Sunday, April 28, 2013

Want to remember something? Clenching your fist doesn't help!

Image Credits: fist and brain.

You might have seen this news story the other day:
Want to remember something? Clench your fists!

Giving a speech and need to remember what to say? Just clench your right fist while rehearsing. Then, when it's time to give the speech, clench your left fist, and voila, you’ll recall what you rehearsed! That's what a new study found, which was published April 24 online at PLOS ONE

Sounds too easy now, doesn't it? And if you're exclaiming, "that's just too good to be true!" then you'd be correct.

The new study by Propper et al. (2013) has unleashed a torrent of criticism on Twitter, including this starter by @js_simons.

What motivated such a study in the first place? I'll try to run through the authors' rationale here, starting with statements from the abstract, which are followed by my commentary.

  • Unilateral hand clenching increases neuronal activity in the frontal lobe of the contralateral hemisphere.
It's true that unilateral hand movement is executed via motor cortex activity in the opposite hemisphere, so the right hemisphere controls the left hand and vice versa.

  • Such hand clenching is also associated with increased experiencing of a given hemisphere’s “mode of processing.”
This statement is based on EEG studies that have looked at alpha power suppression recorded at scalp electrodes over left and right frontal cortex (Harmon-Jones, 2006). The hypothesis is that left hand contractions "activate" (i.e., suppress alpha waves in) the unhappy right hemisphere, thereby producing negative affect, while right hand contractions activate the happy left hemisphere, which results in positive affect. The affective "modes of processing" aspect of this research isn't directly relevant to the Propper et al. (2013) paper, and further discussion is beyond the scope of this post. I'll just say that attributing EEG activity to a specific cortical region is a dicey proposition, because the spatial resolution of the technique isn't great.1

  • Together, these findings suggest that unilateral hand clenching can be used to test hypotheses concerning the specializations of the cerebral hemispheres during memory encoding and retrieval.
Here the EEG research on emotion is being applied to memory.

  • We investigated this possibility by testing effects of unilateral hand clenching on episodic memory. The hemispheric Encoding/Retrieval Asymmetry (HERA) model proposes left prefrontal regions are associated with encoding, and right prefrontal regions with retrieval, of episodic memories.
The Hemispheric Encoding/Retrieval Asymmetry (HERA) model of Tulving et al. (1994) postulates that the left prefrontal cortex encodes information into memory, while the right prefrontal cortex retrieves information from memory. This was back in ye olden days of PET using block designs with 40 seconds of one condition subtracted from 40 seconds of another condition. In other words, poor temporal resolution.

The HERA model was revisited and confirmed by its proponents using fMRI data (Habib et al., 2003), but the evidence against it was considerable (Owen, 2003). The general consensus is that HERA has been discredited. In fact, noted memory researcher Dr. Jon Simons posted a comment at the PLOS ONE website explaining why the underlying hypothesis of Propper et al. is problematic (among other issues).

  • It was hypothesized that right hand clenching (left hemisphere activation) pre-encoding, and left hand clenching (right hemisphere activation) pre-recall, would result in superior memory. 
Here we're expecting to see better memory in the R/L condition than in the control condition. There is no mention that the other fist-clenching conditions would result in worse performance than in the control condition.

  • Results supported the HERA model.
Results did NOT support the HERA model, and I'll explain why below (and you can read the PLOS ONE comment).

In the experiment, participants studied a list of 36 words, engaged in a filler task, and then recalled as many words as possible. Approximately 10 subjects participated in each of 16 conditions, only five of which are reported in the paper. These involved squeezing a small pink ball in one hand (2 sets of 45 sec) before the encoding and the retrieval phases of the study. The control condition did not involve clenching, but the participants held a small pink ball in each hand.2

The five conditions are shown below, named by the hand used during encoding/retrieval. You'll notice that the number of participants in each group (n) is pretty small. I calculated standard deviations from the standard error values to determine effect sizes using this effect size calculator. 3

Although the authors reported the total number of words written down (correct or not) and the number of correct words in Figs. 1 and 2 respectively, the important result is shown in Fig. 3, which takes into account the false alarms, or incorrectly recalled words.

Figure 3 (Propper et al., 2013). Corrected scores as a function of hand clench condition. [NOTE: NENR = None Encoding/None Recall, or control.]

The one-way ANOVA for this comparison "did not reach traditional significance" (p=.08), but two of the post hoc comparisons did (uncorrected for multiple comparisons involving 16 groups). The p<.09 bar in the figure is in the wrong place. Below is a table I made using the effect size calculator (ESC) for Cohen's d, compared to what was reported in the paper.

The key HERA condition (R/L) did not differ from the control condition, so the predicted hand clenching improvement in memory did not materialize. The superiority of R/L over the other two clenching configurations was due to worse performance in the latter. In other words, if you squeeze a ball with your left hand before encoding, you'll do worse than if you didn't (all statistical objections aside) and the L clenching before retrieval didn't help. The authors stated otherwise, however:
Individuals who encoded language-based information immediately following right hand clenching (left hemisphere activation), and recalled such information immediately following left hand clenching (right hemisphere activation), demonstrated superior episodic memory compared to the other hand clenching conditions. It is noteworthy that this condition was also superior to the no hand clenching control condition, though not significantly so.

The difference between the two rightmost bars in Fig. 3 above is not terribly close to being significant (as far as I can tell), so the major hypothesis was not supported here.

Clench Your Fist to Get a Grip on Memory? I don't think so.

Thanks to blog commenter Lew for first pointing out this study.

ADDENDUM (May 1, 2013): An Erratum to Figure 3 has been posted in the Comments section of the PLOS ONE article. It might have been in response to my comment in a previous thread, but this comment wasn't addressed directly.

ADDENDUM #2 (May 5, 2013): There has been a formal correction to Figure 3, which can be downloaded here (as a TIF). This doesn't affect any of the other points I made in this post, which were never addressed.


1 The fist-clenching activation of motor cortex is supposed to spread to dorsolateral prefrontal cortex, or so it goes (Harmon-Jones, 2006).

2 @neuromusic noted that the authors measured ear temperatures: "Immediately following pre-clenching condition, participants’ ear temperatures were taken (to be reported elsewhere)..." This sounded a little bizarre, but I found this publication by Propper and Brunyé, Lateralized difference in tympanic membrane temperature: emotion and hemispheric activity, which must explain the concept.

3 @rogierK thought the reported effect sizes were way too large.


Habib R, Nyberg L, Tulving E. (2003). Hemispheric asymmetries of memory: the HERA model revisited. Trends Cogn Sci. 7(6):241-245.

Harmon-Jones E. (2006), Unilateral right-hand contractions cause contralateral alpha power suppression and approach motivational affective experience. Psychophysiology 43(6):598-603.

Owen AM. (2003). HERA today, gone tomorrow? Trends Cogn Sci. 7(9):383-384.

Propper, R., McGraw, S., Brunyé, T., & Weiss, M. (2013). Getting a Grip on Memory: Unilateral Hand Clenching Alters Episodic Recall PLoS ONE, 8 (4) DOI: 10.1371/journal.pone.0062474

Tulving E, Kapur S, Craik FI, Moscovitch M, & Houle S (1994). Hemispheric encoding/retrieval asymmetry in episodic memory: positron emission tomography findings. Proceedings of the National Academy of Sciences of the United States of America, 91 (6), 2016-20 PMID: 8134342

Subscribe to Post Comments [Atom]


At April 28, 2013 5:16 AM, Anonymous Anonymous said...

This is a great analysis, thank you. Have you thought about posting it as a comment at PLoS?

At April 28, 2013 5:31 AM, Blogger The Neurocritic said...

Thanks. Can I register as first name 'The' last name 'Neurocritic' at PLoS?

At April 28, 2013 8:59 AM, Anonymous Anonymous said...

nice post, thanks.
The first point on the misuse of eeg source localization is correct, but not for the reason advanced here. Source estimation of scalp-recorded eeg has reached an good spatial resolution from several years already (e.g. PMID: 15351361). That said, what can definitely not be done, and will always be wrong, is to infer the localization of neural sources based on where on the scalp the effect manifested: electrode locations is not equal to brain region: having an EEG effect on a frontal electrode does not mean that the underlying effect comes from frontal sites.

At April 28, 2013 3:57 PM, Blogger trisbek said...

If Neuroskeptic can publish papers as such

you can certainly send a commentary to PLoS as Neurocritic. At this point I think you should. and PLoS will ask for a response from the authors.

good luck,


At April 29, 2013 1:27 AM, Anonymous Anonymous said...

Yes, you certainly can submit a commentary to PLoSONE as neurocritic!

At April 29, 2013 4:53 AM, Blogger Rogier said...

To clarify: My comments about effect sizes concern the danger of inferring effect sizes with such small sample sizes. Based on Cohen's admittedly arbitrary criteria, effect sizes of .3 are considered small, .5 are moderate and .8 are large. If these reported effect sizes are real then we should all be permanently clenching our hands: a Cohen's d of 1 would be equivalent to gaining 15 iq points based on this manipulation. But more importantly, in a study with such small N it wouldn't be 'possible' to observe a significant yet small effect size, so the authors' argument that the effect sizes are large is virtually irrelevant (any significant difference with N=9 has to be 'large'). One of the problems with small N is that effect sizes vary widely under the null hypothesis (which is why the p-values can be non-significant event for 'large' effects). In their rejoinder on the PLoS website it is clear that the authors fail to appreciate this when they state 'and the non-significant p values reflect small sample sizes, not small effects.' and later 'Furthermore, as we point out, the effect sizes are quite large, regardless of sample size, indicating that replications with larger samples may show larger effects.'. The latter is actually the reverse of what one would expect, e.g. Ioanniddis (2008) 'First, theoretical considerations prove that when true discovery is claimed based on crossing a threshold of statistical significance and the discovery study is underpowered, the observed effects are expected to be inflated.' Ioannidis, J. P. (2008). Why most discovered true associations are inflated. Epidemiology, 19(5), 640-648.

At April 29, 2013 8:55 AM, Blogger The Neurocritic said...

Rogier - Thank you for making the point about effect sizes and n so clearly. I thought you might be concerned that the calculations were done improperly, so that's why I did them myself. And we can see that Figure 3 must be wrong, because the effect size for the R/L vs. control comparison is only 0.284 (and non-significant with small n's).

At April 29, 2013 2:01 PM, Anonymous Anonymous said...

I just noticed the study was funded by the U.S. Army. Why would the U.S. Army fund this kind of stuff? Good old boys network?...

At April 29, 2013 2:03 PM, Blogger The Neurocritic said...

I registered at PLOS ONE and posted a comment of my own.

There has been an active comment thread, with new comments by Dr. Propper and Dr. Simons (among others).

At April 29, 2013 2:49 PM, Blogger Unknown said...

Rogier wrote that "one of the problems with small N is that effect sizes vary widely under the null hypothesis". To clarify: the *sample effect size* varies widely for small N, but that has nothing to do with the null-hypothesis; or, under the null-hypothesis, the *population effect size* is zero, but that doesn't vary widely then.
When the authors' state that "replications with larger samples may show larger effects", they seem to mean, perhaps, that larger N may show larger significance (not larger effect size). Whether that is true depends on whether there is indeed an effect of the size that the authors found. But the crucial point is that that could actually be zero, given that the null-hypothesis wasn't rejected. When the group size is increased, the confidence interval around the true effect size will shrink, but it is by no means clear if that will converge to the effect size the authors report, or towards zero. It is precisely because this is uncertain that the current study did not reach significance.
Unfortunately the "we would have been able to show a significant effect if only our group would have been bigger" is an often encountered mistake.

At April 29, 2013 3:34 PM, Anonymous Rogier said...

Dave, you're right, I was referring to sampling variability not population/'true' variability, should have emphasized that more clearly. I think interpreting their 'larger effects' as 'more significant' is a charitable reading, given that it follows shortly on a mention of effect sizes, but you might be right. But p-values are imperfect tools as they are for deciding between hypotheses, let's hope the authors didn't want them to carry the burden of reflecting large vs small effects also :)

At April 29, 2013 3:47 PM, Blogger Right Mind Matters said...

Like Propper and colleagues, I'm very interested in handedness research. I found an article in which they make a new distinction between consistent handedness (CH) and inconsistent handedness (ICH), rather than left-. right- or mixed-handed.

Their point is that extreme left-handers don't differ that much from extreme right-handers, whereas those who are mixed or inconsistent, have the advantage of bilateral (increased right-hemispheric input with a wider corpus callosum). The advantages are better episodic memory retrieval, younger age for earliest memory retrieval, superior face memory, greater susceptibility to the placebo effect, better word retrieval in foreign languages, more creativity via divergent thinking.

On the downside, depending on how you look at it, ICH are more prone to magical ideation (beliefs in
ESP and astrology, for example).and exhibit more inaction inertia, probably because they're uncertain of the benefit of moving ahead. Fleck et al have also argued that mixed-handers are more prone to magical ideation and Sommer and Kahn say this type of atypical lateralization is more likely to lead to psychosis than extreme left-handedness.

I have argued in an article (in press) that Sylvia Plath's brain dominance was bilaterally lateralized, based on external signs, not EEG. I am a literary scholar, not a neuroscientist.

At June 12, 2013 7:45 AM, Anonymous Anonymous said...

Interesting analysis of the article. I believe the most powerful test of this phenomenon would be very similar to this published paradigm, but with a within-participants design.

In any case, I also wanted to note that the Title of your post "Clenching your fist doesn't help!" is almost as egregious as the original paper. Unfortunately there is not enough data to support either conclusion - helping or not.

At June 12, 2013 9:27 AM, Blogger The Neurocritic said...

Anonymous - Thanks for your comments. You're correct that a within-subject design would be better. The authors could start over with a new (much larger) group of participants, with the four major fist-clenching conditions administered in a counterbalanced fashion (with appropriate rest in between).

As for the title, I meant to counter the strong assertions in the media (and the paper) that the study had proven the R/L condition improves memory. But you're right that we don't really know for sure either way on the basis of this one pilot study.


Post a Comment

<< Home

eXTReMe Tracker