Information technology, efficiency and productivity: evidence from Korean local governments

Abstract

This study evaluates the performance of Korean local government by measuring their technical efficiency (TE) and total factor productivity (TFP) growth and, more importantly, examining the impact of information technology (IT) on this performance. The study is different from received analysis in that a unique measure of the state of IT–the Informatization Index–is used to investigate the impact of IT on both TE and TFP growth. Empirical analyses are conducted on data from 222 Korean local governments for the period 1999 to 2001. In particular, data envelopment analysis techniques are applied to calculate TE scores and TFP growth rates for sampled local governments. The empirical findings confirm the positive impact IT has in improving technical efficiency and accelerating productivity growth. The estimated coefficients are correctly signed (with other regional characteristics controlled for) when TE scores and TFP growth rates are regressed on the Index. In addition, the findings indicate that economies of density are present in the production of local public services.

Acknowledgements

This work was supported by a Korean Research Foundation Grant (KRF-2003-041-B00142). The author gratefully acknowledges valuable comments from Choon-Geol Moon at Hanyang University, Gary Madden at Curtin University of Technology and an anonymous referee. Also, I would like to thank Bo Honoré for kindly uploading the PANTOB program at his Web site.

Notes

¹ A summary of the early studies on Solow's productivity paradox appears in Brynjolfsson (1993) and David (2000).

² Milana and Zeli (2002), relying on Italian firm data, is an exception.

³ Brynjolfsson and Hitt (2000) indicate a significant component of the value of IT is the ability to enable complementary organizational investment such as business process and work practices. These investments lead to increased productivity via cost reductions and by enabling firms to increase the quality of output.

⁴ Another source of productivity growth is economies of scale. For example, a technically efficient DMU can improve productivity by moving upward along the production frontier and exploiting scale economies. This study does not consider scale economies and the main concern is the impact of IT on differences in efficiency and change in productivity.

⁵ Similarly, an input distance function can be used to define an input-oriented measure of TE.

⁶ The period-t Malmquist TFP change index can be defined in a similar manner.

⁷ This two-stage method has some advantages in considering the environment that is beyond management control and influence the efficiency of a DMU. For instance, it allows accommodation of statistical tests of significance. For a detailed explanation on the methods in which environmental variables can be accommodated in a DEA analysis, see Coelli et al . (1998) and Cooper et al . (2000).

⁸ Some statistical programmes supporting a censored model cannot be applied without this transformation. For example, Honoré's PANTOB program assumes that the censoring is limited to lower censoring at zero. The transformation, however, generate few changes in the regression results. For instance, when the TE scores were used instead of the TIE scores in Table 8, there were only minor changes in the parameter estimates and their t-values. Detailed estimation results will be provided on request.

⁹ The DEA have been increasingly applied in recent times to compare the economic performance of other public sector organizations. To cite some examples, especially papers published in Applied Economics, both Diez-Ticio and Mancebon (2002) and Drake and Simper (2002) use the DEA to measure the efficiency of Spanish and English police service, respectively. Hammond (2002) derives technical and overall efficiency scores for UK public library systems by using the DEA. Parkin and Hollingsworth (1997) and Giuffrida and Gravelle (2001) apply the DEA to data from Scottish acute hospitals and data from the UK Health Service Indicators database, respectively.

¹⁰ The study ignores the fact that some big cities have districts. This is inevitable because of lack of appropriate data on these districts belonging to cities.

¹¹ The study eliminates four cities and six counties as their data are inconsistent or unreliable.

¹² Also, I attempted to evaluate each local government only within the group as the third alternative, e.g. a city government with reference to city governments. Not surprisingly, this method produced the highest efficiency scores. The all-year and all-government average is 0.768 for Method 3, which is higher than 0.753 and 0.572 in the Method 1 and Method 2, respectively, in Table 3.

¹³ Should these factors have no systematic relationship with independent variables then the parameter estimates are consistent. If the assumption is not true, however, the explanatory power of exogenous variables is likely to be small. For example, the management skills of an elected chief of a local government affect the performance of the government and may be correlated with the state of IT. Then the exclusion of this variable may generate omitted variable biases.

¹⁴ The Z-scores in one year are not highly correlated with those in another year. The correlation coefficient between the all-items Z-scores in 1999 and 2001 (2002) is only 0.176 (0.216), and that between the all-items Z-scores in 2000 and 2001 is 0.600. The sub-indexes for either ‘Investment and Equipments’ or ‘Usage and Applications’ have a strong and positive correlation across years, while those for ‘Human and Organizational Factors’ have a weak correlation.

¹⁵ A majority of studies using the DEA technique rely on cross-sectional data. A reason for this reliance is that the efficiency scores cannot be directly compared across years because the reference technology differs from year to year. Mahlberg and Url (2003) do apply the fixed and random effects Tobit model. They include time dummy variables to deal with this problem.

¹⁶ I used Honoré's PANTOB, a Gauss program which estimates a censored Tobit model with fixed effects model as in Honoré (1992). The polynomial loss function is chosen for optimization.

¹⁷ The population criterion for these dummy variables can be modified. On the other hand, whatever criterion is used, their coefficients are always negative and mostly statistically significant.

¹⁸ In Korea, data on gross regional product for districts, cities and counties are not publicly available. Thus the amount of collected local tax has been often used as a proxy for gross regional product (GRP) in empirical studies.

¹⁹ To examine the robustness of these empirical findings, the same model specifications and estimation methods are applied to yearly cross-sectional data. Then the coefficients of the ITID are always significant and also, had the expected sign when the 1999 data was used. The application of the same model specifications and estimation method to the 3-year averages produced almost the same results. Also, when the efficiency scores from other methods are used, there are few differences in the results.

²⁰ Similarly despite changes in the data and estimation methods, the results are similar to those in Table 10. In particular, when the 2-year averages are used, the coefficients of the ITID are always positively signed.