# International Competition, R&D Investment Patterns, and Endogenous Sunk Costs in Canada and the United States, 1987–2002

The views and opinions expressed in the research paper are those of the author alone and do not represent, in any way, the views or opinions of the Department of Industry or of the Government of Canada.

Footnote *

## Acknowledgements

I would like to thank Marc Duhamel of Industry Canada and Mauricio Varela from North-western Kellogg for their very helpful comments.

## Introduction

The causal relationship between market structure, innovation, and competition intensity has been contested in economic literature, resulting in contradictory empirical estimates of the magnitude, and even the direction, of linkages between competition intensity and research and development (R&D) spending.Footnote 1

This paper contributes to the literature of competition intensity and innovation by using John Sutton's (1998) sunk cost theory to provide empirical evidence as to the impact on R&D investment of competition intensity increases via large scale trade liberalization for countries with heterogeneous productivity levels. Holding relative productivity differences between the two countries constant, sunk costs theory predicts that an increase in competition intensity between two countries due to the reduction in trade barriers produces the following effects (in an R&D intensive industry): at first, R&D expenditures in both countries escalate, but eventually the country with higher productivity raises its R&D intensity relative to the second country, for a given industry. This results in firms in the country with lower productivity either dropping out of the industry entirely, or remaining in the industry and concentrating on producing goods that are less R&D intensive and that incorporate less value added. Using a difference-in-difference econometric specification and industry level data for the 1987–2002 period, this study finds empirical evidence of these effects in the period following the North American Free Trade Agreement (NAFTA) between Canada and the United States (US). Although R&D investment increased for both countries in absolute terms, Canadian industries' R&D intensity declined relative to their United States competitors.

The paper proceeds as follows. The first section describes the theory of sunk costs and presents its predictions regarding the impact of an increase in competition intensity on innovation. The second section discusses the econometric model employed and the data used in the estimation process. The third section then presents the results, and the fourth section concludes.

### 1.1 Competition Intensity and Innovation

The literature of competition intensity and innovation is fraught with contradictory theoretical hypotheses, and inconsistent empirical evidence. Schumpeter (1934) states that monopoly rents create an incentive for firms to invest in innovation, thereby producing a negative relationship between the intensity of market competition and firm innovation. Other models predict the opposite.Footnote 2 Yet another set of models finds an unstable "inverted U" relationship between competition intensity and innovation.Footnote 3 One of the main reasons for these disparities is that most of the literature fails to account for important methodological issues such as the measurement of competition,Footnote 4 or for the inherent endogeneity between market structure, competition intensity and innovation.

Concerns about these issues have led to the development of theories that offer less specific, though more certain, predictions. One such relatively new approach is the game theoretic model of endogenous sunk costs, as developed in Sutton (1991, 1998), which largely avoids causal econometric estimation in favour of robust reduced form relationships that describe the evolution of R&D spending in conjunction with competition intensity and market structure. Sunk cost theory has been proven to be robust for a number of different market types in Sutton (1998), Lyons, Matraves, and Moffat (2001), and Symeonidis (2002).

### 1.2 Sunk Cost Theory Framework

This study adopts the theoretical framework of sunk costs from Sutton (1991, 1998). Sunk cost theory rests on the concept of an equilibrium configuration in a multi-stage oligopoly game played in quantities (a la Cournot) after firms pay an initial sunk cost to enter the market.Footnote 5 The equilibrium profits of any firm present in the final stage of the game can be defined as a linear function of market size, and as some function of the number of firms in the market and of the degree of competition intensity. Thus, the equilibrium configuration can be expressed as a condition where the profits of firms that have paid the sunk cost to enter the market must be non-negative and cover their sunk cost of entry, but not so large as to generate new entry in equilibrium.

### 1.3 Endogenous and Exogenous Sunk Cost Industries:

Sutton (1991) broadly divides industries into "exogenous" and "endogenous" sunk cost industries. Exogenous sunk cost industries have structural characteristics which do not allow for a firm to increase its demand by spending more on demand shifters such as R&D or advertising. Since such an industry does not have significant entry barriers (besides the initial fixed sunk cost), as the market becomes larger the number of firms it can sustain also increases as profits increase proportionally. A strictly negative relationship therefore exists between concentration and market size, and the concentration tends towards zero as market size tends towards infinity.

In endogenous sunk cost industries, however, firms can invest in producing higher value added products (through R&D), or in advertising, and shift their demand outward, increasing their market share.Footnote 6 When market size is relatively small, the benefits to investing in demand shifters are also small, and an endogenous sunk cost industry behaves in the same way as an exogenous sunk cost industry in equilibrium. However, as market size passes a certain threshold, the rewards for increasing market shares grow much higher than the costs of investment and firms compete in demand-shifter spending prior to the final stage game, creating an additional entry barrier (besides the initial sunk cost investment) which, as Sutton (1998) proves, produces a constant minimum level of equilibrium concentration in the market regardless of its size (Figure 1).

### Figure 1: Lower Bounds in the Presence of Endogenous Sunk Costs

Description of Figure 1

Figure 1 compares the relationships between firm concentration (vertical axis) and market size (horizontal axis) for an exogenous sunk cost industry and for an endogenous sunk cost industry. For an exogenous sunk cost industry, a downward sloping curve shows the strictly negative relationship that exists between its concentration and market size, and the concentration tends towards zero as market size tends towards infinity. For an endogenous sunk cost industry, a straight line beginning just to the right, and from about midway down, of the negative sloping curve for the exogenous sunk cost industry, and running parallel to the horizontal axis, shows that as the endogenous sunk cost industry’s market size passes a certain threshold, the rewards for increasing market shares grow much higher than the costs of investment and firms compete in demand-shifter spending prior to the final stage game. This creates an additional entry barrier (besides the initial sunk cost investment) which produces a constant minimum level of equilibrium concentration in the market regardless of its size.

### 1.4 Impact of an Increase in Competition Intensity on R&D Spending: Exogenous Case

The basic case of firms in industries where R&D expenditures do not play a large role as demand curve shifters (that is, in non-technologically intensive industries) is the following. The intensification of competition via trade creates two contradictory effects that operate simultaneously: (a) the market size effect, whereby firm access to larger markets and greater potential sales creates new incentives to innovate and increase R&D spending;Footnote 7 and (b) the competition effect, whereby an increase in competition intensity squeezes profit margins, making firms less able to engage in innovation and reduces R&D spending. The predicted outcome is ambiguous since for some industries the market size effect will far outweigh the competition effect (possibly due to easier commercialization or better differentiation for their product), while for others, the situation would be opposite.

### 1.5 Impact of an Increase in Competition Intensity on R&D Spending: Endogenous Case

The market size and competition effects still apply in endogenous sunk cost industries, but there is also the added impact of R&D competition. Following an increase in trade, products become closer substitutes across countries, intensifying competition in the joint product market as firms in the industry now compete directly against both domestic and international rivals. As Sutton (1998) predicts, this also intensifies competition of spending on demand shifters as the incentives to gain increased shares of a much larger market are greater, and firms escalate R&D spending and increase the entry costs in the market.

Some firms are inevitably unable to maintain the escalation process for a long time (due to financial constraints, or to a lower efficiency of production), and proceed to reduce their R&D expenditures relative to their more efficient competitors.Footnote 8 On an international scale, if a majority of firms in a given industry in one country are better able to sustain the escalation due to some structural asymmetries such as higher productivity, the escalation and contraction process will be inherently unbalanced. Namely, firms in the more productive country, who continue to escalate their spending, would increase their R&D expenditures relative to firms in the less productive country. As a result, firms in the less productive country, which are unable to pay the sunk costs required to stay in the industry, would either drop out of the market entirely or move to producing goods that require less R&D and embody less value added.Footnote 9

It should be noted that the contraction process does not necessarily leave firms in the less productive country with a lower innovation intensity compared to the start of the process; theory only predicts that the paths of high-productivity and low-productivity countries diverge following the contraction process, expanding the relative innovation gap between them (See Figure 2). Similarly, theory does not require the low-productivity country to initially have lower R&D intensity than the high productivity country;Footnote 10 the only assured outcome is that their paths diverge following the competition increase. Lastly, these theoretical predictions are based on the assumption that productivity differences between the countries are constant over the period examined – that is, it only takes into account the demand impact of R&D spending and innovation.

### Figure 2: Escalation and Contraction Stages Following an Increase in Competition Intensity

Description of Figure 2

Figure 2 illustrates a predicted relative innovation gap between high productivity and low-productivity countries following an increase in competition intensity.  R&D/Sales (or R&D investment) is on the vertical axis, and competition increase (over time) is on the horizontal axis. The horizontal axis is split into three parts, an unnamed section on the left, and to its immediate right, an escalation stage and then a contraction stage, for innovation/competitive intensity.

The high productivity and low productivity countries follow the same path along the first section and the escalation stage, as represented by a horizontal straight line beginning just over midway up the vertical axis and proceeding parallel to the competition increase axis where the escalation stage begins, then proceeding at a positive slope upwards to the beginning of the contraction stage. Then, the countries' paths diverge, indicating the expansion of the relative innovation gap between them. The high productivity country path is a fairly positive slope along the contraction stage, while the two potential paths of the low-productivity country are a slightly negative slope and a more pronounced negative slope, along the contraction stage.

### 2.1 Empirical Model

It is difficult to empirically measure sunk entry costs and model the equilibrium configuration of firms in the market. Bresnahan and Reiss (1991) do so for a number of professional industries, although they use hand-collected data and are forced to limit their investigation to small and isolated geographic markets. In Sutton (1991, 1998), empirics were done via illustrative individual case studies showing the impact of an external competition intensity shock on a given industry, either globally or locally.

The empirical analysis in this paper is modeled after Symeonidis (2002), the only known large scale econometric study using this theory. Symeonidis examines the impact of ending collusive practices in Britain in the 1950s on market structure and innovation.Footnote 11 In this study, we use the North American Free Trade Agreement between Canada and the US, a wholesale reduction in trade barriers in a large number of industries, as an external policy shock allowing us to measure the impact of a change in competition intensity on R&D spending intensity.

Using standard difference-in-difference methodology, we take the difference in R&D intensity between Canadian and US industries before and after NAFTA, as well as the total difference in R&D intensity over time to obtain a measure of the impact of NAFTA on R&D intensity in Canadian industries (the experimental group) relative to the US industries (the control group). We run two regressions – one for a group of R&D intensive industries and one for a group of non-R&D intensive industriesFootnote 12 – since the characteristics governing R&D spending outcomes for endogenous and exogenous sunk cost industries are markedly different (see Sections 1.3 and 1.4 for more details).

The estimating equation (for industry i, at time t, and in country c) is, therefore:

${\mathrm{RD}}_{\mathrm{tc}}^{i}$ = β0 + β1 ${\mathrm{lnMES}}_{\mathrm{tc}}^{i}$ + β2 ${\mathrm{CANADA}}_{c}^{i}$ + β3 ${\mathrm{NAFTA}}_{t}^{i}$ + β4 CANADA*${\mathrm{NAFTA}}_{\mathrm{tc}}^{i}$+ Di+ ∂itc

Where ${\mathrm{RD}}_{\mathrm{tc}}^{i}$ is R&D intensity,Footnote 13 ${\mathrm{CANADA}}_{c}^{i}$ is a country dummy (which is 1 for Canadian industries), ${\mathrm{NAFTA}}_{t}^{i}$ is a dummy for whether the industry is in the period before or after NAFTA, Footnote 14 CANADA*${\mathrm{NAFTA}}_{\mathrm{tc}}^{i}$ is the interaction term between the two, ${D}^{i}$ is a vector of industry fixed effects, and ∂ is the residual term, which in particular may reflect measurement error across the sample.Footnote 15

${\mathrm{InMES}}_{\mathrm{tc}}^{i}$ is the logarithm of median firm sales divided by total sales in the market. This variable controls for the relative size of Canadian and US firms in a given industry. It is an important control, since there are heterogeneities in firm and market size which should play an important role in the R&D spending escalation and contraction process following NAFTA.Footnote 16

β4 is the coefficient which measures the 'difference in differences' of R&D intensity for the experimental group of Canadian industries relative to their United States counterparts after the overall effects of NAFTA have been accounted for. As such, it is the variable of interest here. Since numerous studies show that Canadian industries lag their US counterparts in productivity and productivity growth,Footnote 17 we expect this coefficient to be negative for endogenous sunk cost industries.

### 2.2 Data Sources and the Sample of Industries

This study uses a sample of R&D intensive industries and a sample of non R&D intensive industries in Canada and the US to create two unbalanced panels covering the 1987–2002 period. By choosing this time-span, we are able to compare between the 1987–1994 and 1995–2002 sub-periods, which gives us ample sample space to estimate differences in R&D spending trends before and after NAFTA.

Following the selection process of Symeonidis (2002), seven industries with an R&D spending to sales ratio greater than 1% and with the highest number of observations for Canada and the US were selected as the endogenous sunk cost group for this study.Footnote 18 Seven other industries where R&D spending as a share of firms' sales rarely moved above the 1% threshold were selected to be the exogenous sunk cost group.

#### Industry Sample

R&D Intensive Industries
Industry NAICS
Fibers Manufacturing 3252
Pharmaceutical Manufacturing 3254
Computer Equipment Manufacturing 3341
Communications Equipment Manufacturing 3342
Semiconductor and Other Electronic Components Manufacturing 3344
Navigation, Measuring and Electromedical Instrument Manufacturing 3345
Aerospace Product and Parts Manufacturing 3364
Non R&D Intensive Industries
Industry NAICS
Textile Product Mills Manufacturing 314
Paper Manufacturing 322
Petroleum and Coal Products Manufacturing 324
Non-Metallic Mineral Manufacturing 327
Iron and Steel Mills and Alloys Manufacturing 3311
Alumina and Aluminum Production and Processing 3313
Forging and Stamping Industry 3321

Due to data limitations, it was only possible to obtain complete data for R&D intensive industries at the 4 digit NAICS level for Canada and the US. Data was even more limited for the non-R&D intensive group, resulting in the use of a combination of 3 and 4 digit industries in the sample. This is a much higher aggregation than the one Sutton (1998) uses – a 5 digit SIC industry – although Symeonidis (2002) successfully uses similarly aggregated three and four digit SIC industries.

Three primary data sets are used for data on sales and R&D expenditures in industries: The annual Research and Development in Canadian Industry survey from Statistics Canada, the Survey of Industrial Research and Development from the National Science Foundation and the US Census Bureau, and the OECD's Structural Analysis STAN database. Additional data, such as the median firm size relative to total industry sales came from the Bureau of Economic Analysis in the US and from Statistics Canada.Footnote 19

### 3.1 Descriptive Statistics

R&D Intensity MES (Median Firm/Mkt Size) R&D Intensive Industries Non-R&D Intensive Industries 7.61 0.58 3.E−03 3.E−03 4.06 0.39 1.E−02 1.E−02 8.51 0.47 3.E−04 6.E−03 5.38 0.24 7.E−03 1.E−03 8.06 0.53 2.E−03 2.E−03 4.78 0.33 7.E−02 7.E−02 223 218 224 224

Aggregate summary statistics suggest that R&D intensity has increased for R&D intensive industries between the two periods, thereby tentatively confirming the sunk cost theory predictions of an overall escalation in R&D spending after an increase in competition intensity. Conversely, for non-R&D intensive industries, R&D intensity declined between the two periods. As well, it appears that the MES for both R&D intensive and non-R&D intensive industries declined between the two periods, although this decline has been more pronounced for R&D intensive industries. However, not knowing more about the distribution of firm size, it is difficult to make conclusive remarks about the impact of NAFTA on the consolidation of firms in North America.

Canada United States R&D Intensity MES (Median Firm/Mkt Size) R&D Intensity MES (Median Firm/Mkt Size) R&D Intensive Industries Non-R&D Intensive Industries 8.00 0.47 6.E−03 6.E−03 7.23 0.68 8.E−05 6.E−04 4.64 0.30 1.E−02 1.E−02 3.41 0.44 2.E−04 1.E−03 8.52 0.39 6.E−04 1.E−03 8.51 0.55 5.E−05 2.E−04 6.12 0.20 1.E−03 1.E−03 4.57 0.25 9.E−05 3.E−04 8.26 0.43 3.E−03 3.E−03 7.87 0.61 7.E−05 4.E−04 5.41 0.26 1.E−02 1.E−02 4.06 0.37 1.E−04 1.E−03 111 106 112 112 112 112 112 112

When separated by country, summary statistics show that Canadian industries' R&D intensity growth is 10% lower than their US counterparts, perhaps suggesting that less productive Canadian firms are unable to maintain the spending escalation at the United States rate.Footnote 20 It also appears that the MES in Canada is larger than that of the United States although this likely only reflects Canada's smaller market size.

An issue which comes out of the descriptive statistics is that Canada appears to have higher R&D intensity than the United States for the group of R&D intensive industries, which runs counter to the common perception (as confirmed by aggregate statistics) that Canadian firms have lower R&D intensity than United States firms. This may indicate problems with the data, or that a sample selection bias exists in the analysis. However, both of these are unlikely. Previous studies, such as ab Iorweth (2005), also find that Canada has a higher R&D intensity than the United States for multiple R&D intensive industries, but cannot explain this or find any evidence that these industries are markedly different than other similarly structured Canadian industries. Additionally, as stated before, sunk cost theory does not require that industries in low productivity countries should have lower R&D spending than industries in high-productivity countries. In fact, the absolute values are irrelevant, since the predictions center on relative trends.

### 3.2 Regression Results

Regression Results
R&D Intensive Industries
Coeff
(std. err.)
Non-R&D Intensive Industries
Coeff
(std. err.)
ln(MES) −1.60Footnote **
(0.49)
−0.07Footnote **
(0.02)
(2.28)
0.04
(0.09)
NAFTA 1.79Footnote *
(0.73)
−0.07
(0.45)
(1.20)
−0.07
(0.07)
R2 0.396 0.4698
N 223 218

Regression results largely confirm the theoretical predictions and observations from the descriptive statistics. For the regression of non-R&D intensive industries, the CANADA*NAFTA dummy is slightly negative and not statistically significant. As well, from the NAFTA dummy it seems that R&D intensity on the whole has fallen in both countries in the period following NAFTA. This is consistent with theory, since an escalation in R&D spending is not profitable in industries where it has a negligible impact on consumers' willingness to pay.

The dummy representing the overall change over time is positive and highly statistically significant for R&D intensive industries, showing that when focusing on all industries in both countries, R&D intensity increased on average in the period after NAFTA. This is consistent with the theoretical predictions of escalation. Observing the Canada-US differences across industries, the dummy representing the change of Canadian industries relative to US industries over time (CANADA*NAFTA) is negative, and statistically significant at the 1% confidence level. In fact, taking the total derivative of R&D intensity with respect to NAFTA shows that Canadian industries experienced an average net loss of 1.1% in R&D intensity, while US firms experienced an average net gain of 1.8%. Consistently with theory, it appears that despite an absolute increase in R&D investment, Canadian R&D intensive industries lost some ground compared to similar US industries in the period after NAFTA.

## 4. Conclusions

Using Canadian and US industry level data for 1987–2002, the findings of this paper provide some systematic empirical evidence on the impact of an increase in competition intensity via globalization on innovation intensity in countries with heterogeneous productivity levels (holding relative productivity levels constant). The evidence is consistent with the theoretical model of endogenous sunk costs and suggests that an exogenous increase in competition in the post NAFTA period produced an absolute increase in R&D investment in R&D intensive industries in Canada and the US. However, the less productive Canadian industries fell relative to their US counterparts in R&D intensity levels.

These results potentially suggest some answers to the generally accepted but unexplained stylized fact, that despite Canada's extensive R&D support programsFootnote 21, the R&D intensity gap between Canada and the United States has persisted over the last twenty years. These results are also consistent with recent studies which find Canadian firms to be less innovative in their strategic outlook compared to their US counterparts.

Of course, it should be noted that this study is a simple difference-in-difference comparison between two periods; as such, it does not present evidence of a definitive causal link between NAFTA and innovation investment patterns, but merely a suggestion regarding the form such a link might take. In addition, the economic impact of trade on firms varies depending on factors such as their products, location, and size (Melitz 2003). This study uses industry level data, and as such it fails to incorporate firm heterogeneity into the analysis – particularly important since evidence suggests that large firms (500+ employees) explain much of the R&D investment gap between Canada and the US (Songsakul, Lau, and Boothby 2008). Future research should use firm level data in examining changes in the dynamic investment process of firms following an increase in competition intensity. Sutton's theory suggests that firms which lag in R&D investment in R&D intensive markets are forced to either exit the markets entirely, or to focus on producing lower value added goods which require less technological inputs. It is unclear whether this occurred in the period following NAFTA, and it could be examined further using appropriate data.

This paper also assumes that productivity differences between the two countries are constant over the period examined – it does not take into account the supply side impact of investment in innovation. Expanding the study to incorporate the possibility that investing in innovation does not only expand demand but also improves efficiency could produce further insights into the interaction between innovation, competition and market structure following globalization.

## Bibliography

• Acemoglu, D. and J. Linn (2004) "Market Size in Innovation: Theory and Evidence from the Pharmaceutical Industry," The Quarterly Journal of Economics, MIT Press, vol. 119(3), pp. 1049-1090.
• ab Iorweth, A. (2005) "Canada's Low Business R&D Intensity: the Role of Industry Composition," Working Papers-Department of Finance Canada 2005-03, Department of Finance Canada.
• Aghion P, N. Bloom, R. Blundell, R. Griffith and P. Howitt (2005) "Competition and Innovation: An Inverted-U Relationship," The Quarterly Journal of Economics, MIT Press, vol. 120(2), pp. 701-728.
• Boone, J. (2000) "Competitive Pressure: The Effects on Investments in Product and Process Innovation," RAND Journal of Economics, vol. 31(3), pp. 549-569.
• Bresnahan, T. and Peter C. Reiss (1991) "Entry and Competition in Concentrated Markets," The Journal of Political Economy, vol. 99(5), pp. 997-1009.
• Chen, Z. (2006) "Rivalry, Market Structure and Industrial Competitiveness: Issues and Evidence," Industry Canada Mimeograph.
• Demsetz, H. (1973), "Industry Structure, Market Rivalry and Public Policy," Journal of Law and Economics, vol. 16(1), pp. 1-9.
• Griliches, Z. (1957) "Hybrid Corn: An Exploration in the Economics of Technological Change", Econometrica, 25, pp. 501-522.
• Kamien, L. and N. L. Schwartz (1982) Market Structure and Innovation. Cambridge: Cambridge University Press.
• Lyons, B.R, Catherine Matraves, and Peter Moffat (2001) "Industrial Concentration and Market Integration in the European Union," Economica, vol. 68(269), pp. 1-26.
• Melitz, M.J. (2003) "The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity, " Econometrica, vol. 71 (6), pp. 1695-1725.
• National Science Foundation, Division of Science Resources Studies (1999) Research and Development in Industry: 1995–96, NSF 99-312.
• National Science Foundation, Division of Science Resources Statistics (2006)
Research and Development in Industry: 2003. NSF 07-314.
• OECD (2003) OECD Economic Outlook, Volume 2003, Issue 1, Paris: OECD.
• OECD (2006) OECD Science, Technology, and Industry Outlook, Paris: OECD.
• OECD (2008) Main Science and Technology Indicators, Volume 2008, Release 2, Paris: OECD.
• Rao, S. and Andrew Sharpe (2002) eds., Productivity Issues in Canada, Calgary:University of Calgary Press.
• Rao, S. Jianmin Tang, and Weimin Wang (2004) "Measuring the Canada-US Productivity Gap: Industry Dimensions" International Productivity Monitor, 9, pp. 3-14.
• Scherer, F.M. (1967) "Market Structure and the Employment of Scientists and Engineers." American Economic Review, vol. 57(3), pp. 524–531.
• Schumpeter, J. A. (1934) The Theory of Economic Development. New Brunswick, NJ: Transaction Publishers.
• Shaked, A. and John Sutton (1987) "Product Differentiation and Industrial Structure." Journal of Industrial Economics, vol. 36(2), pp. 131-146.
• Songsakul, Thitima, Bernice Lau and Daniel Boothby (2008) "Firm Size and Research and Development Expenditures: A Canada-U.S. Comparison". Mimeo, Industry Canada
• Statistics Canada (2000) Industrial Research and Development: 1999 Intentions, Catalogue no. 88-202-XIB.
• Statistics Canada (2006) Industrial Research and Development: 2005 Intentions, Catalogue no. 88-202-XIE.
• Sutton, J. (1991) Sunk Costs and Market Structure, Cambridge, Mass: MIT Press.
• Sutton, J. (1998) Technology and Market Structure, Cambridge, Mass: MIT Press.
• Symeonidis, G. (2002) The Effects of Competition: Cartel Policy and the Evolution of Strategy and Structure in British Industry, Cambridge, Mass: MIT Press.
Date modified: