Q- sorting of photographs
7. Measuring ‘the fear of crime’
7.3 Reliability, scales & dimensions of concepts
To enable the use of structural equation modelling (SEM) in the explanatory part of this chapter (section 7.4), we will now search for coherence among items that were intended to measure elements of theoretical concepts. This was done by checking on their correlation coefficients and reliability (by calculating Cronbach’s alpha).
The next step was factor analysis. Principal Axis Factoring (PAF) was the exploratory factor extraction method of our choice, for the following reasons: (I) the factors were expected to have fairly simple patterns based on relatively few indicators (De Winter & Doudou 2011:695); (II) furthermore, we expected the factors to be correlated (ibid:708) due to their theoretical similarity; and (III) our goal with factor analysis was to find out whether or not there were substantial differences
(ibid:708; Thompson 2004) between the factors. Inter-correlation between the factors was expected due to them being the outcomes of the same ‘underlying factors’ or ‘hypothetical constructs’ (Kim & Mueller 1978:15). This made oblique rotation the most suitable strategy for our factor rotation.
7.3.1 Personal vs. situational fear of crime
The items used to measure ‘personal fear of crime’ (V1A; V1B; V4A; V4B_HIC; V7; V8;
V10.1; V11.1; V12B & V12C) and ‘situational fear of crime’ (V18.1 to V18.7 & V19.1 to 19.4) showed significant correlations at at least the .05 level. But the strongest correlation coefficients were within the separate categories. Their combined reliability is rather good (α =. 84), but this was primarily due to the items related to
‘situational fear of crime’, since those related to ‘personal fear of crime’ seemed to add little reliability to the scale.
Principal Axis Factoring (PAF) showed four eigenvalues >1 (6.43; 2.16; 1.22; 1.10). With theoretical considerations in mind, two factors were interpreted. These two factors explain 37.96% of variance and their correlation coefficient is .23.
MEASURING ‘THE FEAR OF CRIME’
The first factor (tab. 16) shows relatively low loadings on the items of ‘personal fear of crime’ and high loadings on the items of ‘situational fear of crime’, while the second factor shows roughly equivalent, but mirrored loadings. This presents a strong empirical argument for treating ‘personal fear of crime’40 and ‘situational fear of crime’41 as separate concepts.
For the concept of ‘personal fear of crime’, reliability is relatively poor with an alpha of .68. But this is due the fact that it comprises an assembly of very diverse cognitive and affective elements.
7.3.2 Neighbourhood and societal fear of crime The items used to measure ‘neighbourhood fear of crime’ (V2A.1; V5A.1; V5B.1_HIC; V6.1) and ‘societal fear of crime’ (V2A.3; V5A.3; V5B.3_HIC; V6.3; V10.2; V11.2; V13B;
V13C ) show significant correlations at the .05 and .01 levels. But, again, the strongest correlation
coefficients are within the separate categories.
Principal Axis Factoring (PAF) showed five eigenvalues >1 (3.22; 1.55; 1.31; 1.95; 1.05). All five factors were interpreted. Together these five factors explain 49.03% of variance. Correlations between the five factors (tab. 17) are not high in general, except for the moderate correlations of factor 1 with factors 3 and 5.
The first factor (tab. 18) shows relatively low values on the items of ‘neighbourhood fear of crime’
and high values on the items of
‘societal fear of crime’. The second factor is roughly equivalent to the first. The third factor only has relatively high loadings on the affective aspects of ‘societal fear of crime’ (V13B & V13C). And the fourth factor only has high loadings on thinking about high-impact crime at both levels of the
40 Significant correlation between all items at the .05 and .01 level; α = .68.
41 Significant correlation between all items at the .01 level; α = .92.
1 2
Tab. 17 – Correlations matrix for the factors of the items for neighbourhood and societal
fear of crime.
Tab. 18 – Factors for items of neighbourhood and societal fear of crime.
KEEPING TROUBLE AT A SAFE DISTANCE
neighbourhood (V5B.1_HIC) and society (V5B.3_HIC). The final factor primarily has high loadings on the standard items of both neighbourhood (V2A.1) and societal fear of crime (V2A.3).
This factor analysis leads to the conclusion that while cognitions and feelings of unsafety at the level of the neighbourhood and Dutch society may co-vary for some respondents, they are essentially separate concepts in the experience of
respondents. Therefore, neighbourhood fear of crime42 and societal fear of crime43 will be treated as separate concepts, albeit in the knowledge that they correlate on some aspects for some respondents.
7.3.3 Avoidance behaviour
The items used to measure ‘avoidance behaviour’ (V20.1 to V20.8) show significant correlations between all variables at the .01 scale. Combined, their reliability is rather good (α= .81), but some items (V20.5 to V20.8) contributed relatively little to this scale.
Principal Axis Factoring (PAF) showed two eigenvalues >1 (3.02; 1.26). These two factors explain 46.91% of variance and their correlation coefficient is .48.
Relatively high loadings are seen for the first factor (tab. 19) on the first four items (V20.1 to V20.4) and for the second factor on the last four items (V20.5 to V20.8). These factors are fairly straightforward to explain from a theoretical point of view, since the first is about avoidance behaviour, while the second is more about a psychological state of mind when dealing with the risk of crime in the public sphere. For the concept of ‘avoidance behaviour’44, the choice was made to separate the items from the second factor (V20.5 to V20.7).
7.3.4 Social disorganisation
The items used to measure ‘social disorganisation’ in neighbourhoods (V17.1 to V17.5) show significant correlations between all variables at the .01 scale. Combined, their reliability is quite good (α= .70) and factor analysis (PAF) showed that they formed a single factor, explaining 33.07% of variance.
7.3.5 Societal discontent
The items used to measure ‘societal discontent’ (V22.1 to V22.10) show significant correlations at the .01 scale. Combined, their reliability is good (α= .88) and factor analysis (PAF) showed that they also form a single factor, explaining 43% of variance.
42 Significant correlation between all items on the .01 level; α = .62.
43 Significant correlation between all items on the .01 level; α = .74.
44 Significant correlation between all items on the .01 level; α = .82.
1 2
MEASURING ‘THE FEAR OF CRIME’
7.3.6 Conservatism, authoritarianism & dispositional fear
The correlation matrix for the items used to measure
‘authoritarian sentiments’ (A. - V21.1 & V21.3), ‘societal conservatism’ (C. - V21.4 & V21.5) and ‘dispositional fear’
(D. - V14.1 & V14.2) show significant correlations at the .01 level.
Principal Axis Factoring (PAF) showed two eigenvalues
>1 (3.15; 1.37). These two factors explain 61.92 % of variance and their correlation coefficient is .29. The first factor (tab. 20) has high loadings on the items of
‘societal conservatism’ and ‘authoritarian sentiments’
with low loadings on the items of ‘dispositional fear’.
The second factor mirrored the first.
This advocates ‘dispositional fear’45 to be a separate concept in the experience of respondents. The option of treating the items of ‘societal conservatism’ and
‘authoritarian sentiments’ as separate concepts is more theoretical, since, empirically speaking, co-variation suggests they are part of the same factor. But looking at the correlation matrix for these items (tab. 21), we see that the correlations between each pair of items (in bold) is relatively high.
So, ‘authoritarian sentiments’46 and ‘societal conservatism’47 will be used as separate concepts in the following, although it is clear that these concepts are aligned.
7.3.7 Psychological defenses
The items used to measure psychological defenses (V14.7 to V14.12) have significant correlations at the .01 level for all items. Principal Axis Factoring (PAF) showed three eigenvalues >1 (1.59; 1.17; 1.03). These three factors explain 33.57% of variance.
Correlations between the factors are negative (tab. 22).
45 Significant correlation between all items at .01 level; α = .68.
46 Significant correlation between all items to .01 level; α = .88.
47 Significant correlation between all items to .01 level; α = .85.
1 2 ** Correlation is SIGNIFICANT at the 0.01 level.
Tab. 21 – Correlations between the items used to measure ‘authoritarian sentiments’ and ‘societal
KEEPING TROUBLE AT A SAFE DISTANCE
For the first factor (tab. 23), we see that two general indicators of defenses - ‘tending to ignore unpleasant facts’ (V14.8) and ‘being frequently told not to show feelings’ (V14.9) - are connected with negative loads on the crime-specific defenses of ‘displacement’
(V14.11) and ‘suppression’ (V14.12).
For the second factor, we see a crucial role for the crime-specific defense of ‘rationalisation’ (V14.10) and relatively smaller roles for ‘displacement’ (V14.11) and
‘suppression’ (V14.12), with low and even negative loadings on the general indicators of defenses (V14.7-V14.9).
For the third factor, we see coherence between the general indicator of defenses of ‘being good at keeping problems out of one’s mind’ (V14.7) and ‘tending to ignore unpleasant facts’ (V14.8) with the crime-specific defense of ‘suppression’ (V14.12). In contrast with previously discussed concepts, all elements of psychological defenses must be seen as separate cognitive phenomena that can co-exist (Sandler 1985, Cramer 2006; Valliant 1992). This is supported by the general weak reliability of the items used (α = .08). But since the nature - being unconscious - and the function - bolstering the self-image – of these phenomena are of the same order, it is justified to treat them as a latent construct together.
7.3.8 Vulnerability
The items used to measure psychological (V16.1) and physical vulnerability (V16.2) have a correlation coefficient of .44, which is significant at the .01 level. Their reliability is rather low (α= .61). But Principal Axis Factoring (PAF) showed that together they formed a single factor, explaining 44.28% of variance. Therefore they are held to form one concept of ‘vulnerability’.
Now that we have a clear inferential position for each of the relevant concepts, we can turn to the actual explanatory models. The observed items used to compose the latent variables in the structural equation models correspond with the dimensions for these concepts, as explored in the previous section.