Date post: | 15-Apr-2017 |
Category: |
Law |
Upload: | royal-statistical-society |
View: | 122 times |
Download: | 1 times |
Quantitative
Methods in
Socio-Legal
Studies: A
Methodology Clinic
Dr. Michelle Cowley
CSLS, University of Oxford
Overview
– What do we mean by a quantitative research process?
– Recent trends in socio-legal research… (e.g., Nuffield Foundation)
– What to be mindful of: variables and measures
– Experimental designs
– Example
– Survey research
– Example
– Secondary datasets
– Example
The Quantitative Process:
A deductive paradigm
1. Theory
2. Hypothesis
3. Research design- testing hypotheses
4. Devise measures of concepts
5. Select research site(s)
6. Select research participants
7. Collect data
8. Process data
9. Analyse data
10. Findings
11. Write-up
Let’s Brainstorm – What are your
research questions? Have you
thought about your methodology?
See Whiteboard & Blank Pages for Working
Illustrative Example: Evaluating PC
admissibility in Child Abuse
Proceedings
– Recent changes to evidence law regarding prior convictions (CJA, 2003)
– … In assessing the probative value of prior conviction evidence for the purposes of subsection (1)(b) the court must have regard to the following factors (and to any others it considers relevant)—
– (a) the nature and number of the events, or other things, to which the evidence relates;
– (b) when those events or things are alleged to have happened or existed;
– (c) where—
– (i) the evidence is evidence of a person’s misconduct, and
– (ii) it is suggested that the evidence has probative value by reason of similarity between that misconduct and other alleged misconduct…
How could we evaluate whether this law
positively or negatively affects the
balance of justice?
– Are there any social theories which predicts how this law change may affect the decision making process in the courtroom?
– Given these theories what would we hypothesise…
– How would we test these hypotheses?
– Let us think of some ways to obtain data to address this question… research designs…
Let’s Brainstorm –
How could we test this
research question?
See Whiteboard & Blank Pages for Working
Measures and data
– Nominal/categorical
– Scales (Interval and Likert scales)
– Reasons for choice…
– Non-parametric and parametric data
Let’s Brainstorm – Can you
think of any examples of
these measures being used in
your research literature?
See Whiteboard & Blank Pages for Working
Research design
– A framework for the collection of data…
– Experimental design- manipulation
– Quasi-experimental design
– Cross-sectional design
– Longitudinal design
– Case study design
Let’s Brainstorm – Can you
describe any research you
have read drawing on any of
these methodologies?
See Whiteboard & Blank Pages for Working
Sorts of quantitative analyses
– Univariate analyses (e.g., means)
– Bivariate analyses (e.g., correlations)
– Multivariate analyses (regression)
– Statistical significance (p <. 05)
– In real-life we need to plan the sorts of analyses at the same time as our design (a-priori vs post-hoc)
Let’s Brainstorm – Are there
examples in your thesis
literature of any of these
analyses – how might they help
us to evaluate theories?
See Whiteboard & Blank Pages for Working
Reliability and Validity
– Reliability- repeatable…
– Validity- integrity of the conclusions…
– Measurement validity
– Internal validity (e.g., cause and directionality)
– External validity (e.g., maps to real world contexts)
– Replication
– Sample representativeness is often the key
– Power of the study
Let’s Brainstorm – Can you
think of why reliability and
validity need to be embedded
in your research design?
See Whiteboard & Blank Pages for Working
An experiment: Previous convictions and
underlying ratings of guilt
Logical v Probabilistic Theoretical Frameworks… (Cowley & Colyer, 2010)
Logic Prediction 1: Participants will conclude that the defendant is ‘not guilty’ regardless of the number of previous convictions.
Probability Prediction 1: Participants will conclude that the defendant is ‘guilty’ or more guilty than chance (.5) when a previous conviction is present.
Logic Prediction 2: Participants’ underlying ratings of guilt will not be increased by the presence of previous convictions between the absolute values of not guilty (zero) and guilt (one).
Probability Prediction 2: Participants’ underlying ratings of guilt will be increased by the presence of previous convictions, and this increase will occur between the absolute values of not guilty (zero) and guilt (one).
Experiment 1: Method
Fifty one participants: 8 men and 43 women
Mean age 20.64years, range from 19 to 33 years
Design: 1 x 3 (control, one previous, two previous)
Materials: Reasoning about a scenario created from a real life case of a child who was killed by a man with two previous convictions for similar offences.
On January 2, 2006, David Baxter had been arrested. He had been accused of killing 18-month-old Joanna Connolly. Joanna’s skull had been fractured when she received a physical blow to the head. She was the daughter of Susan Connolly, the woman who David Baxter had been seeing.
Materials Template
Knowledge of previous convictions (one; two; none):David Baxter had previously served a three year sentence for being physically abusive towards an ex-girlfriend’s three year old girl in 2003.
Please answer the following questions: Q.1 Please tick whether you think:
David Baxter is guilty __David Baxter is not guilty __You cannot decide __
Q.2 On a scale of 1 to 10, circle the number that you think best reflects how guilty you think David Baxter is:
(0 represents Not Guilty, 10 represents Guilty)
Not | | | | | | | | | | | Guilty
Guilty
Experiment 1
0
10
20
30
40
50
60
70
80
90
100
no previous one previous two previous
Number of previous convictions
Perc
en
tag
e o
f verd
ict
typ
e
guilty
not guilty
cannot decide
Fig 1. Percentage of verdict type per condition. Regardless of the number of
previous convictions people could not decide if the defendant was guilty.
Chi2= 75.444 (2), p <.0005
Experiment 1
0
1
2
3
4
5
6
7
8
9
10
no previous one previous two previous
Number of previous convictions
Un
derl
yin
g r
ati
ng
of
gu
ilt
Fig 2. Underlying ratings of guilt were higher when a previous conviction was
present than when a previous conviction was not present regardless of the
number of previous convictions.
Kruskal-Wallis chi2= 16.162 (2), p <.0005
Experiment 2: Method
Seventy-two participants, 24 men and 48 women. Age range 18- 53years, mean 22.4years
Design 3 x 2 between subjects (left-handedness, right-handedness, no handedness) x (previous conviction, no previous conviction) [6 conditions]
Materials: The same scenario and measures either with or without a previous conviction and sort of handedness:
Forensic evidence showed that the blow was delivered by a left-handed person. David Baxter is left-handed
or
Forensic evidence showed that the blow was delivered by a right-handed person. David Baxter is right-handed
Experiment 2
0
2
4
6
8
10
12
Control RH LH PC PC and RH PC and LH
Presence of previous conviction (PC) and left (LH) or
right handedness (RH)
Nu
mb
er
of
pe
op
le p
er
ve
rdic
tguilty
notguilty
cannotdecide
Fig 3. The number of people who came to a ‘guilty’ ‘not guilty’ or ‘cannot decide’ verdict
in the presence of evidence of handedness (RH or LH) and/or previous convictions (PC).
(p < . 564)
People tend to significantly choose ‘cannot decide’ more often except when a previous
conviction is accompanied by evidence of a left handedness match (PC and LH).
Experiment 2
0123456789
10
Control RH LH PC PC and
RH
PC and
LHPresence of previous convictions (PC) and
left (LH) or right handedness (RH)
Un
derl
yin
g r
ati
ng
of
gu
ilt
Control
RH
LH
PC
PC and RH
PC and LH
Fig 4. Underlying ratings of guilt were higher when a previous conviction was
present than when a previous conviction was not present regardless of the number
of previous convictions (Kruskal Wallis, chi2= 12.282 (5), p <.031).
Interpreting the findings…
Even though on the surface it appears that people cannot decide, their underlying ratings of guilt are biased by knowledge of previous convictions. Previous convictions tended to bias verdicts in the presence of small amounts of confirming forensic evidence.
Perhaps indicating that people reason probabilistically towards logical conclusions.
Possibly due to similarity (e.g., PC) and not necessarily the result of a lower probability
Indicating that we next need to understand how the transition from coincidence to evidence takes place (Griffiths & Tenenbaum, 2006).
Survey/Questionnaires
– What is survey research? Data collected primarily by questionnaire or structured interview methods (big samples usually)
– Aim: patterns of association (spina-bifida and eating habits during pregnancy)
– Be aware- response rate! Power calculations are recommended.
– Often when we design surveys we wish to ask specific questions and also give room to participants to qualitatively reflect on their responses. We will focus on the first sort of question.
Let’s Brainstorm – How
might surveys help us to
advance legal empirical
research?
See Whiteboard & Blank Pages for Working
Survey
Structured interview Self completion questionnaire
Face to face Telephone Supervised Postal Internet
Paper
& pencilCAPI Paper
& pencil
…CATI
Constructing survey questions:
Likert scales
– Items (statements)
– Items must relate to the same object of study
– Inter-related items on a scale
– Equal number of positive and negative items
– Usually five point scale
– Example:
Strongly
agree
agree Neither agree
/disagree
disagree Strongly
disagree
Let’s construct some
examples of survey questions
(10mins)… Do they measure
what we expect them to?
See Whiteboard & Blank Pages for Working
Secondary datasets
– Secondary analysis of data collected by other researchers
– Secondary analysis of official statistics (i.e., stats collected by government departments)
– E.g., Ministry of Justice
– Home Office UK
– ESRC
Let’s Brainstorm – Would
secondary data sets be useful to
test your research question? Would
there be advantages or
disadvantages to using them?
See Whiteboard & Blank Pages for Working
Advantages of secondary analysis
– Cost and time
– High quality data (e.g., representative samples)
– Possibility of longitudinal analyses
– Possibility of subgroup analyses
– Possibility of new interpretations
– Unobtrusive method
Disadvantages of secondary
analysis
– Lack of familiarity with the data
– Complexity of the data
– Data quality control (knowledge of refined parameters for data mining)
– Useful- UK data archive www.data-archive.ac.uk
Example of Critical Appraisal - Crime in England and Wales
Home Office Statistical Bulletin January 2006
The risk of being a victim of crime (23%); lowest since the survey (BCS) began in 1981
Violent crime is stable compared with the previous year (BCS) Domestic burglary recorded by the police decreased by 7% (remember
this percentage)
The ‘Dark Figure’ (unreported crime which we will show still exists even in the 2006 statistics…)
Percentage risk of being a victim based on British Crime Survey
interviews in the 12 months to September 2005 compared with the
previous 12 months [adapted from the Home Office Statistical
Bulletin, (Kara & Upson, Jan 2006)]. See infographic below:
0
5
10
15
20
25
30
Interviews in Oct
2003 to Sept 2004
Interviews in Oct
2004 to Sept 2005
All household crime
Domestic burglary
All vehicle thefts
All personal crime
With injury
With no injury
All BCS crime
‘All personal crime’ dropped significantly at the p = .05 level, two tail tests
Domestic burglary didNot drop 7% according To the BCS! Multiple sources of official statistics…
Does some crime ‘disappear’ from official
statistics (i.e. police reports) or ‘remain pending’
from the event to conviction? Simmons Report
(2000)
Step 1. Does the victim notice the crime? Victimless crime Unnoticed crime
Step 2. Will the victim report the crime? Fear (humiliation, being blamed) Beliefs about police competency
Step 3. Will police record the crime? Notifiability (e.g., Ford vs Ferrari wing mirror) Change in policy over time (e.g., Steven Lawrence Inquiry) Political issues (e.g., Gudjonsson et al., 2004) Missing persons are in the inquiry stage Policing decision processes (e.g., malicious accusations)
Step 4. Will the offender be caught? Not all reported crimes are solved
Decision process (e.g., 15-16yr old shoplifting)
Step 5. Will the offender be prosecuted? Seriousness of offence
Strength of the evidence (e.g., Billy Jo Jenkins)
Enough evidence to go to trial (i.e., Crown Prosecution Service decisions)
Step 6. Will the offender be found guilty if he/she is in fact guilty?
Justice is a balancing of many complex considerations
Manslaughter versus Intent
Time to discuss! What are the
main issues and debates
likely to be in legal empirical
research?
See Whiteboard & Blank Pages for Working
General discussion…
– Are there differences between the social and natural worlds?
– … Precision at a price?
– Are there generalisations at the expense of the individual?
– … Individual or contextual factors?