• No results found

4. Prior LEED research

4.5 Regression estimates

The graphical analysis in the previous section neatly summarises the main findings of the paper. Crichton et al (2005) also presented results from regression analysis. This was undertaken to control for variables, such as recent prior benefit receipt, not used in the matchings. This estimated the impact of injuries on future employment, benefit, and earnings experiences. This was looked at by duration of compensation and the characteristics of the individual. The regression specification used in the paper was as follows:

i

l

Duration injury duration l X

Y

 

()*[ * ( )]

 

(1)

Y is the outcome of interest (ie employment or benefit receipt).

α is the model intercept.

Injury is a dummy variable indicating whether they are from the injured or matched control group.

Duration is a dummy variable indicating the length of the individual’s ACC spell.

X is a vector of variables to control for other factors influencing the outcome. These are age, sex, region, months employed and earnings levels prior to injury, benefit receipt prior to injury, firm size, industry, and month of injury.

μ is an error term to capture unobserved effects.

The key coefficient of interest for Crichton et al (2005) is

Duration(l) which represents the effects of injury spells of varying duration on outcome Y. All results are based on ordinary least squares (OLS) regressions.

Even controlling for these variables, Crichton et al (2005) found higher rates of benefit receipt among the injured during the 6–48 months before injury, and lower rates of

35 employment. It seems likely that these differences would remain post injury. To control for the differences in prior benefit receipt and employment rates between the injured and non-injured populations, a difference-in-difference estimator was used. This compares the difference in, say, benefit receipt rates for the two groups prior to injury, with the difference post injury (ie the difference between two differences).

Table 4 in Crichton et al (2005) summarises the regression estimates for the effect of injury on employment rates, for variety of specifications. Those given below are those that control for the covariates and include the difference-in-difference estimator. They show the effects of injury on employment six months and 18 months after leaving ACC, compared to

employment rates 18 months prior to being injured.

Table 30–Regression estimates of effect of injuries on employment Individual match with covariates

(from Crichton et al, 2005 table 4) Effect of ACC spell of:

Six months after vs. 18 months before

10–12 month duration -0.096 -0.100

13–24 month duration -0.116 -0.162

Observations 730,720 426,080

As Crichton et al (2005) state, injuries with one to two months of compensation now have almost no impact on employment six months after compensation ends (column 7). There is still strong evidence that longer duration injuries have negative impact on the employment rates (though this effect has been dampened by using the difference-in-difference estimator).

For example, an ACC spell of more than seven months reduces the likelihood of employment six months after compensation ends, compared with 18 months before, by around 10 percent.

Column 11 captures the change in employment status 18 months after leaving ACC versus eighteen months prior to injury. The estimated impacts are similar to those after six months.

Crichton et al (2005) state that this suggests that injuries have long-term effects on individual labour market prospects.

Similarly to table 30, the effects of injury on benefit receipt six months and 18 months after leaving ACC, compared with 18 months prior, are shown in the table 31. The results

suggest, as Crichton et al (2005) state, that longer duration injuries increase the likelihood of receiving benefits six months after leaving the ACC spell, compared to 18 months prior to injury. For example, a six month spell on benefit increases the likelihood by 4 percent, and longer spells by around 5 percent. These are large increases compared with the benefit receipt rate of only 7 percent for the non-injured population. Column 9 shows there is no evidence that these impacts decline over time.

36 Table 31–Regression estimates of effect of injuries on benefit receipt

Individual match with covariates (from Crichton et al, 2005 table 5) Effect of ACC spell of:

Six months after vs. 18 months before

10–12 month duration 0.022 0.035

13–24 month duration 0.051 0.086

Observations 730,720 426,080

The paper also tests the hypothesis that the impact of injuries differs for different groups of individuals. Table 32 below summaries the effects of injury on changes in employment 12 months after leaving ACC, compared with 18 months prior to injury. The results are classified by sex, age, prior industry, prior earnings, prior benefit receipt, and prior employment.

Table 32–Regression estimates of effect of injuries on employment 12 months after versus 18 months before

(from Crichton et al, 2005 table 6)

Duration on ACC 1–2 months 3–4 months 5–7 months 8–24 months

(1) (2) (3) (4)

Manufacturing -0.005 -0.032 -0.063 -0.104

Transport, et al -0.016 -0.020 -0.087 -0.115

Construction 0.004 -0.031 -0.016 -0.087

Wholesale trade -0.016 -0.029 -0.041 -0.099 Retail trade -0.003 -0.034 -0.004 -0.105

Accommodation, restaurants -0.009 0.003 -0.097 -0.195 Finance, business, property -0.009 -0.047 -0.066 -0.136 Other services -0.018 -0.052 -0.183 -0.158

Education -0.025 -0.047 -0.108 -0.112

Health, community services -0.007 0.007 -0.026 -0.155 6mth prior earnings < 1750 -0.006 -0.023 -0.067 -0.178 6mth prior earnings 1750–2500 -0.004 -0.016 -0.056 -0.118 6mth prior earnings 2500–3250 -0.005 -0.044 -0.026 -0.106 6mth prior earnings >= 3250 -0.013 -0.039 -0.060 -0.095

37 As Crichton et al (2005) state, there is very little difference in the impact of short-duration injuries (ie 1–2 and 3–4 months) across the different groups of workers. However, there are noticeable differences in the impact of longer-duration injuries. For example, having an ACC spell of between eight and 24 months reduces employment 12 months after (as opposed to employment 18 months before) the spell by:

 11 percent for males; 14 percent for females.

 9 percent for those whose previous job was in construction; 20 percent for accommodation and restaurants.

 33 percent for those that were employed in 1–2 months out of the preceding 6 months; 12 percent for those employed in 5–6 months.

However, the authors found no difference in the impact of longer duration injuries (on subsequent benefit receipt, employment, and income) between those that had received income-tested benefits in the six months prior to the ACC spell, and those who had not. This result controls for differences in industry, age, gender, and benefit receipt in the preceding six months, as well benefit receipt in the previous 18 months through the use of the difference-in-difference estimator.

Those who had only been employed for 1-2 months of the preceding six months had worse employment outcomes than those that had been employed for 3-6 months.

Table 7 in Crichton et al (2005) presented similar estimates for different groups of individuals, this time looking at the impact on total income 12 months after leaving ACC, compared with 18 months prior to injury. The results in this table are less consistent than the previous one, but generally tell a similar story, with differences arising as ACC spell tenure lengthens34.

Related documents