Nature of respondents
In total, 15 of 20 individuals (75%) who took part in the demonstration of the software agreed to complete the questionnaire; a further two provided qualitative comments only. The main reason for declining to complete the questionnaire was lack of time (four respondents); one respondent said that they preferred not to complete the questionnaire but would give unstructured verbal feedback, which was recorded on a blank questionnaire form.
Of the respondents, 10 (66%) reported being a psychiatrist, five of whom (33%) were specialised forensic psychiatrists. Five other respondents (33%) identified themselves as psychologists, of whom three were specialised forensic psychologists and the remaining two were clinical psychologists. Of the remaining two respondents, one was as an occupational therapist and one was a probation officer.
Quantitative responses
Responses to the quantitative aspects of the questionnaire were in general positive, with all items being rated on average ≥ 3; these responses are summarised in .
Mean responses to DSVM-P questionnaire items with SDs as error bars.
The most highly rated items were item 3 (‘I thought the questions asked by the software were relevant to risk management’; mean 4.50, SD 0.52), item 6 (‘The software recommended similar targets for management to those that I would have’; mean 4.27, SD 0.90) and item 7 (‘The intervention targets identified were helpful’; mean 4.25, SD 1.14). The lowest rated items were item 8 (‘I thought that using the software might save me time in my professional practice’; mean 3.39, SD 1.36) and item 10 (‘I thought that the software was more useful than SPJ tools like the HCR-20’; mean 3.46, SD 1.57). There seemed to be a higher degree of uncertainty about the less positively rated items, as evidenced by higher SDs for item responses.
All respondents endorsed item 3 (relevance of items) and item 4 (‘I thought the software asked all the right questions to enable violence risk management’) with a score of ≥ 3 (‘a fair amount’) and all except one respondent (93% of the sample) rated item 9 (useful information provided) as ≥ 3. In total, 80% of respondents rated the accuracy of the judgement about the risk of the case entered as representing their own judgement ‘a fair amount’ or better.
Qualitative responses
The responses to the unstructured questions provided substantial information about professional perceptions of the potential merits of and problems with the network.
Positive comments included:
Far more useful than existing risk assessment tools.
An easy and simple method to get likelihood of reconviction.
Minimises the need for lengthy assessments of personality (PCL-R etc.).
[It is] fairly quick to answer the risk questions.
[It is helpful] being able to see reconviction change over different time periods.
The ability to make predictions post intervention; we cannot normally estimate drop in risk after individuals complete interventions normally.
Positive comments seemed broadly to endorse the time taken to complete the DSVM-P measure and the high level of information and accuracy it provided. Some of the comments were not entirely accurate, for example the DSVM-P does specifically require a PCL-R and clinical assessments of personality to be completed and relies on these for improved accuracy.
Feedback which suggested that further work might be required was:
What about measurement of institutional violence?
Consider risk scenarios; should consider a sexual offence version.
Can you get the interface right?
Numbers confuse me; explanation of the risk % figure would be helpful.
[What about] circumstances of crime, or trigger?
A limited number of interventions modelled.
I was sceptical about the reliability of the model to make specific predictions of violence for an individual, e.g. 67% likelihood of reconviction is higher than most other predictive tools.
Lack of evidence for medical treatment; no medication.
One individual had also written the following in the margin of the questionnaire next to item 10 asking whether or not the software was more useful than SPJ tools like the HCR-20: ‘Only if completed instead of, not in addition to’.
Some of these comments reflected the developers’ own concerns, specifically with respect to the unfriendliness of the interface and the possible need for additional interpretation of raw probabilities of reoffending (e.g. categorical ratings such as high, medium or low risk). The suggestion that the number of interventions modelled should be expanded, possibly to include medication, was mostly accurate, although the modelling of ‘psychiatric treatment’ (described in Chapter 20) does relate primarily to psychopharmacological treatment rather than psychological interventions.
Other comments seemed to relate to other areas of need within forensic settings that were intentionally not addressed by the DSVM-P, for example suggestions of alternative networks covering institutional violence and/or sexual reoffending. However they do appear to illustrate an enthusiasm for the approach adopted. One comment relating to the inclusion of specific violence triggers and contexts was well taken, but modelling work on this has not yet proved possible with existing data.464
