Analyzing Cost Effectiveness

Analyzing Cost-Effectiveness of Organizations: The Impact of Information Technology Spending
SABYASACHIMITRA AND ANTOINE KARIM CHAYA
SABYASACHI MITRA is an Assistant Professor at the Dupree School of Management at the Georgia Institute of Technology. His research interests include strategic information systems, the economic impacts of information technology, software outsotircing, and telecommunications. His education includes a Ph.D in infonnation systems from the University of Iowa, and a B.Tech. in mechanical engineering from the Indian Institute of Technology, India.
ANTOINE KARIM CHAYA is a Ph.D candidate at the Dupree School ofManagement at the Georgia Institute of Technology. His education includes an M.S. in management from Georgia Tech, and a B.S. in civil engineering from the American University of Beirut. His research interests are focused on the economic impacts of information technology, justification of information technology expenditures, and strategic information systems. The performance impacts of infonnation technology (IT) investments in organizations have received considerable attention in recent years. In this research, we investigate the cost factors that are affected by such investments. We analyze a data set containi ng the information technology budgets of over 400 large and mediumsized U.S. corporations. We find that higher IT investments are associated with lower average production costs, lower average total costs, and higher average overhead costs. We also fmd that larger companies spend more on information technology as a percentage of their revenues than smaller companies. We do not find any evidence that infonnation technology reduces labor costs in organizations. We explain our findings, which are often counterintuitive but interesting, using basic microeconomic theory ofthe firm.
ABSTRACT:
business value of information technology, impact of information technology, information technology investments, information technology payoff.
KEY WORDS AND PHRASES:
THE LAST TEN YEARS HAVE WITNESSED A DRAMATIC INCREASE
in investments in information technology (IT) by U.S. businesses. Brynjolfsson [8] reports a tenfold
Acknowledgments: We would like to thank Computerworld for making their database of IT budgets available to us. We would also like to thank two anonymous referees and the guest editors of the special section for their valuable comments about an earlier draft of this paper. Both authors contributed equally to the project.
Joumat of Management Information Systems I Fall 1996, Vol. 13, No. 2, pp. 29-57 Copyright 1996 M.E. Sharpe, Inc.
30
SABYASACHI MITRA AND ANTOINE KARIM CHAYA
increase in IT investments between 1971 andl990. Accordingto [25], IT expenditure now accounts for 50 percent of capital investments and about 4 percent oftotal revenue for many U.S. businesses. By all accounts [6,28,30,32], investments in information technology in the U.S. economy have been staggering, and the upward spending trend continues. With such large investments in IT, researchers and managers have been struggling to quantify the benefits that firms derive from information technology. Much ofthe research in this area has focused on increases in white-collar productivity. The results of this research have generally been disappointing. Roach [28], in his study of information workers, documents a large increase in IT capital dedicated to information workers from 1970 to 1986. During the same time period, production worker productivity grew by 16.9 percent, while information worker productivity decreased by 6.6 percent. The period under investigation also witnessed a large increase in the number of information workers, leading Roach to conclude that computers had not yet contributed to the productivity ofinformation workers. Similar results were obtained in [15], which found that IT was associated with a sharp drop in capital productivity and with stagnation in labor productivity. Banker and Kauffinan [3], in a study ofthe banking industry, found no significant relationship between the number of ATM machines owned by a bank and its share of the local demand deposits and savings. Several others [ 12,24,27,30] failed to document the productivity impact ofinformation technology. This has given credibility to the term "productivity paradox," which is succinctly described by economics Nobel laureate Robert Solow: "we see computers everywhere except in the productivity statistics" [6]. Other evidence on the payoff from information technology investments has been more encouraging. Brynnjolfsson and Hitt's [7] study of 380 large firms between 1987 and 1991 found that the return on investments for IT capital was over 50 percent per year and that the return to spending on IT labor was also very high. Hitt and Brynjolfsson [18] found that computers have led to higher productivity and have created substantial value for consumers. Alpar and Kim [ 1 ], in their analysis of a large number of banks, found that IT investments were associated with a decrease in total costs. Harris and Katz [17] analyzed forty insurance companies and concluded that high IT spending was associated with lower growth in operating expenses. Similarly, Bender [4] reports that high IT spending results in improved cost efficiency in the insurance industry. However, many of these studies focused on specific industries, and their results are hard to generalize for other segments ofthe economy. Some evidence of IT payoff is weak and inconclusive. Mahmood and Mann [25], using the Computer World Premier 100 list, found that IT investments were weakly related to financial performance measures such as return on investments, return on assets, sales growth, productivity, and market-to-book value, only when grouped together and analyzed using canonical regression. Dos Santos, Peffers, and Mauer [ 13] found that the stock market reacted favorably when firms announced innovative IT investments, while noninnovative investments did not change the market value ofthe firm. Weill [31 ] investigated thirty-three valve manufacturing firms over a period of six years and found that investments in transactional IT were significantly and
ANALYZING COST-EFFECTIVENESS IN ORGANIZATIONS
31
consistently associated with higher retum on assets, sales growth, and nonproduction labor productivity. However, use of strategic IT had no effect on perfonnance in the long run and a slightly negative effect in the short run. Although a clear and consistent relationship between IT investments and financial performance is yet to emerge, companies have continued their large investments in infonnation technology. This is so in spite ofthe dissatisfaction expressed by general managers in mesisuring the value of IT to their organization [32]. If managers have been rational in such investments, it is possible that research may not have captured the true benefits that companies derive from IT.
Research Summary and Methodology

[10,11 ], WE OBSERVED AN INVERSE RELATIONSHIP between IT investments and the cost of operations of firms. In other words, higher spenders and more effective users of information technology achieved lower cost of operations when compared with low spenders on IT. Our results corroborated industry-specific fmdings [1,17]. In this paper, we investigate in more detail the cost factors that are influenced by IT investments. We analyze a data set containing IT budget data for over 400 mediumsized to large U.S. companies. We find that high spenders on information technology achieve lower cost of production and lower total operating costs, while their overhead costs are higher. We find no evidence that higher IT investments are associated with lower labor costs. Further, we find that larger firms spend more on information technology (as a percentage of total revenue) than smaller firms. The results are consistent with our belief that the primary benefit of information technology lies in the information and control that IT provides to management, and less in the automation aspects (replacing labor with IT capital) that are often considered the primary source of cost savings from IT use. As firms grow in size, they achieve greater economies of scale in their production processes. IT enables management to organize their production of goods and services better, to manage, track, and control their operations better, and to grow and achieve the benefits of economies of scale in their production facilities without the corresponding diseconomies of scale in their monitoring and control costs. We do not suggest that the automation benefits of IT are insignificant. At the present time, it would be difficult to find a large U.S. business that has not yet automated many of its clerical processes. Thus, when one examines a sample of large U.S. firms, it would be difficult to find differences in operating costs among these firms that can be attributed to their differences in levels of automation. In fact, firms that have a larger number of information (white-collar) workers would be likely to have both higher overhead and higher IT budgets because information workers are the primary consumers of IT. This view of information technology deviates from much ofthe existing literature on the "productivity paradox" that focuses more on the automation impacts of IT such as labor-cost savings obtained through automated order entry, invoicing, general
IN EARLIER RESEARCH
32
ledger, and payroll systems. However, it has wide support in the trade literature. A recent editorial in Business Week [9], which ran a cover story on 7%e Racefor Bigness, says: "the great irony is that much of this race to bigness is being propelled by the same infonnation technology that futurist Alvin Toffler and other techno-libertarians predicted would lead to smallness Corporate America is discovering what the military has known for at least a decade: that information technology allows for much more effective command and control of size and complexity." Paul Strassman, in his widely cited book on the business value of computers [30], suggests that management is the primary consumer of information technology, and we concur with him on this issue. We caution the reader that we establish no cause-effect relationships between the IT and performance variables examined in this paper. The distinction between the information and automation effects discussed above (and explained in more detail later) is one possible interpretation of our results. There can be several logical non-IT-based explanations as well. However, the results suggest that more research should be focused on assessing the information, control, and decision-making impacts of IT in an effort to understand the payoff from IT investments.
Conceptual Framework and Hypotheses

on the cost of operations of a firm. However, the process by which IT reduces operating costs is not well understood. For the purpose of this analysis, we envision two separate effects of IT: one is an automation effect; the other is an information effect. This distinction is important in understanding the impact of IT in organizations, specifically on their operating expenses. The automation effect of IT refers to the replacement of clerical labor with IT capital. Examples of such effects include automated order entry, invoicing, payroll, and other such systems whose primary purpose is to replace clerical labor with cheaper IT capital and to achieve lower costs in the process. The automation effect of IT is better understood, easier to quantify in terms of labor savings, and more local in its impact. The information effect of IT refers to the better control, monitoring, and decision making that IT provides to management. Such effects are harder to identify and quantify because their impact may be felt in various areas ofthe organization as well as in performance indicators that may seem unrelated to IT. One may argue that the primary benefit of automated transaction processing lies in the information that it provides to managementinformation that would otherwise have been too costly to compile. This allows management to track costs, identify and eliminate unprofitable lines of business, track performance of subordinates, and grow the company without losing control. Strassman [30] attempts to capture these benefits of IT in his Retumon-Management performance metric. The information effect is often ignored in productivity research.
THERE IS AMPLE ANECDOTAL EVIDENCE OF THE IMPACT of IT
33
A
V
e r a g e C
0
w w
LATCl
LACC /
LATC"
V
\
A
*1
^.^,^-^
" ' ' ' ^
/1/ // /
i.
q2
i
/
/ /
;
/ ' LACC"
,'
) LAPC
-^^=: --'
s t
^ Quantity
LATC LAPC LACC
Long Run Average Total Cost Long Run Average Production Cost Long Run Average Control and Monitoring Cost
Figure I. The Effect of Infortnation Technology on Long-Run Average Total Costs
Elemetitary Microecotiomic Analysis

Figure 1 shows the familiar long-run average total cost (LATC) curve of a manufacturing firm. The average total cost is the total cost per unit of output. The curve initially slopes downward, indicating economies of scale, and eventually tums upward, indicating diseconomies of scale as the firm grows. The increasing retums to scale that are responsible for the declining portion ofthe average total cost curve (left of point A) primarily reflect technological factors, while the increasing portion of the curve reflect monitoring and control costs [5], which increase as the firm gets larger. We draw the long-run average production cost (LAPC) curve as sloping downward, indicating typical economies of scale that firms achieve in their production processes. The larger the capacity ofthe plant, the cheaper the production output.' However, as firms grow larger, their control and monitoring costs increase, and the long-run average control and monitoring cost (LACC) curve slopes upward. The total cost curve, therefore, has the familiar U shape. Figure 1 also shows the effect of information technology on the average total cost curve. We hypothesize that IT reduces the average control and monitoring costs and shifts the LACC curve down and to the right, as shown in LACC". As a result, the total cost curve (LATC) shifts down and right to LATC". This results in the new optimal point (B), with a higher quantity and lower average total cost per unit output. The distinction between production costs and control and monitoring costs is admittedly difficult to draw; an example might make that distinction clear. Consider a manufacturing company that produces some end-product (WIDGET). The cost of
34
production includes the cost of direct labor, machinery, raw materials, warehousing, electricity, fuel, plant, and any other costs that are directly incurred to produce the WIDGET. The cost of monitoring and control, on the other hand, includes indirect costs. Such costs will include the salary of managers, clerical labor, analysts, personnel officers, market researchers, planners, and coordinators. They also include the cost of inventory outages, wastage, mismanagement, bad decisions, labor problems, agency costs, and other costs related to the complexity of the organization. When the firm starts as a small entrepreneurial enterprise with few employees and a single set of books managed by the entrepreneur, its coordination and control costs are low because of the simplicity of its operations. As it grows in size (produces more WIDGET and other related products), it achieves better economies of scale in its production facilities through better utilization of the plant and machinery, volume purchases of raw materials, and lower fixed costs per unit of output. However, as the complexity ofthe business increases, the firm needs more people to manage and coordinate its activities. Further, the complexity-related costs mentioned above also increase. The use of MRP II (Manufacturing Resource Planning) software illustrates the effect of IT on such organizations. A typical medium-sized to large manufacturing enterprise must stock, control, and ensure the availability of an average of 10,000 items (endproducts, subassemblies, and raw materials) [29]. Further, production of subassemblies and parts and the purchase of raw materials must be coordinated to ensure that the firm meets its delivery promises and production plans. MRP II software (with its ABC classification, automated reordering, stock status reports, master production scheduling, automated bill of materials, invoicing, and order-processing modules) makes it possible for the firm to control complexity effectively. This reduces its control and coordination costs and causes a shift in the LACC curve. In the MRP II example described above, it may also be difficult to distinguish between the information and automation effects of IT. We contend that the MRP II system primarily illustrates the information impact of IT. This is because its primary purpose is to provide control and information to management, and not to replace expensive labor with cheaper IT capital. Even if the MRP II system did not provide labor-cost savings, the complexity of the task would require large manufacturing companies to use automated inventory control and order processing. Microeconomics aside, the basic argument is as follows: Information technology provides greater control over dispersed operations, better information for decision making, and better tracking of costs and profits, and therefore increases the effectiveness of a firm's control and monitoring processes. This enables the firm to grow and to achieve the benefits of economies of scale in its production processes, while incurring a smaller increase in its overhead costs. In the process, the firm lowers its average total cost per unit of output. The above discussion leads us to the first and rather intuitive hypothesis on information technology investments and the average total cost per unit of output. HI: High spenders on information technology achieve a lower average total cost per unit of output than low spenders on IT.
35
Information Technology and the Cost of Production

If we exclude from our evaluation investments in shopfloor automation and robotics (which arguably constitute a small percentage of IT investments), information technology is unlikely to have a direct effect on a firm's cost of production. However, there are several ways in which IT can indirectly affect production costs. One such indirect effect is through reduced control and monitoring costs, illustrated in figure 1. Notice that, as a result ofthe shift in LACC to LACC", the productionrelated costs at point B are lower than at point A. Once again, the key argument based on the diagram is that IT enables a firm to grow and to achieve economies of scale in its production facilities, without a corresponding equal or greater increase in its monitoring and control costs. Another parallel and corroborating argument for such indirect effects is that IT provides better information that leads to better investment decisions. Thus, managers are able to identify the firm's less profitable lines of business and decrease their investments in such endeavors. According to Drucker [14], the "availability of information transforms the capital investment decision from opinion into a rational weighing of alternative strategies and assumptions, thereby allowing managers to focus on more fruitful courses of action and avoid costly mistakes." Better investment decisions, in tum, lead to better utilization of their resources and reduce their average production costs per unit of output. This effect of IT can be explained through figure I by including the cost of bad decisions and mismanagement (which increases as the firm grows in size and complexity) within the cost of monitoring and control (LACC). IT systems such as inventory control and shopfloor scheduling are also likely to affect production costs indirectly. The MRP II software discussed in the previous section can lead to better utilization of production facilities and a need for fewer people to manage the production process. IT is also the enabling technology for just-in-time inventory systems, which reduce a firm's costs for carrying inventory. For example. General Motors' Saturn plant in Tennessee includes a manufacturing database that is accessible to all its suppliers, who can check inventory and production status at GM. The suppliers are responsible for maintaining adequate stock of their products at GM, and GM pays for a part when it is actually used [16]. This reduces GM's control and monitoring costs for such products because GM has effectively outsourced a part of its control and monitoring activities to its suppliers. Through a shift in the LACC curve, such a reduction affects GM's cost of production. IT can also have a direct effect on a firm's cost of production. Computer-integrated manufacturing and robotics may directly affect production costs. Companies may use information technology to reduce the cost of raw material purchases through centralized buying. For example, Hewlett Packard allows each of its independent operating divisions to make its own purchasing decisions but stores all purchasing information in a centralized database that is used to negotiate volume discounts [21]. This reduces Hewlett Packard's cost of raw material purchases and consequently their production costs.
36
Our next hypothesis may at first appear counterintuitive, but quite naturally follows from figure 1 and the above discussion. H2: High spenders on information technology achieve a lower average cost of production per unit of output than low spenders on IT.
Information Technology and Overhead Costs

Analysis of figure 1 leads us to a rather surprising conclusion. Notice fTom the figure that the average cost of monitoring and control is greater at point B than at point A. The key is that the firm has increased its control and monitoring costs but has achieved a lower average cost of production and, consequently, a lower average total cost. Of course, without IT and the shift in the LACC curve, the control and monitoring cost at this higher level of output would have been even higher (the LACC curve is above the LACC" curve), which leads to a higher average total cost per unit output. Notice also from the figure that the average cost of control and monitoring at point B is not necessarily higher than at point A and depends on the relative positions and slopes ofthe LACC and LACC" curves. However, we believe, for a variety of reasons, that firms that spend more on information technology will have higher overhead costs than others that spend less. Information workers are the primary consumers of IT [30]. A firm with more information workers (such as managers, financial analysts, accountants, personnel staff, market researchers, planners, and coordinators) is likely to have more overhead expenses. Also, such a firm would require more IT to support these information workers. Another reason for higher overhead in firms that use more IT is that IT-related costs often appear as part of corporate overhead in financial statements.^ Although the costs of developing and maintaining inventory control systems and shopfloor scheduling systems can be thought of as production-related costs, it is often difficult to separate the cost of such systems from others developed by the MIS department. In such situations, it is natural to allocate the costs of the data center as part of corporate overhead. This leads us to the next hypothesis: H3: High spenders on information technology have higher average overhead costs per unit of output.
Information Technology Investments and Firm Size

\i automation effects were the primary effects of information technology, we would expect larger firms to spend less on IT as a percentage of their revenue because economies of scale in IT dictate that a larger firm could obtain the same level of automation as a smaller firm by spending a lower percentage of their revenues on information technology. The information effects of IT, however, dictate that larger firms have a greater need
37
for information technology than smaller firms because their control and monitoring costs are higher. If the primary benefit of IT lies in reducing such costs, we would expect larger fmns to spend more on IT as a percentage of their revenue than smaller firms, which derive little benefit from a reduction in such costs. This leads us to the next hypothesis: H4: Larger firms spend more on information technology as a percentage of their revenues than smaller firms.
Information I'echnology and the Cost of Labor

The previous hypotheses are mostly concerned with the information impact of IT. The next hypothesis is based on the automation impact of IT. It is often argued that IT reduces labor costs by replacing clerical workers with infonnation technology in the accounting, payroll, order-processing, distribution, and purchasing functions. Osterman [27] reports that, in his sample of forty firms in the manufacturing and service industries, a 10 jjercent increase in computing capital was associated with a 2 percent decrease in clerical employment. Several firms have also used information technology and communication networks to lower labor costs by relocating certain jobs offshore and to other cheaper sites. Using satellite lines for data communieations, voice communications, and teleconferencing. Citibank of New York moved its credit card operations to South Dakota during the 1980s because ofthe high labor costs in New York [26]. Other such examples include software development companies in the United States that employ programmers in Third World nations such as India, connecting them to the client's mainframe through electronic communication channels. Information technology has made many job functions' location independent. This brings us to the next hypothesis: H5: High spenders on information technology achieve lower average clerical labor costs than low spenders on IT.
Data Analysis
THIS SECTION DESCRIBES THE DATA ANALYSIS USED TO EXAMINE our hypotheses. We
describe the proxies used for the IT budget and cost measures, the data set used, and the analysis performed.
Proxies for IT Investments

Determining an appropriate measure of a firm's IT investment has been one ofthe greatest challenges for researchers [25]; many IT investment measures have been used in prior research. Single measures, such as IT expense as a percentage of total operating expense [4] and IT expense as a percentage of sales [31], have been used. Other researchers have used multiple measures to represent total IT investments [25].
38
This paper uses the IT budget as a percentage of sales (ITB/S), averaged over a period of time, as a proxy for the level of IT investment made by the firm. This measure has been used by other researchers [25,31 ], but it has many disadvantages [30]. The IT budget, which typically represents the annual budget for the IS department, is not the total IT investment made by the firm in that year. Some researchers have estimated that as much as 33 percent of IT expenditures is borne by end users and is not included in the IT budget figure [25]. Further, the measure does not include the IT investments made by the firm in the years prior to the period under investigation. It also does not take into account the effectiveness ofthe IT department, infrastructure, and personnel. Sales is used in the IT investment measure to account for the size ofthe firm. Using sales as a proxy for firm size has advantages and disadvantages. Among the disadvantages may be their potential volatility over time. However, sales has an advantage over some other measures used as surrogates for the size ofthe firm, such as number of employees or total assets. It is not sensitive to the capitallabor mix ofthe inputs. IT budget as a percentage of sales is an intuitive and easily understood measure that is widely used in research and in practice [25, 31]. A more complete discussion ofthe advantages and disadvantages of using such financial ratios can be found in [23]. Averaging the IT budget for a number of years is crucial in capturing any investment cycles that firms may adopt. However, it also introduces some bias in the analysis, since the average value based on the years with available data is used as an estimate for the firm's investment levels in other years. Another approach would be to include only those firms that had investment data for each ofthe five years ofthe study, but the resulting sample size would be too small. As a compromise, we include only those companies that had at least three years of budget data available. In using this average ITB/S value as a proxy for the firm's IT infrastructure, we made two assumptions: first, the IT budget in any given year is proportional to the total IT investment in that year; second, the average IT investment made by the firm during the period of the study is proportional to the total IT investments made by the firm in the past. Given these assumptions, it is reasonable to use the average ITB/S value as a proxy for the firm's total IT investment.
Proxies for Average Costs and Firm Size

For the purpose of this study, four cost measures need to be captured through routinely reported accounting data available fTom public sources such as Compustat: (1) the average total cost per unit of output, (2) the average cost of production per unit of output, (3) the average overhead cost per unit of output, and (4) the average clerical labor cost per unit of output. In addition, hypothesis 4 requires us to use a proxy for firm size. We chose operating expenses (OEXPdefined as the sum of COGS and SG&A) as a proxy for the firm's total cost of operations. Costs due to depreciation, interest expense, extraordinary items, and taxes are excluded ft-om operating expense. This better isolates the results from special non-recurring situations, and precludes the costs incurred due to the adoption of different accounting and financing methods. Operating
39
expense is selected because it is the most general and encompassing measure of a firm's total cost of operations. In this analysis, we represent operating expense (OEXP) as a percentage ofthe firm's total sales (OEXP/S). The average total cost per unit of output during a certain period of time is equal to the total cost of all units produced, divided by the number of units produced during that period. The number of units produced during a given period of time is proportional to the sales plus the ending inventory minus the beginning inventory. When we assume that fluctuations in inventory levels are insignificant relative to sales, the number of units produced during a given period of time is proportional to Ihe sales during that period. Thus, the average total cost per unit of output is proponaonal to the total cost of operations divided by sales. Cost of goods sold (COGS) and selling, general, and administrative (SGA) costs were chosen to represent the total production and overhead costs, respectively. These are the generally accepted accounting measures for the costs of production and overhead expenses. We admit that these are not perfect measures because they are affected by the finn's accounting practices. Using the same reasoning employed for OEXP above, we divided both COGS and SGA by sales in order to capture the average cost of production per unit of output and the average overhead cost per unit of output, respectively. The average cost of clerical labor is difficult to calculate from published financial data. Firms sometimes report their total labor costs, but that figure represents manufacturing, clerical, and managerial labor. Nevertheless, we used labor costs (LBR) divided by sales as a proxy for the firm's average clerical labor costs per unit of output. In the data analysis, we attempted to correct this bias by considering industries (such as banks and insurance companies) where clerical labor accounts for a significant portion ofthe firm's labor costs. Hypothesis 4 requires us to use a proxy for the size of the firm. Several measures are possible for firm size, such as total revenue or sales, number of employees, and total assets. In this study, we chose sales as the proxy for firm size for reasons explained earlier.
Accounting for Differences Based oti Industry

Certain industries by their nature spend more on IT than others. Further, the various cost-sales ratios may differ significantly across industries. One possible approach is to perform the analysis separately for each industry classification. The resulting small sample sizes, however, would undermine the reliability ofthe results. We therefore calculate a standardized score for the cost-sales and ITB/S values for each firm that takes into account the characteristics ofthe industry in which the firm operates. Several approaches may be followed to calculate such a standardized score. We may determine the rank or percentile score for the cost-sales ratios and ITB/S values for each firm within its own industry. Or we may measure the absolute deviation ofthe ITB/S and cost-sales values from the industry mean or median. A better approach would be to calculate the Z scores [22] for the IT and cost measures for each firm based
40
on the industry. The Z score is a measure of position (similar to percentile) based on the mean and standard deviation of the data. In essence, the Z score measures the number of standard deviations between the data point and the mean. The Z score for any data valuex is defmed as:Z = (x-x)/s, where x is the mean and 5 is the standard deviation ofthe population. The Z score is intuitive. It allows us to pool data from various industries, thereby providing for a larger sample size than is otherwise possible. For large samples, the distribution ofthe Z scores will converge to the normal distribution (central limit theorem). This allows us to perform the statistical tests described below. Thus, for each company in our sample, the following standardized scores (denoted as Z_costs/sales and Z_ITB/S) were calculated for each year: ,/ I cost/sales (company) - mean cost/sales (industry) Z_cost/sales = ^ j ^ -;;^^-^ ; standard deviation cost/sales (industry)
7 ITR/'i -
ITB/S {industry) standard deviation ITB/S (industry)
The Z_cost/sales values were calculated for each ofthe four costs measures (OEXP, COGS, SGA, and LBR) described in the previous section. Details about the industry classifications and the approximations used to calculate the industry means and standard deviations are explained later.
The Data Set

The data for the IT budgets were obtained from the editors of Computerworld, a leading trade publication. Computerworld conducts an annual survey of large, publicly traded companies, and collects IT budget data, among other items. In spite of its shortcomings, several researchers have agreed that the data set is the among best currently available [25]. The Computerworld database used for this research spans a period of five years, from 1988 to 1992. The accounting data used to compute the performance measures were obtained from the CompuStat database, which provides annual and quarterly operating and financial infonnation on 15,600 publicly traded companies for the last twenty years. The original database used in this study consisted of 609 companies that we could match in Compustat, with sporadic IT and financial data over the period between 1988 and 1992. After removing banks, insurance companies, and a few utilities'* in the sample, the database contained 457 companies. Of these, we removed an additional nine companies because their revenues reported in Compustat did not match (with even a wide margin of error) the revenues reported in the Computerworld database. Thus, we were left with 448 companies that we were able to match with a high level of confidence in Compustat.' Many of these companies were in the service sector and did not report their cost ofgoods sold. Further, these companies had sporadic IT data over the period of the study. The reason for the incomplete data was that the set of respondents to the Computerworld survey changed from year to year. As a result, the
41
number of companies for which both financial and IT data were available for an acceptable number of years required to perform the analysis was significantly smaller. These numbers are reported later when we present our analysis results.
Calculating Industry Means and Standard Deviations

The Standard Industry Classification (SIC) codes were used to group companies in our sample into different industries. SIC codes are four-digit numbers that classify firms according to their primary activity. The classification scheme is such that the primary activities of two firms that have the same first few digits of the code are more related than those that do not. The greater the number of common first digits, the more similar the primary activities of the firms are. When calculating the mean and standard deviation of the industry for the Z_ITB/S values, we had to rely on our database of 448 companies because IT budget data were not available foi- other firms. Thus, for the IT budget data, we used a two-digit SIC classification to obtain reasonable sample sizes within each industry group, in order to reliably calculate the industry parameters. In other words, each two-digit SIC code represented one industry, and we calculated the mean and standard deviations of the ITB/S ratios based on all firms with that two-digit classification. Admittedly, a four-digit classification would have better captured the nuances of IT investments across industries, but it would have resulted in too few companies within each group, thereby reducing our confidence in the parameters estimated. The industry means and standard deviations for the cost/sales ratios were calculated from the Compustat database. There were a total of 1,794 large companies in the Compustat database (with annual sales greater than $200 million, the same category of companies as those in the Computerworld database) with available cost/sales data. Choosing a two- or a four-digit classification scheme represents a tradeoff. Classifying firms based on the two-digit SIC codes results in more firms within each industry group and increases our confidence in the parameters estimated. A four-digit classification results in many industry groups with fewer than ten companies, even when using the Compustat database. We observed from our sample that the standard deviation for the OEXP/S ratio was around 7 percent, while the standard deviation for the COGS/S and SGA/S ratios was much higher (18 percent). This is because firms within the same two-digit classification may follow very different accounting methods for allocating costs to COGS and SGA, while the total cost figure is less affected by such policies. A four-digit classification reduced this variance within each group. Thus, we used a four-digit classification for the SGA/sales and COGS/sales variables and a two-digit classification for the OEXP/sales variable. We grouped all firms in the Compustat database (with sales more than $200 million) based on their four-digit or two-digit SIC code, and we calculated the mean and standard deviation for each group for every year between 1988 and 1992.
42
Calculating the Average Z Scores

After calculating the industry means and standard deviations for each industry group for the ITB/S and the various cost-sales ratios as explained in the previous section, we calculated the Z_ITB/S score and the Z_cost/sales scores for each year between 1988 and 1992 for every company in our sample. For the two-digit SIC code classification, we required a minimum often companies within each industry group to calculate the mean and standard deviation for that group. For the four-digit classification used to calculate the Z_cost/sales ratios, we required a minimum of five companies within each group to calculate the industry parameters. We also tried various other values for these arbitrary cutoffs in our computational experiments. Thus, a company may have a "null" value for the Z_ITB/S or Z_cost/sales score for any specific year if the corresponding IT budget or cost/sales data were not available for that year, or if a sufficient number of companies within the two-digit or four-digit SIC code classification was not available to calculate the corresponding Z scores. For every company, we determined an average Z_ITB/S and Z_cost/sales score for each ofthe cost measures discussed earlier. In calculating that average, we required at least three years (out of five) of the corresponding Z score for that company. We also used other values of this arbitrary cutoff in our computational experiments. Calculating the average is important in capturing the investment cycles that firms may adopt in their spending on IT, and in reducing the effect of an uncharacteristic economic downtum that a firm may face in a specific year. The data on IT budgets were very sporadic because the respondents to the survey changed from year to year. Many firms had only a single year of budget data available. Such companies were used to calculate the industry parameters but had to be omitted from the sample. Further, some industries were omitted because we could not reliably calculate the industry parameters, as explained above. Also, many ofthe 448 companies did not have all their cost values reported in Compustat. As a consequence, the sample sizes decreased considerably, but they were still large enough to give us confidence in the results. The sample sizes are reported in Tables 113.
Statistical Analysis and Results

FOR EACH OF THE HYPOTHESES FORMULATED ABOVE, WE PERFORMED very similar analyses. In this section, we explain the basic methodology. In the next section, we present the results from the statistical tests used to support the hypotheses. To identify the high and low spenders on information technology, as required by the hypotheses, companies in the sample were sorted in descending order of their average Z_ITB/S values. The resulting sample was split at the 50 percent level (midpoint). The top 50 percent represented high IT spenders, while the bottom 50 percent represented low IT spenders. One-tailed, two sample Mests of means and one-tailed, two-sample Mann-Whitney tests were performed on the various average Z_cost/sales values of the two subsamples to test for significant differences in their means and medians. The sample was also split at the 33 percent and 25 percent levels. In the first case, the top
43
33 percent of the .sample represented high IT spenders while the bottom 33 percent of the sample represented low IT spenders. In the second case, the top 25 percent of the sample represented high IT spenders while the bottom 25 percent of the sample represented low IT spenders. One of the criticisms of the above method is its undue sensitivity to outliers in the sample. Unusual values for the ITB/S variable may be due to inaccurate reporting (since the IT budget data are not audited and are self-reported), unusually high spending during the years under investigation by a firm, or sudden changes in revenue (used in the denominator) in a particular year. Outliers for the cost-sales ratios will mainly result firom unusual financial situations in a specific year during the period under investigation, such as sudden changes in revenue. To control for these effects, we created three related samples. The first was the original untrimmed sample. The second sample was obtained by trimming the original sample 2.5 percent at the top and bottom based on the average Z_ITB/S values. The third sample was obtained from the original samjjle after trimming 2.5 percent at the top and bottom based on the average Z_cost/sales values. Each of these three samples was then split at the 50 percent, 33 percent, and 25 percent level as explained earlier, resulting in a total of nine samples to test each hypothesis. The trimming methods represent an attempt to determine whether the results obtained from the statistical analyses are primarily driven by a few high or low spenders on IT and by extreme performers. Trimming removes such extreme points from the data set. It also reduces the effect of incorrect values for the Z_ITB/S and Z_cost/sales scores (such as an erroneous matching of the company in the Compustat database), since such values are likely to be at the extremes. Consistent results with and without trimming increase our confidence in the analysis. A more complete discussion of such trimming methods is found in [23].
Hypothesis 1
Our first hypothesis has been extensively examined in our earlier research reported in [ 10,11 ]; we reproduce only the summarized results here. In addition to the basic results reported here, the earlier study [11] also reports extensive sensitivity analysis to test the robustness of the results and examines the possibility of lagged benefitsfi-omIT. Table 1 reports the results from the /-test and Mann-Whitney tests on each of the nine samples described above. As shown in the table, each of the three samples (untrimmed, trimmed on the ITB/S values, and trimmed on the OEXP/S values) were split at three levels (50 percent, 33 percent, and 25 percent), resulting in nine separate runs for each test. The difference in the means of the two groups (high and low spenders) is reported in the table, along with the corresponding/? value (in parentheses). Several important observations can be madefi-omthe table. For each of the nine cases reported, the results were consistent with the hypothesis. That is, the high IT spenders had a lower average value for the Z_OEXP/sales variable than the low spenders (all values in the difference column are positive). Further, many results were significant at the 1 percent level, several were significant at the 5 percent level, and
44
Table 1. Mean and Median Comparisons for High and Low IT Spenders; Minimum of Three Years of Financial and IT Budget Data; Minimum of Ten Companies for Each Two-Digit Classification, Sample Size = 120
Trim var. Sample Sample means Split size % Bot Top Bot Top 50 33 25 50 33 60 Difference (p values) Sample medians Bot Top Difference (p values) 0.0775* 0.0202* 0.1044 0.0612* 0.0179* 0.217
0.235 0.0524* 0.1943
None None None

ITB/S ITB/S ITB/S
60 0.0368 -^.2088 0.2456 0.0405" 0.015 -0.134 0.149
39 40 0.1431 -0.2706 0.4137 0.0085*" 0.093 -0.134 0.227 30 30 0.0700-0.3106 0.3806 0.040" 0.013 -0.113 0.126 0.022 -0.147 0.169 0.096 -0.121 0.217 0.014 -0.202 0217
0.004 -0.106 0.110 0.064 -0.113 0.177 0,006 -^,106 0.112
57 57 0.0371 -0.2253 0.2624 0.036" 38 39 0.1573-0.2483 0.4055 0.011"
2.5 28 29 0.072 -0.3676 0.4396 0.024"

57 37 28 57-0.0172-0.1326 38 0.0899-0.166 29 0.0211-0.1875 0.1154 0.1735 0.2558 0.0265" 0.2086 0.1015*
OEXP/S 50 OEXP/S 33 OEXP/S 25
Means (medians) reported here for the high (top) and low (bot) IT spenders are for the average Z_OEXP/S value. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample Mest of means or for the one-tailed, two-sample Mann-Whitney test. Significant at the 1% level; significant at the 5% level; significant at the 10% level.
almost all results were significant at the 10 percent level. The median values for the two groups, the difference in the median values, and the p values from the MannWhitney tests are also reported in Table 1. Once again, all values in the difference column are positive (supporting our hypothesis). Based on the table, we find that companies with higher IT investments have lower operating costs.
Hypothesis 2
We performed extensive analyses to support hypothesis 2. Table 2 shows the results of those analyses on a sample of ninety-eight firms. For the results reported in Table 2, we required a minimum often companies within each two-digit SIC code classification to calculate the Z_ITB/S values, and a minimum of five companies within each four-digit classification to calculate the COGS/sales values. We also required both financial and budget data for each firm for a period of at least three years between 1988 and 1992. The ninety-eight companies thus included constitute a sample in which we have the most confidence. We later vary these parameters to obtain larger sample sizes of 145 and 190 companies but introduce some "noise" in the process.
45
Table 2. Comparisons of COGS/S for High and Low IT Spenders; Minimum of Three Years of IT Budget and Financial Data; Minimum of Ten Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 98
Sample Sample means Split size % Bot Top Bot Top
50 33 25 50 33 25 49 32 24 47 31 23 49
Trim var. None None None ITB/S ITB/S ITB/S
Difference (p values)
Sample medians Bot Top
Difference ip values) 0.0091" 0.0047** 0.0101" 0.0112" 0.004*** 0.0101" 0.009*** 0.002*** 0.004***
0.1470-0.2266 0.3736 0.008"* 0.076 -0.106 0.182 0.139 -0.106 0.245 0.139 -0.106 0.245
33 0.2569 -0.2499 0.5068 0.005*" 25 46 31
0.2287 -0.2867 0.5154 0.017**
0.1421 -0.2338 0.3759 0.009*** 0.076 -0.119 0.195 0.2492 -0.2708 0.5200 0.005*** 0.118 -0.165 0.283 0.5238 0.019** 0.118 -0.165 0.283
23 0.2016-0.3222 47 31 24
COGS/S1 50 46 COG/S 33 COG/S 25

31 23
0.1202-0.2231 0.3433 0.008*** 0.069 -0.106 0.175 0.2607 -0.2461 0.5068 0.002*" 0.118 -0.106 0.224
0.2170-0.2976 0.5146 0.006*** 0.118 -0.136 0.254
Means (medians) reported here for the high (top) and low (bot) IT spenders are for the average Z_COGS/S value. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported aie for one-tailed, two-sample f-test of means or for the one-tailed, two-sample Mann-Whitney test. Significant at the 1% level; **significant at the 5% level; *significant at the 10% level.
The results in Table 2 strongly support hypothesis 2. All values in the difference column (for both the medians and means) are positive, indicating that high spenders had a lower COGS/sales ratio than low spenders. Almost all the results are significant at the 1 percent level. It is difficult to translate the difference in Z scores between high and low spenders into actual dollars. An approximate analysis, however, is possible. For the sample of 448 companies, the mean COGS/sales value is approximately 70 percent, and the standard deviation is 18 percent (the standard deviation within each four-digit SIC code classification varies between 3 and 25 percent, with an average of approximately 12 percent). From Table 1, a difference of 0.4 in the Z scores between the high and low spenders translates to an approximate difference of 4.8 percent in their COGS/sales ratios. We also performed an ordinary least squares (OLS) regression on the nine samples with the average Z_ITB/S as the independent and the average Z_COGS/S as the dependent variable. As before, the three samples (untrimmed, trimmed on ITB/S, and trimmed on COGS/S) were split at the 50 percent, 33 percent, and 25 percent level. We then combined the top and bottom subsamples and performed the regression on
46
Table 3. Trim var.

NONE NONE NONE ITB/S ITB/S ITB/S COGS/S COGS/S COGS/S
Ordinary Least Squares Regression for ITB/S and COGS/S Variables Split %
50 33 25 50 33 25 50 33 25
Sample Size ZJTB/SCoeff

98 65 49 93 62 46 93 62 47 -0.241756 -0.244060 -O.223028 -0.318651 -0.321473 -0.297397 -0.246192 -0.256282 -0.233660
p value
0.007* 0.010* 0.028* 0.003* 0.005* 0.017* 0.003* 0.003* 0.009* 0.05062 0.06777 0.05591 0.06675 0.09081 0.07795 0.06849 0.10553 0.09807
Sample obtained after combining the top and bottom 50%, 33%, or 25% as indicated. The p value reported here is for a one-tailed test. ***Significant at the 1% level; **significant at the 5% level; 'significant at the 10% level.
the resulting sample. Thus, splitting the original sample at the 50 percent level implies that we performed the regression on the original sample (after appropriate trimming), while splitting at the 33 percent and 25 percent levels implies that the middle 33 percent and 25 percent ofthe data points were ignored in the analysis. This was done to see if removing the middle "noise" had any effect on the analysis. The results are reported in Table 3. Notice that the coefficients ofthe average Z_ITB/S variable are negative, and all the results are significant. In order to further support these initial results, we conducted a sensitivity analysis on the arbitrary parameters used in the above analysis. For the results shown in the following tables, we used different values for the minimum number of years of financial and budget data required and the minimum number of companies that are required for two-digit and four-digit SIC classifications for the ITB/S and COGS/S variables, respectively. The results are shown in Tables 4 and 5. All the results in the tables support hypothesis 2, and most are significant at the 10 percent level. We conclude that high spenders on IT have lower average production cost per unit of output than low spenders. Table 6 shows the distribution of SIC codes and industry names in the sample of 98, 145, and 190 firms in Tables 2,3,4, and 5. It also shows the number of companies in each SIC code classification in each sample.
Hypothesis 3
An analysis identical to the one discussed in the previous section was performed with the cost measure SGA/sales as the proxy for the average overhead cost per unit of output. The results are shown in Tables 7,8,9, and 10. Once again, the results support hypothesis 3, and most are significant at the 10 percent level. Wefindfromthe analyses that high spenders on IT have higher overhead costs per unit of output than low spenders. The results are interesting because they are counterintuitive, but they can be explained as discussed earlier.
47
Table 4. Comparisons of COGS/S Means and Medians for High and Low IT Spenders; Minimum of Three Years of IT Budget and Financial Data; Minimum of Five Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 145 Trim
var. None None None ITB/S ITB/S ITB/S Sample Sample means Split size Top % Bot Top Bot 50 33 25 50 33 25 72 48 36 69 46 35 68 45 34 Difference (p values) Sample medians Bot Top 0.007 Difference (p values) 0.072 0.116 0.167 0.042** 0.185 0.018** 0.069 0.127 0.162 0.05** 0.180 0.022** 0.080 0.10* 0.145 0.067* 0.173 0.029*'
73 0.1007-0.0761 0.1768 49 0.1188-0.1595 0.2783 37 0.1870-0.2275 0.4145 69 0.0761 -0.0944 0.1705 46 0.0995-0.1697 0.2692 35 0.1747-0.2404 0.4151 70 0.0873-0.0717 0.1590 47 0.1030-0.1179 0.2209 35 0.1929-0.1477 0.3406
0.088* 0.079
0.046** 0.094 -0.073 0.015** 0.112 -0.073 0.100* 0.076 0.007
0.057* 0.094 -0.068 0.016** 0.118 -0.062 0.094* 0.079 -0.001 0.07* 0.083 -0.062
COGS/S 50 COGS/S 33 COGS/S 25
0.023** 0.112 -0.062
Means and medians reported here for the high (top) and low (bot) IT spenders are for the Z_COGS/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample r-test of means or for the one-tailed, two-sample Mann-Whitney test. ***Significant at the 1 % level; **significant at the 5% level; *significant at the 10% level.
We emphasize that the above results do not imply that information technology increases overhead costs. In fact, causality is likely to be in the opposite direction. That is, larger firms with higher overhead costs have more information workers (managers, planners, strategists, accountants, analysts). Consequently, they require more IT to support these information workers.
Hypothesis 4
Hypothesis 4 does not involve any average cost measure. To test hypothesis 4, we calculate the ZJSales score for each year for each company as: _, _ Sales (company) mean sales {industry) I. Sales = ,,.. ;,. , ^ ~ std. deviation sales {industry) The Z_Sales value represents the number of standard deviations the sales of the company lies above or below the mean of the industry. The reasoning behind
48
Table 5. Comparisons of COGS/S Means and Medians for High and Low IT Spenders; Minimum of Two Years of IT Budget and Financial Data; Minimum of Five Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 190 Trim var. None
None None ITB/S ITB/S ITB/S
!Split
50 33 25 50 33 25
Sample size Sample means % Bot Top Bot Top 95 63 37 91 60 45 90 60
Sample medians Bot Top
Difference (p values) 0.098 0.097* 0.196 0.035** 0.186 0.040** 0.099 0.127 0.162 0.063* 0.145 0.077* 0.061 0.234 0.156 0.078* 0.128 0.090*
95 0.0622-0.0941 0.1564 64 0.1133-0.1413 0.2546 38 0.1556-0.1472 0.3028 90 0.0525 -0.0953 0.1478 60 0.0975-0.1379 0.2354 45 0.1473-0.1223 0.2696 91 0.0136-0.0546 0.0682 61 0.0932-0.0764 0.1696
0.091* 0.071 -0.027 0.046** 0.106 -0.090 0.040** 0.118 -0.068 0.114 0.071 -0.018
0.070* 0.094 -0.068 0.070* 0.118 -0.027 0.266 0.111 0.052 -0.009 0.094 -0.062
COGS/S 50 COGS/S 33
COGS/S 25 45 46 0.1689-0.0328 0.1997
0.100* 0.118 -0.010
Means and medians reported here for the high (top) and low (bot) IT spenders are for the Z_COGS/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample ;-test of means or for the one-tailed, two-sample Mann-Whitney test. ***Signiflcant at the 1 % level; **significant at the 5% level; *significant at the 10% level.
calculating the Z score is the same as for the cost measuresnamely, to account for natural differences in firm size across various industries. Once again, we use a four-digit classification to calculate the Z_Sales values. The Z_ITB/S values are calculated in the same way as before, using the two-digit SIC classification. To test hypothesis 4, we sorted the sample (of 113 resulting companies) based on the average Z_Sales value (not on the Z_ITB/S value as before). We then trimmed and split the sample as explained earlier; this resulted in nine samples. We compared the mean Z_1TB/S values for the top and bottom segments (representing large and small companies, respectively) of each sample. The results are shown in Table 11. The results support our hypothesis. All values in the difference column are positive, indicating that larger companies had higher ITB/S ratios when compared with smaller companies. Most results are significant at the 10 percent level. We did find evidence that larger companies spend a larger percentage of their sales on information technology than smaller companies. The results of sensitivity analysis were similar and are omitted from this presentation. A possible concern regarding the results presented in Table 11 is that our sample
49
Table 6.
SIC Codes and Industry Names Included in the Study Samples

Sample 145 2
1 1 1
SIC code 2000 2011 2015 2030 2040 2060 2080 2086 2211 2320 2330 2510 2621 2631 2670 2711 2731 2800 2810 2821 2834 2840 2844 2851 2860 2911 3089 3312 3330 3334 3410 3490 3510 3523 3540 3561 3640 3711 3714 3724 3728 3742 3812 3841 3861 4011 4813 4833 4841 5122 5140
Industry name Food and Kindred Products Meat Packing Plants Poultry SIghtr. & Process Can/Prsrvd Fruit/Veg Grain Mill Products Sugar & Confec. Products Beverages Bttld/Can Soft Drnks/Wtr Brdwoven Fabr. Mill, Cttn Means, Boys Furn. & Cloth Women's Outerwear Household Furniture Paper Mills Paperboard Mills Convrt Papr, Paperboard Newspaper: Pub. & Print Books: Pub. and Printing Chemicals and Allied Prod. Indl Inorganic Chemicals Plastics, Resins, Elastom. Pharm. Preparations Soap, Deterg., Toilet Ppr Pert, Cosm., Toilet Prep. Paints, Varnishes, Lacq Indl Organic Chemicals Petroleum Refining Plastic Products, Nee StI Works & Blast Furn. Refin Nonfer Metals Prim Prod of Aluminum Ctlry, Handtis, Gen Hrdwre Misc. Fabricated Mtl Prod. Engines and Turbines Farm Machinery and Equip. Metalwork Mach. & Equip. Pumps and Pump Equip Elctrnic Light, Wire Equip Motor Veh. and Car Bodies Motor Veh Parts, Ace. Aircrft [Engines & Parts Aircraft Parts, Aux Equip Railroad Equipment Srch, t3et, Nav, Aero Sys Surg., Medical Inst, Appar. Photo Equip & Supplies Railroads, Line-Haul Oper Phone Comm, Ex Radio Television Broadcast Stn Cable fi Other Pay TV Ser. Drugs & Proprietary, Whis Groc. & Rltd Prod., WhIs
190 2 3 1 3 4 2
4 1
98 2
1 1 1
4 2
1
3 1 3 3 10 3 2 6 2
6 1 5 10 2 3
8 2 2 6
1 5 1
3 2 8 2 2
5 1
2 3
16
3 9 2 1 2
1 15 4 5
3 9 2
1
2
1 14
4 6 2 3 1 2 5 2
1 1
2 2 1 2 5 2
1 1
5 2
1 1
3 2 5
4 1 1 4
2 2
4 4 1 4
2 2 2 5 2 1 3 2
2 2 2
4
2 4 4 1 3 2
1 4
2 1 1 2
50
Table 6.
Continued Sample 145 1 3 5 8 2 4
SIC code 5141 5150 5311 5331 5411
Industry name Groc, Gen. Line Wholesale Farm Prod., Raw Mtrl. WhIs Department Stores Variety Stores Grocery Stores
190 1 2 5 6 9
98
Table 7. Comparisons of SGA/S Means and Medians for High and Low IT Spenders; Minimum of Three Years of IT Budget and Financial Data; Minimum of Ten Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 98 Sample Split size Sample means % Bot Top Bot Top 50 33 25 50 33 25 50 33 25 49 32 24 47 31 23 46 31 23 49 -0.2590 33 -0.3509 25 -0.4357 46 -0.2568 31 -0.3386 23-0.4147 47-0.1995 31 -^.2360 24 -0.2764 Sample medians Bot Top 0.024 0.024 0.121 0.073 0.121 0.150 0.023 0.023 0.072
Trim var. None None None ITB/S ITB/S ITB/S SGA/S SGA/S SGA/S
Difference (p values) -0.308 -0.382 -0.489 -0.357 -0.476 -0.516 -0.273 -0.364 -0.434 0.010* >* 0.007* ' 0.010* '* 0.016* 0.011* 0.018* 0.046* 0.042*
0.051*
0.1635-0.4225 0.1676-^.5185 0.1263-0.5620 0.1511 -0.4079 0.1596-0.4982 0.1418-0.5565 0.0472 -0.2467 0.0540 -0.2900 0.0342-0.3106
0.005** -0.284 0.004** -0.358 0.0008***-0.368 0.0008***-0.284 0.006** -0.355 0.012** -0.366 0.045** -0.250 0.041** -0.341 0.057* -0.362
Means and medians reported here for the high (top) and low (bot) IT spenders are for the average Z_SGA/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample t-test of means or for the one-tailed, two-sample Mann-Whitney test. Significant at the 1% level; **significant at the 5% level; *significant at the 10% level.
consisted mainly of medium-sized to large firms. The average sales for all firms in the sample (448 companies) was approximately $5 billion. However, the standard deviation was approximately $ 10 billion, which indicates sufficient variation in firm size within the sample.
Hypothesis 5
Companies do not routinely report their clerical labor costs as a separate item on their financial statements. We therefore used the total labor cost as a percentage of sales (LBR/Sales) as a proxy for their clerical labor costs per unit of output. However, there were two problems with this approach. First, only a small set of companies reported their total labor costs in Compustat, resulting in a small sample size of twenty-six and
51
Table 8. Trim var None None ITB/S ITB/S ITB/S SGA/S SGA/S SGA/S
Ordinary Least Squares Regression for ITB/S and SGA/S Variables Split % 50 25 50 33 25 50 33 25 Sample size ZJTB/Scoeff 98 49 93 62 46 93 62 47 0.243821 0.242482 0.288949 0.291956 0.288205 0.155286 0.156716 0.150249 p value 0.010*** 0.015** 0.011** 0.009*** 0.018** 0.045** 0.035** 0.053* Adj;?^ 0.04474 0.07703 0.04555 0.07459 0.07544 0.02028 0.03726 0.03601
Table 9. Comparisons of SGA/S for High and Low IT Spenders; Minimum of Three Years of IT Budget and Financial Data; Minimum of Five Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 145 Sample Split size Sample means % Bot Top Bot Top Sample medians Bot Top
Trim var. None None None ITB/S ITB/S ITB/S SGA/S SGA/S SGA/S
Difference (p values) 0.057* 0.033** 0.009*** 0.057* 0.026** 0.013** 0.174 0.078* 0.043**
50 72 73-0.1743 0.0593-0.2336 0.043**-0.290 -0.082 -0.198 33 48 49-0.1828 0.1328-0.3156 0.031**-0.290 25 36 37-0.3451 0.1584-0.5035 0.004**^.364 0.121 -0.411 0.121 -0.485
50 69 69-0.1567 0.0815-0.2382 0.046**-0.284 -0.091 -0.193 33 46 46-0.1706 0.1656-0.3362 0.028**-0.290 25 35 35-0.2784-0.2138-0.4922 0.006**M).355 0.136 -0.426 0.150 -0.505
50 68 70-0.1336-0.0332-0.1004 0.201 -0.250 -0.115 -0.135 33 45 47-0.1567 0.0442-0.2009 0.087*-0.284 25 34 35-0.2536 0.0409-0.2945 O.038**-0.358 0.080 -0.364 0.080 -0.438
Means and mediansreportedhere for the high (top) and low (bot) IT spenders are for the average Z_COGS/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample r-test of means or for the one-tailed, two-sample Mann-Whitney test. Significant at the 1% level; significant at the 5% level; significant at the 10% level. introducing bias in the analysis. Second, the labor-cost value includes the cost of manufacturing, clerical, and management labor. In industries where clerical labor is a small fraction ofthe total labor force, this proxy does not adequately capture their average clerical labor costs. We performed the same analysis reported in earlier sections using the LBR/Sales ratios. The results were not significant, and we did not find any evidence to support hypothesis 5. For brevity of presentation, the results are not shown here.
52
Table 10. Comparisons of SGA/S Means and Medians for High and Low IT Spenders; Minimum of Two Years of IT Budget and Financial Data; Minimum of Five Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 190 Sample Sample means Trim Split size Sample medians Difference Difference var. % Bot Top Bot Top (p values) Bot Top (p values)
None None None ITB/S ITB/S ITB/S SGA/S SGA/S SGA/S 50 33 95 63 95-0.1037 0.1103 0.2140 0,037"-0.176 64-0.1521 0.1769 0.3290 0.014"-0.217 48-0,2029 0.1474 0.3502 0.016**-0.296 90-0,0935 0,1196 0.2131 0.044**-0,176 0.080 0.167 0.136 0.101 0,167 0,150 0.023 0.121 0.191 0.256 0.384 0.432 0.277 0.375 0,367 0.164 0.262 0,318 0.050* 0.019** 0.034** 0.586* 0.029** 0.083* 0,201 0.088* 0.115
25 47 50 33 91
60 60-0,1296 0.1901 0.3198 0,020**-0.208 45-0,1662 0,1332 0,2994 0,044**-0,217 91 -0.0509 0.0154 0,0663 0,235 -0,141 61-0.0931 0,0733 0.1664 0,097*-0.141 46-0.1274 0.0575 0.1849 0.104 -0.217
25 45 50 33 25 90 60 45
Means and medians reported here for the high (top) and low (bot) IT spenders are for the average Z_COGS/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample t-test of means or for the one-tailed, two-sample Mann-Whitney test. ***Significant at the 1% level; **signiflcant at the 5% level; *signiflcant at the 10% level.
To partially correct for the problems discussed above, we attempted to identify an industry that routinely reports total labor costs and where clerical labor represents the majority of the work force. The original sample had about sixty banks (SIC code 60XXDepository Institutions) that we had not considered in the analysis because their cost structure is very different from manufacturing companies, and they do not report their cost of goods sold. However, banks routinely report their labor costs. Further, since they do not have manufacturing labor, clerical labor would be a significant part of their work force. The results ofthe analysis on the LBR/Sales ratio for the banks in our sample are shown in Table 12. Although all the results are in the right direction (the difference column is positive), they are not significant. fVe therefore do notfind any evidence to support hypothesis H5.
The Effect of Firm Size on Cost Factors

Using our sample of 448 companies, we performed an ordinary least squares regression between the Z_Sales scores on the one hand, and the various Z_Cost/Sales ratios
53
Table 11. Comparisons of ITB/S Means and Medians for Large and Small Companies; Minimum of Three Years of IT Budget and Financial Data; Minimum of Ten Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size =113
Trim var. Sample Sample means Split size Top % Bot Top Bot
50 33
Sample medians Bot Top 0.087 0.106 0.096 0.086 0.183 0.250 0.086 0.096 0.013
Difference (p values) 0.338 0.386 0.458 0.337 0.451 0.639 0.343 0.376 0.327 0.077* 0.131 0.040** 0.092* 0.120 0.057* 0.089* 0.201 0.133
None None None Sales Sales Sales ITB/S ITB/S ITB/S
56 37 28
54
57-^.1697 0.0682 0.2379 0.059* -0.251
38 -0.2074 0.0703 0.2777 0.070* -0.280 29-0.3102 0.1002 0.4105 0.029**-0.362

54-0.1574 0.0534 0.2207 0.075* -0.251 36 -0.2065 0.0649 0.2714 0.073* -0.268 27 -0.2776 0.0881 0.3657 0.045* -0.389 55-0.1838 0.0037 0.1875 0.855* -0.257 37-0.1983-0.0400 0.1582 0.169 -0.280 28 -0.2371 0.0513 0.1858 0.148 -0.314
25 50 33
25 50 33 25
36
27 53
35 26
Means and medians reported here for the high (top) and low (bot) IT spenders are for the average Z_ITB/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample /-test of means or for the one-tailed, two-sample Mann-Whitney test. Significant at the 1% level; **signiflcant at the 5% level; *significant at the 10% level.
on the other hand. The results are shown in Table 13. The results indicate no relationship between firm size and COGS/Sales or SGA/Sales. However, the association between firm size and lower total average costs is marginally significant. Given that larger firms also invest more in IT, it is not possible to predict from our analysis whether these firms achieve lower total costs through the use of IT or through other means not captured in the analysis. Given that larger companies do not necessarily achieve lower cost of production as shown in Table 13, the strong relationship between IT spending and lower production costs is interesting and intriguing.
Discussion and Conclusions

IN THIS RESEARCH, WE INVESTIGATED THE IMPACT OF I T investments on the various cost factors of an organization. We found that higher investments in information technology were associated with lower average production costs, lower average total costs, and higher average overhead costs. Further, we found that larger companies spent more on information technology as a percentage of their sales than smaller
54
Table 12. Comparisons of LBR/S Means and Medians for High (Top) and Low (Bot) Spenders on IT for SIC Code (60XX^Depository Institutions); Minimum of Three Years of IT Budget and Financial Data; Minimum of Ten Companies for Each Two-Digit SIC Code Classification; Minimum of Five Companies for Each Four-Digit SIC Code Classification, Sample Size = 44
Trim var. None None ITB/S ITB/S ITB/S Sample Sample means Split size Top % Bot Top Bot 50 25 50 33 25 22 11 21 14 10 Difference (p values) Sample medians Bot Top Difference (p values) 0.065 0.082 0.025 0.251 0.082 0.084 0.175 0.167 0.180 0.311 0.224 0.053* 0.370 0.110 0.060* 0.113
22-0.1585-0.4166 0.2581 0.176 -0.105 -0.170 11 -0.4452 0.4699 0.0247 0.474 -0.107 -0.189 20-0.1254-0.3140 0.1886 0.499 -0.103 -0.078 14-0.1542-0.4677 0.3135 0.159 0.004 -0.255
10-0.4045-0.3132 0.0913 0.402 -0.088 -0.170
BLBR/S 50
20 21 -0.1281 -0.4835 0.3553 0.068* -0.105 -0.189
BLBR/S 33 13 14-0.1020-0.5146 0.4127 0.063 -0.070 -0.255 BLBR/S 25 10 10-0.2080-0.5546 0.3465 0.140 -0.088 -0.255
Means and medians reported here for the high (top) and low (bot) IT spenders are for the average Z_LBR/S variable. Difference reported here is the difference between means or medians for the high and low IT spenders (Bot-Top). P values reported are for one-tailed, two-sample ;-test of means or for the one-tailed, two-sample Mann-Whitney test. Significant at the 1% level; significant at the 5% level; significant at the 10% level.
companies. We found no association between information technology investments and lower labor costs. Our results do not establish a causal relationship between IT investments and the various cost factors investigated. We suggested one possible explanation of our results through a distinction between the infonnation and automation effects of IT. If we assume that the information effect is the primary impact of IT investments, we would expect larger firms to spend more of their revenue on IT, consistent with our results, since larger firms have higher control and monitoring costs owing to the complexity of their operations. Further, a decrease in their control and monitoring costs through the use of IT makes larger companies (with higher IT expenditures) more cost-effective, as observed in Tables 1 and 12. Through its infonnation impact, IT enables a company to grow and achieve economies of scale in their production facilities without a corresponding increase in control and monitoring costs. This allows such firms to achieve lower average production costs, as observed from Tables 2, 3,4, and 5. Fur-
55
Table 13. Regression Analysis for Sales versus COGS/Sales, Sales versus SGA/Sales, and Sales versus OEXP/Sales Observations Sales coefficient Vs. COG/Sales Vs. SGA/Sales Vs. OEXP/Sales 249 249 249 -^.0446 0.0117 -0.0848 p value 0.4255 0.8362 0.1186 Adjusted K -O.0015 -^.0039 0.0058
ther, the automation impact of IT would dictate that firms with higher IT spending would achieve lower labor costs, which is not supported by our results. In addition, the information impact of IT implies that a firm with more information workers (who are the primary consumers of such information) would require more spending on information technology, as evidenced in Tables 7, 8, and 9. This study is limited, however, by the reliability of the data used. The IT budget data were provided by the editors of Computerworld, who collected the information through mail and phone interviews. In some cases, when the data were not available, they estimated the data based on "consulting and industry sources." Researchers have agreed, however, that the Computerworld data are among the best available [25]. Other limitations concem possible biases in the data. Companies in the data set are large and medium-sized, and the results may only be valid for other companies of similar size. Furl:her, the analysis was performed on mainly manufacturing firms and excluded financial and other firms in the service sector. It is also difficult to define IT investments. Companies may include different items in their IT budgets so that its definition may not be consistent across the sample. Another potential source of bias is the exclusion of a large number of companies from the sample because of the methodology used. However, unless there are systematic differences in the excluded companies, this may not be a major concem.
NOTES 1. Eventually, the LAPC may slope upward, as shown in the diagram. It is not important to our analysis because firms will not typically operate at that range. For example, many studies on the long-run costs of electricity production indicate significant economies of scale in electricity generation, but diseconomies of scale in the cost of power transmission, which causes the LAPC eventually to curve upward. Electric-power firms tend to operate on the left of the minimum point (see [19] for more detail). 2. We have ignored some other costs, such as the cost of capital. However, as long as the LACC curve shifts to the right as shown, the LATC curve will shift down and to the right. 3. Paul Strassman [30] attributes this practice to the days when central MIS delivered "pre-programmed and pre-defmed output to users, and all costs were made part of the fixed overhead." With the advent of end-user computing, this practice may become obsolete. 4. Banks, insurance companies, and other financial institutions do not report their cost of goods sold. Utilities are usually excluded from such studies because of the highly regulated environment in which they operate. 5. We extensively checked all extreme values for the ITB/S and the cost-sales ratios. When in doubt, we consulted Moody's Industrial Manuals for verification. 6. The Mann-Whitney test is a nonparametric test that does not assume underlying normal
56
populations and is less sensitive to outliers than the /-test. Similar to the Wilcoxon signed rank test (which is for paired samples), it calculates the sum of ranks of the two subsamples. The Mann-Whitney procedure essentially tests for differences in position between the two subsamples. In the tables in this paper, we report the one-tailed/; vaiues from the Mann-Whitney test. We also report the medians ofthe two samples (high and low IT spenders) to give the reader an idea ofthe actual difference in location between the two samples.
REFERENCES
1. Alpar, P., and Kim, M.A. A microeconomic approach to the measurement ofinformation technology value. Journal of Management Information Systems, 7, 2 (Fall 1990), 55-69. 2. Baily, M., and Chakrabarti, A. Electronics and white-collar productivity. Innovation and the Productivity Crisis. Washington, DC: The Brookings Institute, 1988. 3. Banker, R.D., and Kauffrnan, R.J. Strategic contributions ofinformation technology: an empirical study of ATM networks. Proceedings of the Ninth International Conference on Information Systems, Minneapolis, December 1988, pp. 141-150. 4. Bender, D.H. Financial impact of information processing. Journal of Management Information Systems, 3, 2 (Fall 1986), 22-32. 5. Browning, E.K., and Browning, J. Microeconomic Theory and Applications. New York: HarperCollins, 1992. 6. Brynjolfsson, E. The productivity paradox ofinformation technology. Communications ofthe ACM, 36, 12 (December 1993), 67-77. 7. Brynjolfsson, E., and Hitt, L. Paradox lost? Firm level evidence on the returns to information systems spending. Management Science, 42, 4, (April 1996), 541558. 8. Brynjolfsson, E.; Malone, T.; Gurbaxani, V.; and Kambil, A. Does information technology lead to smaller firms? Management Science, 40, 12 (December 1994), 16281644. 9. Business Week. The race for bigness. September 11,1995. 10. Chaya, A., and Mitra, S. Exploring the relationships between IT investments and organizational performance: preliminary empirical evidence. Proceedings ofthe 28th Hawaii International Conference on System Sciences, January 1996, pp. 271280. 11. Chaya, A., and Mitra, S. Empirical evidence on the relationship between information technology investments and cost of operations. Working Paper, School of Management, Georgia Institute of Technology, Atlanta, February 1996. 12. Cron, W., and Sobol, M. The relationship between computerization and performance. Information and Management, 6 (1983), 171-181. 13. Dos Santos, B.; Peffers, K.; and Mauer, D. The impact of information technology investment announcements on the market value ofthe firm. Information Systems Research, 4 1 (March 1993), 1-23. 14. Drucker, P. The coming ofthe new organization. Harvard Business Review, 66, 1 (January-February 1988), 45-53. 15. Franke, R.H. Technological revolution and productivity decline: computer introduction in the financial industry. Technological Forecasting and Social Change, 31,2 (\9S7), 143-154. 16. Hammer, M., and Champy, J. Reengineering the Corporation: A Manifesto for Business Revolution. New York: HarperCollins, 1993. 17. Harris, S.E., and Katz, J.L. Organizational performance and information technology intensity in the insurance industry. Organization Science, 2, 3 (1991), 263-295. 18. Hitt, L., and Brynjolfsson, E. The three faces of IT value: theory and evidence. Proceedings ofthe Fifteenth International Conference on Information Systems, Vancouver, British Columbia, December 1994, pp. 263-277. 19. Hirshleifer, J. Price Theory and Applications. Englewood Cliffs, NJ: Prentice-Hall, 1980. 20. Huettner, D. Plant Size, Technological Change, and Investment Requirements. New York: Praeger, 1974. 21. Keen, P. Shaping the Future: Business Design Through Information Technology. Boston: Harvard Business School Press, 1991. 22. Kvanli, A.H. Statistics: A Computer Integrated Approach. New York: West Publishing 1988.
57
23. Lev, B., and Sunder, S. Methodological issues in the use of financial ratios. Journal of Accounting and Economics, I (1979), 187210. 24. Loveman, G. An assessment of the productivity impact of information technology. Working Paper 88-054WP, Sloan School ofManagement, MIT, 1988. 25. Mahmood, M.A., and Mann, G. Measuring the organizational impact of information technology investment: an exploratory study. Journal ofManagement Information Systems, 10, 1(1993), 97-122. 26. O'Brien, J.A. Management Information Systems: Managing Information Technology in the Networked Enterprise, 3d ed. Chicago: Irwin Publishers, 1996. 27. Osterman, P. The impact of computers on the employment of clerks and managers. Industrial and Lahor Relations Review, 39 (1986), 175-186. 28. Roach, S.S. Services under siegethe restructuring imperative. Harvard Business Review, 69, 5 (September-October 1991), 82-91. 29. Smith, S.B. Computer-Based Production and Inventory Control. Englewood Cliffs, NJ: Prentice-Hall, 1989. 30. Strassman, P. The Business Value ofComputers. New Canaan, CT: Information Economics Press, 1990. 31. Weill, P. The relationship between investments in information technology and firm performance: a study of the valve manufacturing sector. Infromation Systems Research, 3, 4 (December 1992), 307-333. 32. Wilson, D. Assessing the impact of information technology on organizational performance. 5/ra/eg/c//i/brmfl;io/i Technology Management.HaxTishuTg, PA: Idea Group Publishing, 1993.

Analyzing Cost Effectiveness

Uploaded by

Copyright:

Available Formats

Analyzing Cost Effectiveness

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analyzing Cost Effectiveness

Uploaded by

Copyright:

Available Formats

Analyzing Cost-Effectiveness of Organizations: The Impact of Information Technology Spending

SABYASACHIMITRA AND ANTOINE KARIM CHAYA

THE LAST TEN YEARS HAVE WITNESSED A DRAMATIC INCREASE

SABYASACHI MITRA AND ANTOINE KARIM CHAYA