Ejecución de los programas 2014-2020

Having analysed the effect of the enforcement proxy variables on the derived clusters and having gained further insight into the variation that exists, the analysis now moves on to look at the effects of enforcement on the regional clusters. Regional groupings were produced and these are listed in Table 6.3.1. The two-level multilevel models for regional clusters follow the same procedures used for the derived clusters, with PFA‟s as the level one variable and regional clusters, rather than derived clusters, as the level two variable.

6.6.1 Quarter 3 Multilevel Models using ZFPN_1000’s on Regional Clusters

In Table 6.6.1.1 the results of multilevel model development on regional clusters are detailed. The methodology follows that used to investigate the derived clusters producing initially three models - a null model, a variance components model and a third model examining the effects of ZFPN_1000‟s on the KSI rate. Detailed in Table 6.6.1.1 are the results of these models.

Table 6.6.1.1: Multilevel Negative Binomial Models of Effect of ZFPN_1000’s on Regional Clusters in Quarter 3

The variance components model, Table 6.6.1.1, has a significant variation between regional clusters. Detailed in the NB model are the results when the effects of ZFPN_1000‟s are added. Here the fixed effect of ZFPN_1000‟s are statistically significant, showing that any increase in enforcement, as measured by ZFPN_1000‟s, leads to a decrease in the KSI rate. There is also significant regional variation between clusters but there is no significant random variation, at the 5% level, related to the effect of ZFPN_1000‟s, although it is approaching significance with a P = 0.06.

The variation between regional clusters relating to the effect of ZFPN_1000‟s, is shown in Figure 6.6.1.1. This suggests that, for the regional clusters, there is a

trend suggesting lower KSI rates are associated with higher levels of enforcement.

Figure 6.6.1.1: Effect of Enforcement, ZFPN _1000’s, on Regional Clusters in Quarter 3

The fixed effects of ZFPN_1000‟s are significant in seven of the nine regional clusters, see Table 6.6.1.2, where clusters are ordered by ascending p-value.

The clusters which do not have significant effects, at the 5% level, would be significant at the 10% level and allied to the significance of the other seven regional clusters suggests that there is a general trend associating an increase in police enforcement with a decrease in KSI rates.

Table 6.6.1.2: Parameter Estimates and p-values for Fixed Effects of ZFPN_1000's on Regional Clusters in Quarter 3

Models based on

6.6.2. Quarter 3 Multilevel Models using ZLag1_FPN_1000’s and ZLag1_FPN_1000’s on Regional Clusters

Lagged multilevel models are developed on the regional clusters with the proxies for enforcement being ZLag1_FPN_1000‟s and ZLag2_FPN_1000‟s, equivalent to ZFPN_1000‟s lagged by one and two quarters respectively. The results of model development, for both lagged variables, are shown in Table 6.6.2.1. Here it can be seen that there are significant fixed and random effects to be found for all proxy variables, with both lagged variables having a

significant effect on the KSI rate.

Table 6.6.2.1: Multilevel Negative Binomial Models of Effect of ZLag1_FPN_1000’s and ZLag2_FPN_1000’s on Regional Clusters in Quarter 3

The variation in the fixed effect of the lagged proxy variables on each cluster is shown in Figures 9 and 10, Appendix 6b.

The fixed effects relating to both lagged variables are shown in Tables 9 and 10, Appendix 6c. For both variables seven out nine clusters are associated with significant effects of increased enforcement which is linked to a decrease in the KSI rate. This adds to the evidence suggesting a general trend associating an increase in police enforcement with a decrease in KSI rates.

6.6.3 Quarter 3 Multilevel Models using ZFPN_G16_1000’s on Regional Clusters

The proxy for enforcement used in this section is ZFPN_G16_1000‟s. Once more three models are produced – see Table 6, Appendix 6a. In the variance components model significant variation is found between clusters in relation to KSI rates. When the effect of enforcement is added, ZFPN_G16_1000‟s in the NB model, a significant fixed effect is seen where an increase in the number of ZFPN_G16_1000‟s leads to a decrease in the KSI rates. The variation in the

fixed effects between clusters is shown in Figure 11, Appendix 6b. There is however no significant variation, at the 5% level, between clusters in relation to ZFPN_G16_1000‟s. The parameter estimates for the fixed effects of

ZFPN_G16_1000‟s are shown in Table 11, Appendix 6c, where all but one of the nine clusters have significant effects related to enforcement.

6.6.3.1 Quarter 3 Multilevel Models using ZLag1_FPN_G16_1000’s and ZLag2_FPN_G16_1000’s on Regional Clusters

Detailed results from modelling with the lagged proxy variables

ZLag1_FPN_G16_1000‟s and ZLag2_FPN_G16_1000‟s, are given in Table 7, Appendix 6a. For the fixed effect part of both models, both lagged variables have a significant effect linked to a decrease in KSI rates. For random effects neither lagged variable has any significant effect, at the 5% level,. Once again there is significant variation between clusters, in both the fixed and random part of the models. This is not unexpected as the clusters were developed in order to produce maximum variation between clusters. The variation between clusters in relation to fixed effects can be seen in Figures 12 and 13, Appendix 6b, and the parameter estimates showing the effect of the proxy variables are given in Tables 12 and 13, Appendix 6c. Both lagged proxy variables have a significant effect in seven out of nine clusters although the effect is seen in different clusters for each proxy.

6.6.4 Quarter 4 Multilevel Models using ZFPN_1000’s on Regional Clusters

Model development for regional clusters Quarter 4 data follows the same procedure as for derived cluster Quarter 4 data. As before a null model, a variance components model and a Negative Binomial model are produced and results are given in Table 6.6.4.1. In Table 6.6.4.1, across all models, there is significant variation between clusters in relation to both fixed and random effects. There is also a significant fixed effect associated with ZFPN_1000‟s in the NB model and significant variation across clusters in the effect of

ZFPN_1000‟s.

Table 6.6.4.1: Multilevel Negative Binomial Models of Effect of ZFPN_1000’s on Regional Clusters in Quarter 4

This variation between clusters is shown in Figure 14, Appendix 6b. The parameter estimates and associated p-values are given in Table 6.6.4.2, where six of nine clusters have a significant effect indicating that enforcement is linked to decreasing KSI rates.

Table 6.6.4.2: Parameter Estimates and p-values for Fixed Effects of ZFPN_1000's on Regional Clusters in Quarter 4

Models based on

6.6.4.1 Quarter 4 Multilevel Models using ZLag1FPN_1000’s and ZLag1_FPN_1000’s on Regional Clusters

Results for multilevel models using Quarter 4 data, with variables

ZLag1_FPN_1000‟s and ZLag2_FPN_1000‟s, are shown in Table 8, Appendix 6a. In this table, for fixed and random effects, both lagged variables are

significant in relation to the KSI rate. There is also significant variation between clusters in relation to both fixed and random effects. The variation between clusters relating to the fixed effect of the lagged proxy variables is shown in Figures 15 and 16, Appendix 6b. For ZLag1_FPN_1000‟s has a significant effect on seven out of nine regional clusters, see Table 14, Appendix 6c. The variable ZLag2_FPN_1000‟s has a significant effect on the KSI rate in six out of the nine clusters; this is shown in Table 15, Appendix 6c. The effect in for both variables is associated with a decrease in the KSI rate.

6.6.5 Quarter 4 Multilevel Models using ZFPN_G16_1000’s on Regional Clusters

Following the methodology of previous sections three models are developed using the proxy for enforcement ZFPN_G16_1000‟s and the results can be seen in Table 9, Appendix 6a. There is a significant variation between clusters in the variance components model and this fixed effect between clusters can be seen in Figure 17, Appendix 6b. When the effect of enforcement is added, in the NB model, Table 9, Appendix 6a, this also has a significant fixed effect, relating an increase in the number of ZFPN_G16_1000‟s to a decrease in the KSI rates. There is no significant random variation found between clusters, at the 5% level, in relation to ZFPN_G16_1000‟s. Parameter estimates for the fixed effects of ZFPN_G16_1000‟s are shown in Table 16, Appendix 6c, where seven of the nine clusters have a significant effect related to enforcement.

6.6.5.1 Quarter 4 Multilevel Models using ZLag1_FPN_G16_1000’s and ZLag2_FPN_G16_1000’s

The results from the final set of models, using the lagged proxy variables ZLag1_FPN_G16_1000‟s and ZLag2_FPN_G16_1000‟s are shown in Table 6.6.5.1.

Table 6.6.5.1: Multilevel Negative Binomial Models of Effect of

ZLag1_FPN_G16_1000’s and ZLag2_FPN_G16_1000’s on Regional Clusters in Quarter 4

Here, for both fixed and random effects, there is significant variation between clusters. This variation is shown in Figures 18 and 19, Appendix 6b. Both lagged variables have significant fixed effects linked to a decrease in KSI rates. However, only ZLag1_FPN_G16_1000‟s has a significant random effect.

Parameter estimates for the effect of ZLag1_FPN_G16_1000‟s and

ZLag2_FPN_G16_1000‟s on individual clusters are given in Tables 17 and 18, Appendix 6c, and in both cases six of the nine clusters have significant effects related to the enforcement proxies.