國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
12
Deephouse used two thought experiments to demonstrate why a moderate degree of strategic similarity increases performance under strong competitive and institutional forces (Deephouse, 1999, p.153–155). The first thought experiment applies the differentiation and conformity propositions to three types of firms with different level of strategic similarity: high, moderate, and low. The second applies the two propositions to a firm as its strategic similarity decreases from high to low. The results of these experiments revealed that moderate strategic similarity leads firms not only to avoid legitimacy challenges but also to reduce competition. Firms can achieve maximum performance when the gains from reduced competition are equal to the costs of legitimacy challenges. Conversely, a higher or lower degree of strategic similarity will erode performance. Thus, the replication of the strategic balance proposition is presented as follows:
Replication Proposition: A moderate degree of strategic similarity will increase performance.
SECTION 2.2 METHODS Replication procedures
To compare with Deephouse’s (1999) study in detail, I adhered to the core procedures of the original study (Tsang and Kwan, 1999). Our replication procedures were conducted as follows. First, I adopted the same measures, model specification, and analysis, but used different measures for strategic similarity, market share, and market growth. As these three variables were specific to commercial banking industry used in the Deephouse’s study, I had to adjust the measures to fit manufacturing industries. I used sales instead of deposits to calculate market share and market growth. Regarding strategic similarity, I kept the original concept of Deephouse’s calculation but adopted the measurement developed by Finkelstein and Hambrick
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
13
(1990). These two measures of strategic similarity both use multiple resource allocation strategies and the same calculation method, but Deephouse’s measure is specific to the banking industry whereas Finkelstein and Hambrick’s measure is a general measure for all industries.
Second, I used the same method as Deephouse (i.e., weighted least squares estimation) did and adjusted the regression coefficients to make the results of the two studies comparable. I divided these regression coefficients by 100 because Deephouse multiplied relative return on assets (ROA) by 100 to simplify presentation (Deephouse, 1999, p.157). Then, I further divided the regression coefficients by the standard deviation (0.007) of relative ROA due to the difference between our measure of relative ROA and his (for details, see the section on dependent variables).
Third, I conducted an improved analysis, the Driscoll-Kraay estimation, to obviate the potential bias (i.e., cross-sectional or spatial dependence) and then compared the results with those from the second step. Fourth, I conducted three robustness tests including (1) incorporating three potential common factors, (2) reconstructing two modifications of strategic similarity, (3) and adding ten industry-level variables to test the robustness of the results. Fifth, I implemented an exploratory analysis by examining the difference of industry-level variables such as industry concentration across industries with different patterns of strategic similarity-performance relationship using oneway ANOVA. Sixth, I executed an additional analysis to explore the influences of firms’ capabilities against environmental forces on strategic balance perspective by examining strategic similarity-performance relationship in two subgroup samples, small firms and large firms.
Data and sample
Whereas Deephouse (1999) focused on competition between U.S. commercial
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
14
banks in a single market from 1985 to 1992, I focused on 159 manufacturing industries (identified by three-digit Standard Industrial Classification (SIC) codes) by extracting a panel dataset from the Annual Industrial Survey Database (2003 to 2007) of the Chinese National Bureau of Statistics (CNBS). Using CNBS data to conduct replications of the Deephouse’s study simultaneously has two advantages.
First, CNBS has the most comprehensive information on manufacturing firms in China (Tian, 2007) and has adequate representation of the target population.
Manufacturing firms with sales above RMB 5 million are required to submit their basic information and financial reports to CNBS annually. Firms in CNBS account for nearly 90 percent of total industrial sales in China (of that 90%, state-owned enterprises account for 22% while the others account for 68%). Numerous replications in diverse circumstances are beneficial for verifying and falsifying theories (Tsang and Kwan, 1999), fitting the principle of replicability, and enlarging the scope of theories (Hubbard, et al., 1998). CNBS includes firms across 159 industries, allowing us to simultaneously investigate the strategic similarity–performance relationship across nearly all manufacturing industries. Second, CNBS differs from the original study’s dataset in various aspects such as country, industry, and time frame. The benefits of external validity are enhanced by conducting a replication study in a different or independent context from the original study (Rosenthal, 1991).
Although CNBS provided unique dataset for numerous replications in diverse circumstances, the difference of competition scope between commercial banks and manufacturing industries might affect the definition of competitors and peers.
Commercial banks serve local customers and pursue localized competition, whereas manufacturing firms serve national clients and compete nationwide. Due to the property of localized competition, Deephouse defined competitors as local
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
15
commercial banks in Twin Cities area in the U.S. According to the feature of nationwide competition, I defined competitors as manufacturing firms in a given industry in China in order to fit empirical context.
For our samples, I selected firms that operated in the same industry from 2003 to 2007. Based on the above criteria, our original samples included a total of 107,900 firms from 159 industries and covered 31 provinces or province-equivalent municipal cities in China. I set 2003 as the starting point of our study because CNBS adjusted the SIC coding system in 2002. The adjustment made it extremely difficult to identify whether a firm was in the same industry between 2002 and 2003. Because firms perceive each other as competitors when they operate in the same industry for a period of time, identifying a firm’s presence in the same industry is a critical criterion for our sample. Therefore, I excluded the data before 2003. Additionally, I included the data up to 2007 because most firms faced the global financial crisis in 2008. Firms chose to reduce costs against decreasing demand during the crisis period, which implies that most firms during the financial crisis adopted similar strategies, intending to respond to environmental shock rather than conform to industrial norms (Lieberman and Asaba
,
2006). As a result, the relationship between strategic deviation and firm performance may be distorted during this period and thus I tested our hypotheses for the data between 2003 and 2007.To obtain a more valid sample and rule out outliers, I followed Cai and Liu’s (2009) procedures and deleted the following observations from the original sample population: (1) firms missing any key indicators such as profits, total assets, number of employees, or sales; (2) firms with fewer than eight employees (these firms lack reliable accounting systems); (3) firms for which one of the following was true: (a) total sales were lower than sales of the primary business, (b) total assets were smaller
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
16
than net fixed assets, (c) accumulated depreciation was lower than current depreciation, (d) net fixed assets were smaller than or equal to zero, or (e) total costs were smaller than sales and general administrative expenses (SGA expenses); and (4) firms with values of key variables either above the 99.9 percentile or below the 0.1 percentile) Finally, I excluded state-owned enterprises (SOEs) from our sample because SOEs’ strategies and goals are different from those of firms in the private sector (Ju and Zhao, 2009; Tan, 2002). After the aforementioned procedure, I obtained four years of panel data with 226,946 observations from 57,644 firms across 159 industries.
Measures
Dependent variables
To be consistent with the measure used by Deephouse, I adopted the relative ROA as a performance indicator. The relative ROA indicates how well a firm performs relative to its competitors. Whereas Deephouse computed the relative ROA for each year by mean centering a bank’s ROA in a given year, I further divided it by its standard deviation; that is, standardized ROA, to compare across different industries.
Strategic deviation
I followed Deephouse’s method to measure strategic similarity using strategic deviation. The strategic deviation was the degree of deviation of a firm’s strategy from the average strategic profile of its competitors in the same industry. I used six strategic indicators for manufacturing industries (following Finkelstein and Hambrick, 1990) to create a composite measure of strategic deviation.
The six strategic indicators were: (1) advertising intensity (advertising expenses/sales), (2) research and development intensity (R&D expenses/sales), (3)
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
17
capital intensity (fixed assets/total sales), (4) non-production overheads (SGA expenses/sales), (5) inventory levels (inventories/sales), and (6) financial leverage (debt/total assets). The following equation illustrates the calculation for the strategic deviation for firm j in industry i:
Strategic Deviation
where Vaij is the value of strategic dimension a for firm j in industry i, ABS is the absolute value function, M(Vai) is the mean of strategic dimension a in industry i, and
SD(V
ai) is its standard deviation.To capture the conception of strategic similarity, the operationalization must meet the underlying assumptions of strategic balance perspective which includes (1) managers can select and implement strategies that they think will lead to higher performance; (2) firms can resist or attempt to influence CP and IP; (3) other determinants of competition, legitimacy and performance are assumed to be constant (Deephouse, 1999, p.149). Accordingly, I chose these six strategic dimensions based on the following criteria. First, top managers can control the allocation of resources to each one for increasing firm performance. Finkelstein and Hambrick (1990) employed these six strategic dimensions to examine how top management team tenure affects firms’ strategies (i.e. strategic persistence and strategic conformity). They did find top managers’ tenure increases both strategic persistence and strategic conformity, indicating that tenured top managers tend to follow industrial norm strategies and the firm’s past strategies. It suggests that top managers can decide the resources allocation on these six strategic dimensions. Second, stakeholders impose expectations and norms on firms in these six strategic dimensions and conforming helps firms to gain legitimacy (advertising intensity, capital intensity, and financial leverage: Miller, Le
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18
Breton-Miller, and Lester, 2013; R&D intensity: Chen and Miller, 2007;
non-production overheads, and inventory levels: Buchko, 2011). For example, Buchko (2011) found U.S. automobile suppliers tend to behave similarly on several strategic dimensions such as inventory levels and non-production overheads due to institutional isomorphic forces. Third, outdoing competitors on these dimensions ameliorates firms’ positions in the market (by improving their reputation, for example) so that they can better withstand CP (advertising intensity: Russo and Fouts, 1997;
R&D intensity: Baysinger and Hoskisson, 1989; capital intensity: Koch and McGrath, 1996; inventory levels: Slone, 2004; financial leverage: Vicente‐Lorente, 2001;
non-production overheads: Kotha, Rajgopal, and Rindova, 2001). For instance, Miller, Le Breton-Miller, and Lester (2013) showed that family firms could employ several of these strategic dimensions including R&D intensity, advertising intensity, capital intensity, and financial leverage not only to gain legitimacy by conforming stakeholders’ expectation, but also to obtain competitive advantage by deviation from industry norms due to pursuit of idiosyncratic family agenda. Additionally, these dimensions are complementary and each focuses on a particular aspect of a firm’s strategic profile. They are also most visible to external stakeholders, so that firms usually compete and compare with each other on these dimensions. Thus, these six strategic dimensions are especially relevant to strategic balance perspective.
As R&D expenses were not included in the CNBS database until 2005, I created strategic deviation 1 (SD1) by summing all six indicators and created strategic deviation 2 (SD2) by excluding R&D intensity (i.e., summing the other five indicators). The correlation coefficient of SD1 and SD2 was extremely high (r = 0.92).
I used SD1 and SD2 to conduct the analyses and the results were similar. For simplicity, I only presented the results of SD2.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
19
Control variables.
In accordance with Deephouse’s study, I used one-year lagged dependent variables and included market share, total expense ratio, and market growth as control variables. The lagged dependent variables were included to control for omitted variables. Market share, reflecting firms’ competitive position and operating efficiency, was measured by the ratio of a firm’s sales to total sales in the relevant subsector. The total expense ratio, reflecting cost reduction, was measured by the ratio of total expenses to total sales. Market growth, reflecting changes in the resource environment that may affect competition and performance, was operationalized by the annual percentage change in total sales in the relevant subsector.
Recent studies have suggested that firm age, firm size, and organizational slack should also be controlled when investigating the relationship between a firm’s conduct and performance (Miller et al., 2007; Miller and Chen, 1996). Firm age was calculated by the focal year minus the funding year of the firm. Firm size was measured as the natural logarithm of employees. Financial slack, a specific class of organizational slack flexibly redirecting a firm’s strategic decisions (Natividad, 2013), was measured by working capital (i.e., the difference between current assets and current liabilities).
Replication Hypothesis
Given the measures for strategic similarity and firm performance, the following hypothesis was tested:
Replication Hypothesis: The relationship between strategic deviation and relative ROA is inversely U-shaped.
Analysis
I followed the estimation procedure used by Deephouse and adopted the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
20
weighted least squares (WLS) estimation to test the replication hypothesis for each industry. Autocorrelation and heteroscedasticity is a potential concern when using lagged dependent variables and panel data (Deephouse, 1999). Deephouse used Durbin’s h-statistics to test for autocorrelation (and found it was not a problem) and chose WLS in the case of heteroscedasticity. I conducted the Wooldridge test (Wooldridge, 2002), to test first-order autocorrelation and the results revealed that there was an autocorrelation problem in the model (F = 6899, p < 0.001). I used the Sargan test and Modified Wald test (Greene, 2008) to test for heteroscedasticity, and found that heteroscedasticity was a problem in our panel data.
In addition, cross-sectional or spatial dependence neglected by Deephouse may lead to invalid statistical inferences (Driscoll and Kraay, 1998). Given that social norms and psychological behavior patterns, such as social learning and herd behavior, typically enter panel regressions as unobservable common factors, complex forms of spatial and temporal dependence may arise even under randomly sampled conditions (Hoechle, 2007). In particular, the independent variable, strategic deviation, is determined not only by the strategic position of the focal firm but also the strategic positions of the other firms in the same industry. Social learning or herd behavior across firms may affect the strategic position of the focal firm and the strategies of other firms in the same industry. This implies substantial spatial correlations in the disturbances of our panel models and Deephouse’s (1999). I therefore conducted a Pesaran cross-sectional dependence test (Hoechle, 2007) and found significant cross-sectional dependence (p<0.001). In the presence of autocorrelation, heteroscedasticity, and cross-sectional dependence, the feasible generalized least squares (FGLS) estimation and the Driscoll-Kraay estimation (a special case of the fixed effect model) were both reliable. However, I chose the Driscoll-Kraay
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
21
estimation for two reasons. First, I performed the standardized Hausman test (Wooldridge, 2002) and the result (p<0.001) suggested that the model should be estimated using a fixed effects (within) regression. Second, FGLS is typically inappropriate for large-scale microeconometric panels (Hoechle, 2007). I used STATA statistical software for all aforementioned analyses.