Consumer Attitudes Toward Mobile Advertising:

Proceedings of the Third Workshop on Knowledge Economy and Electronic Commerce

Estimating IT Productivity Via Semi-Parametric Models Jong-Rong Chen Graduate Institute of Industrial Economics National Central University [email protected]

Ting-Kun Liu Graduate Institute of Industrial Economics National Central University [email protected]

Cliff J. Huang Department of Economics Vanderbilt University, U.S.A. [email protected]

Abstract This paper explores the issue of whether information technology (IT) investment brings about the Solow productivity paradox. In order to take the improvement of product quality caused by IT investment into account, a proper hedonic price index is used to deflate the IT variable. Besides the general specification of Cobb-Douglas production function, this paper considers the impact of augment labor productivity through IT by applying a semi-parametric smooth coefficient model. We employ the manufacturing firm-level data of Taiwan in 1991, and empirical results show that IT investment provides a significant contribution to productivity. Besides, the “share” of IT in production is not constant and the IT has non-neutral impact on the productivity of labor. We also find that when IT reaches a certain threshold, it starts to have a spillover impact on the productivity of labor. In fact, the spillover impact “takes over” the direct impact of IT on output. That is, the direct impact of IT on output is decreasing when spillover impact on the productivity of labor is increasing. Keywords: Information technology; Hedonic price index; Semi-parametric smooth coefficient model; Spillover JEL classification: D24; L60

1. Introduction The progression of information technology (IT) has brought about the so-called third industrial revolution for the world’s population. With the application of Internet and E-commerce, the function and efficiency of IT have diffused rapidly. In the past two decades, most industrial countries have invested a huge amount into IT in order to create, accumulate, storage, and transfer knowledge and also to improve competition and profit. For example, the share of IT in a firm’s total investment in equipment has jumped from 7% in 1970 to 40% in 1996 (Economist, 1996). Dewan and Kraemer (2000) quoted a statistical survey by the International Data Corporation (IDC) in 1997, showing that global IT expenditure has increased from $162 billion in 1985 to $630 billion in 1996 and with a continuing rising trend. Jorgenson (2001) found that the decline in IT price provides enterprises powerful economic incentives for the substitution of IT investment for other forms of inputs. The value of the net stock of IT capital equipment approached $900 billion by the end of 1999 in the U.S. (Stiroh, 2001). Pakko (2002) analysed official statistics from the U.S. Bureau of Economic Analysis (BEA) and wrote that IT expenditure has risen to account for over one-third of a business’ total fixed investment by the year 2000. 404


Taiwan plays a key position in the production and manufacturing of IT equipment for the world. Taiwan has an important role in producing IT-related equipment such as that in the semiconductor, computer, and telecommunications fields. In addition to the IT manufacturing industry, other industries in Taiwan also have made massive investment in IT so as to face the competition in this age of the knowledge-based economy. Taiwan’s IT expenditure approached NT$118.4 billion in 2000 and reached to NT$132.1 billion in 2003. 1 According to the survey that was conducted by the electronic data-processing center of the Directorate General of Budget, Accounting and Statistics (DGBAS), the share of the manufacturing industry’s IT expenditure to the overall industrial sector amounted to 85%. 2 Furthermore, the IT expenditure of the manufacturing industry rose from NT$20.1 billion in 2000 to NT$27.1 billion in 2003. Some economists have discovered that the IT revolution does not necessarily result in productivity growth. They also point out that the increase in IT investment does not necessarily help to promote productivity. Solow (1987) called this phenomenon the “Productivity Paradox”. Berndt and Morrison (1995) explored relationships between industry performance measures and investments in high-tech office and IT capital for two-digit manufacturing industries from 1968 through 1986, and they found that increases in IT are negatively correlated with multi-factor productivity. Loveman (1994) exploited business unit level data for U.S. and Western Europe manufacturing companies from 1978 to 1984 to explore the productivity impact of IT investment, writing that no matter whether having full sample or sub samples, the output elasticity for IT ranges from -0.12 to 0. This shows the result that the value of marginal product (MP) is not only zero, but also negative. Therefore, the above empirical studies display the phenomenon of the productivity paradox. However, the empirical studies of later scholars present entirely different conclusions. Brynjolfsson and Hitt (1996) used firm-level data for 1987-1991, which included 367 U.S. large firms, to explore the contribution of output from IT. Their results indicate that IT spending has a substantial and statistically significant contribution to firm output. They also found that the gross MP for IT capital averaged 81%. Lehr and Lichtenberg (1998) examined the impact of IT on productivity in the public sector, showing that there is a strong positive relationship across federal agencies between productivity growth and computer-intensity growth during the period 1978-1992. Therefore, according to the above studies the productivity paradox phenomenon is non-existent. Barua and Lee (1997), Dewan and Min (1997), Stiroh (1998), Gera, et al. (1999), Thatcher and Oliver (2001), and Jorgenson (2001) all showed that firms investing into IT will not bring about the phenomenon of productivity paradox. Therefore, opinions of productivity paradox are mixed among those in the literatures above. Furthermore, intensive use of computer is likely to raise the labor and nonIT capital productivity through input substitution. Dewan and Min (1997) indicate that IT capital is a net substitute for both ordinary capital and labor. Stiroh (1998) also finds that noncomputer input growth decrease as the use of computer capital services increased in these computer intensive sectors, and cheap computers substituted for other inputs, including labor. Therefore, a suitable production function is needed to answer the non-neutral impact of IT on the productivity of conventional inputs. Besides incorrect methodologies and measurement errors, some economists have tried to find 1

See Table 1 and Table 2 for the details. The survey method regards the questionnaire as the main mode, effectively having a total number of observations at 7,225. The sample data is used to figure out total population, and so the value is an estimated value and is not a physical observation number.

2

405


out the rational explanations for these mixed empirical findings. Belleflamme (2001) pointed out that the productivity paradox might result from oligopolistic competition and product differentiation, but not from pursuing efficiency. Thatcher and Oliver (2001) suggested that these mixed empirical findings might result from incorrect methodology, data error, but not reflect the real IT contribution, especially from the ability of IT to improve product quality as indicated by Brynjolfsson and Hitt (1996), Barua and Lee (1997), and Lehr and Lichtenberg (1998). They indicated that quality improvements are realized when a technology investment leads to the creation of new products, new features, or existing products, which directly increase human desire to consume those products. For example, patient-tracking systems, data mining tools, interaction television technologies, and decision support systems all help IT equipment to promote product quality. Furthermore, Chun and Nadiri (2002) mentioned that productivity growth could take place in the improvement of output quality (product innovation). In particular, improvement in output quality is one of the most prevailing characteristics in the IT production such as microprocessor speed, the capacity of storage devices and memory, etc. They noted that an increase in product variety and quality should properly be counted as part of the value of output, and then the IT contribution to output will not be underestimated. Hence, IT plays an important role in improving and promoting product and service quality. When the issue of whether IT investment brings about a productivity paradox has been debated in the academic circle, we fail to find any study about the impact of IT on productivity in Taiwan, except Yang (1998), due to a lack of full and integrity-related data. 3 Suffering from the restriction of the available data, Yang’s paper can only study the relationship between IT investment and financial revenue and does not really reflect the contribution of IT on productivity. This paper also faces the problem of acquiring empirical data in the early stage, and the difficulty of matching a firm’s IT investment, output, and other variables. We finally were able to acquire computer equipment quantity survey data and other related variables from a manufacturing sampling survey data of the industry, commerce, and service census in Taiwan by the DGBAS in 1991. According to the studies of Brynjolfsson and Hitt (1996) and Lehr and Lichtenberg (1998), we can define personal computer and terminal equipment as the IT investment variable. Due to the findings from Dewan and Min (1997) and Stiroh (1998), except the general specification of Cobb-Douglas production function, this paper considers the impact of augment labor productivity through IT by applying a semi-parametric smooth coefficient model 4 to construct an empirical model and appraise the impact of IT on productivity. Therefore, the end result of this on production is a non-neutral shift in the observed output. Not only will the productivity of inputs change, but also will the marginal rate of technical substitution (Huang and Liu, 1994). In order to take the improvement of product quality caused by IT investment into account, this paper not only uses firm-level data to reflect the impact of IT on product quality (Hitt and Brynjolfsson, 1996), but also adopts a deflator to deflate IT variables. 5 The quality-adjusted price index used in our paper is a computer hedonic price index. 6 We thus can be free from the 3

4

See Yang (1998) for the details. See Li, et al. (2002) for the details.

5

Loveman (1994), Barua and Lee (1997) and Gera, et al. (1999) indicated that each variable’s real term can be derived from deflation by using the proper price indices, and this can be used as the method of reflecting the product quality recondition. 6 See Jang, et al. (1996) for the details.

406


problem of too much deflation due to deflate the PC related variables, and turn IT variables from nominal terms to real terms. 7 The computer hedonic price index adopted by this paper considers the character variables of quality, such as brand, CPU, screen, memory, hard drive, and time dummy variables, etc. Furthermore, the computer hedonic price index is estimated by the hedonic regression method to reflect the improvement of output quality. The rest of the paper is organized as follows. In section 2 we discuss our empirical model. Section 3 describes the data sources and the selection of variables used in this study. The analyses of empirical results are presented in section 4. The conclusion is presented in section 5.

2. Empirical framework Although the empirical models of existing literature have different samples and periods, most of them (e. g. Loveman, 1994; Berndt and Morrison, 1995; Hitt and Brynjolfsson, 1996; Lehr and Lichtenberg, 1998; Gera, et al., 1999) adopt the Cobb-Douglas production function as an empirical model. Hence, we apply the Cobb-Douglas production function as our basic model. This paper assumes that the production function can be approximated by a general CobbDouglas function form: Model (1): CD specification Qi = exp(α 0 ) (ITi )α 1 ( Labi ) β 2 ( NonITi ) β 3 , (1) where Q is value added, IT, Lab, and NonIT are the IT capital, labor input, and non-IT physical capital, respectively. The subscripts i refer to firm i. We can take the logarithms of equation (1) and obtain a linear regression in order to implement the estimation of the Cobb-Douglas function as shown below: Yi = α 0 + α1Ti + β 2 Li + β 3 N i , (2) where the upper-case letters denote the logarithms of the variables i.e., Y=ln(Q); T=ln(IT); L=ln(Lab); N=ln(NonIT), and α1 (the elasticity of value added with respect to IT capital), β 2 and especially β 3 are the parameters in which we are particularly interested. We can use the results of equation (2) to verify and compare with the results of Brynjolfsson and Hitt (1996), Hitt and Brynjolfsson (1996) and Dewan and Min (1997), whether marginal product (MP) could be a proper productivity indicator of the contribution of IT to output. This paper uses the estimated coefficients of equation (2) to calculate MP in the following way: Y ∂Y ∂Y X i ⋅ Y = = δi MPi = , (3) ∂X i ∂X i Y ⋅ X i Xi where X i through X n are the variables that measure NONIT, LAB and IT, and δ i is the output elasticity of X i . The MP of X i ( MPi ) is simply the elasticity multiplied by the ratio of output to X i input as shown in equation (3). Besides the general specification of Cobb-Douglas production function, this paper considers the impact of augment labor productivity through IT by applying a semi-parametric smooth coefficient model to detect the non-neutral impact of IT on the productivity of conventional inputs. Then we formulate in the following semi-parametric way: 7

Loveman (1994) used a BEA quality-adjusted price index as the deflator of IT variable. However, Loveman’s IT data contained not only computers, but also: 1. computers and peripheral equipment, 2. office, computing, and accounting, 3. other office computing equipment, 4. communication equipment, 5. instruments, science and engineering, and 6. photocopy and related equipment. Barua and Lee (1997) used the same data and indicated that the study of Loveman (1994) results in too much deflation and will cause a productivity paradox.

407


Q = β1 ( IT )( Lab) β 2 ( IT ) ( NonIT ) β 3 , (4) where β1(IT) and β2(IT) are some unknown function of IT. This implies that the role of IT is not simply as a factor of production through β1(IT), it serves as factor of augmenting labor productivity through β2(IT) as well. Since NonIT and IT are both “capital” inputs, we can’t think of reason that the productivity of NonIT is affected by IT. Hence, in the paper, we assume that the productivity of the NonIT does not depend on the (IT). Our other three related models specification as shown below: Model (2): Semi-parametric specification (Robinson type) 8 Q = exp(β1 (T ) ) ( Lab) β 2 ( NonIT ) β 3 , (5) Yi = β1 (T ) + β 2 Li + β 3 N i , (6) Model (3): Modified CD specification Q = exp(α 0 ) (IT )α1 ( Lab) β 2 +α 2T ( NonIT ) β 3 , (7) Yi = α 0 + α1T + ( β 2 + α 2T ) L + β 3 N , (8) 9 Model (4): Semi-parametric specification (Li/Huang/Li/Fu type) Q = exp(β1 (T ) ) ( Lab) β 2 (T ) ( NonIT ) β 3 , (9) Yi = β1 (T ) + β 2 (T ) Li + β 3 N i , (10)

3. Data sources and definitions of variables Our source of sample data is the manufacturing sampling survey data from the industry, commerce, and service census of Taiwan in 1991 with 1,174 samples of firm-level data. In equation (1) of this paper, due to the absence of certain materials in the model, we use valueadded (Q) as the proxy for the dependent variable. According to the MOEA’s (Department of Statistics, Ministry of Economic Affairs) definition, we measure Q based on the following specification: Q is equal to total output sales less intermediate inputs. Total output sales are deflated using a wholesale price index and intermediate inputs are deflated using a price index of intermediate inputs. With respect to the explained variables of the right-hand side equation, NonIT capital (NonIT) is obtained by subtracting the IT capital from the fixed capital and subtracts accumulated depreciation. We use a capital price index to adjust for inflation, while labor (Lab) refers to the number of employees. According to the search of Hitt and Brynjolfsson (1996) and Lehr and Lichtenberg (1998), PCs and terminals can be defined as IT investment (IT) items. In order to take the improvement of product quality caused by IT investment into account, we use firm-level data, and adopt a proper computer hedonic price index to deflate personal computers. In other words, the qualitative data on IT capital is adjusted through price and then deflated by the computer hedonic price index for empirical study. 10 In accordance with Hitt and Brynjolfsson (1996), when an 8

Note: Model (1) is a special case of Model (2), i.e., β1(T) = α0 + α1 T . Note: Model (1), (2), and (3) are special cases of Model (4), i.e., For Model (1): β1(T) = α0 + α1 T; β2(T) = β2; For Model (2): β2(T) = β2; For Model (3): β1(T) = α0 + α1 T; β2(T) = β2 + α2 T.

9

10

Since computer equipment price data is hard to collect, we finally acquired the available average price of PCs and

408


increase in product variety and quality is properly counted as part of the value of output, the contribution to output will then not be underestimated. Table 3 provides sample statistics for our key variables.

4. Empirical results The previous equation (2) is regarded as the starting point of the analysis. The second column of Table 4 shows that the labor coefficient is higher than the capital coefficient, although both have a significant impact on the level of productivity. The IT capital’s parameter ( α1 ) is significantly positive at the 1% statistical level, however its impact on output is small than NonIT and Lab. Then we estimate MP of each input follow the method provided by Brynjolfsson and Hitt (1996) as specified in equation (3) to verify and compare with the results of Brynjolfsson and Hitt (1996), Hitt and Brynjolfsson (1996) and Dewan and Min (1997). The results show that MP of each input presents the MPIT> MP NONIT >MPLAB pattern. This estimate pattern is some different from the empirical results in Brynjolfsson and Hitt (1996), Hitt and Brynjolfsson (1996), and Dewan and Min (1997), which have MPLAB> MPIT> MP NONIT pattern. All these studies reflect a contradiction between the magnitude of output elasticity and MP of IT even smaller than labor, although its MP larger than nonIT capital. Owing to the fact that the small share of IT to output, by applying equation (3), the estimation of MP would be larger. This implies that MP could not be a proper productivity indicator. In order to consider the impact of augment labor productivity through IT, this paper apply a semi-parametric smooth coefficient model to detect the non-neutral impact of IT on the productivity of labor. We construct other three related models and the polynomial and semiparametric estimation results are shown in Table 4. Test statistic Jn in Model (1) tests H0: β1(T) = α0 + α1 T; against H1: β1(T) ≠ α0 + α1 T, i.e., H0: Model (1) is the true model; H1: Model (2) is the true model. The test statistic Jn has asymptotic standard normal distribution. In Model (1) Jn = 91.8516 > 1.96 at the 5% significant level, Model (1) (the CD model) is rejected. Similarly, Jn = 2.5811 > 1.96, Model (3) is rejected at the 5% level when it is tested against Model (4).Based on the Jn test statistic, Model (1) and Model (3) are rejected. Furthermore, Model (2) is rejected when it is tested vs. Model (4). This implies that the “share” of IT in production is not constant and the IT has non-neutral impact on the productivity of labor. We have the 95% confidence bands for β1(T) and β2(T). Because the semi-parametric estimators of the smooth coefficients are function of IT, we plot the coefficients in Figure 1-6. Comparing the coefficient on IT in the Model (4) with other three models, we see that the coefficient on IT in the Model (4) is relatively undulate, whereas it is monotone increasing in IT in other three models. Therefore in the Cobb-Douglas model, the coefficient on labor is restricted to be constant and thus any effect of IT investment on output could be overestimated according to the monotone increasing in IT as shown in the Figures. Hence we allow the coefficient on labor to change with IT, in Figure 6 we can see that the coefficient on labor is increasing in IT, when T reaches T = 1 (i.e., IT = exp(1) = 2.7 million NT$). This implies that when IT reaches a certain threshold, it starts to have a spillover impact on the productivity of labor. Looking at the plots of β1(T) and β2(T) against T, they show a very interesting result. In fact, the spillover impact “takes over” the direct impact of IT on output, i.e., β1(T) is increasing when β2(T) is constant, and β1(T) is decreasing when β2(T) is increasing. This implies that firms invest in IT investment have the positive contribution on productivity, but terminals as the IT equipment price. See Jang, et al. (1996) and Pelaia (1993) for the details.

409


when IT reaches a certain threshold, over investment could lead the direct impact of IT on output to decline. On the other hand, when IT reaches a certain threshold, the contribution of the spillover impact on the productivity of labor would replace and counteract the direct impact of IT on output. This result could be an explanation why some studies’ finding that IT has productivity paradox, and this could be due to the reason that most of them neglect the non-neutral or spillover effects from IT on the productivity of labor or other inputs.

5. Conclusions Over the past decade, quite a few studies have discussed and debated the issue of the Solow IT productivity paradox. Some of these studies applied the Cobb-Douglas production function to estimate the MP of IT, and use it as the contribution of IT to productivity. Nevertheless, few of them focus on the impact of IT on non-neutral impact of IT on the productivity of conventional inputs. For this reason, this study uses the manufacturing firm-level data of Taiwan in 1991 to explore the issue of whether IT investment brings about the Solow productivity paradox. In order to take into account the improvement of product quality caused by IT investment, a proper hedonic price index is used to deflate the IT variable. Besides the general specification of CobbDouglas production function, this paper considers the impact of augment labor productivity through IT by applying a semi-parametric smooth coefficient model. Empirical results show that IT investment provides a significant contribution to productivity. Besides, the “share” of IT in production is not constant and the IT has non-neutral impact on the productivity of labor. We also find that when IT reaches a certain threshold, it starts to have a spillover impact on the productivity of labor. In fact, the spillover impact “takes over” the direct impact of IT on output. That is, when the direct impact of IT on output is decreasing, then spillover impact on the productivity of labor is increasing. This result could be used to explain why some studies’ finding that IT has productivity paradox, and this could be due to the reason that most of them neglect the non-neutral or spillover effects from IT on the productivity of labor or other inputs.

References Barua, A. and Lee, B. “The Information Technology Productivity Paradox Revisited: A Theoretical and Empirical Investigation in the Manufacturing Sector,” The International Journal of Flexible Manufacturing Systems, (l), 1997, pp. 145-166. Belleflamme, P. “Oligopolistic Competition, IT Use for Product Differentiation and the Productivity Paradox,” International Journal of Industrial Organization, 19, 2001, pp. 227248. Berndt, E. R. and Morrison, C. J. “High-tech Capital Information and Economic Performance in U.S. Manufacturing Industries: An Exploratory Analysis,” Journal of Econometrics, (65), 1995, pp. 9-34. Brynjolfsson, Erik and Hitt, L. M. “Paradox Lost? Firm-level Evidence on the Returns to Information Systems Spending,” Management Science, (42:4), 1996, pp. 257-279. Chun, H. and Nadiri, M. I. “Decomposing Productivity Growth in the U.S. Computer Industry,” National Bureau of Economic Research Working Paper, (9627), 2002, pp. 1-36. Dewan, S. and Min, C. K. “The Substitution of Technology for Other Factors of Production: A firm Level Analysis,” Management Science, (43:12), 1997, pp. 1660-1675. Dewan, S. and Kraemer, K. L. “Information Technology and Productivity: Evidence from

410


Country-Level Data,” Management Science, (46:4), 2000, pp. 548-562. Gera, S., Gu, W. and Lee, F. C. “Information Technology and Labor Productivity Growth: An Empirical Analysis for Canada and the United States,” Canadian Journal of Economics, (32:2), 1999, pp. 385-407. Hitt, L. M. and Brynjolfsson, E. “Productivity, Business Profitability, and Consumer Surplus: Three Different Measures of Information Technology Value,” MIS Quarterly; Management Information Systems, (20:2), 1996, pp. 121. Huang, C. J. and Liu, J. T. “Estimation of a Non-Neutral Stochastic Frontier Production Function,” Journal of Productivity Analysis, (5), 1994, pp. 171-180. Jang, S. L., Liu, B. J. and Yang, S. Z. “Constructing a Quality-adjusted Price Index for Computers in Taiwan,” Taiwan Economic Review, (2:4), 1996, pp. 557-577. Jorgenson, D. W. “Information Technology and the U.S. Economy,” The American Economic Review, (9:1), 2001, pp. 1-32. Lehr, W. and Lichtenberg, F. R. “Computer Use and Productivity Growth in US Federal Government Agencies,” The Journal of Industrial Economics, (16:2), 1998, pp. 257-279. Loveman, G. W. “An Assessment of the Productivity Impact of Information Technologies’” in T.J. Allen and M. S. Scott Morton (Eds), Information Technology and the Corporation of the 1990s: Research studies, Oxford University Press, 1994, pp. 84-109. Pakko, M. R. “The high-tech investment boom and economic growth in the 1990s: Accounting for quality,” Review - Federal Reserve Bank of St. Louis, 2002, pp. 84:2. Pelaia, E. (1993) “IBM Terminal Prices,” Personal Communication with IBM Representative. Li, Q., Huang, C. J., Li, D., and Fu, T. T. “Semiparametric Smooth Coefficient Models,” Journal of Business & Economic Statistics, (20:3), 2002, pp. 412-422. Solow, R. M. “Technical change and the aggregate production function,” Review of Economic Statistics, (39:2), 1957, pp. 312-320. Solow, R.M. “We’d better watch out,” New York Times (July 12), Book Review, 36, 1987. Stiroh, K. J. “Computers, productivity and input substitution,” Economic Inquiry, (36:2), 1998, pp. 175-191. Thatcher, M. E. and Oliver J. R. “The Impact of Technology Investments on a Firm’s Production Efficiency, Product Quality, and Productivity,” Journal of Management Information System, (18:2), 2001, pp. 17-45. The Economist “Paradox Lost,” The Economist, (340:7985), 1996, pp. 13-16. Yang, C. Y. “Productivity Paradox of Information Technology: A Study on Organization in Taiwan,” Graduate Institute of Information Management, National Sun Yat-Sen University, an unpublished master thesis, 1998.

Table 1: IT Expenditure and Distribution of Taiwan (2000) (NT$ million)

411


Sector and Industry Agr., For., Fis. & Herds Industrial sector Mining & Grounding Manufacturing Hydroele. & Natgas. Construction Services sector Whole., Retail & Meals Tan.,Ware.& Commun. Fin., Insur. & Realestat. Industry & Bus. service Social & Per. service Total

Total Expenditure Hardware 44 23,539 26 20,105 877 2,532 94,819 20,808 3,616 26,289 5,676 24,263 118,403

40.45 40.44 50.34 38.86 46.73 52.10 45.69 40.65 46.03 38.40 63.43 53.73 44.52

Share (%) Software

Telecommunication

Personnel

Other

41.03 31.88 39.5 33.30 16.58 24.90 25.27 26.14 30.41 27.74 21.69 23.24 26.75

7.49 7.78 6.63 7.68 4.08 10.07 8.11 11.89 9.31 8.77 5.23 5.41 8.04

2.30 14.95 2.52 14.82 28.22 11.22 15.57 15.45 11.69 19.76 8.36 12.76 15.42

8.72 4.95 1.02 5.34 4.38 1.70 5.36 5.87 2.55 5.33 1.29 4.85 5.27

Source: The DGBAS electronic data - processing center in 2000. Note: The first sector contains agriculture, forestry, fishery and herd industries. Industrial sector contains mining & grounding, manufacturing, hydroelectric & natural gas and construction industries. Services sector includes wholesale, Retail & Meals, transportation, warehousing & communication, finance, insurance & real estate, industry & business service and social & personal service industries. The share is a ratio of expenditure of each item to total expenditure. Table 2: IT Expenditure and Distribution of Taiwan (2003) (NT$ million) Total Sector and Industry Expenditure Hardware Agr., For., Fis. & Herds Industrial sector Mining & Grounding Manufacturing Hydroele. & Natgas. Construction Services sector Total

194 31,700 81 27,098 1,264 3,257 100,240 132,133

30.42 23.63 20.60 22.60 31.63 29.24 33.66 31.25

Share (%) Software Telecommunication 21.67 20.87 22.40 21.60 12.42 17.98 21.06 21.02

23.49 8.28 5.33 8.14 4.23 11.07 7.56 7.76

Personnel 14.74 40.12 44.11 40.71 43.08 33.96 29.08 31.71

Source: The DGBAS electronic data - processing center in 2003. Table 3: Statistics on Variables (After Deflation) (NT$ Million) Variable Name Mean (S.E.) Value added 992.416 (6406.222) Q Non-IT capital 162.821 (1,791.495) NonIT Number of employees 554.064 (1,617.800) Lab IT capital 2.234 (6.173) IT Note: The numbers in parentheses are standard errors.

Table 4: Polynomial and Semi-parametric Estimation

412

Other 9.67 7.10 7.57 6.95 8.65 7.76 8.63 8.27


Model (1): Polynomial specification 1.282 (10.348) a 0.126 (6.227) a

α0 α1 β1(T) α2

Model (2): Semi-parametric specification

Model (3): Polynomial specification 1.276 (10.134) a 0.112 (2.122) a

0.968 (1347.338) a

Model (4): Semi-parametric specification

1.023 (232.534) a 0.002 (0.269) a

β2 β2(T)

0.545 (20.394) a

β3 Sigma R-square Test statistics: Jn

0.387 (22.297) a 0.37717 0.84115 91.8516

0.592 (65.064) a

0.545 (20.345) a 0.577 (734.950) a

0.394 (27.283) a 0.38079 0.83962

0.386 (22.284) a 0.37715 0.84412 2.5811

0.386 (22.284) a 0.37892 0.84041

Note: (1) Figures in parenthesis are t-value, and a, b, and c represent the 1%, 5%, and 10% significance levels, respectively. (2) β1(T) and β2(T) in Model (2) and (4) are function of T. The figures in the table are sample average of β1(Ti) and β2(Ti) over the sample i = 1, 2, …, 1174 observations. Figure 1: α0 + α1 T against T on the horizontal axis Model (1): α 1 + α 2Τ 2.00 1.80 1.60 1.40 1.20 1.00

-3.00

-2.00

0.80 -1.00 0.00

1.00

2.00

3.00

4.00

5.00

6.00

Figure 2: β1(T) against T on the horizontal axis Model (2): β1(T) 1.40 1.30 1.20 1.10 1.00 0.90

-3.00

-2.00

0.80 -1.00 0.00

1.00

2.00

3.00

4.00

5.00

Figure 3: α0 + α1 T against T on the horizontal axis

413

6.00


Model (3): α 0 + α 1T 2.00 1.80 1.60 1.40 1.20 1.00

-3.00

-2.00

0.80 -1.00 0.00

1.00

2.00

3.00

4.00

5.00

6.00

Figure 4: β2 + α2 L against T on the horizontal axis Model (3): β2 + α 2 L

-3.00

-2.00

0.560 0.555 0.550 0.545 0.540 0.535 0.530 0.525 0.520 -1.00 0.00

1.00

2.00

3.00

4.00

5.00

Figure 5: β1(T) against T on the horizontal axis

414

6.00


Model (4): β1(T) 1.40 1.20 1.00 0.80 0.60 0.40 0.20 -3.00

-2.00

0.00 -1.00 0.00

1.00

2.00

3.00

4.00

5.00

6.00

Figure 6: β2(T) against T on the horizontal axis Model (4): β2(T) 0.80 0.75 0.70 0.65 0.60 0.55

-3.00

-2.00

0.50 -1.00 0.00

1.00

2.00

415

3.00

4.00

5.00

6.00