Unpacking insanity defence standards: An experimental study of rationality and control tests in criminal law

Participants included 477 undergraduate students from a large United States university that contains both a public, state university and a private university within a single administrative structure. Students participated in the study for course credit. We examined the demographics of our sample to give an indication of the population that they represent. The sample was 68.8% female with an average age of 19.27 ( SD = 1.17). The majority of our sample was of non-Hispanic White descent (57.7%).

All participants received the same case summary based on one used by Roberts, Sargent, and Chan (1993) and later used by Skeem and Golding (2001) and Gurley and Marcus (2008) . The case summary describes the murder of a postal official. The defendant was found with the murder weapon, which was linked to him via fingerprint and blood matching. The defendant has been found to have killed the postal official and participants are asked to decide whether he is guilty or not guilty by reason of insanity.

Participants were randomly assigned to one of six conditions resulting from a fully-crossed factorial design involving two variables: mental disorder of the defendant (rationality defect or control defect), and legal test to be applied (Model Penal Code, McNaughten, or McNaughten + ).

The rationality deficit defendant and the control deficit defendant had many of the same symptoms–the defendant was socially isolated, had vague and rambling speech and an MRI scan showed some abnormality which may have been the cause of his abnormalities. However, in addition to these similarities, the defendant with a rationality defect believed that the victim was part of a group of aliens conspiring to take over the world. The defendant with a control defect was stated to have trouble controlling criminal or otherwise anti-social conduct due to feeling compelled to act a certain way despite knowing that he should not.

Participants allocated to the Model Penal Code and McNaughten conditions made one determination–whether the defendant was guilty or not guilty by reason of insanity. Participants allocated to the McNaughten+ condition first made a determination regarding whether the defendant was guilty or not guilty by reason of insanity (a complete defence), and then went on to make a determination as to whether the defendant lacked control (a partial defence).

In addition to a demographic questionnaire, participants completed the Individualism and Hierarchy scales from Kahan and Braman (2008) . The Individualism scale contained 17 items and measured the relative priority that subjects assigned to group and individual interests (e.g., ‘The government should do more to advance society's goals, even if that means limiting the freedom and choices of individuals’) (α = .86). The Hierarchy scale consisted of 13 items assessing subjects’ attitudes toward socially stratified roles (e.g., ‘We have gone too far in pushing equal rights in this country’) (α = .86).

We also measured subjects’ perceptions of various facts by asking them to indicate whether they agreed or disagreed with various propositions concerning facts that a juror would be likely to consider when making the determination of guilty or not guilty by reason of insanity (e.g., it is likely that the defendant would not have killed the victim had it not been for his mental illness; because the defendant suffered from a mental illness he was unable to stop himself from killing the victim).

47.8% of participants rendered a verdict of not guilty by reason of insanity (NGRI). The other 52.2% rendered a verdict of guilty.

We ran an initial ANOVA with insanity decision (the determination under the model penal code in the first legal test condition and the determination under McNaughten in the second and third legal test conditions) as the dependent measure and legal test and defendant mental state as factors. We ran the ANOVA with and without gender as a factor: the significant effects were the same in both ANOVAs and there was no main effect of gender ( p = .952) or interactions with gender. Results from the ANOVA without gender are reported in Table 1.

The results of our initial ANOVA show a significant main effect of legal test F(2, 2.38) = 9.86, p < .001, ηp 2 = .04 (see Figure 1 ). The subjects who were given the McNaughten+ control instructions were roughly 40% more likely to return guilty verdicts using the rationality test, a relatively large effect (difference between MPC and McNaughten+: d = 0.4; difference between McNaughten and McNaughten+: d = 0.45). There was no main effect of mental state and the interaction between legal test and mental state was not significant.

Figure 1

Proportion of Guilty Verdicts by Legal Test. Error bars show ± 1 Standard Error.

In order to further probe decisions under each legal test, we conducted individual ANOVAs for each legal test with verdict as the dependent measure and mental state of the defendant as the independent measure. The results of these ANOVAs are reported in Table 2 . In ANOVAs for the Model Penal Code and the McNaughten test, there was no main effect of mental state, with NGRI rates being similar for defendants who lacked rationality or lacked control ( p = .876 and p = .737 respectively). In the ANOVA for the McNaughten+ test, there was a marginally significant main effect of mental state F(1, 0.75) = 3.36, p = .069, ηp 2 = .02, showing that participants categorised defendants who lacked control guilty more often than defendants who lacked rationality.

We ran a follow up ANOVA that investigated the number of participants who granted the defendant at least a partial excuse. This includes participants who found the defendant NGRI in all three legal test conditions, and also the participants who found the defendant to have a partial excuse based on lack of control in the third legal test condition. In this ANOVA, granting a complete or partial defence was the dependent measure and the independent measures were legal test and defendant's mental state.

Overall, 64.6% of participants rendered a verdict giving the defendant either a complete or partial defence. The other 35.4% rendered a verdict of guilty, with no defence.

The results of this ANOVA are reported in Table 3 . Again, the results show a significant main effect of legal test F(2, 4.03) = 18.79, p < 0.001, ηp 2 = .074. In this case, subjects in the McNaughten+ condition allowed some kind of defence more often than those in the Model Penal Code or McNaughten conditions. This is illustrated in Figure 2.

Figure 2

Proportion of Guilty Verdicts with no Partial Defence by Legal Test. Error bars show ± 1 Standard Error.

In order to probe differences in the decision-making process between the legal tests, we split the data by legal test and ran binary logistic regressions to see what factors significantly predicted final verdict. We put each of our fact perception questions into the regression as well as participants’ individualism and hierarchy scores. The results of each of these regressions are reported in Table 4.

Model penal code . For the participants who made their decision based on the model penal code, four predictors were significant: Hierarchy Score ( B = 1.330, p = .025) and three fact perceptions: it is likely that the defendant would not have killed the victim had it not been for his mental illness ( B = 1.651, p = .001), because the defendant suffered from a mental illness he was unable to stop himself from killing the victim ( B = 1.576, p = .001), and the defendant did not know what he was doing when he killed the victim ( B = 1.121, p = .029).

McNaughten. For the participants who made their decision based on the McNaughten standard, three fact perceptions were significant: it is likely that the defendant would not have killed the victim had it not been for his mental illness ( B = 1.099, p = .019), because the defendant suffered from a mental illness he was unable to stop himself from killing the victim ( B = 1.132, p = .018), and the defendant was not in control of his actions when he killed the victim ( B = 0.980, p = .049).

McNaughten plus an independent control test. For the participants who made their decision based on the McNaughten standard (but who knew an independent control test would follow), three fact perceptions were significant: because the defendant suffered from a mental illness he was unable to stop himself from killing the victim ( B = 1.001, p = .018), the defendants MRI results show that he is mentally disabled ( B = 1.331, p = .009), and the defendant did not know what he was doing when he killed the victim ( B = 1.276, p = .004).

For all of the legal tests, the same fact perceptions were significant when included with or without the hierarchy and individualism scores.

The results of this study support the contention that jurors’ decisions in insanity cases are not affected by whether they are asked to decide based on the Model Penal Code test (with a rationality limb and a control limb) or on the McNaughten test (based entirely on rationality), even when considering a defendant suffering from a clear control disorder. This suggests that jurors are making decisions based on who they think is insane rather than on the specific legal standard they are given and is consistent with existing literature showing that jurors tend to use their own conceptions of insanity rather than legal definitions when making determinations ( Finkel & Handel, 1989; Skeem & Golding, 2001 ).

This study built on the existing literature by introducing a new method that shows promise in reducing bias in this area. Our results show that this indifference to legal instructions changes when rationality and control are included separately (with rationality as a complete defence and subsequently control as a partial defence). Using this latter legal test, a higher proportion of defendants received either a complete or partial defence when using McNaughten+, but not to the extent that all defendants were found not guilty (supporting the idea that it is possible to include a control prong without opening the floodgates to excuse any seemingly “caused” behaviour, as discussed by Morse, 1995 ). In addition, when using the control prong as only a partial defence, significantly fewer defendants were given a complete defence than under the MPC or McNaughten standards.

Whether lack of control should be a complete defence or a partial defence is something to be debated and further investigated, but importantly, this study suggests that including the two separately led to more accurate categorisations of defendants. Under the Model Penal Code (MPC) and the McNaughten standard, defendants with rationality disorders and defendants with control disorders were classified as NGRI at roughly the same rates, which should be the case under the MPC but not under McNaughten. Using the McNaughten+ legal test, more defendants with rationality deficits were classified as having a rationality deficit than defendants with control deficits, showing that the test improved accuracy in categorisation. This is consistent with traditional dual process theory, which suggests that decision-making based on specific rules that require conscious processing is the most accurate (see Evans & Stanovich, 2013 ). Our results provide support for the idea that including a separate control test may increase accuracy in juror decisions and provide an opportunity for the legal system to increase accuracy and adherence to legal standards.

The McNaughten+ standard may appear similar to the current defence of guilty but mentally ill utilised in some U.S. states (for example, Alaska, Arizona, and Georgia) (see Robinson, 2014 ), but focused specifically on lack of impulse control. However, the fact that it recognises control as a partial excuse (partially excusing the action rather than simply acknowledging mental illness) should avoid problems associated with the guilty but mentally ill verdict, where defendants are typically assigned longer sentences and do not receive any treatment ( Desmond & Lenz, 2010).

This suggestion is more similar to the current system in the Netherlands, where there are different grades of insanity representing different levels of severity and impacting the outcome for the defendant (see Radovic et al., 2015 ). Although these grades in the Netherlands are not based on rationality and control, they do show that such a system can be effective and workable. Treating control in this way may provide a compromise in terms of recognising accepted psychological deficiencies but also recognising some personal responsibility of a rational agent.

Our results indicate that people's opinions relating to hierarchy and individualism only play a role in the verdict under the Model Penal Code test (where hierarchy plays a role). Under the other two tests, hierarchy and individualism attitudes did not predict when included in a regression with fact perceptions. This suggests that bias and pre-existing attitudes may play more of a role under the Model Penal Code test. This is in accordance with our prediction that extra legal attitudes will play more of a role under the Model Penal Code due to the broader test giving more room for interpretation than with a more specific test based on either a single cognitive consideration or a temporally specified sequence of cognitive followed by control considerations. This is also supported by research showing than even clinicians struggle to agree on who is insane under the Model Penal Code standard, with evaluators reaching agreement in only 55.1% of cases in Hawaii (although it should be noted that clinician agreement has not been tested under the McNaughten test so we cannot directly compare the tests) ( Gowensmith, Murrie, & Boccaccini, 2013 ).

Particularly relevant to the issue of control is that participants’ answer to the question “the defendant was not in control of his actions when he killed the victim” was predictive of verdict in the McNaughten test. This is essentially the jurors’ perception of the defendant's control, which should not be relevant in the McNaughten test. Including a separate control test after the rationality determination, as is accomplished in McNaughten+ test, reduces the influence of this fact perception and it is no longer a significant predictor. This is important, as by creating an independent test for control we can take control out of consideration in the rationality test.