At this point in the analysis we have seen no evidence to suggest that exposure to the SARP programme had any effect (be it positive or negative) on the likelihood of residents reporting LLTI relative to comparator area residents. In addition, and much the same as the results from the previous chapter, no evidence has been observed to suggest that the likelihood of experiencing LLTI (increased or decreased) between different migrant groups and those who remain in regeneration areas relative to residents who remain in comparator areas from 1991 to 2001.
Therefore, with this final empirical section of this chapter I employ the difference in difference (DiD) approach to investigate the impact of the SARP programme on the likelihood of experiencing morbidity from a further longitudinal angle which attempts to account for national trends in morbidity prevalence.
The full description of the DiD approach is provided in Chapter 3. However, to provide a brief recap with relevance for this chapter, the aim of the DiD method is to measure the average impact of a policy programme on a specific outcome. The approach differs from the previous cross sectional and longitudinal selective migration analyses by comparing the difference in the likelihood of morbidity among residents in SARP areas before and after the programme with that for comparator area residents. It therefore takes account of the changes occurring both in the treated group and comparator group, in effect the national trend, to identify whether the programme has had any net
effects on the treated group minus the general change reflected in the comparator group. In this case, it is expected that between 1991 and 2001, even without the SARP programme, the level of LLTI would change in the treated and comparator areas reflecting the Scotland-wide change, which as was noted earlier, appeared to increase from 1991 to 2001. Explanations for this may well be due to a change in wording of the question in 2001 (Boyle 2004) or other aspects such as the consequences of selective migration, (which this analysis seems to show is not significant) or hidden unemployment (Marshall 2011).
The rationale for employing the DiD method in addition to the analytical techniques used above is that it appears to offer key additional advantages over these techniques in controlling for unobserved confounders. For example, it can remove any possible unobserved external differences in the SARP and comparator groups that may lead to better outcomes. These might be, for example, some of the factors listed above such as migration, hidden unemployment or indeed wider factors that may impact on morbidity such as, economic growth, or decline.
175 Here I will return to the first of the two research questions investigated in this chapter, which is set up to investigate whether living in SARP areas have had a positive or negative impact on the likelihood of suffering morbidity compared with living in similarly deprived comparator areas that did not receive the programme. However, despite the above mentioned advantages of the DiD method, neither of the analyses in the previous two empirical sections demonstrated any impact of the SARP programme in a positive or negative regard. I therefore hypothesise (hypothesis 3) that the results here will also be unable to demonstrate that the SARP programme has had a net positive or negative impact on likelihood of experiencing morbidity for residents in the treatment group.
To investigate hypothesis 3, the difference in difference technique is applied using two sets of fixed- effect regression models where LLTI is the dependent variable firstly, followed by hospital admissions in the second set (see chapter 3 for full description of DID technique). The analysis investigates whether SARP had any effects on the likelihood of experiencing morbidity in the SARP group net of the general change reflected in the comparator group over the ten year period between 1991 and 2001. As with the previous analyses, progressively more explanatory variables are added to help identify those factors that affect the likelihood of LLTI:
Model M: Examines the net effect of the SARP programme with no control for individual or household characteristics.
Model N: Estimates the net effect of the SARP programme on LLTI by controlling for individual explanatory variables that I expect to impact on the likelihood of an individual experiencing LLTI. These variables are marital status and social class.
Model O adds additional explanatory variables that may act as confounders in the relationship between the SARP programme and the likelihood of individuals experiencing LLTI. These are educational qualifications, housing tenure, car ownership and household type.
This sequence is repeated for hospital admissions (Models P, Q and R):
Table 5-7 below presents the results of the DiD analysis for LLTI whilst Table 5-8 presents the results for hospital admissions. Model M shows that after the implementation of the SARP, residents living in SARP areas were significantly more likely to experience LLTI than residents living in the comparator areas (OR= 8.23 p<0.001). However, the additional models demonstrate that after we account for individual and household characteristics, hypothesis 3 is upheld. The result for Model M again therefore underlines the importance of controlling for individual characteristics. The results here therefore reinforce the findings from the repeated cross section and selective migration analyses as even when the national trend is accounted for over the ten year period the SARP
176 programme is again shown to have had no effect (positive or negative) impact on likelihood of experiencing LLTI for SARP area residents relative to comparators.
In regards to the analysis for hospital admissions, Model P shows that after the implementation of the SARP, residents living in SARP areas were significantly more likely to be admitted to hospital than residents living in the comparator areas (OR= 1.50 p<0.001). However, the additional models demonstrate that after we account for individual and household characteristics, hypothesis 6 is confirmed. The results here therefore reinforce the findings from the repeated cross section and selective migration analyses as even when the national trend is accounted for over the ten year period the SARP programme is again shown to have had no positive or negative impact on likelihood of being admitted to hospital for SARP area residents relative to comparators
The results from Model N for LLTI demonstrate similar findings to those observed in the earlier analysis in regards to economic status, and demonstrate that in this model, economic status is a key driver in regards to likelihood of experiencing LLTI. Thus, beyond the somewhat obvious finding that the permanently sick are the most likely of the economic status categories to report LLTI compared to those in full-time employment (OR=1128 p< 0.001), we observe that the retired are the next most likely group to experience LLTI (OR=6.44 p< 0.001). As mentioned earlier in this chapter, advancing age will play a major contributory factor in this result. We then observe the more surprising result that students are approaching five times more likely than full-time workers to report LLTI (OR= 4.71 p< 0.001). Students on the whole tend to be younger individuals, thus one would have expected this group to be less likely or not significantly different to the full-time group in regards to likelihood of experiencing LLTI. However this result attenuates and becomes non-significant with the addition of variables pertaining to educational qualifications, housing tenure, household type, overcrowding in the household and car ownership. Economically inactive individuals are reported as just over four times more likely than full-time workers to experience LLTI (OR= 4.22 p< 0.001) whilst the unemployed are just under four times as likely (OR= 3.77 p< 0.001, results that are expected following the results from the modelling above.
One of the key findings from Model O (beyond the fact that there are once again no net positive or negative impacts on the likelihood of experiencing LLTI for SARP residents relative to comparators) is that the strength of the impact of economic status as a driver of LLTI remains and intensifies for some categories. These are permanently sick (OR=1151.33 p<0.001) retired (OR=6.56 p< 0.001) and other inactive (OR=4.36 p< 0.001) categories. In addition, we find that in terms of household type the difference in difference results also accord with results from the cross sectional and selective
177 migration modelling in that lone parents (OR=0.40 p< 0.05) and couples with children (OR=0.57 p< 0.05) are found to be less likely to report LLTI than couples with children.
Model O therefore demonstrates that the characteristics most likely to increase the odds of experiencing LLTI are being retired and living as part of a couple with no children, a result which is heavily influenced by ageing. In summary, the DiD estimation confirms hypothesis 3 by showing that the SARP programme had no net positive or negative impact on the likelihood of experiencing LLTI for regeneration area residents relative to residents in comparator areas net of all other variables.
Finally, based on the Models Q and R, several conclusions can be drawn in regards to what affects the likelihood of being admitted to hospital in disadvantaged areas. For example, residents who are divorced are more likely than single residents to be hospitalised (Model Q: OR=2.00 p<0.001; Model R: OR=1.80 p<0.01) whilst those who are retired are more likely to be hospitalised than those in full time employment (Model Q: OR=1.67 p<0.001; Model R: OR=1.82 p<0.001). The extended model (Model R) also demonstrates that those who did not state their qualifications were almost two and a half times (OR=2.76 p<0.001) more likely than those with no qualifications to be hospitalised.
In addition to the above, the sensitivity analyses for LLTI (Appendix 7) showed one difference in that the result for one parent families became non-significant when the permanently sick category was dropped from the economic status category. Furthermore, the sensitivity analysis for Hospital admissions A (Model R Appendix 8) also showed one difference on the lone parent category. However in this case the sensitivity model showed that one parents were significantly more likely to be admitted to hospital. Despite these differences on independent variables the sensitivity analyses did not show any differences in terms of net impact, thus demonstrating the robustness of the results reported here.
The DiD analysis is the most sophisticated of the three quantitative analytical techniques employed in this thesis. The results here therefore provide the most rigorous assessment of the impact of the programmes on morbidity for those who lived in SARP areas for the duration of the study period.At the end of the previous section I focused on (as a prelude to this section) the implications of finding no positive regeneration effect on likelihood of suffering morbidity for those who remained in SARP areas throughout the study period. I stated that this group were extremely important as they had the greatest exposure to the programmes. Thus finding that there had been no improvement in morbidity outcomes for this group indicated the programmes had been unsuccessful. I have with this analysis therefore went one step further in analytical sophistication to further attempt to uncover regeneration effects on this group by assessing the unique impact of the programmes by
178 using a technique that eliminates the influence of any unobserved and fixed (over time) effects on likelihood of morbidity. The results have again shown that the SARP programmes did not improve morbidity outcomes for those who remained in the regeneration areas throughout the study period and therefore lend further strength to the supposition that the SARP programmes have been unable to improve morbidity outcomes for residents. The DiD estimation therefore confirms hypothesis 3 by showing that the SARP programme had no net positive or negative impact on the likelihood of being admitted to hospital for SARP area residents relative to residents in comparator areas net of all other variables. The concluding section will reflect further on these findings.
179 Table 5-7 Fixed effects regression models predicting the net impact of suffering from LLTI in SARP areas by 2001 relative to comparator areas
LLTI Difference in Difference
Variable Category Model M (n= 7850) Model N (n= 7810) Model O (n=7622)
OR 95% CI OR 95% CI OR 95%CI
Net impact of suffering from LLTI in SARP areas by 2001 relative to comparator areas 8.23 *** 7.25, 9.35 1.27 0.86,1.46 1.14 0.87, 1.49 *Dummy variable 7.22 ** 5.80, 9.00 6.83 *** 5.31, 8.80 *Treatment variable 0.77 0.36,1.65 0.54 0.28,1.31
Marital Status Single (reference) 1 1
Married 1.4 0.60, 3.28 1.81 0.71, 4.61
Widowed 1.39 0.51, 3.78 2 0.71, 5.68
Divorced 1.68 0.65, 4.30 2.44 0.89, 6.70
Economic Status In full-time employment (reference) 1 1
In part-time employment 1.37 0.83, 2.25 1.43 0.87, 2.34 Self-employed 2.02 0.82, 4.99 1.66 0.66, 4.20 Unemployed 3.77 *** 2.31, 6.15 3.48 *** 2.11, 5.73 Student 4.71 *** 2.79, 4.95 2.51 0.99, 6.39 Permanently sick 1128.35 *** 426.78,2983.16 1151.33 *** 431.12, 3074.66 Retired 6.44 *** 4.05, 10.25 6.56 *** 4.07, 10.56 Other inactive 4.22 2.78, 6.40 4.36 2.83, 6.72
180
*** ***
Social Class Professional (reference) 0.31 0.16,1.61
Managerial 0.49 0.12,2.59
Skilled and non-manual 0.35 0.10,2.61
Skilled-manual 0.51 0.12,2.64
Partly-skilled 0.28 0.08, 1.64
Unskilled 0.37 0.11, 1.72
Never worked 0.38 0.06, 2.09
Qualifications No qualification and NCR Persons under 18 (reference)
1
Sub-degree 0.58 0.29, 1.18
Degree and higher degree 1.31 0.53, 2.40
Over 75 with a qualification 0.91 0.57, 1.44
Not stated 1.35 0.94, 1.93
House Tenure Owner occupied (reference) 1
181
Private renting 1.29 0.72, 2.30
Household type Married and unmarried couples with no dependent children (reference) 1
Unmarried adult 0.76 0.42, 1.37
One parent families with dependent children 0.40
*
0.19, 0.82
Married and unmarried couples with dependent children 0.57
*
0.37, 0.88
Car ownership 0 cars (reference) 1
1 cars 0.64 0.47, 0.87
2 cars 0.72 0.43, 1.19
3 cars 0.55 0.23, 1.27
Log Likelihood -419.58805 -845.72063 -807.09691
*Dummy variable represents the likelihood of suffering LLTI over time (2001 vs 1991) * p<0.05, **p<0.01, ***p<0.001 *Treatment variable is a dummy variable for living in the SARP areas or in comparator areas through time
182 Table 5-8 Fixed effect regression models predicting the net impact of being admitted to hospital in SARP areas by 2001 relative to comparator areas
Hospital admissions Difference in Difference
Variable Category Model P (n= 13898) Model Q (n= 13758) Model R (n=13378)
OR 95% CI OR 95% CI OR 95%CI
Net impact of likelihood of hospitalisation in SARP areas by 2001 relative to comparator areas
1.50 *** 1.41, 1.60 0.93 0.84,1.03 0.95 0.85, 1.06 *Dummy variable 1.42 *** 1.30, 1.55 1.28 *** 1.15, 1.41 *Treatment variable 1.42 ** 1.10,1.82 1.39 ** 1.08,1.79
Marital Status Single (reference) 1 1
Married 1.04 0.81, 1.33 1.25 0.94, 1.66 Widowed 0.91 0.65, 1.26 1.11 0.78, 1.58 Divorced 2.00 *** 1.44, 2.77 1.80 ** 1.26, 2.57
Economic Status In full-time employment (reference) 1 1
In part-time employment 1.08 0.92, 1.26 1.13 0.96, 1.33 Self-employed 0.82 0.60, 1.12 0.87 0.63, 1.19 Unemployed 0.91 0.76, 1.09 0.93 0.7 7, 1.11 Student 1.06 0.87, 1.28 1.20 0.86, 1.77 Permanently sick 1.23 * 1.04, 1.45 1.36 *** 1.14, 1.61 Retired 1.67 *** 1.42, 1.96 1.82 *** 1.54, 2.15 Other inactive 1.18 1.01, 1.37 1.37 1.17. 1.60
183
* ***
Social Class Professional (reference) 1
Managerial 0.86 0.55,1.37
Skilled and non-manual 0.83 0.52,1.32
Skilled-manual 0.87 0.55,1.38
Partly-skilled 0.87 0.54,1.38
Unskilled 0.85 0.53,1.38
Never worked 0.76 0.48,1.21
Qualifications No qualification and NCR Persons under 18 (reference) 1
Sub-degree 0.94 0.74, 1.20
Degree and higher degree 0.92 0.71, 1.19
Over 75 with a qualification 1.20
*
1.00, 1.43
Not stated 2.76
***
2.24, 3.40
House Tenure Owner occupied (reference) 1
Social renting 1.09 0.97, 1.23
184
Minimal household unit Married and unmarried couples with no dependent children (reference)
1
Unmarried adult 1.04 0.85, 1.28
One parent families with dependent children 0.82 0.63, 1.07
Married and unmarried couples with no dependent children 0.92 0.59, 1.43
Car ownership 0 cars (reference) 1
1 cars 0.9 0.81, 1.02
2 cars 0.87 0.73, 1.03
3 cars 0.81 0.61, 1.07
Log Likelihood -4727.1408 -4553.8169 -4368.2738
*Dummy variable represents the likelihood of being admitted to hospital over time (2001 vs 1991) * p<0.05, **p<0.01, ***p<0.001 *Treatment variable is a dummy variable for living in the SARP areas or in comparator areas through time
185