D activity
Yüklə 401,38 Kb.
|
We have now seen that Invest NI clients experienced greater levels of net job creation than other non-assisted private sector plants in Northern Ireland. However, what we do not know is whether this job increase would have happened anyway in these Invest NI assisted firms. This section summarises the findings of our analysis of a firm-level/plant-level longitudinal database for Northern Ireland constructed from official business surveys30 and linked to Invest NI client records (see Chapter 3). This new panel dataset contains data on around 16,000 firms of which around 1,400 are Invest NI clients.4.4.1 How do Invest NI Assisted Firms Compare with Non-Assisted Firms?What determines the likelihood that a firm in Northern Ireland is a client of Invest NI? We examine this probability using a probit model and our findings show that Invest NI clients are different from non-assisted firms in Northern Ireland in the following ways – they are:
This different profile reflects a potential selection issue – either Invest NI select firms with these characteristics or firms with these characteristics approach Invest NI for assistance – any attempt to model the impact of Invest NI financial assistance to individual firms needs to take account of this issue. 4.4.2 Trends in GVA and Turnover per Employee: 2001-2008 We now focus on a sub-group of firms which have annual data on employment, turnover and GVA over the period 2001 to 2008 (Figure 3). In total, we track 480 firms (253 Invest NI clients and 227 non-assisted firms). Since 2004/05 there has been an upward trend in GVA per employee in Invest NI clients compared to non-assisted firms.In contrast to GVA, we can see that turnover per employee remained virtually unchanged between the start and end points for both assisted and non-assisted firms (Figure 4). Over the period the trend for Invest NI clients appears to be slightly cyclical whilst it is more stable for non-assisted firms. 4.4.3 What Impact has Invest NI had on the Private Sector since 2001? We answer this question by using the panel datasets discussed in Chapter 3 to estimate three econometric growth models for employment, sales and GVA. We start by adopting our preferred method of a 2-stage Heckman model which, as explained in the previous section, controls more effectively for the selection bias inherent in this type of assistance. The sole objective is to isolate the net effects of Invest NI assistance after controlling for the differing characteristics of the two groups of firms (i.e., Invest NI clients and other non-assisted firms). In other words the ‘counterfactual’ is embedded within the analysis.In summary, the results show that, ceteris paribus, Invest NI assistance in the period 2001-08 was found to have:
The detailed results supporting these summary conclusions are set out in Tables 2-4 below. Table 2: Treatment Model containing first stage probit (determinants of being an INI client) and second stage OLS (determinants of GVA growth)
Probit
OLS
Table 3: Treatment Model containing first stage probit (determinants of being an INI client) and second stage OLS (determinants of Turnover growth)
Probit
OLS
Table 4: Treatment Model containing first stage probit (determinants of being an INI client) and second stage OLS (determinants of Employment growth)
Probit
OLS
4.5 Alternative Econometric Techniques 4.5.1 Difference in Differences Estimation The impact of a policy/treatment on an outcome can be estimated by computing a difference in differences (DID) model. It works by estimating a difference over time (before and after treatment) and a difference across subjects (between beneficiaries and non-beneficiaries) and produces an estimate of the impact of the treatment. If we were to just measure the difference in outcomes between beneficiaries and non- beneficiaries, measured after the intervention has taken place we may leave ourselves open to selection bias. However by incorporating data on the outcome variable for beneficiaries and non-beneficiaries observed before the intervention takes place we can then subtract the pre-intervention difference in outcomes from the post-intervention difference to eliminate selection bias related to time-invariant individual characteristics. Thus, the DID estimator works on the principle that if what differentiates beneficiaries and non-beneficiaries is fixed in time, subtracting the pre-intervention differences eliminates selection bias and produces a plausible estimate of the impact of the intervention. In our case we want to identify the effects of Invest NI assistance on performance measures (GVA growth, turnover growth and employment growth). We first identify firms that have received assistance and then compare changes in GVA (for example) for assisted firms to non-assisted firms, and across two periods (pre- and post-assistance). The resulting DID estimator is an unbiased estimate of the effect of the assistance if, without the assistance, the average change in GVA would have been the same for treatment and controls. Obviously the key element of DID estimation is this latter assumption, known as the ‘parallel trend’ i.e. that the counterfactual trend is the same for treated and non- treated units. This, along with other elements suggest that DID estimation may be limited in terms of its usage for certain evaluations. A pre-condition of the validity of the DID assumption is that the program is not implemented based on the pre-existing differences in outcomes and therefore it is only appropriate to use when the interventions are as good as random; conditional on time and group fixed effects. Use of this method on Invest NI assistance may, therefore, not be suitable due to the fact that we are not certain that assistance is assigned on a random basis (as opposed to a set criteria based on performance measures) and in fact a Heckman selection model has shown that selection is not on a random basis. Also the assistance data we have does not cover a specific set period of time, rather it can be ongoing or sporadic and also firms (both assisted and non-assisted) may have received similar interventions in the ‘before’ period, under a previous intervention programme. We therefore proceed with the econometric estimation with caution. As Invest NI came into existence in 2001 we use the post-2001 period as the ‘after’ intervention period and the pre-2001 period as the ‘before’ intervention period. However, as we only have data from 1998 onwards we restrict the ‘after’ period to end in 2004 so that the periods before and after the intervention cover an identical time scale. The resulting difference in difference estimates for the period 1998-2004 are set out in Table 5. Table 5: DID Estimates 1998-2004
Discussion of Results: Turning firstly to the GVA growth model, the results show that:
The results for turnover growth show that:
The employment growth model displays a different set of results from the previous two, it shows that:
The negative and significant findings for the INI interaction term in the employment growth model are not what we would expect a priori; as was stated earlier this could be due to the fact that in the model we assume that assistance was given in 2001. However, our data do not actually correspond to this – INI clients may have received assistance in any year, or combination of years, post-2000 so the 2001 cut off is not strictly accurate. In addition they may have received assistance prior to 2001. 4.5.1 Propensity Score Matching (PSM) The central idea of propensity score matching is to use a control group to mimic a randomized experiment. It uses information from a pool of units that do not participate in an intervention to identify what would have happened to participating units in the absence of the intervention. By comparing how outcomes differ for participants relative to observationally similar nonparticipants, it is possible to estimate the effects of the intervention. Yüklə 401,38 Kb. Dostları ilə paylaş:
|