《Innovation, Diversity and Diffusion：A Self-Organisation Model》11 Brush-1998-Comparing Alternative Methods to Estimate Corporate and Industry Effects

KRANNERT GRADUATE SCHOOL OF MANAGEMENT Purdue university West lafayette. Indiana Comparing alternative Methods to Estimate Corporate and Industry effects b Thomas h. brush Constance R James Philip bromiley Paper No. 1118 Date: 1998 Institute for Research in the Behavioral economic and Management sciences

KRANNERT GRADUATE SCHOOL OF MANAGEMENT Purdue University West Lafayette, Indiana Comparing Alternative Methods to Estimate Corporate and Industry Effects by Thomas H. Brush Constance R. James Philip Bromiley Paper No. 1118 Date: 1998 Institute for Research in the Behavioral, Economic, and Management Sciences

Under review at Strategic Management Journal COMPARING ALTERNATIVE METHODS TO ESTIMATE CORPORATE AND INDUSTRY EFFECTS Thomas h. Brush Krannert School of Management Purdue universi West Lafayette, Indiana 47906 (765)494444 (765)494-1526fax Constance R. James Pepperdine University Malibu. Califomia 90265 (310)317-7371 (310)456-0437fax Philip Bromiley Carlson School of management University of Minnesota Minneapolis, MN 55402 (612)624-5746 (612)626-1316fax

COMPARING ALTERNATIVE METHODS TO ESTIMATE CORPORATE AND INDUSTRY EFFECTS Abstract Recent studies of the relative size of corporate and industry effects have used ANOVA, Variance Components Analysis and Simultaneous Equations(Roquebert, Phillips and Westfall 1996 McGahan and Porter, 1997a; 1997b, Brush, Bromiley and Hendrikx, forthcoming ). This paper provides a comprehensive evaluation of the advantages and disadvantages of these techniques for evaluating the relative importance of effects. Using a Monte Carlo approach, we empirically compare these techniques. Based on bias and precision of estimation, the simultaneous equation estimates and particularly standardized beta provide the best estimates of effect size

COMPARING ALTERNATIVE METHODS TO ESTIMATE CORPORATE AND INDUSTRY EFFECTS Introduction In recent years, a controversy has arisen over the relative importance of corporate, business unit or industry effects on business unit profitability (Schmalensee, 1985; Rumelt, 1991 Powell, 1996; Roquebert, Phillips Westfall, 1996; McGahan Porter, 1997a). The underlying issues center on differing approaches to explaining performance. Strategic management scholars have emphasized corporate portfolio and corporate management along with organizational capabilities at the business unit level( Chandler, 1991, 1992; Prahalad& Hamel, 1990). In contrast, work drawing most heavily from industrial organization economics and the structure-conduct-performance paradigm emphasizes the importance of industry positioning as a determinant of performance(Bain, 1951, 1956; Porter, 1980) Efforts to assess the relative importance of corporate, industry, and business effects have relied on three statistical techniques -analysis of variance(ANOVA), variance components analysis(VCA), and simultaneous equations systems. Although controlling for sample haracteristics reduces the differences in findings(Bowman Helfat, 1998), some remaining differences in findings appear to depend on the researchers choice of statistical technique (McGahan and Porter, 1997a; McGahan and Porter, 1997b; Rocquebert, et al, 1996). We use a Monte Carlo simulation to compare the three approaches and to determine which approach(es) provide the best estimates of relative importance anova and VCa have been most widely used in empirical studies assessing effect size Both techniques have problems when interpreted as measuring the relative importance of the effects of industry, corporate and business units. Brush and bromiley(1997) question the metric

of importance inherent in VCA and note other metrics have been used in social science research For example, researchers in other areas often discuss importance of a variable as the expected change in the dependent variable for a one standard deviation change in the independent variable (the standardized beta rather than VCA's explained variance. Brush and Bromiley(1997)also argue that under many circumstances, VCA lacks the power to find effects even when they are imposed to be present in the data. Another issue revolves around the interpretation of variance component effects-according to Brush and Bromiley(1997)the square root of variance components should be used when interpreting the relative importance of effects. ANOVA presents difficulties because corporate effects must be entered into the model before business unit ffects which gives an upper bound on the relative importance of corporation (Rumelt, 1991) For example, Rumelt(1991)finds a substantial corporate effect when he enters corporation before business unit. In addition to the debate over corporate, industry, and business unit effects, the underlying issue of estimating importance of an effect has wide impact in business research Brush, Bromiley and Hendrickx(forthcoming)use a simultaneous equation model to assess relative importance. They claim this method provides reliable estimates of effects and solves some of the difficulties posed by assumptions in the other estimation approaches(Brush, Bromiley and Hendrikx, forthcoming). While Brush, Bromiley and Hendrickx(forthcoming) focus on the use of continuous performance variables to estimate corporate and industry effects, they still use dummy variables to control for business unit effects We use Monte Carlo simulation to compare ANOVA, VCA and simultaneous equations techniques in their ability to estimate the relative importance of effects. The comparisons evaluate both bias and precision of the estimators

The Alternative Techniques: Variance Components analysis versus anova Two schools of thought contest the source of business unit profitability. The classical school of industrial organization economists uses the structure-conduct-performance paradigm Bain 1951,1956), while a school of revisionists argue for firm efficiency(Demsetz, 1973 Conner, 1991). The classical school argues that firms earn abnormal profits due to their industry structure and market power(Bain 1951, 1956); the revisionist school argues that efficient and well-managed firms grow to dominate industries(Wernerfelt, 1984; Rumelt, 1984, 1987) Most empirical studies testing whether industry or firm effects matter more for business unit performance use two techniques, ANOVA and variance components analysis (VCA) Several studies include both, but the authors demonstrate a preference for one method over the other by relying more heavily on one than the other in deriving conclusions. These studies are descriptive rather than positive; they seek to identify the magnitude of a particular effect rather than test an explanation for the effects. These studies often use broad measures, such as dummy variables, to capture industry and firm effects. While industry effects have been modeled consistently across the studies, differing representations of corporate and business effects have been presented Schmalensee(1985) provides the first major study of industry, corporate, and business effects, using both VCA and ANOVA, though clearly preferring vCA for determining importance of effects. He attempts to determine the relative importance of industry or firm on 1974 FTC line-of-business data for manufacturing firms. Using ANoVa to measure the significance of each effect, he finds that firm effects(the same as Rumelt's corporate effects)are insignificant and concludes that firm management does not matter. Using VCa to compare the relative importance of each effect, he finds that corporate effects do not exist, important industry

effects existed and explained 19% of the variance in rates of return, and market share effects exist but are trivial in magnitude. Schmalensee(1985: 349)says the absence of a corporate effect"merely means that knowing a firm's profitability in market a tells nothing about its likely profitability in a randomly selected market B. His study supports the classical view of industry structure, but provides no details into the characteristics behind these effects Schmalensee's use of vCA to measure importance presents at least two potential methodological problems. First, the management and economics literatures are largely silent on how to interpret VCA. Second, Schmalensee uses the variance of each component rather than the more standard practice of measuring importance with an estimated parameter Intrigued by the non-existence of a corporate effect, Kessides(1987)and Wernerfelt and Montgomery(1988)use similar methods on Schmalensee's data in follow-up studies. Their results agree with Schmalensee's on industry effects, but not on the corporate effect. By excluding corporations with fewer than three business units, Kessides(1987)finds a larger corporate effect than Schmalensee s Using Tobin s q to measure performance and adding diversification focus, Wernerfelt and Montgomery(1988)find a small corporate effect in the form of diversification focus strategies which are measured with a continuous variable Rumelt(1991)respecifies Schmalensee's(1985) model to decompose 1974-1977 FTC line-of-business profitability variance over time using both VCA and ANOVA. Rumelt (1991) adds a different specification of business unit effects(industry corporate interactions),year effects and industry-year interaction effects to Schmalensee's(1985)model. RumeIt's model allows him to identify business- units as an independent effect rather than Schmalensee's method that used market share as a proxy for business-unit effects. Both Schmalensee and rumelt use VCA and ANOVa to estimate their models, but while Schmalensee emphasizes the role of 6