Life in the fast lane: Origins of competitive interaction in new vs, established markets, страница 11

We control for demand diversity because diversity of consumer demand increases available opportunities and, thus, may trigger diversity in competitive moves (Miller and Chen, 1994) while lessening the frequency of specific moves. We measured demand diversity using a Herfindahl-type index per round: 1 − ∑(P_a/P_T)², where P_a equals the number of products in segment a, and P_T equals the total number of products in all the segments in the (Sonite or Vodite) market. The measure ranges from 0 to 1 where higher values indicate greater diversity.

We control for the frequency of competitor moves. When competitors make many moves, especially relative to a focal firm, this hypercompetitive activity by rivals is likely to spur more moves by the focal firm (D'Aveni, 1994). We measured competitor moves as the number of moves made by all firms less those made by the focal firm. We compute this measure for each firm in each round and for each product market (i.e., Sonite or Vodite).

We control for the growth of the industry sector using the sector growth variable. Because slowly growing demand in a sector intensifies competition as firms increasingly compete for the same customers, we control for it. We measured industry sector growth as the percentage change in total industry revenue (sum of Sonite and Vodite markets) in each round.

We collected longitudinal data for each of the first six rounds of the Markstrat simulation. Although the simulation has seven rounds, we omitted the final round data to eliminate any possible end game actions. All statistical models consist of round-by-round panel data. To establish correct causal relationships, we use a lagged variable design. We recorded antecedents of moves (e.g., performance) at round r and predicted move frequency in round r + 1.

Statistical methods

Since our dependent variable, move frequency, is a count variable, we use Poisson regression. To account for firm heterogeneity, we use the Generalized Estimating Equations (GEE) method. The GEE method accounts for autocorrelation that arises because each firm (team) is measured repeatedly across multiple rounds of competition (Liang and Zeger, 1986; Haveman and Nonnemaker, 2000). The standard errors are derived from the Huber/White robust estimator of variance that is insensitive to the correlation structure in the GEE method.

We also predict diversity of moves, using Generalized Least Squares (GLS) regression. Since GLS models control for firm-specific variability in time series data, they do not produce biased estimates as OLS models might. Specifically, the GLS model corrects for autocorrelation and heteroscedasticity that arise in pooled time series data (Sayrs, 1989). Our data are subject to autocorrelation since each firm is measured repeatedly across multiple rounds and subject to heteroscedasticity because move diversity may increase over time as markets develop.

RESULTS

1.Top of page

2.Abstract

3.INTRODUCTION

4.THEORETICAL BACKGROUND

5.HYPOTHESES: ORIGINS OF COMPETITIVE MOVES

6.METHODS

7.RESULTS

8.DISCUSSION

9.CONCLUSION

10.Acknowledgements

11.APPENDIX

12.REFERENCES

Table 1 contains descriptive statistics and correlations. On average, firms made 2.4 competitive moves in each round in the established Sonite market and 1.3 competitive moves per round in the new Vodite market. Across both markets, firms generally made more R&D moves (1.4 Sonite and 0.8 Vodite) than market moves (0.9 Sonite and 0.5 Vodite). Market share variation is greater in the new Vodite market, consistent with work showing greater performance volatility in markets where advantages are more temporary (Thomas and D'Aveni, 2009).