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.
SECTION 2.3 RESULTS
Table 2-1 reports the means, standard deviations and correlations among variables. Table 2-2 and Table 2-3 present the results from our replication study, WLS estimates and Driscoll-Kraay estimates for the aggregated sample and for 155 industries2 respectively, in addition to the original results reported in Deephouse (1999).
--- Insert Table 2-1 about here --- --- Insert Table 2-2 about here --- --- Insert Table 2-3 about here ---
2
Four of the one hundred and fifty night industries had fewer than forty firms, causing no estimates. I therefore did not conduct individual analyses for these four industries, so the results from only one hundred and fifty five industries are presented in Table 2-2. However, all one hundred and fifty night industries were included in the aggregation analysis.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
22
WLS estimation (exact replication)
The results of the WLS estimation are presented in Table 2-2. It depends on whether the squared term of strategic deviation is negative and significant to exam the replicative hypothesis due to I standardized all variables in the analyses. In analyzing our aggregated sample, the coefficient for the squared term of strategic deviation was negative and significant (β = -0.003, p < 0.001). In the individual analyses for each industry, the curvilinear relationship between strategic deviation and firm performance was significantly negative for only 34 out of 155 industries (21.9%).
The coefficients of squared strategic deviation for the other 121 industries showed no significant (78.1%). The results merely provided slight support for the strategic balance perspective and the replication hypothesis. Although the relationships were not significant in 121 industries, most of them were in line with the hypothesis.
I then conducted a more rigorous test, the uTest, which was developed by Lind and Mehlum (2010), to test for the presence of an inverted U shape (or U shape) by assessing whether the slope is positive at low values and negative at high values, and checking that the estimated turning point is within the data range. I found that the inverted U relationship was significant in 41 of the 155 industries3 (26.5%). The results also showed that the strategic deviation-performance relationship was inverted-U shape but not significant for 94 industries and that the relationship was insignificantly U-shape for 15 industries. For the remaining 5 industries, the turning points were out of data range, so that it is unable to conduct uTest.
3
The number of industries (41/155) presenting significantly inverted-U relationship between strategic
deviation and relative ROA is larger than the number of industries (34/155) where the coefficients of
square strategic deviation are significantly negative. It is because the uTest employs one-tailed test by
default but I conducted WLS estimation with two-tailed test. Since uTest is an advanced analysis for
testing U or inverted-U relationship, I identified the strategic deviation-performance relationship firstly
based on the results of uTest and secondly depending on the results of WLS analyses.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23
According to the results of uTest and the coefficients of both first order and second order term of strategic deviation from WLS analyses, I identified the strategic deviation-performance relationship for each industry4, also presenting in Table 2-2. In addition to 41 industries presenting inverted-U shaped relationship, five industries showed significantly positive and linear relationship between strategic deviation and relative ROA while 109 industries displayed no significant relationship. Overall, the results of our exact replication study barely support the replication hypothesis and Deephouse’s strategic balance perspective.
Driscoll-Kraay estimation (improved replication)
The results of the Driscoll-Kraay estimation are presented in Table 2-3. In the analysis of the aggregated sample, the coefficient for the squared term of the strategic deviation was negative and significant (β = -0.002, p < 0.001). In the analysis for each industry, the curvilinear relationship between strategic deviation and firm performance was significantly negative for 79 out of 155 industries (51%). The curvilinear relationship was not significant for 62 industries (40%) while it was positive and significant for the remaining 14 industries (9%), which is opposite to Deephouse’s results.
Based on uTests, I found that the inversely U-shaped relationship was significant in 69 out of 155 industries (44.5%), whereas the significantly U-shaped relationship presented in 8 industries (5.2%). The results also showed that the strategic
4
The procedure of identification included three steps. Firstly, I identified whether the strategic
deviation-performance relationship is U or inverted-U shaped based on the results of uTest. Secondly, I
identified whether the relationship is increasing/decreasing concave-up/down or positively/negatively
linear based on the coefficients of both the first order and second order term of strategic deviation from
WLS analyses. For example, I identified the relationship as increasing concave-down if the squared
strategic deviation was significantly negative and the strategic deviation was positive (even if it is not
significant). I identified the relationship as positively/negatively linear if the squared strategic deviation
was not significant and the strategic deviation was significantly positive/negative.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
24
deviation-performance relationship was inverted-U shape but not significant for 43 industries and that the relationship was insignificantly U-shape for 17 industries. For the remaining 18 industries, the turning points were out of data range, so that it is unable to conduct uTest.
The strategic deviation-performance relationship I identified also presents in Table 2-3. In addition to 69 industries presenting inverted-U shaped relationship and 8 industries presenting U shaped relationship, eight industries showed significantly positive and linear relationship and eleven industries showed significantly negative and linear relationship between strategic deviation and relative ROA while 45 industries displayed no significant relationship. Besides, there were seven industries presenting increasing concave-down relationship, five industries presenting decreasing concave-down relationship, one industry presenting increasing concave-up relationship, and one industry presenting decreasing concave-up relationship. The results of the uTests partially support the replication hypothesis, suggesting that moderate strategic similarity enhance firm performance in almost half of the manufacturing industries. It also demonstrated that divergent patterns of the strategic deviation-performance relationship presented across industries.
Comparing the results of the WLS estimations to the Driscoll-Kraay estimations, the number of industries presenting significantly negative coefficients of the squared strategic deviation increased from 34 to 79 and the number of industries presenting significantly positive coefficients of the squared strategic deviation increased from 0 to 14, whereas the number of industries with insignificant coefficients of the squared strategic deviation decreased from 121 to 62. Regarding the patterns of the strategic deviation-performance relationship (please see to comparison in Table 2-4), the numbers of industries presenting inverted-U shape increased from 41 to 69, those
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
25
presenting U shape increased from 0 to 8, those presenting positively linear increased from 5 to 8, those presenting negatively linear increased from 0 to 11, those presenting increasing concave-down shape increased from 0 to 7, those presenting decreasing concave-down shape increased from 0 to 5, and those presenting decreasing and increasing concave-up shape both increased from 0 to 1.
--- Insert Table 2-4 about here ---
The comparison revealed that the WLS estimation may underestimate the influence of strategic deviation on firm performance. The differences between the two estimations are rooted in either an unobserved individual effect or spatial dependence.
This suggests that the fixed effect model with spatial dependence correction is more appropriate for examining the strategic balance perspective (Greene, 2008; Hoechle, 2007).
By conducting a more rigorous test for an inverted U relationship, the uTest, with an improved model, the Driscoll-Kraay estimation, our replication study showed that the strategic balance perspective was supported in 69 out of 155 industries. This indicates that Deephouse’s study is found reliable, replicable in only limited contexts (less than half). The results also reveal that the patterns of the strategic deviation-performance relationship may vary depending on the industrial contexts.
Sensitivity analysis
To test the robustness of the results, I conducted several robust tests. First, I incorporated three potential common factors, firm age, firm size, and financial slack, as additional control variables. The comparison between with and without additional controls is shown in Table 2-4. For WLS estimation, it only led one industry with
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
26
positively linear relationship turned into no relationship. Regarding D&K estimation, it increased the number of industries with no relationship from 45 to 48, the number of industries with positively linear from 8 to 11, the number of industries with U relationship from 8 to 9, the number of industries with increasing concave-down from 7 to 10, whereas it decreased the number of industries with negatively linear from 11 to 8, the number of industries with inverted-U relationship from 69 to 68, the number of industries with decreasing concave-down from 5 to 1, and the number of industry both with increasing concave-up and decreasing concave-up from 1 to 0. Overall, the results only slightly changed after adding in the three additional controls, confirming the robustness of the results. Additionally, I found that firm age, firm size, and financial slack were significantly related to both strategic deviation and performance.
Given the potential bias of firm age, firm size, and financial slack, it is necessary to control for these variables when examining strategic balance perspective.
Second, I employed two modified measures of strategic similarity to test the sensitivity. One modification of strategic deviation was reconstructed by excluding inventory level and financial leverage as SD3 using absolute deviation method because there might be an optimal value of inventory level and financial leverage for firm performance. For example, logistics and supply chain discipline suggests the lower the inventory costs, the better the firm performance. Therefore, stakeholders may not expect firms to behave the same as industry average on these two dimensions.
If this logic is true, the results representing no relationship between strategic deviation and firm performance will turn into inverted-U relationship. The other modification was recalculated strategic similarity by Euclidean distance as ESD using the same strategic dimensions. Although Euclidean distance and absolute deviation both were used to measure strategic similarity, these two measures represented different mental
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
27
model of stakeholders to evaluate firms’ strategic behaviors. Employing Euclidean distance as measurement suggests that stakeholders evaluate firms’ action with a multi-dimensional space model. Whereas, using absolute deviation indicates that stakeholders evaluate firms’ strategic behaviors dimension by dimension and each dimension weighs equally. However, in the literature there is no discussion about which method better fits stakeholders’ mental model. If Euclidean distance is better (than absolute deviation), it is expected that Euclidean distance has higher explanatory power. The correlation coefficients between SD2 and SD3 and ESD are 0.82 and 0.97, respectively.
The comparison of SD3 and ESD modifications with both WLS and D&K estimations is presented in Table 2-5. Regarding the results of SD3 modification, the number of industries with no relationship for both WLS estimation and D&K estimation was increased (128 and 66, respectively), and the number of industries with inverted-U relationship was decreased (WLS estimation: 19; D&K estimation:
32). These results showed that exclusion of inventory level and financial leverage turned industry inverted-U relationship into no relationship which is contrary to the original expectation. It indicated that original measure is a better operationalization than SD3 and revealed that there are industry norms and public expectations on firms’
inventory level and financial leverage. With regard to the results of Euclidean distance (ESD) modification, the number of industries with no relationship for both WLS estimation and D&K estimation also was increased (135 and 66, respectively), and the number of industries with inverted-U relationship was decreased (WLS estimation: 14;
D&K estimation: 27). This result represented that Euclidean distance modification had no higher explanatory power than absolute deviation measure of strategic similarity. However, the results also showed that the patterns of strategic
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
28
similarity-performance relationship have a significant change, especially for negatively linearity and U shape (according to D&K estimation, the number of industries with negatively linear relationship increased from 11 to 19 and 16; the number of industries with U relationship increased from 8 to 19 and 23), suggesting that the patterns of the relationship is sensitive to the operationalization of strategic similarity.
--- Insert Table 2-5 about here ---
Third, I added several industry-level variables including market size, industry average ROA, industry average firm age, industry concentration, number of firms, ratio of foreign firms to total firms, market uncertainty, ratio of export, and ratio of new product sales as control variables in the aggregation analysis due to the aggregation sample were crossing industry in nature and industry-level variables potentially affecting the strategic similarity-performance relationship. Market size was measured as accumulated sales of all firms in a given industry. Industrial average ROA was calculated by the mean value of all firms’ ROA in a given industry. Industry average firm age was calculated by the mean value of all firms’ age in a given industry. Industry concentration was measured by Herfindahl index, which was calculated as the sum of the squared market share of firms in a given industry.
Number of firms was measures by the number of firms in a given industry, while ratio of foreign firms was computed by the number of foreign firms divided by the number of firms in a given industry. Market uncertainty was measured by the standard deviation of market size between 2004 and 2007 in a given industry. Ratio of export
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
29
was calculated by total value of export product divided by accumulated sales of all firms in a given industry, while ratio of new product sales was measured by total sales of new product divided by accumulated sales of all firms in a given industry, The coefficients of the first-order and the second-order strategic deviation were still positively significant and negatively significant, but these two coefficients both decreased from 0.036 to 0.019 and from -0.003 to -0.008. The uTest of presence of inverted-U relationship was also statistically significant (p < 0.001), suggesting the robustness of aggregation analysis.
Exploratory analysis
Since the patterns of strategic similarity-performance relationship were much diverse than expected and it should be attributed to industrial contexts, I further conducted an exploratory analysis to investigate what industry variables potentially affect the patterns of strategic similarity-performance relationship in order to provide additional information for future research. Specifically, I created a nominal variable, pattern of strategic similarity-performance relationship, based on the results of D&K estimations with additional controls in Table 2-4. After that, I conducted one-way ANOVA on industry-level to examine whether industry-level characteristics differ across patterns. The results of one-way ANOVA were presented in Table 2-6 and the descriptive statistics and correlations between industry-level characteristics and main variables were shown in Table 2-7. It shows that most industry-level variables correlate with each other, especially for ratio of foreign firms and ratio of export (r
=0.85).
--- Insert Table 2-6 about here ---
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
30
--- Insert Table 2-7 about here ---
In Table 2-6, the extent of industry concentration in industries with no relationship (0.016) was significantly larger than that in industries with inverted-U relationship (0.010) and those in industries with other patterns were also larger, though not significantly, than that in industries with inverted-U relationship. It revealed that industry concentration might be the boundary condition of strategic balance theory. When industry concentration is large enough, the competition is less severe and the benefits of differentiation may be weakened. In this situation, strategic balance on competition and legitimacy may not be the optimal strategic choice.
Market size in industries with inverted-U relationship was found significantly larger than that in industries with other patterns. This result indicated that strategic balance perspective may be limited in industries above certain level of market size. Average ROA, market growth, and ratio of new product sales were similar across patterns of relationship. Average firm age in industries with positive linear relationship was significantly larger than that in industries with negative linear relationship. However, average firm age in industries with other patterns was not significantly different from that in industries with these two patterns of relationship. Number of firms in industries with inverted-U relationship was significantly larger than that in industries with other patterns. While number of competitors and industry concentration both were proper proxy of competition in an industry, this result produced similar implication with industry concentration, suggesting strategic balance perspective only valid in industries with fierce competition.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
31
Industries with negative linear relationship was found to have larger portion of foreign firms than those with positive linear relationship, while industries with no relationship was found to have larger ratio of foreign firms than those with inverted-U relationship. This result implied that low ratio of foreign firms may be one of the boundary conditions of strategic balance perspective because foreign firms in an industry may form a different social group from local firms. Industries with inverted-U relationship and positive linear relationship faced higher market uncertainty than those with no relationship, U relationship, and increasing concave-down relationship. Industries with negative linear relationship and increasing concave-down relationship had higher ratio of export than other patterns.
Additional analysis
Although this study primarily focused on industry level analysis, firm-level variables, especially for firms’ capability to against environmental threads, may alter the strategic similarity-performance relationship. For example, successful firms may be easier to gain legitimacy than small firms. Because legitimacy contributes small firms’ survival the most, conforming to gain legitimacy is more critical for them than for large firms. Hence, it is worth to explore the role of firms’ capability in strategic balance perspective. Since firm size is a good proxy of firm capability, I separated the sample firms into small and medium enterprises (SMEs) and non-SMEs by mean firm size in a given industry. After that, I conducted D&K estimation for both SMEs and non-SMEs groups and presented the results in Table 2-5. Table 2-5 shows that overall patterns of strategic similarity-performance relationship between SMEs and non-SMEs group are similar. The results provided no support for the idea that strategic balance perspective is more valid for SMEs than large firms.