your mind’s directive for safety. Allow the penny to drop. Don’t force an understanding.