Social Behaviour in Police Interviews: Relating Data to Theories - Conflict and Multimodal Communication: Social Research and Machine Intelligence - page 314

Table 16.1 The ten items
loading highest on the first
suspect factor

Item Factor loading
“Building pressure” 0.96
“Interruptions” 0.95
“Aggressive behaviour” 0.94
“Angry behaviour” 0.94
“Steering a conversation” 0.93
“Accusing the other” 0.92
“Attacking behaviour” 0.92
“Cutting the other off” 0.91
“Worked-up behaviour” 0.90
“Raised voice” 0.88
This factor was interpreted as dominant and
opposed

step 2 (see Fig. 16.1 ). The fragments were scored by asking the rating questions
explained above for both the police officer and the suspect, resulting in every
question (corresponding to a term) being scored 28 times.
We performed a factor analysis to find a clustering of correlated questions which
indicated which categories of questions—and, by extension, which terms—are
related (step 6 from Fig. 16.1 ). Questions that were scored with no variation—
that is, they always received the same score—for either the police officer or the
suspect were excluded from analysis. This resulted in nine questions being removed
(two were excluded from analysis for the subject, seven for the police officer). The
excluded terms were found to be very role-specific; for example, crying is something
the police never does. The factor analyses (extraction method: Principal Component
Analysis, rotation method: Varimax with Kaiser Normalisation) revealed 13 factors
for both the suspect and the police, see Tables 16.2 and 16.3 .
Based on the related questions, we determined which terms loaded strongly
(having a correlation of more than 0.50) per factor. The observers used these terms
to interpret the corresponding factors (step 7 from Fig. 16.1 ). For example, the
first factor (explaining 19:4 % of the variance) for the suspect was interpreted as
dominant and opposed . In Table 16.1 , we show only the first 10 (of 54) items with
factor loadings for the first factor of the suspect.
A subgroup of four of the original six observers interpreted all factors. The
consensus on keywords describing the strongly loading factors of the suspects and
the police officers is reported in Tables 16.2 and 16.3 , respectively. In general,
the observers' interpretations were similar. For example, one of the observers
interpreted the first factor of the suspect as “negative, confrontational and domi-
nant”, while another observer interpreted it as describing “dominant behaviour and
frustrations”. Discussing the interpretations among the observers generally resulted
in agreement on the meaning of the factors. Some factors (suspect factors 9, 11,
and 13 and Police factor 8) remain unclear as the observers were unable to reach
consensus. We attribute this confusion to the few and diverse items that load on
these factors.