DataPDF Available

Abstract

Data appendix to the article "Managerial attention to exploitation versus exploration"
1
Appendix to:
Managerial attention to exploitation versus exploration:
Toward a dynamic perspective on ambidexterity
This appendix reports the geographical distribution of the sample and several robustness tests
conducted to assess the reliability of the results in “Managerial attention to exploitation versus
exploration: toward a dynamic perspective on ambidexterity.” (This appendix is made available
to the reviewers. In case the paper is published, the document will be available from the authors
as well as made accessible via a public website.)
Geographical distribution of the final sample
Country
Number of firms
United States of America
66
Germany
7
France
3
Great Britain
3
The Netherlands
3
Norway
1
Sweden
1
Spain
1
Swiss
1
SUM
86
The length of the letters to shareholders
The length of the letters can possibly influence the independent variable; for example, shorter
letters can be expected to exhibit more extreme exploitation-exploration (EE) ratios. That is,
finding one additional keyword in a shorter text, that is likely to contain a relatively small
number of keywords, may have a greater influence on the EE ratio than finding one additional
keyword in a longer text. To test for this possible confounding effect, we took the absolute value
of 0.5 (the mean of EE ratio scale), subtracted the EE ratio and correlated the result with the
number of characters appearing in a letter. This procedure effectively tests whether fewer
characters result in more extreme EE ratios. This robustness check produced a non-significant
2
relationship (r = 0.010, p > .1); therefore, the length of the letters in our sample appears to have
no significant effect on the EE ratio distilled.
Robustness tests with respect to the analysis
We also assessed the robustness of the key findings. More specifically, we tested the model:
(a) by varying the number of instruments (Tables I and II);
(b) with a two and six quarters delay in how managerial attention affects firm performance
(rather than four quarters) (Table III);
(c) with return on assets as dependent variable (rather than net profit) (Table IV);
(d) with a quarterly varying measure of the EE ratio (i.e., a moving average over 3
observations) (Table V);
(e) by means of competing pooled OLS and random-effects GLS regression (allowing for
time-invariant variables) (Table VI).
Overall, these tests serve to demonstrate sufficient levels of robustness of our main findings. The
outcomes of these tests are reported in the remainder of this appendix.
3
a. Varying the number of instruments
Tables I and II denote the results of these robustness tests. All tests conducted confirm the
robustness of the results.
Table I.
Robustness tests with respect to the number of lags (instrument matrix) (Part 1)
Dependent variable:
Profit margin
1–1 lag (min)
1–2 lags
1–4 lags
Coeff.
(SE) b
Coeff.
(SE) b
Coeff.
(SE) b
b1 – EE ratio t 4
3.36
(1.63)*
3.67
(1.50)**
3.36
(1.35)**
b2 – (EE ratio)2 t 4
–2.28
(1.17)*
–2.51
(1.08)*
–2.31
(.96)**
b3 – Recovery dummy
1.35
(.70)*
1.46
(.65)*
1.10
(.51)*
b4 – EE ratio t 4 × Recovery dummy
–4.06
(2.29)*
–4.39
(2.09)*
–4.09
(1.84)*
b5 – (EE ratio)2 t 4 × Recovery dummy
2.81
(1.76)+
3.04
(1.61)*
2.80
(1.44)*
b6 – Profit margin t 1 a
.16
(.08)*
.15
(.09)*
.18
(.09)*
b7 – Net income a
.70
(.24)**
.70
(.24)**
.67
(.22)**
b8 – Total assets a
–.55
(.34)+
–.55
(.31)*
–.46
(.25)*
b9 – R&D–Sales ratio a
–.01
(.10)
.01
(.10)
.01
(.08)
b10 – R&D missing dummy
.16
(.27)
.17
(.24)
.33
(.20)+
b11 – Firm size a
.04
(.27)
.04
(.24)
–.03
(.19)
b12 – Firm age a
–.14
(.20)
–.18
(.13)+
–.16
(.09)*
b13 – GICS 4510 dummy
.14
(.16)
.16
(.15)
.11
(.14)
b14 – GICS 4520 dummy
.14
(.15)
.13
(.13)
.13
(.13)
b15US location dummy
–.09
(.15)
–.11
(.14)
–.09
(.12)
b16 – Constant
–1.14
(.52)*
–1.23
(.50)**
–1.13
(.47)**
Hansen test of over-identification
1.00
1.00
1.00
Arellano Bond AR(1) c
–2.87
**
–2.91
**
–2.94
**
Arellano Bond AR(2) c
–.03
–.05
.08
a Standardized value; b The standard errors are robust to heteroskedasticity and arbitrary patterns of autocorrelation within agents
(Roodman 2009a); c z values reported; + p < .10; * p < .05; ** p < .01; *** p < .001. Time dummy variables were included in all
models but are omitted from these results. Onetailed significance levels are reported.
4
Table II.
Robustness tests with respect to the number of lags (instrument matrix) (Part 2)
1–16 lags
1–20 lags (max)
All endogenous
Coeff.
(SE) b
Coeff.
(SE) b
Coeff.
(SE) b
3.16
(1.27)**
3.12
(1.26)**
3.35
(1.43)**
–2.18
(.91)**
–2.16
(.90)**
–2.28
(1.01)*
1.44
(.55)**
1.14
(.51)*
1.27
(.53)**
–3.62
(1.67)*
–3.52
(1.66)*
–3.86
(1.80)*
2.52
(1.31)*
2.46
(1.31)*
2.70
(1.38)*
.21
(.09)**
.21
(.09)**
–.22
(.09)**
.60
(.19)**
.60
(.19)**
.57
(.19)**
–.39
(.19)*
–.38
(.19)*
–.36
(.18)*
–.02
(.07)
–.02
(.08)
–.02
(.07)
.02
(.10)
.01
(.10)
.06
(.08)
–.04
(.13)
–.04
(.12)
–.03
(.12)
–.11
(.05)*
–.10
(.05)*
–.08
(.05)+
0.17
(.12)+
.17
(.12)+
.14
(.11)+
0.13
(.11)
.13
(.11)
.12
(.11)
–.08
(.10)
–.08
(.10)
–.05
(.10)
–1.32
(.47)**
–1.03
(.44)*
–1.42
(.54)**
1.00
1.00
–3.09
**
–3.10
**
–3.15
**
.26
.28
.33
a Standardized value; b The standard errors are robust to heteroskedasticity and arbitrary patterns of autocorrelation within agents
(Roodman 2009a); c z values reported; + p < .10; * p < .05; ** p < .01; *** p < .001. Time dummy variables were included in all
models but are omitted from these results. One-tailed significance levels are reported.
5
b. Varying the delay in the effect of managerial attention on firm performance
In the paper, we assumed the effect of senior management’s attention to exploitation-exploration
on the profit margin involves a delay of four quarters. To test for robustness, we also ran the
model with delays of two quarters and six quarters. The results reported in table III are similar,
especially with a delay of two quarters. This also supports our key assumption that, in the IT
industry, performance effects of different EE ratios arise over a relatively short time span (e.g.,
compared to the electronics or pharmaceutical industry).
Table III.
Robustness tests with respect to lagging the EE ratio (8/20 lags used)
Dependent variable:
Profit margin
EE ratio lagged 2
quarters
EE ratio lagged 4
quarters (presented)
EE ratio lagged 6
quarters
Coeff.
(SE) b
Coeff.
(SE) b
Coeff.
(SE) b
b1 – EE ratio t x
2.36
(1.19)*
3.39
(1.33)**
.84
(.75)
b2 – (EE ratio)2 t x
–1.57
(.87)*
–2.35
(.95)**
–.59
(.59)
b3 – Recovery dummy
.63
(.51)
1.56
(.61)**
.39
(.55)
b4 – EE ratio t x × Recovery dummy
–2.46
(1.55)+
–3.93
(1.82)*
–.92
(1.72)
b5 – (EE ratio)2 t x × Recovery dummy
1.51
(1.19)
2.72
(1.41)*
.95
(1.34)
b6 – Profit margin t 1 a
.20
(.09)*
.20
(.09)*
.20
(.09)*
b7 – Net income a
.62
(.20)**
.62
(.20)**
.60
(.20)**
b8 – Total assets a
–.39
(.21)*
–.40
(.22)*
–.39
(.21)*
b9 – R&D–Sales ratio a
–.01
(.07)
–.00
(.07)
–.03
(.07)
b10 – R&D missing dummy
.15
(.14)
.17
(.17)
.16
(.13)
b11 – Firm size a
–.04
(.15)
–.04
(.16)
–.03
(.15)
b12 – Firm age a
–.13
(.06)*
–.14
(.07)*
–.12
(.06)*
b13 – GICS 4510 dummy
.14
(.13)
.15
(.13)
.10
(.12)
b14 – GICS 4520 dummy
.10
(.12)
.12
(.12)
.08
(.11)
b15US location dummy
–.09
(.11)
–.09
(.11)
–.07
(.11)
b16 – Constant
–.77
(.41)*
–1.39
(.49)**
–.50
(.28)*
Hansen test of over-identification
1.00
1.00
1.00
Arellano Bond AR(1) c
–2.90
**
–3.03
**
–2.90
**
Arellano Bond AR(2) c
.43
.18
.42
a Standardized value; b The standard errors are robust to heteroskedasticity and arbitrary patterns of autocorrelation within agents
(Roodman 2009a); c z values reported; + p < .10; * p < .05; ** p < .01; *** p < .001. Time dummy variables were included in all
models, but are omitted from these results. One-tailed significance levels are reported.
6
c. ROA and Tobin’s Q as dependent variable
The model was also tested with return on assets (ROA). Table IV shows the results, which are
highly similar to the main model.
Table IV.
Robustness tests regarding the dependent variable (8/20 lags used)
Dependent variable:
Profit margin
(presented)
ROA
Coeff.
(SE) b
Coeff.
(SE) b
b1 – EE ratio t x
3.39
(1.33)**
3.85
(1.63)**
b2 – (EE ratio)2 t x
–2.35
(.95)**
–2.68
(1.17)*
b3 – Recovery dummy
1.56
(.61)**
1.24
(.60)*
b4 – EE ratio t x × Recovery dummy
–3.93
(1.82)*
–4.41
(1.96)*
b5 – (EE ratio)2 t x × Recovery dummy
2.72
(1.41)*
3.00
(1.49)*
b6 – Autocorrelation t 1 a
.20
(.09)*
.14
(.05)**
b7 – Net income a
.62
(.20)**
.60
(.19)***
b8 – Total assets a
–.40
(.22)*
-
-
b9 – R&D–Sales ratio a
–.00
(.07)
–.07
(.07)
b10 – R&D missing dummy
.17
(.17)
.29
(.21)
b11 – Firm size a
–.04
(.16)
–.38
(.11)***
b12 – Firm age a
–.14
(.07)*
–.18
(.06)**
b13 – GICS 4510 dummy
.15
(.13)
.11
(.14)
b14 – GICS 4520 dummy
.12
(.12)
.03
(.13)
b15US location dummy
–.09
(.11)
–.22
(.12)*
b16 – Constant
–1.39
(.49)**
–1.26
(.59)*
Hansen test of over-identification
1.00
1.00
Arellano Bond AR(1) c
–3.03
**
–2.79
**
Arellano Bond AR(2) c
–.18
–.36
a Standardized value; b The standard errors are robust to heteroskedasticity and arbitrary patterns of
autocorrelation within agents (Roodman 2009a); c z values reported; + p < .10; * p < .05; ** p < .01; *** p
< .001. Time dummy variables were included in all models but are omitted from these results. Onetailed
significance levels are reported.
7
d. Quarterly varying measure of attention to exploitation-exploration
As a robustness test, we developed and tested another measure of managerial attention to
exploitation-exploration, one that varies quarterly, by taking the moving average of the annual
EE ratio (over three t’s). The results of the model run with this alternative EE ratio are highly
similar with our main findings (see Table V).
Table V.
Robustness tests regarding yearly versus quarterly
operationalization of the EE ratio (8/20 lags used)
EE ratio (yearly)
(presented)
EE ratio as moving
average
Coeff.
(SE) b
Coeff.
(SE) b
3.39
(1.33)**
4.92
(1.90)**
–2.35
(.95)**
-3.56
(1.41)**
1.56
(.61)**
1.98
(.77)**
–3.93
(1.82)*
-5.60
(2.41)*
2.72
(1.41)*
4.06
(1.89)*
.20
(.09)*
.22
(.10)*
.62
(.20)**
.66
(.21)**
–.40
(.22)*
-.38
(.25)+
–.00
(.07)
.02
(.08)
.17
(.17)
.04
(.14)
–.04
(.16)
-.05
(.19)
–.14
(.07)*
-.13
(.06)*
.15
(.13)
.20
(.14)+
.12
(.12)
.17
(.13)+
–.09
(.11)
-.12
(.11)
–1.39
(.49)**
-1.85
(.68)**
1.00
1.00
–3.03
**
-2.83
**
–.18
.54
a Standardized value; b The standard errors are robust to heteroskedasticity and arbitrary patterns of
autocorrelation within agents (Roodman 2009a); c z values reported; + p < .10; * p < .05; ** p < .01; *** p
< .001. Time dummy variables were included in all models but are omitted from these results. One-tailed
significance levels are reported.
8
e. Competing pooled OLS and random-effects GLS regression
Table VI denotes the results of competing pooled OLS and random-effects GLS regression
model (bootstrapped 100 iterations). Once more, the results are highly comparable to our main
findings.
Table VI.
Robustness test with respect to method
Dependent variable:
Profit margin
Pooled OLS
Random-effects
GLS
(bootstrapped;
100 iterations)
Coeff.
(SE) b
Coeff.
(SE)
b1 – EE ratio t 4
2.61
(.79)***
2.61
(1.52)*
b2 – (EE ratio)2 t 4
-1.82
(.60)**
-1.82
(1.07)*
b3 – Recovery dummy
1.10
(.41)**
1.10
(.53)*
b4 – EE ratio t 4 × Recovery dummy
-3.28
(1.45)*
-3.28
(1.69)*
b5 – (EE ratio)2 t 4 × Recovery dummy
2.19
(1.18)*
2.19
(1.29)+
b6 – Profit margin t 1 a
.24
(.07)***
.24
(.10)**
b7 – Net income a
.62
(.14)***
.62
(.27)**
b8 – Total assets a
-.39
(.12)**
-.39
(.22)*
b9 – R&D–Sales ratio a
-.01
(.04)+
-.01
(.12)
b10 – R&D missing dummy
.06
(.04)
.06
(.09)
b11 – Firm size a
-.03
(.05)
-.03
(.15)
b12 – Firm age a
-.04
(.03)+
-.04
(.04)
b13 – GICS 4510 dummy
.13
(.07)*
.13
(.13)
b14 – GICS 4520 dummy
.14
(.08)*
.14
(.15)
b15US location dummy
-.05
(.06)
-.05
(.11)
b16 – Constant
-.94
(.28)***
-.94
(.55)**
a Standardized value; + p < .10; * p < .05; ** p < .01; *** p < .001. One-tailed significance levels
are reported. b Robust standard errors.

File (1)

ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
T he critical role of research and development (R&D) and advertising in the marketing strategy of the firm is well established. This paper conceptually and empirically examines why and how much the effectiveness of these two marketing instruments differs between times of economic expansions versus periods of economic contractions—and whether these results depend on the cyclicality of the industry in question. We consider a key marketing metric (market share) and a key financial metric (firm profit). Our empirical setting is 1,175 U.S. firms across a time period spanning over three decades. We find that R&D and advertising contribute to firm performance but that their effectiveness is not constant across the business cycle. Increasing advertising share in contractions has a stronger effect on profit and market share than increasing advertising share in expansions. Likewise, investments in R&D in contractions lead to higher gains in market share and profit than R&D investments in expansions, albeit only in subsequent years. If in contractions the firm faces tight budget constraints and has to choose between either maintaining R&D or advertising, our simulation results show that maintaining R&D is associated with better company performance. We find that advertising effectiveness, in general, and in contractions, in particular, is systematically moderated by the degree of cyclicality of the industry in which the firm operates. In relatively stable industries, advertising effects are small or even nonsignificant, and they do not go beyond the year the firm advertises. However, in highly cyclical industries, advertising effects are long-lasting, its total effect being 50% larger (market share) and 200% larger (profits) than in industries of average cyclicality. The effect of industry cyclicality on advertising effectiveness is especially pronounced in contractions. Collectively, these findings provide valuable and actionable insights into how firms should respond to contractions in order to grow profits and market share.
Article
Full-text available
The growth of private labels over the past decades has been attributed to various factors. This article formally addresses the link between private-label success and economic expansions and contractions using recently developed time-series/econometric techniques. The findings confirm conventional wisdom that a country's private-label share increases when the economy is suffering and shrinks when the economy is flourishing. However, asymmetries are found in the extent to and speed with which private-label share changes in cyclical up-versus downturns. Consumers switch more extensively to store brands during bad economic times than they switch back to national brands in a subsequent recovery. In addition, the switch to private-label brands is faster than the opposite movement to national brands after the recession ends. Finally, not only are consumers more prone to buy private labels during economic downturns, but some keep buying them when bad economic times are long over as well, leaving permanent " scars " on national brands' performance level. The authors argue that national-brand manufacturers can mitigate the effect of an economic downturn on their shares by intensifying their marketing-support activities in recessions. Such a proactive strategy is not often observed. On the contrary, available evidence suggests that many manufacturers exacerbate their predicament by cutting back on their marketing expenses when the economy turns sour. Most retailers invest more strongly in their private-label program when the economy deteriorates, making it even more difficult for national brands to catch up with the share lost during contractions.
Article
The difference and system generalized method-of-moments estimators, developed by Holtz-Eakin, Newey, and Rosen (1988, Econometrica 56: 1371–1395); Arellano and Bond (1991, Review of Economic Studies 58: 277–297); Arellano and Bover (1995, Journal of Econometrics 68: 29–51); and Blundell and Bond (1998, Journal of Econometrics 87: 115–143), are increasingly popular. Both are general estimators designed for situations with “small T, large N″ panels, meaning few time periods and many individuals; independent variables that are not strictly exogenous, meaning they are correlated with past and possibly current realizations of the error; fixed effects; and heteroskedasticity and autocorrelation within individuals. This pedagogic article first introduces linear generalized method of moments. Then it describes how limited time span and potential for fixed effects and endogenous regressors drive the design of the estimators of interest, offering Stata-based examples along the way. Next it describes how to apply these estimators with xtabond2. It also explains how to perform the Arellano–Bond test for autocorrelation in a panel after other Stata commands, using abar. The article concludes with some tips for proper use.
Book
1. The Entrepreneur in Economic Theory. 2. Limited Attention. 3. Allocating Limited Entrepreneurial Attention. 4. Innovation, Firm Size and Growth. 5. Career Choice. 6. Optimally Incomplete Contracts. 7. The Internalization of Transactions. 8. The Role of the Venture Capitalist. 9. Independent Contractors. 10. Conclusion. Appendix. Name Index.
Article
The “difference” and “system” generalized method of moments (GMM) estimators for dynamic panel models are growing steadily in popularity. The estimators are designed for panels with short time dimensions (T), and by default they generate instruments sets whose number grows quadratically in T. The dangers associated with having many instruments relative to observations are documented in the applied literature. The instruments can overfit endogenous variables, failing to expunge their endogenous components and biasing coefficient estimates. Meanwhile they can vitiate the Hansen J test for joint validity of those instruments, as well as the difference-in-Sargan/Hansen test for subsets of instruments. The weakness of these specification tests is a particular concern for system GMM, whose distinctive instruments are only valid under a non-trivial assumption. Judging by current practice, many researchers do not fully appreciate that popular implementations of these estimators can by default generate results that simultaneously are invalid yet appear valid. The potential for type I errors—false positives—is therefore substantial, especially after amplification by publication bias. This paper explains the risks and illustrates them with reference to two early applications of the estimators to economic growth, Forbes (2000) on income inequality and Levine, Loayza, and Beck (LLB, 2000) on financial sector development. Endogenous causation proves hard to rule out in both papers. Going forward, for results from these GMM estimators to be credible, researchers must report the instrument count and aggressively test estimates and specification test results for robustness to reductions in that count.
Article
This study examines how CEO cognition, organizational capabilities, and organizational incentives interacted to shape firm strategy during the fiber-optic revolution. I tested for these factors' association with subsequent investment in optical technologies using longitudinal data from 71 communications firms. Results show that each is separately important in shaping outcomes, and their alignment toward the adoption of a new technology leads to the greatest levels of change. In addition, cognition can compensate when organization-level factors are lacking. Considering cognition, capabilities, and incentives together contributes to a more contingent view of the conditions under which CEO cognition matters for firm strategy.
Article
This analysis considers the impact of the top managers in an organization on the organization's outcomes, specifically strategic choices and performance levels. The focus is not on the chief executive alone, but rather on the entire top management team. Using a macro view, these organizational outcomes are perceived to be related to the values and cognitive bases of those high-power individuals in the organization. In developing the model, emphasis is on the background characteristics of the top managers as opposed to the psychological dimensions. A series of propositions that should be tested to support the upper echelons theory are presented. The topics of these propositions include age, functional track, other career experiences, education, socioeconomic roots, financial position, and group characteristics. The creation of this model is just the beginning of the work that is necessary to evaluate and understand the upper echelons theory. Further input is needed from areas such as the executive recruiting industry. Additionally, clinical and statistical studies are both necessary to fully develop this theory. (SRD)