Five cluster result
Of the seventeen firms in SIC 36, only one firm had a common strategy for a four year period and only two additional firms had a common strategy for as long as three years. Similar results were found for SIC 38 and SIC 73. Out of the sixty-two firms only an additional thirteen firms had consistent two year strategies. Thus, the preponderance of firms appeared to change strategy on an annual basis. Therefore it appears that most firms do not have stable strategies.
A somewhat different result was found in a recent study in the banking industry, which concluded that six years after the IPOs, the initial strategies were still in evidence (Bamford et. al, 1996). These findings were the result of a longitudinal study, in an industry which likely has had less revolutionary changes during that time period than healthcare (SIC 38) and microelectronics (SIC 36 and 73) firms have undergone due to healthcare reform and the introduction of Windows 95, respectively.
Hypothesis Three: Faster growth firms have distinctly different strategies than firms with slower growth
The third hypothesis was tested using two statistical techniques: multiple discriminant analysis and multiple regression analysis. For the discriminant analysis the data were again separated by the three SIC codes and each group was tested separately. Within each SIC category, the rate or percentage of gain (or loss) in sales for each two year time period was calculated for each firm.
The reported corporate strategies were associated not with the current year's sales gain, but with the percent gain (or loss) in the following year. This time lag was used to measure the long term affect of strategy selection. The choice of the percentage sales gain (or loss) variable is not inconsistent with the literature which suggests that the use of gain is appropriate as new ventures focus more on sales gain than profitability (Timmons, 1994).
The gain in year (t+1) could not be computed for any firm's last observation. The gains were normalized and any gain in excess of 2.5 standard deviations above the average gain was removed from all subsequent analysis. Five observations fell outside the 2.5 standard deviation limit.
The sets of firms were then separated into three sets of approximately equal numbers, representing high, medium and low gains for all firms for all years - not by the average annual gain for each firm. Since the analyses were used to discriminate between the high and low performers, the middle groups were omitted from the discriminant analysis. The resulting data set had a total of 47 observations for SIC 36, 55 observations for SIC 38 and 69 observations for SIC 73. Table VI contains the break points in gains for high and low performers.
Break points for high and low performers
It was possible for a firm to have been classified as a high performer in one year and to have been classified as a low performer in another year. For the three SIC codes together, the discriminant analyses of high and low gain data, resulted in sixty-nine percent of the data being correctly classified as a high gain or a low gain firm. The results with the highest correct classifications occurred when the data was split into the three SIC groupings and the high and low performers were analyzed. The results of the discriminant analyses are shown in Table VII, which show that firms in higher gain years have identifiably different strategies than firms in lower gain years. This result confirms the hypothesis that high growth firms have different strategies than low growth firms.
Discriminant analysis for high and low gain firms
However, this is an interesting, but somewhat weak test, because the dependent variable-high gain vs. low gain is dichotomous and the analysis was performed only on the top third and bottom third of the observations. Multiple regression is a stronger test of the relationships, because the dependent variable is continuous. Further, the results of a multiple regression analysis by SIC code would allow for the definition of the relationship between gain and the discrete strategy variables through the beta coefficient, thus linking a change in performance to discrete strategies.
The dichotomous strategy variables for the 62 firms for the three to seven years of firm data (62 firms with 317 observations less the five outliers), were used as the independent variables, and the percentage gain from the base year to the next year was used as the dependent variable. A stepwise multiple regression analysis using backward elimination was used with a one-tailed significance test of the results which are shown in Table VIII.
To determine if the year of measurement had any effect on the results, an additional regression was run. The years the data were generated, were coded as dummy variables, and regressed against the percentage gain (or loss). The year of data collection proved to be inconsequential except for SIC 7372, prepackaged software firms. In this case, 1992 proved to be significant in a bivariate regression, but in a stepwise procedure the variable did not enter the equation.
As expected, the regression analyses (Table VIII) resulted in a number of both positive and negative betas (Stearns et. al, 1995, Carter et. al., 1994). What was not expected were the negative betas on the formal planning variable. Formal planning appears to be a reaction to competitive challenges. A regression analysis of gain in period (t=0) on formal planning resulted in a negative relationship significant at the 0.004 level for SIC 73, thus supporting the contention that strategic planning may have been a reactive response to poor performance in a prior year.
Other strategy variables also had negative betas. Those strategies produced less than average performance. It could be argued that those strategies required more time to produce a positive result. It could also be argued that those strategies could have been poorly implemented or were the wrong strategies for the firms in their then current environment.
The regression of the strategy variables on gain revealed that significant relationships existed between certain strategy variables and performance. For SIC 36: Semiconductors, Related Devices, Components, cost leadership, innovation and product development strategies resulted in above average gain, while high price, joint venturing/licensing and market development strategies produced below average gains. For SIC 38: Medical Equipment, Surgical Equipment, Medical Supplies, innovation and market focus resulted in above average gain, and product differentiation lead to below average gain. For SIC Code 73: Prepackaged Software, differentiation, joint venture/licensing, and quality strategies lead to above average gain, while uniqueness and cost leadership strategies resulted in below average gains. Therefore, the third hypothesis: "Higher growth firms have distinctly different strategies than firms with slower growth" was strongly supported. Six of the strategy variables had no significant relationship to gain.
© 1997 Babson College All Rights Reserved
Last Updated 03/03/98