Sunday, March 31, 2013

Are Cognitive Factors Related to Criminal Reoffending?

Image from Graphic Sociology

Can Brain Activity Predict Criminal Reoffending?  The previous post discussed a functional MRI study suggesting that the level of error-related activation in the anterior cingulate cortex (ACC) might have value in predicting whether a recently released prisoner will be rearrested within 4 years (Aharoni et al. 2013):
The odds that an offender with relatively low anterior cingulate activity would be rearrested were approximately double that of an offender with high activity in this region, holding constant other observed risk factors. These results suggest a potential neurocognitive biomarker for persistent antisocial behavior.

However, using ACC activity as a dichotomous variable misclassified 40% of low ACC participants who did not reoffend and 46% of high ACC participants who did commit crimes after release, not exactly the odds you'd want for making parole decisions. Even the senior author was doubtful that an fMRI test would ever be useful for risk assessment purposes on a case by case basis.

Since Aharoni and colleagues made their individual subject data available as supplementary material (Download Dataset_S01, XLSX), I was interested in how some of the demographic and performance variables might be related to recidivism, since these are obviously cheaper and easier to collect from incarcerated prisoners than MRI scans.

The cognitive task performed during the fMRI experiment required responding to a frequent stimulus presented 84% of the time ("X") and inhibiting responses to a rare stimulus ("K").

Fig. S4. (Aharoni et al., 2013). Go/No-Go task.

The study compared brain activity on incorrect responses to "K" (commission errors) and correct responses to "X" (hits) in a region of interest in the dorsal ACC, which has been implicated in error processing (Simons, 2010), among many other things. The authors framed the results largely in the context of impulse control, but other explanations are possible (as we'll see later).

Are any of the task performance variables related to recidivism? Starting with some very simple-minded t-tests, the rate of commission errors in the group of participants arrested for nonviolent offenses1 (n=40) did not differ significantly from what was seen in those not arrested again (n=56).2

Data from (Aharoni et al., 2013). Commission errors in the Go/NoGo task (% incorrect responses on NoGo trials) and omission errors (% missed responses on Go trials) for inmates that went on to commit nonviolent offenses within 4 years after release (Nonviolent) and those that did not (None). The trend for the reoffenders to commit more errors was not significant (p=.09) even without correcting for multiple comparisons.

Although there were data from a large control group of nonoffenders (n=102) used to set the ACC ROI, we don't have their behavioral results. I consulted an earlier fMRI paper by Kiehl et al. (2000) that used a very similar Go/NoGo task in 14 control participants. Commission errors occurred on 23.7% of NoGo Trials and omission errors on 3% of Go Trials, which is similar to what was seen in the offenders (overall means of 25.04% and 3.44%, respectively).

Reaction times (RTs) did not differ between the two offender groups either, suggesting there wasn't a differential speed-accuracy tradeoff (e.g., if the reoffenders were slower yet making marginally more errors).

Data from (Aharoni et al., 2013). RTs in milliseconds for commission errors (incorrect responses on NoGo trials) and hits (correct responses on Go trials) for inmates that went on to commit nonviolent offenses within 4 years after release (Nonviolent) and those that did not (None). There were no group differences.

Surprisingly, RTs were slower on commission errors (358 ms) than on hits (346 ms), a small but highly significant difference (p=.0005). This is the opposite of what you'd expect if the errors were due to impulsive responses. If the participants were becoming careless and not fully evaluating the NoGo stimulus, they'd be faster on error trials. This is why I'm not convinced the ACC activations are entirely related to behavioral impulsivity. In EEG studies of error processing, the degree of ACC activity3 is related to the emphasis placed on accuracy (Gehring et al., 1993), so if the reoffenders didn't care as much about accuracy, this could account for their low ACC status. One interesting bit of data for the authors to examine would be RT and accuracy on responses following an error, which indicates the amount of behavioral adjustment after making a mistake. Did the reoffenders show a lower propensity to slow down and become more careful? If so, this might reflect a lack of concern about the consequences of their actions.

However, the most puzzling thing to me were scores on Factor 2 of the Psychopathy Checklist-Revised (PCL-R) (Hare, 2003). Factor 2 is thought to reflect impulsivity, stimulation seeking, and irresponsibility (Ermer et al., 2012). The rearrested and not-rearrested groups were significantly different as expected, but in the opposite direction (unless I'm missing something here) — scores were lower in the group that was rearrested, in comparison to those who were not (p=.001).

Data from (Aharoni et al., 2013). Beta-weights from the dACC region of interest and a control region in medial prefrontal cortex (mPFC). PCL-R f2 is score on Factor 2 of the Psychopathy Checklist-Revised, normalized using a log transform (p=.001). [ROIs and PCL-R not measured using the same units, obviously.]

In the paper, Aharoni and colleagues noted that age at release and Factor 2 scores showed predictive effects along with ACC activity. This was only when nonviolent crimes were considered [remember that only nine participants were arrested again for violent crimes]. Some research suggests that the PCL-R may predict violent recidivism, but other work questions this assertion.4 I'm definitely not the expert here, so please weigh in if you have an opinion.

Returning to behavioral performance on the X/K response inhibition task, this did not clearly differentiate between those inmates who would reoffend after release from those who did not. So we cannot conclude that cognitive factors are related to nonviolent criminal reoffending,5 at least from this one experiment that evaluated one specific executive function.


1 There were nine participants arrested for violent offenses, and of these six were arrested for both violent and nonviolent offenses. These latter subjects were an inaccurate bunch (42% commission errors), but it's hard to make much of such a small group.

2 The significance went down further if you controlled for age at release (for instance).

3 As reflected by the amplitude of the error-related negativity component, which is further modulated by motivational incentives and personality factors.

4 From my post on The Disconnection of Psychopaths:
Forensic psychologist Dr. Karen Franklin has written about multiple controversies surrounding the PCL-R, including the failure of Factor 1 to predict violence and Dr. Hare's attempt to block publication of a critical article. Also see this NPR series on Weighing The Value Of A Test For Psychopaths.

5 WAIS scores were not predictive, either.


Aharoni, E., Vincent, G., Harenski, C., Calhoun, V., Sinnott-Armstrong, W., Gazzaniga, M., & Kiehl, K. (2013). Neuroprediction of future rearrest. Proceedings of the National Academy of Sciences. DOI: 10.1073/pnas.1219302110

Ermer E, Cope LM, Nyalakanti PK, Calhoun VD, Kiehl KA. (2012). Aberrant paralimbic gray matter in criminal psychopathy. J Abnorm Psychol. 121(3):649-58.

Gehring WJ, Goss B, Coles MGH, Meyer DE, Donchin E. (1993). A neural system for error-detection and compensation. Psychological Science 4:385–390.

Kiehl, K., Liddle, P., & Hopfinger, J. (2000). Error processing and the rostral anterior cingulate: An event-related fMRI study Psychophysiology, 37 (2), 216-223 DOI: 10.1111/1469-8986.3720216

Simons RF. (2010). The way of our errors: theme and variations. Psychophysiology 47(1):1-14.

Subscribe to Post Comments [Atom]


At March 31, 2013 5:39 PM, Anonymous Brain Molecule Marketing said...

Nice piece. Boy, we are "evil" marketers and have to say we find the whole theme of higher order process and top down effects on behavior fraught with lack of evidence, contradictory evidence from say visual-behavior studies and the overwhelming cultural biases of all these ideas and models.

"Gee whiz, the brain works just like our subjective experience and out cultural beliefs/econ-etc theory tell us it does!!" How kool is THAT!!??

So cognitive models need very hardheaded critiques regarding physiology, duh.

That said, maybe there are some hints here....well, probably not.

We just spend a day with the best and brightest neuro grad students in Chicago -- boy are they capable. They are off of imaging as old hat and "bloobism" (finding s place in the brain for functions using fMRI). They are deciding their careers and all that stuff isn't credible with them anymore.

So this study just seems uninformed and using old models.


At March 31, 2013 7:32 PM, Blogger Unknown said...

Can you reproduce the results in the manuscript using the data they provide? I can't: I get qualitatively similar results, but not close to the same numbers.

For example, in the simple Cox model with only the ACC split, I get a coefficient of 0.5 rather than 0.96

Your comment system won't let me format this correctly but:

> coxph(Surv(MinMonthsDV_noPVs,AnyChargeSinceScanExclPVs)~dACC_14mm_split,data=sd)
coxph(formula = Surv(MinMonthsDV_noPVs, AnyChargeSinceScanExclPVs) ~
dACC_14mm_split, data = sd)

coef exp(coef) se(coef) z p
dACC_14mm_split -0.503 0.605 0.284 -1.77 0.076

Likelihood ratio test=3.18 on 1 df, p=0.0746 n= 96, number of events= 51

At April 01, 2013 5:11 AM, Anonymous Anonymous said...

Nice and detailed analysis, NC. This is what I call a neuroblog worth following. unlike most of the neuroregurgitation blogs out there :)

At April 02, 2013 11:03 AM, Blogger The Neurocritic said...

Anonymous - Thanks for the compliment!

Thomas Lumley - I haven't had a chance to look at that yet...

At April 06, 2013 4:29 PM, Blogger Russ Poldrack said...

I just posted an article to my blog that assesses the out-of-sample predictive accuracy of the Aharoni data:

At April 06, 2013 4:32 PM, Blogger Russ Poldrack said...

This comment has been removed by the author.

At April 06, 2013 4:33 PM, Blogger Russ Poldrack said...

Thomas - if you use dACC_centered instead of the binarized split data, you will reproduce their results.

At April 09, 2013 3:03 PM, Blogger Roger Bigod said...

It would be interesting to see if there are meds to improve ACC reliability. Perhaps Ritalin, or something to stabilize neuronal membranes. Managing the delivery of a chronic drug would be a lot less expensive than dealing with the chronic recidivism.

There is, of course, the risk of producing a cohort of attentive felons.

At April 13, 2013 10:57 AM, Blogger J. F. Aldridge said...

Nice posts. :)
It be neat to create a reintegration program modeled off of the witness protection program. Basically, prisoners would get a new identity, location and occupation. The felony would be hidden from background checks, but become an amplified deterrent, with extreme consequences for reoffending. I wonder what the recidivism rates would be, or what they are now for current witness protection programs.

At April 14, 2013 7:39 PM, Blogger Roger Bigod said...

The only obvious benefit from the witness-protection model is the lack of a stigma resulting from the criminal record. The subject would be free to continue contacts in the previous social environment, if that was criminogenic. And the main stigma appears to be employment. AFAIK, employers are free to ask about a criminal record, and this would imply a legally sanctioned freedom to lie. There would be legal liability if the employee rips off the employer.

But absent these complications, it's not clear that the employment problem is a major factor in recidivism. It looks like most small-time felons have a deficit of executive function -- factors like impulse control, risk evaluation, foreseeing consequences.

A problem with interpreting recidivism rates is that the rate doesn't distinguish between subjects who commit fewer crimes and those who commit crimes but use better judgment and avoid re-arrest.


Post a Comment

<< Home

eXTReMe Tracker