Meta-analysis and evidence synthesis for strategic management

Evidence synthesis in strategic management

Strategic management asks why some firms perform better than others and which decisions explain it. Common topics are corporate diversification, mergers and acquisitions, strategic alliances, the resource-based view, top management team composition, entrepreneurial and market orientation, and innovation. Meta-analyses have been published on many of these, typically pooling correlations or regression-based estimates between a strategic variable and performance.

The evidence has features that call for care. Firms choose strategies and are not assigned to them, so correlations with performance are open to reverse causation and omitted variables. Constructs such as capabilities and orientation are measured with survey scales or proxies. Performance is measured in many ways. And effects depend on industry, country and period, so average effects across all firms may describe no one. Our methods follow meta-analysis and systematic review practice, extended for these features. This page builds on the general guidance for management and business.

Measuring performance

Each measure answers a different question and they are only moderately correlated with each other. A review should not combine them into a single "performance" outcome without showing that the result holds for each, and should report results by type of measure. Studies using perceived performance are particularly prone to inflated correlations when the strategy variable comes from the same survey.

Endogeneity and identification

A firm that diversifies does so for reasons that may be related to its performance, for example because it has surplus resources or because its main market is declining. The observed association between diversification and performance therefore does not show what would happen if a given firm changed its scope. Studies that address endogeneity use instruments, matching, fixed effects, natural experiments or event windows. A review should code the identification strategy and compare estimates. Associations from simple cross-sectional regressions are commonly larger than, or even of different sign from, those from stronger designs, and the review should describe this rather than treating them as equivalent.

Fixed-effects models use within-firm change, which removes stable differences but needs variation over time. Panel studies with short periods give little within-firm variation for slowly changing strategic variables.

Industry, country and period

The performance effect of a strategy depends on the competitive environment. Diversification relates differently to performance in emerging and developed markets; alliances work differently in industries with fast technological change than in stable ones; the value of resources depends on competitors. Meta-regression on industry characteristics and on country institutions can examine this, though the number of independent contexts is small. We recommend reporting the distribution of studies by country, industry and decade, and stating plainly that findings from large listed firms in a few high-income countries may not apply to small or private firms or to other regions.

Period effects are significant: studies from the 1980s describe conglomerates that differ from modern diversified firms. Results should be examined by decade of data, not only by year of publication.

Alliances, mergers and acquisitions

Studies of mergers and acquisitions measure outcomes by announcement returns (short-term market reaction), long-run returns and accounting performance. These give different answers: announcement returns to acquirers are, on average, small and often near zero, while target returns are positive, and long-run measures are noisy and sensitive to benchmarks. A review should present them separately. For alliances, outcomes include survival, perceived success and financial effects. Alliance data are often from surveys of one partner, so the view of the other partner is missing.

Reviews of M&A face the clustering of deals in time and the overlap of samples drawn from the same deal databases, which we address with robust variance methods and sample-level coding.

Resources, capabilities and orientation

Constructs such as dynamic capabilities, market orientation and entrepreneurial orientation are measured with questionnaires, and different scales capture different content. A review of the resource-based view may be tempted to pool a wide range of measures of "resources", but the result then depends on which were included. We code the construct definition and the scale content and pool only measures that match, and we examine whether results differ across measure families. Reviews of orientation constructs have found positive average associations with performance that vary across studies, which is a reason to look at moderators and to be careful with generalizations.

Top management teams and boards

Research relating the characteristics of executives and boards to firm outcomes uses archival data on age, tenure, education, gender and background, and relates them to performance, strategic change and risk. Effects are small on average and vary by context. The key limit is the mismatch between a firm-level outcome and a characteristic of a small group of people. Studies also differ in how diversity is measured. A review should state definitions, use consistent measures and avoid inferring individual behavior from group-level patterns.

Publication bias and theory confirmation

Strategic management journals favor theory-driven, significant findings. We check small-study effects, compare published and unpublished studies (dissertations, conference papers, working papers), and use selection models and p-curve as sensitivity analysis. Because many studies include the tested relationship among several controls and reported correlations, we also examine whether effects that were the hypothesis of the paper differ from incidental ones.

Data, coding and transparency

A coding manual for strategy reviews records the strategic variable and how it was measured, the performance outcome and its source, the sample frame (listed firms, private firms, one industry or many), the country, the period of data, the controls used and the identification strategy. It also records whether the strategy variable and the outcome came from the same source and whether they were measured in the same year. Two coders work independently on a sample, agreement is reported and disagreements are resolved before the full set is coded.

Strategy research often reports many models in one paper, with and without controls. We extract the fully specified model by a rule set in the protocol, record the alternatives and run sensitivity analyses with other choices. Where primary studies share data, as when several papers analyze the same panel of listed firms, we code the sample so that overlap can be handled.

We recommend sharing the coded data set and analysis code with the final report so that others can check and extend it. Reproducibility in this field is helped by stating the version of any commercial database used and the date it was accessed, because such databases are revised.

Common pitfalls we look for

• Combining accounting, market and perceived performance as one outcome.
• Causal language for associations from non-experimental data.
• Pooling very different measures of the same named capability.
• Ignoring industry and period when generalizing.
• Treating announcement returns and long-run returns as the same outcome.
• Counting overlapping deal or firm samples more than once.

Planning a strategic management synthesis

We help define the strategic variable and the performance outcomes, choose the effect measure, plan the search across management, economics and finance sources, and set up coding of measures, identification strategy, industry, country and period. For questions about mediating mechanisms, such as whether alliances affect performance through innovation, we plan a meta-analytic structural model and check whether enough studies report the needed correlations. See the meta-analysis service for scope and process.

Qualitative and configurational evidence

Much strategy knowledge comes from case studies and comparative qualitative work, which cannot be pooled numerically. A systematic review can still map the cases, compare how the studies defined the strategy and its context, and look for patterns using methods such as cross-case synthesis. Where a question is about configurations of conditions that together lead to an outcome, qualitative comparative analysis can be applied to coded case data, though it needs care about case selection and calibration. We describe such work in a structured narrative synthesis and link it to the quantitative findings where they address the same question. See also the pages on qualitative evidence synthesis and mixed-methods reviews.

Reading an average effect

Suppose a review reports a pooled correlation of 0.12 between entrepreneurial orientation and firm performance, with a prediction interval from minus 0.05 to 0.28. These numbers are invented. A correlation of 0.12 is small, and the interval says that in some settings the relationship could be nil or negative. If studies using archival performance data show an average of 0.06 and survey-based studies 0.20, the difference suggests that shared source contributes to the overall estimate. The honest summary is that the evidence points to a modest positive association that depends on how it is measured, and that it cannot be used to predict the result for a given firm.

How we support research projects in this area

Boundaries of this service

A synthesis of strategic management research describes average associations across firms and studies. It does not provide corporate strategy, consulting advice, valuation or evaluation of any firm's decisions. Most evidence is observational and subject to endogeneity, and results depend on industry, country and period, so they should not be read as predictions for a single organization.

Frequently asked questions