pcl r disparity

Patricia Carlin1, Paul Barrett2, and Gisli Gudjonsson3

1&3 Institute of Psychiatry 2Psytech International Ltd. 2University of AucklandDept. of Psychology The Grange Faculty of BusinessDe Crespigny Park Church Road, Pulloxhill Commerce C BuildingDenmark Hill Bedfordshire Symonds StreetLondon SE5 8AF, UK MK45 5HE, UK Auckland, NZ

2University of CanterburyDept. of PsychologyPrivate Bag 4800Christchurch, NZ

Patricia Carlin1, Paul Barrett2, and Gisli Gudjonsson3

1&3 Institute of Psychiatry 2Psytech International Ltd. 2University of AucklandDept. of Psychology The Grange Faculty of BusinessDe Crespigny Park Church Road, Pulloxhill Commerce C BuildingDenmark Hill Bedfordshire Symonds StreetLondon SE5 8AF, UK MK45 5HE, UK Auckland, NZ

2University of CanterburyDept. of PsychologyPrivate Bag 4800Christchurch, NZ

NZ Psych Conference – Aug/Sept 2003

Download from: www.pbarrett.net

The Psychopathy Checklist Revised (PCL-R: Hare, 1991, 2003) provides three scores to be used in the assessment and diagnosis of psychopathy as well for actuarial-predictive risk applications:


a Factor 1 “interpersonal variables” scorea Factor 2 “social deviance” scorea Total score formed from the sum of all

20 items comprising the PCL-R.


It consists of 20 cognitive-behavioural constructs or items, each assigned a 0, 1, 20, 1, 2rating indicating the degree to which an individual matches the description of the item in the test manual. A 0 rating indicates that the item does not apply to the individual. A rating of 1 indicates a partial application, with 2 indicating that an item definitely applies to an individual.


The sum score across all 20 items yields the total PCL_R score (0-40). Two other scores are normally computed, the Factor 1 and 2 scores (F1 & F2) which are the summed scores respectively for the particular items …


Failure to Accept Responsibility for Own Actions

16

Callous/Lack of Empathy8

Shallow Affect7

Lack of Remorse or Guilt6

Conning/Manipulative5

Pathological Lying4

Grandiose Sense of Self Worth2

Glibness/Superficial Charm1

Factor 1: Interpersonal VariablesPCL-R itemreference ID

*For confidentiality reasons, these items do not appear in the web version of the paper


Criminal Versatility *new in 2003 manual20

Revocation of Conditional Release19

Juvenile Delinquency18

Irresponsibility15

Impulsivity14

Lack of Realistic, Long-Term Goals13

Early Behavioural Problems12

Poor Behavioural Controls10

Parasitic Lifestyle9

Need for Stimulation/Proneness to Boredom

3

Factor 2: Social DeviancePCL-R itemreference ID


It is proposed by Hare that the factors represent two distinct but related aspects of psychopathy.

Hare, in the 1991 manual, p.38 states “The “The items that make up the two PCLitems that make up the two PCL--R factors, which R factors, which on average correlate about on average correlate about 0.50.5 with one another”with one another”.

In the 2003 manual, p. 77, “For both the PCL “For both the PCL and PCLand PCL--R the correlation between factors 1 and R the correlation between factors 1 and 2 was about 2 was about 0.50.5””.


Facet 1: Interpersonal1. Glibness/Superficial Charm2. Grandiose Sense of Self-Worth4. Pathological Lying5. Conning/Manipulative

Facet 2: Affective6. Lack of Remorse or Guilt7. Shallow Affect8. Callous/Lack of Empathy9. Failure to Accept Responsibility forOwn Actions

Facet 3: Lifestyle3. Need for Stimulation/Proness to Boredom9. Parasitic Lifestyle13. Lack of Realistic/Long-Term Goals14. Impulsivity15. Irresponsibility

Facet 4: Antisocial10. Poor Behavioural Controls12. Early Behavioral Problems18. Juvenile Delinquency19. Revocation of Conditional Release20. Criminal Versatility

Factor 1 Factor 20.699

0.809

0.691

0.752

0.710

The 2003 2nd Edition model


Given an expected observed correlation between the two factor scores, to what extent might disparities between the two scores exceed “permissible” values?

Legal Context:defence –v- prosecution psychologists rating the same offender – but with markedly different ratings on Factor 1 – leading to quite different total scores.


Evaluate the expected distributions of scores on each factor, conditionalconditional upon the scores of the respective factor using a sample of10,000 simulated F1 and F2 scores which:

possess a correlation of 0.5are drawn from perfectly normally

distributed distributions (bivariate normality as required for a Pearson correlation coefficient)

possess means and SD’s as per 2003 test manual.


Evaluate the expected distributions of scores on each factor, conditionalconditional upon the scores of the respective factor using two samples of actual data:

Forensic psychiatric patientsForensic psychiatric patients (N=217)Prisoners/OffendersPrisoners/Offenders (N=1358)


Forensic psychiatric patientsForensic psychiatric patients (N=217) from two of the UK’s High Security Forensic Psychiatric Hospitals, Ashworthand The State Hospital) and Arnold Lodge (a high security clinic).

Prisoners/OffendersPrisoners/Offenders (N=1358) from several UK prison institutions (some data donated by David Cooke).


The study participants were all male adultsmale adultswho had been assessed using a PCL-R within a personal interview (in addition to collateral file information) - conducted by a formally trained practitioner.


Where missing ratings were encountered, the Factor and Total scores were prorated as per manual instructions.

All scores were created via computer from raw item data, using a custom scoring routine which handled prorating and the computation of F1, F2, and Total scores.


Step 1: Construct two variables, Norm_1 and Norm_2, sampling 10,000 observations from a random normal population distribution with mean of 0.0 and standard deviation of 1.0.

Step 2: Construct a new variable, New_Norm_2, from the two variables in step 1 according to the formula …


( )2_ _2 _1 + _2

1

rNew Norm Norm Normr

⎛ ⎞⎜ ⎟= ⋅⎜ ⎟⎜ ⎟−⎝ ⎠

where:

Norm_1 and Norm_2 are random normally distributed sets of continuous-valued “scores”

r is the desired correlation between Norm_1 & New_Norm_2 (0.54*) *allowing for integer mapping & scaling constraint

attenuation which will bring the correlation back to 0.5

New_Norm_2 is the transformed Norm_2 variable


Step 3: Convert1 the real-valued variables into integer “factor” scores, designed to possess the same mean and standard deviation as provided in Table 4.7 of the PCL-R manual (Hare, 2003). This was achieved separately for Offenders and Patients as their means and SDs are quite different. 1(note:note: this required two kinds of rescaling adjustments, so as to allow for the constrained integer measurement range of the PCL-R).

191020814516

191020229015

191020236314

18920356513

18920172812

18820284111

17820297210

17720011339

16620010318

1652009407

1552008356

1442006705

1432005194

1332004113

1321902922

1221601761

111160890

90th

Percentile Score on Factor 2

10th


Maximum Observed Factor 2

score

Minimum Observed Factor 2

score

Number of cases scoring at this level

Factor 1 Score

Offender Data

15716320020

15716226119

14716235418

14716442417

14716060016

14616269315

13516073414

13516183113

13516088412

12416085811

13416083510

1231607989

1131606108

1021505107

1021504296

911502915

911402494

911201873

811501342

70120641

7090540

90th Percentile Score on Factor 1


Maximum Observed

Factor 1 score

Minimum Observed

Factor 1 score


Factor 2 Score


0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 Score

0

100

200

300

400

500

600

700

800

900

1000

1100

1200

No.

of o

bser

vatio

ns

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Factor 2 Score

0

100

200

300

400

500

600

700

800

900

1000

No.

of O

bser

vatio

ns

Simulated Data, N=10,000 casesWith F1 v F2 correlation of 0.50


Step 1: Look at the data for patients and offenders – to see whether they can be sensibly combined into one sample …


0.500.500.50Factor r

4018182016Maximum

00000Minimum

8.14.44.44.93.7SD

158895Median

15.78.08.08.95.3Mean

Total Old Factor 2

Factor 22003

Factor 1

OffendersN = 1358

Male


0.310.310.28Factor r

3517171916Maximum

10000Minimum

6.93.83.84.34.0SD

201111117Median

19.79.99.910.87.4Mean

Total Score

Old Factor 2

Factor 2Factor 1

PatientsN = 217

Male

Conclusion: Patients and Offenders to be analysed separately


Step 2: Use resampling analysis to provide “population” estimates of F1 and F2 conditional score distributions.


Step 3: Construct 17 datasets of ~1000 randomized scores constructed in line with the observed proportions of each F2 score at each F1 score point (0-16 range -F1 scores).

Step 4: sample N (=total dataset) cases at random (uniform sampling) “with replacement”from F1 scores in the same dataset from which the F2 conditional score distributions have been calculated.


Step 5: For each resampled F1 score, sample an F2 score at random from the proportionate conditional F2 score distribution associated with a particular F1 score magnitude (a secondary conditional resampling)

Step 6: Take 1000 complete resamples of the F1 dataset (1000 x 217 cases in respect of the patients, and 1000 x 1358 cases in respect of the offender data).


Step 7: Compute the overall conditional distributions of F2 scores conditional upon each F1 score, and F1 scores conditional upon each F2 score, using all 1000 samples

18101810704116

1851951197215

1982061685114

1661742824813

1751823310512

1772014974811

1751805473810

173200680339

164201785508

1532011077167

1441801122276

1421701400025

1431801399664

1211801765133

1211701370552

1101501069391

90170892960

90th


10th


Maximum Observed Factor 2

score

Minimum Observed Factor 2

score

Number of cases

scoring at this level

Factor 1 Score

Offender Data Resample

147147709820

148158888019

1441632601518

1431603961517

1331605822716

1241414606615

1231607740914

1031617878313

921409279012

9113010376311

10216110627110

91120886529

91150780708

81150895187

90140872676

80150620855

70130750774

70110611873

70120758532

60110563281

100100390460



Maximum Observed

Factor 1 score

Minimum Observed

Factor 1 score


Factor 2 Score

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 Score

0

100

200

300

400

500

600

700

800

900

1000

1100

1200

No.

of o

bser

vatio

ns

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Factor 2 Score

0

100

200

300

400

500

600

700

800

900

1000

No.

of O

bser

vatio

ns

Simulated Datasets – Normative (manual) Offender Specifications

UK offenders (N=1358) actual data

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 Score - Offender Data

0

20

40

60

80

100

120

140

160

180

200

No.

of O

bser

vatio

ns

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Factor 2 Score - Offender Data

0

10

20

30

40

50

60

70

80

90

100

110

120

No.

of O

bser

vatio

ns

pcl r disparity

Documents