Regional Selective Assistance in Scotland: Does it make a difference to plant productivity?

Abstract

This paper examines whether receipt of a Regional Selective Assistance (RSA) grant has a causal impact on plant total factor productivity (TFP). To tackle the problem of self-selection into the treatment group, propensity score matching is employed. In order to control for the endogeneity of other variables in the model, estimations are performed using the system GMM estimator. The results show that for low technology manufacturing, receipt of an RSA grant leads to a fall in TFP.

Keywords

economic development economics Europe RSA grants

Introduction

Regional Selective Assistance (RSA) is the largest and oldest business support scheme currently operating in Scotland. On average, almost £50 million (2003 prices) of RSA grants were offered and accepted in Scotland per annum between 1973 and 2003. This is far larger than expenditure on other schemes such as the Small Firm Merit Awards for Research and Technology (SMART) and Support for Products under Research (SPUR) schemes for which the average value of grants offered and accepted per annum during the period of operation was £568,000 and £809,000, respectively.

RSA provides grants to plants undertaking investment projects in economically deprived ‘Assisted Areas’.¹ Eligible investment includes expenditure on land, buildings, plant, machinery and software. As a component of regional policy, the scheme is principally designed to safeguard and generate employment in the Assisted Areas. As such, many of the grants are given to help foreign firms establish plants in Scotland.² The amount that can be offered is determined by a number of factors including the location and size of the project and the number of jobs it will create or safeguard. In order to receive an RSA grant, an additionality criterion must be satisfied which requires that awards will only be made if the project could not have proceeded in the same form and time frame without the grant. A displacement criterion must also be met, which demands that the jobs created by the project must not be offset by job losses in other Assisted Areas in Scotland.

This paper examines whether receipt of an RSA grant has a causal impact on plant total factor productivity (TFP) in Scotland using a data set created by merging the Selective Assistance Management Information System (SAMIS), a register of RSA recipients dating from 1972 when the scheme began, into the longitudinal Annual Respondents Database (ARD). This is an important question given the importance of productivity in determining living standards.³ If RSA grants merely support plants that would otherwise be forced to close, these subsidies may impede the Schumpeterian process of ‘creative destruction’ that creates growth in the economy by shifting resources from low to high productivity plants (Schumpeter, 1943). On the other hand, if the resources previously employed by the plant would be redundant after closure of the plant, the subsidy would make a positive contribution to growth in the economy.

TFP is used instead of labour productivity because it is a better measure of efficiency and technology and will therefore be a stronger determinant of plant performance. Furthermore, labour productivity, unlike TFP, is determined by factor input levels in addition to levels of efficiency and technology (Harris, 2005). Receipt of a grant will have a positive impact on the employment of capital and labour because, in the case of capital, grants are only provided after expenditure on capital has occurred and, in the case of labour, ‘clawback’ clauses dictate that grants must be returned if employment targets are not met. This is shown empirically by Criscuolo et al. (2012). The impact of RSA on labour productivity will therefore be determined jointly by its impact on employment, capital and TFP and will therefore be difficult to interpret.

The analysis is conducted separately for high-tech, medium high-tech, medium low-tech and low-tech manufacturing. Such a disaggregation is necessary because different sectors will operate with different production technologies and the impact of RSA is therefore likely to differ across sectors.

RSA may have both positive and negative impacts on TFP. Harris (1991) argues that there are two main channels through which an RSA grant may improve TFP. The first is by allowing the acquisition of modern capital, which requires the reorganisation of the plant along more efficient lines. For example, movement to a new plant, the acquisition of new machinery or investment in information and communication technology may trigger a wholesale reorganisation of production processes that will have an impact on TFP. An RSA grant may also raise TFP by allowing the plant to create a new product that can be produced with greater efficiency than older products.

On the other hand, RSA may have a negative impact on TFP if firms can increase their profits by accepting the grant and increasing their use of capital and labour beyond the point at which they would operate if they did not receive the grant. In this situation, the grant can be regarded as compensation from the government to the firm for the loss of profits that the firm would incur from operating at a point with suboptimally high employment.

Previous studies have generally failed to find a statistically significant impact of receipt of an RSA grant on productivity. One exception is Harris and Robinson (2004) who, using a control group consisting of all untreated plants in Great Britain (GB), find a positive and statistically significant impact on TFP for RSA recipients throughout GB for the period, 1990–1998. However, when the model is estimated using a control group drawn from untreated plants from the assisted areas, they find that it is only RSA recipients in Scotland that received a large TFP boost. Their results also confirm that plants assisted by RSA are a self-selected group of the population of plants that have lower than average productivity before receiving the grant. Criscuolo et al. (2012), using ARD data covering the period 1988–2003, do not find a significant effect of RSA on either labour productivity or TFP in GB using an instrumental variables strategy. Finally, Hart et al. (2008), using the control functions approach on cross-sectional data taken from a telephone survey of Scottish firms, do not find a positive impact of receipt of an RSA grant on labour productivity. While the latter two papers use appropriate methods to tackle the consequences of self-selection into the treatment group, they fail to deal with the endogeneity of other explanatory variables in the model. Harris and Robinson (2004), by contrast, do not employ a sufficiently sophisticated method to control for self-selection but do control for the endogeneity of control variables. The strategy below addresses both sources of bias.

Methodology

Consider the following Cobb-Douglas production function:

Y_{i t} = A_{i t} E_{i t}^{β_{E}} M_{i t}^{β_{M}} K_{i t}^{β_{K}},

(1)

where Y_it is gross output in plant i at time t, E_it represents employment, M_it is intermediate inputs, K_it represents the capital stock and A_it is TFP. Taking natural logs of equation (1) gives:

y_{i t} = β_{E} e_{i t} + β_{M} m_{i t} + β_{K} k_{i t} + a_{i t},

(2)

where the lower case is used to denote the natural logarithm of a variable.

It is postulated that the natural logarithm of TFP can be modelled as follows:

a_{i t} = β_{X} x_{i t} + β_{A T T} D_{i t} + (η_{i} + υ_{i t}),

(3)

where x_it is a vector of variables thought to influence TFP (in which continuous variables are logged) and D_it is a dummy taking the value of 1 if a plant receives an RSA grant in that period or has done so in the past. The error term is composed of η_i, an unobservable, plant-specific, time-invariant effect and ν_it, a TFP shock. The RSA dummy is the key variable in the model as its coefficient, β_ATT, will provide the estimate of the impact of receiving an RSA grant on TFP.

The model is therefore:

y_{i t} = β_{E} e_{i t} + β_{M} m_{i t} + β_{K} k_{i t} + β_{X} x_{i t} + β_{A T T} D_{i t} + (η_{i} + υ_{i t}) .

(4)

The group of observations that receive ‘treatment’ are said to be self-selected when the decision of whether or not to receive ‘treatment’ is taken by the plant. In such a situation, the ‘treatment’ group is not a random sample of the population and will have characteristics that would lead to better or worse performance than observations in the ‘untreated’ group, in the event that neither group received treatment. A comparison of the mean of an outcome variable across the ‘treated’ and ‘untreated’ groups will then not provide an unbiased estimate of the average effect of ‘treatment’ on the ‘treated’ (ATT) (see, for example, Blundell and Costa Dias, 2009, for a more detailed exposition of self-selected bias).⁴

Assuming that all relevant characteristics are observed, differences in characteristics across ‘treated’ and ‘untreated’ groups can be controlled for using a correctly specified regression. However, in practice, finding the correct specification is difficult. This is a serious problem as estimating an incorrectly specified equation will generate biased estimates of the ‘treatment’ effect because the estimate of the dependent variable for ‘treated’ plants, in the event that they did not receive ‘treatment’, is entirely dependent on the specification of the model for values of the covariates for which only ‘treated’ plants are observed (see, for example, Blundell et al., 2005). One solution to this problem is to create a matched sample in which ‘treated’ and ‘untreated’ plants are observed for all values of the covariates. This is done here using propensity score matching (see Dehejia and Wahba, 2002), which creates a sample that includes plants that received a grant and a subset of plants that did not receive a grant but that have characteristics similar to the treated plants. This is accomplished by estimating a probit model of ‘treatment’ status including variables that determine both output and whether a plant receives an RSA grant, and then matching on the estimated predicted values.⁵ This approach will improve the match, or the balance, of the covariates across ‘treated’ and ‘untreated’ groups. The advantage of propensity score matching over other forms of matching is that it overcomes the difficulties of matching on a large number of variables (Zhao, 2004).⁶

The matching estimator, as commonly described (see, for example, Blundell et al., 2005), suggests that, once a matched sample has been constructed, a simple regression of the outcome variable on the ‘treatment’ dummy will provide an unbiased estimate of the ‘treatment’ effect. However, even within a matched sample, it is possible that large differences in the distribution of the covariates across ‘treated’ and ‘untreated’ observations may remain. This suggests that covariates should be included in the outcome regression to control for residual differences in the covariate distribution across ‘treated’ and ‘untreated’ groups. This approach is therefore a combination of matching and regression.⁷

Having constructed a matched sample, difficulties remain in the estimation of equation (4) because of the potential for simultaneity and attrition bias.⁸ Simultaneity bias arises because plants may have some knowledge about the value of the productivity shock and use this knowledge to choose the level of inputs in the production function (Marschak and Andrews, 1944). Attrition bias is present if plants base their exit decisions on their productivity level. As plants with a larger capital stock will be able to withstand lower productivity levels, this will generate a negative correlation between the productivity shock and the capital stock variable. Although the main variable of interest is the RSA dummy, it is essential to deal with the endogeneity of other explanatory variables if an unbiased estimate of the ATT is to be obtained. This point is made forcefully by Frölich (2008) who shows that the asymptotic bias of the estimate of the treatment effect can be large if the endogeneity of other variables in the model is ignored.

Therefore, the coefficients in equation (4) will be estimated using the system GMM estimator developed by Blundell and Bond (1998) (see Bond, 2002, for an introduction). This estimates the equation as a system, using lagged levels and lagged first differences of the endogenous variables as instruments for the equations in first differences and levels, respectively.⁹ The endogenous variables in the model that will be dealt with in this way are employment, intermediate inputs and capital. It is assumed that the variables in x_it and D_it are exogenous.¹⁰

Data

The ARD is a longitudinal data set containing financial variables such as investment, intermediate inputs and gross output collected at the reporting unit (or establishment) level on the basis of a stratified sampling frame. Other variables such as employment are given for all reporting units and plants (or local units) (see Griffith, 1999, and Robjohns, 2006 for more information on the ARD). When a plant cannot provide the full range of financial information necessary for the survey, another plant within the firm will report on their behalf and the reporting unit then consists of more than one plant. When a plant can provide the required information and only reports on behalf of itself, the reporting unit consists of only one plant.

The nature of the ARD presents two issues that need to be addressed when estimating the impact of receiving an RSA grant. First, the reporting unit is not an appropriate unit of analysis for estimating the impact of RSA grants because RSA is awarded to support investment in specific plants rather than throughout the enterprise. Second, the reporting unit is an accounting rather than an economic unit. As such, the number of plants covered by a reporting unit may change as enterprises open and close plants, buy and sell plants, or simply because of changes in the way that an enterprise chooses to report to the ARD (Harris, 2005a). To permit analysis at the more appropriate plant level, it is therefore necessary to ‘spread back’ to the plant those financial variables that are only available in the ARD at the reporting unit. This is done using the plant-level employment data available for all plants using the assumption of constant labour-investment ratios and labour productivity levels within reporting units.¹¹

To estimate the impact of receiving an RSA grant, it is, of course, necessary to identify which plants received assistance in the ARD. The SAMIS data base, which is maintained by the Department of Business, Innovation and Skills, has data on firms in Great Britain that applied for an RSA grant. This includes the postcode, standard industrial classification (SIC) code and employment level of the applicant firm and the date on which the application was made. For successful applications, the date when the payment was made and the value of the grant are also recorded.

Variables common to both SAMIS and the ARD can then be used to identify plants in the ARD that had received an RSA grant. The starting point for the linking process was the linkage achieved by Harris (2005b). Further links were added on the basis of work done by Criscuolo et al. (2012). If a plant in SAMIS had still not been linked, matching was done manually using postcodes, SIC codes, employment levels and application years. In the end, this process achieved a link from SAMIS to the ARD of 91.4% of the manufacturing plants that received an RSA grant in Scotland. This is higher than the level achieved in previous studies using data sets created by combining SAMIS with the ARD (Criscuolo et al., 2012; Harris, 2005b; Harris and Robinson, 2004) and therefore provides a firmer basis for empirical analysis.

Variables

The ARD contains information which allows the construction of a number of variables that can be included in x_it in equation (4).¹² The Herfindahl Index (Herfindahl, 1950) is a measure of the degree of concentration and, under certain assumptions (see, for example, Cabral, 2000), competition within an industry. Intuitively, it would be expected that greater competition (which implies a lower Herfindahl index) will pressure firms into adopting new technologies and operating more efficiently. However, following Schumpeter (1943), it can also be argued that competition will be inversely related to productivity if monopoly rents are required for management to invest in the R&D which leads to innovations and improvements in TFP (see, for example, Aghion and Howitt, 1992).

Spatial spillovers or agglomeration economies are benefits that accrue to plants from being located in the vicinity of concentrations of other plants. According to Duranton and Puga (2004), the mechanisms that give rise to agglomeration economies are the sharing of common pools of assets, matching between firms and workers and the transfer of knowledge between firms, all of which are facilitated by their spatial concentration. Agglomeration economies take two main forms: localisation (or Marshallian) externalities, and urbanisation (or Jacobian) externalities. The former arise from the concentration of plants from the same industry in a given area (Arrow, 1962; Marshall, 1890; Romer, 1986). The latter arise from diversity in the activities of plants in a particular area (Jacobs, 1970, 1986). In the analysis below, localisation externalities are measured by a variable measuring the proportion of industry output located within the local authority. Urbanisation externalities are measured by a variable calculated as the number of different SIC codes within the local authority.

Also included in the model is a foreign ownership dummy. This is justified by the observation that, to make it worthwhile for a foreign firm to incur the costs of setting up or acquiring a plant in the domestic market, foreign firms must possess characteristics that give them a cost advantage over domestic firms (Hymer, 1976). These characteristics may include specialised knowledge about production or better management, both of which would lead to higher TFP. On the other hand, cultural differences may cause a lack of understanding between management and labour (Dunning, 1988) which could reduce TFP in foreign-owned plants. Furthermore, foreign-owned firms may undertake FDI to source technology from the host economy rather than to exploit superior technology from the home country (Driffield and Love, 2007). Plants owned by foreign-owned firms that are motivated by technology sourcing are then likely to have lower TFP than domestically owned plants (Cantwell et al., 2004; Fosfuri and Motta, 1999). Foreign-owned plants would also be expected to have lower productivity if foreign-owned firms keep their high value production at home and leave lower value added operations to their foreign subsidiaries (Doms and Jensen, 1998).

A single-plant firm dummy, equal to 1 if that plant is the only plant owned by the firm, is also included in equation (4). Multiplant enterprises that serve a large geographic market where transport costs are relatively high will benefit from being able to locate plants close to their markets (Harris, 1989). Furthermore, single-plant firms will be at a productivity disadvantage compared with multiplant firms if technology is shared within multiplant enterprises (Jarmin, 1999). Conversely, multiplant firms may be less efficient if principal-agent problems are more severe in multiplant enterprises (Leibenstein, 1966). There may also be significant bureaucratic costs to organizing production over a large number of plants (Chandler, 1962). Finally, single-plant firms may be more innovative and flexible because they have access to a higher level of localised knowledge than multiplant firms (Kelley and Harrison, 1990).

An age variable is included among the covariates to measure whether younger plants use more modern vintages of capital embodying better technology than older plants, or if through learning-by-doing productivity increases as the plant ages (for example Jovanovic and Nyarko, 1996). Note that the measure of the capital stock used here (see Harris, 2005a; Harris and Drinkwater, 2000) is in theory adjusted to take account of vintage effects. Thus the coefficient obtained on the plant age variable should be an estimate of only the learning-by-doing effect. In practice though, it is unlikely that the capital stock estimates are fully adjusted for obsolescence.

Finally, a time trend is included in equation (1) to account for (Hicks-neutral) technical change. This is done to capture the impact on TFP of exogenous improvements in technology common to all plants. Further detail on all the variables used in the analysis is given in Table 1.

Table 1.

Variable definitions.

Variable	Definition	Source
RSA	Dummy variable indicating whether the plant received an RSA grant	SAMIS
Gross output	Plant sales (2000 prices)	ARD
Employment	Number of employees within the plant	ARD
Intermediate inputs	Plant intermediate inputs (2000 prices)	ARD
Capital	Plant and machinery capital stock (Harris and Drinkwater, 2000, updated)	ARD
Age	Age of plant in years	ARD
Time	Time trend calculated from 1984 base year	ARD
Herfindahl Index	Herfindahl Index of industry concentration (calculated at 4-digit SIC level)	ARD
Local authority industry share	Share of (4-digit) SIC industry output within local authority in which plant is located for industry in which plant operates	ARD
Local authority diversification	Number of (4-digit) SIC codes within local authority in which plant is located	ARD
FDI	Dummy variable indicating whether the plant is owned by a foreign enterprise	ARD
Single	Dummy variable indicating whether the plant is a single-plant enterprise	ARD

Table A1 in the Appendix shows the mean of these variables across plants that received an RSA grant and plants that did not receive an RSA grant. This table shows that RSA grant recipients are larger, in terms of output and employment, than plants that did not receive an RSA grant in all four sectors. They also have larger capital stocks and use more intermediate inputs. They operate in less concentrated industries (as shown by the lower value of the Herfindahl Index) and in less diversified areas which have a higher proportion of their industry’s output. RSA grant recipients are also older, less likely to belong to a multiplant enterprise and more likely to be foreign-owned. It is these differences in characteristics for which the combination of propensity score matching and regression is designed to control.¹³

Results

Rather than performing the analysis using the entire sample, the sample is split according to the sophistication of the technology employed in the production process. This is done to avoid the imposition of common coefficients across disparate industries. In particular, it is undesirable to impose common coefficients for labour, intermediate inputs and capital as different industries operate with different technologies. The sample is therefore split into high-tech, medium high-tech, medium low-tech and low-tech manufacturing. Plants were organised into these sectors according to the Eurostat classification of SIC codes (with some minor adjustments).¹⁴

The estimated parameters from the probit model used to generate the propensity scores are given in the Appendix. Table 2 gives the difference in the mean of the variables which will be used in the empirical analysis between plants that received an RSA grant and plants that did not receive an RSA grant. This provides an indication of the extent to which the match, or balance, of the covariates is improved across ‘treated’ and ‘untreated’ groups by moving from the full to the matched sample. In the full sample, the differences in the mean of the covariates are statistically significant at the 99% level for each variable in each sector with the exception of the time variable in the medium low-tech sector. Differences in the mean of the covariates are considerably reduced in the matched sample compared with the full sample to the extent that they are not statistically significant in the medium high-tech and medium low-tech sectors. However, significant differences remain in the high-tech and low-tech sectors. This provides support for the strategy employed of including covariates in the outcome regression to control for residual differences in the distribution of the covariates in the matched sample.

Table 2.

Difference in variable means between RSA-assisted and non-RSA assisted plants for different sectors, Scottish manufacturing 1984–2005.

Sector	High-tech		Medium high-tech		Medium low-tech		Low-tech
Sample	Full	Matched	Full	Matched	Full	Matched	Full	Matched
ln(Employment)	2.09***	−0.12	1.68***	−0.09	1.90***	0.00	1.50***	−0.03
ln(Employment)	(28.95)	(−1.58)	(33.69)	(−1.50)	(32.27)	(−0.05)	(44.62)	(−0.93)
ln(Intermediate inputs)	2.61***	−0.38***	1.82***	−0.08	1.67***	0.05	1.58***	−0.05
ln(Intermediate inputs)	(30.37)	(−3.44)	(32.44)	(−1.07)	(28.20)	(0.70)	(36.70)	(−1.02)
ln(Capital)	3.51***	0.48***	2.77***	−0.06	2.30***	0.12	2.06***	−0.09
ln(Capital)	(30.20)	(4.84)	(31.65)	(−0.67)	(26.29)	(1.28)	(33.35)	(−1.53)
ln(Herfindahl Index)	4.99***	4.83	−0.10***	−0.04	−0.30***	0.07	5.45***	5.58
ln(Herfindahl Index)	(3.25)	(−1.47)	(−4.02)	(−1.16)	(−10.49)	(1.70)	(−6.27)	(0.61)
ln(Local authority industry share)	1.05***	0.34	0.27***	0.05	0.18***	0.02	1.55***	1.27
ln(Local authority industry share)	(9.15)	(−1.50)	(6.30)	(1.08)	(5.29)	(0.52)	(9.73)	(−0.84)
ln(Local authority diversification)	−0.33***	0.12***	−0.10***	0.02	−0.10***	0.04	−0.10***	0.01***
ln(Local authority diversification)	(−12.35)	(3.25)	(−4.19)	(0.82)	(−4.70)	(1.08)	(−3.05)	(0.34)
ln(Age)	0.66***	0.39***	0.67***	0.03	0.55***	0.02	0.50***	0.00
ln(Age)	(18.48)	(9.02)	(23.37)	(0.89)	(18.97)	(0.57)	(25.38)	(−0.15)
Single	0.21***	−0.10***	0.08***	0.02	0.16***	−0.03	0.11***	−0.01
Single	(11.12)	(−2.75)	(5.62)	(1.10)	(10.84)	(−1.14)	(11.69)	(−0.82)
FDI	0.31***	0.05**	0.16***	0.00	0.09***	0.01	0.08***	0.02*
FDI	(16.48)	(2.03)	(11.55)	(−0.09)	(7.78)	(0.40)	(11.45)	(1.73)
Time	0.50***	−0.74*	0.87***	0.11	0.33	0.33	0.44***	0.47**
Time	(1.95)	(−2.42)	(4.40)	(0.41)	(1.56)	(1.16)	(3.17)	(2.55)
Number of RSA grants	89		86		87		196
‘Treated’ observations	746		1129		937		2198
‘Untreated’ observations	2799	388	8198	825	9964	758	24,494	1844

*, **, *** denote significance at the 90%, 95%, 99% levels, respectively.

t-statistics are in parentheses.

Table 3 gives the coefficients obtained from estimation of equation (4) using the matched sample. The instruments are lagged three times to allow for autocorrelation in the error term, which would invalidate instruments that are lagged an insufficient number of times. This proved to be a sufficient number of lags to avoid rejection of the null of valid instruments in the Hansen test for each sector.

Table 3.

Estimated parameters from estimation of an augmented production function using a matched sample, Scottish manufacturing 1984–2005.

Sector	High-tech	Medium high-tech	Medium low-tech	Low-tech
ln(Employment)	0.401***	0.371***	0.412***	0.278***
ln(Employment)	(0.102)	(0.116)	(0.111)	(0.068)
ln(Intermediate inputs)	0.579***	0.565***	0.647***	0.743***
ln(Intermediate inputs)	(0.088)	(0.081)	(0.083)	(0.058)
ln(Capital)	0.005	0.055	−0.029	−0.029
ln(Capital)	(0.079)	(0.062)	(0.033)	(0.036)
ln(Herfindahl Index)	−0.004	0.009	0.013	0.018*
ln(Herfindahl Index)	(0.036)	(0.021)	(0.015)	(0.010)
ln(Local authority industry share)	0.011	0.010	0.018	0.009
ln(Local authority industry share)	(0.029)	(0.021)	(0.019)	(0.009)
ln(Local authority diversification)	0.007	0.006	−0.037*	−0.003
ln(Local authority diversification)	(0.046)	(0.035)	(0.019)	(0.011)
ln(Age)	0.012	−0.051	0.027	0.041
ln(Age)	(0.113)	(0.082)	(0.051)	(0.059)
Single	−0.026	−0.012	−0.011	−0.050
Single	(0.042)	(0.024)	(0.038)	(0.037)
FDI	0.112	0.071*	0.028	0.105***
FDI	(0.075)	(0.040)	(0.039)	(0.034)
Time	0.034***	−0.001	0.003	−0.001
Time	(0.006)	(0.004)	(0.003)	(0.003)
RSA	−0.019	−0.029	−0.034*	−0.044***
RSA	(0.045)	(0.023)	(0.018)	(0.015)
AR(1)	−2.175**	−4.289***	−3.526***	−4.123***
AR(2)	−0.241	−1.690*	−1.957*	−0.139
Hansen test	59.78	56.53	68.46	57.12
Observations	1134	1954	1695	4042

*, **, *** denote significance at the 90%, 95%, 99% levels, respectively.

Standard errors are in parentheses.

Because the creation of the matched sample involves the removal of untreated plants that are dissimilar from treated plants, covariates have a smaller variance in the matched sample that causes their coefficients to become less statistically significant.

The most important coefficient is that associated with the RSA dummy. This is negative but not statistically significant in the high-tech and medium high-tech manufacturing sectors. In the medium low-tech and low-tech manufacturing sectors, the coefficient on the RSA dummy is negative and statistically significant. The size of the coefficient indicates that, in medium low-tech manufacturing, receipt of an RSA grant leads to a fall in TFP of 3.3%,¹⁵ while in low-tech manufacturing, receiving an RSA grant reduces TFP by 4.3%. Therefore, the estimated impact of receipt of an RSA grant is most negative in those sectors that employ the most basic technologies. This may be explained by different rates of obsolescence across sectors. If capital becomes obsolete faster in high-tech sectors, new capital will be needed to restore a plant to the technological frontier. However, in low-tech sectors, where obsolescence occurs at a slower rate, the need for new capital is not so pressing. There is therefore more scope for RSA-supported investments in capital that lead to an inefficiently high level of inputs.¹⁶

Having obtained the results from the basic specification in equation (4), the specification of the ‘treatment’ dummy was modified in two ways to gain a better understanding of how receipt of an RSA grant determines productivity. First, rather than assuming that an RSA grant has an impact on plant TFP for the duration of the plant’s existence, as is implicit in the baseline specification in which the RSA dummy equals 1 from the year in which a grant is received, the RSA dummy was modified so that it only took the value of 1 for a given number of years after the grant was received. The marginal effects obtained using different numbers of years are shown in Figure 1. Using a dummy that takes the value of 1 only in the period in which the grant is received yields a positive and significant coefficient for high-tech plants. However, all other specifications of the RSA dummy give coefficients that are not statistically significant, which shows that an RSA grant has no long-term impact on plant TFP in this sector.

Figure 1.

Estimated impact of receipt of an RSA grant on TFP using an RSA dummy that takes the value of 1, for 1 to 33 years after receipt of the grant, Scottish Manufacturing, 1984–2005. Squares denote that the coefficient is statistically significant at the 10% level. Note that the SAMIS data base contains information on RSA recipients back to 1972.

Regardless of the specification used, there is no significant impact of receipt of an RSA grant on TFP in medium high-tech manufacturing while the impact is always negative and significant in low-tech manufacturing. The most important finding of Figure 1 is that the impact of an RSA grant on plants in the medium low-tech sector is not robust to different specifications of the RSA dummy. Using a dummy that equals 1 for 18 years and more after a grant is received always gives a negative and statistically significant coefficient. However, the estimated coefficient is not significant when fewer years are used (the exceptions being 4 and 14 years). As it would be expected that the impact of an RSA grant on TFP would become apparent well before 18 years after the grant is received if the coefficient is truly measuring a causal impact, it cannot be said that receipt of an RSA grant has a significant impact on TFP in this sector.

Second, differences in the impact of receipt of an RSA grant between UK- and foreign-owned plants were tested for by including an interaction variable between the FDI dummy and the RSA dummy. The results are provided in Table 4. The RSA dummy was not statistically significant for any of the sectors used which suggests that domestically owned and foreign-owned plants experience similar impacts on TFP from receiving an RSA grant.

Table 4.

Estimated parameters from estimation of an augmented production function using a matched sample including an interaction between RSA and FDI, Scottish manufacturing, 1984–2005.

Sector	High-tech	Medium high-tech	Medium low-tech	Low-tech
ln(Employment)	0.393***	0.369***	0.412***	0.279***
ln(Employment)	(0.105)	(0.119)	(0.112)	(0.069)
ln(Intermediate inputs)	0.584***	0.567***	0.646***	0.743***
ln(Intermediate inputs)	(0.088)	(0.082)	(0.083)	(0.058)
ln(Capital)	−0.005	0.055	−0.031	−0.031
ln(Capital)	(0.072)	(0.063)	(0.034)	(0.036)
ln(Herfindahl Index)	−0.001	0.009	0.014	0.018*
ln(Herfindahl Index)	(0.036)	(0.021)	(0.015)	(0.010)
ln(Local authority industry share)	0.013	0.010	0.019	0.010
ln(Local authority industry share)	(0.029)	(0.021)	(0.019)	(0.009)
ln(Local authority diversification)	0.002	0.006	−0.037*	−0.003
ln(Local authority diversification)	(0.045)	(0.035)	(0.020)	(0.011)
ln(Age)	0.021	−0.050	0.031	0.045
ln(Age)	(0.099)	(0.082)	(0.054)	(0.059)
Single	−0.028	−0.012	−0.012	−0.052
Single	(0.042)	(0.024)	(0.038)	(0.037)
FDI	0.070	0.067*	0.041	0.121***
FDI	(0.071)	(0.040)	(0.038)	(0.039)
Time	0.033***	−0.001	0.003	−0.001
Time	(0.006)	(0.004)	(0.003)	(0.003)
RSA	−0.044	−0.031	−0.030	−0.041***
RSA	(0.047)	(0.025)	(0.019)	(0.015)
RSA × FDI	0.073	0.005	−0.016	−0.026
RSA × FDI	(0.080)	(0.042)	(0.045)	(0.037)
AR(1)	−2.127**	−4.279***	−3.518***	−4.116***
AR(2)	−0.252	−1.691	−1.961	−0.160
Hansen test	59.55	56.58	69.05	56.89
Observations	1134	1954	1695	4042

*, **, *** denote significance at the 90%, 95%, 99% level, respectively.

Standard errors are in parentheses.

Conclusion

This paper has sought to establish the existence of a causal impact of receipt of an RSA grant on plant TFP in Scotland using a data set created by merging SAMIS with the ARD. Estimation was performed using a matched sample to deal with the consequences of self-selection into the ‘treatment’ group. For high-tech and medium high-tech manufacturing, the estimated coefficient on the RSA dummy was not statistically significant. This was also the case in Criscuolo et al. (2012) and Hart et al. (2008). However, for medium low-tech and low-tech manufacturing, receiving an RSA was found to reduce TFP although the result for medium low-tech manufacturing was found to be not robust to changes in the specification of the RSA dummy.

This suggests that RSA grants lead plants in low-tech manufacturing, the sector that received the highest number of grants, to employ an inefficiently high level of inputs. Without such grants to compensate them for employing a suboptimally high level of inputs, they would employ fewer inputs but have higher levels of TFP. The existence of such a productivity cost is not surprising as, according to the rules of the scheme, grants are provided to get firms to do something that they would not otherwise do: create or safeguard employment. Whether this negative contribution to aggregate productivity should be regarded as a necessary evil for the extra employment in the economy is a moot point that is worthy of future research.

Footnotes

Appendix

Table A2.

Estimated parameters from probit model used to calculate propensity scores, Scottish manufacturing 1984−2005.

Sector	High-tech	Medium high-tech	Medium low-tech	Low-tech
ln(Employment)	0.008	0.205***	0.368***	0.370***
ln(Employment)	(0.047)	(0.033)	(0.032)	(0.018)
ln(Intermediate inputs)	0.072**	0.035	−0.060**	−0.013
ln(Intermediate inputs)	(0.035)	(0.028)	(0.029)	(0.014)
ln(Capital)	0.256***	0.071***	0.019	0.005
ln(Capital)	(0.031)	(0.016)	(0.016)	(0.010)
ln(Herfindahl Index)	−0.006	−0.195***	−0.114***	−0.095***
ln(Herfindahl Index)	(0.041)	(0.024)	(0.022)	(0.016)
ln(Local authority industry share)	0.025	0.039*	0.035	0.054***
ln(Local authority industry share)	(0.024)	(0.020)	(0.024)	(0.014)
ln(Local authority diversification)	−0.364***	−0.157***	−0.136***	−0.108***
ln(Local authority diversification)	(0.050)	(0.037)	(0.033)	(0.019)
ln(Age)	0.188***	0.397***	0.341***	0.224***
ln(Age)	(0.054)	(0.038)	(0.039)	(0.024)
Single	0.700***	0.219***	0.222***	0.373***
Single	(0.062)	(0.043)	(0.044)	(0.028)
FDI	0.411***	0.343***	0.104*	0.223***
FDI	(0.065)	(0.045)	(0.056)	(0.038)
Time	0.031***	0.047***	0.041***	0.036***
Time	(0.006)	(0.004)	(0.004)	(0.002)
Log-likelihood	−1163	−2617	−2510	−6247
Pseudo-R²	0.362	0.240	0.214	0.177
Observations	3545	9327	10,901	26,692

*, **, *** denote significance at the 90%, 95%, 99% level, respectively.

Standard errors are in parentheses.

Acknowledgements

This work contains statistical data from the Office for National Statistics (ONS) which is Crown copyright and reproduced with the permission of the controller of HMSO and Queen’s Printer for Scotland. The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. This work uses research data sets that may not exactly reproduce National Statistics aggregates. This paper is a substantially revised version of a chapter of the author’s PhD Thesis at the University of Glasgow. Thanks are due to Professor Richard Harris for his helpful comments on earlier versions of this paper

Funding

The PhD was jointly funded by the Economics and Social Research Council and the Scottish Government.

Notes

References

Aghion

Howitt

(1992) A model of growth through creative destruction. Econometrica 60: 323–351.

Arrow

(1962) The economic implications of learning by doing. The Review of Economic Studies 29: 155–173.

Baumol

(1984) On productivity growth in the long run. Atlantic Economic Journal 12: 4–10.

Blundell

Bond

(1998) Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 87: 115–143.

Blundell

Costa Dias

(2009) Alternative approaches to evaluation in empirical microeconomics. Journal of Human Resources 44: 565–640.

Blundell

Dearden

Sianesi

(2005) Evaluating the effect of education on earnings: Models, methods and results from the National Child Development Survey. Journal of the Royal Statistical Society: Series A (Statistics in Society) 168: 473–512.

Bond

(2002) Dynamic Panel Data Models: A Guide to Theory and Practice. Cemmap Working Paper CWP09/02, Institute for Fiscal Studies.

Cabral

(2000) Introduction to Industrial Organization. Cambridge, MA and London: MIT Press.

Cantwell

Dunning

Janne

OEM

(2004) Towards a technology-seeking explanation of U.S. direct investment in the United Kingdom. Journal of International Management 10: 5–20.

10.

Chandler

(1962) Strategy and Structure: Chapters in the History of the American Industrial Enterprise. Cambridge, MA and London: MIT Press.

11.

Criscuolo

Martin

Overman

. (2012) The Causal Effects of an Industrial Policy. Science and Engineering Research Council Discussion Paper 98.

12.

Dehejia

Wahba

(2002) Propensity score-matching methods for nonexperimental causal studies. Review of Economics and Statistics 84: 151–161.

13.

Doms

Jensen

(1998) Comparing wages, skills, and productivity between domestically and foreign owned manufacturing establishments in the United States. In: Baldwin

Lipsey

Richardson

(eds) Geography and Ownership as Bases for Economic Accounting. Chicago, IL: Chicago Press, pp. 235–258.

14.

Driffield

Love

(2007) Linking FDI motivation and host economy productivity effects: Conceptual and empirical analysis. Journal of International Business Studies 38: 460–473.

15.

Dunning

(1988) Multinationals, Technology and Competitiveness. London: Unwin Hyman.

16.

Duranton

Puga

(2004) Micro-foundations of urban agglomeration economies. In Henderson

Jacques-François

(eds) Handbook of Regional and Urban Economics. Amsterdam, Oxford: Elsevier, pp. 2063–2117.

17.

Fosfuri

Motta

(1999) Multinationals without advantages. Scandinavian Journal of Economics 101: 617–630.

18.

Frölich

(2008) Parametric and nonparametric regression in the presence of endogenous control variables. International Statistical Review 76: 214–227.

19.

Griffith

(1999) Using the ARD establishment level data to look at foreign ownership and productivity in the United Kingdom. Economic Journal 109: 416–442.

20.

Harris

(1989) The Growth and Structure of the UK Regional Economy 1963–85. Aldershot: Avebury.

21.

Harris

(2005a) Economics of the workplace: Special issue editorial. Scottish Journal of Political Economy 52: 323–343.

22.

Harris

(2005b) Longitudinal Microdata Study of Business Support Programmes. Report to the Department for Trade and Industry.

23.

Harris

Drinkwater

(2000) UK plant and machinery capital stocks and plant closures. Oxford Bulletin of Economics and Statistics 62: 243–265.

24.

Harris

Moffat

(2011) Plant-level Determinants of Total Factor Productivity in Great Britain, 1997–2006. Science and Engineering Research Council, Discussion Paper 64.

25.

Harris

Robinson

(2004) Industrial policy in Great Britain and its effect on total factor productivity in manufacturing plants, 1990–1998. Scottish Journal of Political Economy 51: 528–543.

26.

Harris

RID

(1991) Regional Economic Policy in Northern Ireland 1945–1988. Aldershot: Avebury.

27.

Hart

Driffield

Roper

. (2008) Evaluation of Regional Selective Assistance (RSA) in Scotland 2000–2004. Report to Scottish Government.

28.

Herfindahl

(1950) Concentration in the US steel industry. PhD Thesis, Columbia University.

29.

Hymer

(1976) The International Operations of National Firms: A Study of Direct Foreign Investment. Cambridge, MA and London: MIT Press.

30.

Imbens

Wooldridge

(2009) Recent developments in the econometrics of program evaluation. Journal of Economic Literature 47: 5–86.

31.

Jacobs

(1970) The Economy of Cities. London: Jonathan Cape.

32.

Jacobs

(1986) Cities and the Wealth of Nations. Harmondsworth: Penguin.

33.

Jarmin

(1999) Government Technical Assistance Programs and Plant Survival: The Role of Plant Ownership Type. Working Papers 2, Center for Economic Studies, US Census Bureau.

34.

Jovanovic

Nyarko

(1996) Learning by doing and the choice of technology. Econometrica 64: 1299–1310.

35.

Kelley

Harrison

(1990) The subcontracting behavior of single vs. multiplant enterprises in US manufacturing: Implications for economic development. World Development 18: 1273–1294.

36.

Krugman

(1997) The Age of Diminished Expectations: US Economic Policy in the 1990s. Cambridge, MA and London: MIT Press.

37.

Leibenstein

(1966) Allocative efficiency vs. ‘x-efficiency’. The American Economic Review 56: 392–415.

38.

Leuven

Sianesi

(2003) PSMATCH2: Stata Module to Perform Full Mahalanobis and Propensity Score Matching, Common Support Graphing, and Covariate Imbalance Testing. Available at: http://ideas.repec.org/c/boc/bocode/s432001.html.

39.

Marschak

Andrews

(1944) Random simultaneous equations and the theory of production. Econometrica 12: 143–205.

40.

Marshall

(1890) Principles of Economics. London: Macmillan.

41.

Moffat

(2010) Regional Selective Assistance in Scotland: Does it make a difference to plant performance? PhD Thesis, Economics, University of Glasgow.

42.

Organization for Economic Cooperation and Development (2003) The Sources of Economic Growth in OECD countries. Available at: http://www.oecd.org/dac/ictcd/docs/otherdocs/OtherOECD_eco_growth.pdf.

43.

Robjohns

(2006) ARD2: The new Annual Respondents Database. Economic Trends 630: 43–51.

44.

Romer

(1986) Increasing returns and long-run growth. Journal of Political Economy 94: 1002–1037.

45.

Roodman

(2005) XTABOND2: Stata Module to Extend Xtabond Dynamic Panel Data Estimator. Available at: http://fmwww.bc.edu/repec/bocode/x/xtabond2.ado.

46.

Schumpeter

(1943) Capitalism, Socialism, and Democracy. London: George Allen & Unwin Ltd.

47.

Scottish Enterprise (2011) Regional Selective Assistance: 2010/11 Annual Summary. Available at: http://www.scottish-enterprise.com/∼/media/SE/Resources/Documents/PQR/RSA%20Reports/2010–2011-Annual-Summary.ashx.

48.

Van Beveren

(2007) Total Factor Productivity Estimation: A Practical Review. LICOS Discussion Papers 18207, Katholieke Universiteit Leuven.

49.

Zhao

(2004) Using matching to estimate treatment effects: Data requirements, matching metrics, and Monte Carlo evidence. Review of Economics and Statistics 86: 91–107.