42.7 Conclusion and Recommendations

Sensitivity analysis is not optional in modern empirical research, it’s essential. The techniques described in this chapter provide a toolkit for assessing the robustness of your findings:

  1. Start with specification curve analysis (starbility) to visualize how your results vary across defensible specifications

  2. Apply Oster’s method (robomit) to assess sensitivity to omitted variable bias using coefficient stability and \(R^2\) movement

  3. Use konfound analysis (konfound) to quantify how much bias would be needed to overturn your inference

  4. Conduct additional tests relevant to your context: placebo tests, subsample analysis, outlier diagnostics, etc.

  5. Present results clearly: Use both tables and figures, provide verbal interpretation, and be transparent about which specifications you view as most credible and why

Remember: The goal is not to show that your result is robust to everything, but to demonstrate that it’s robust to reasonable alternative choices and plausible threats to identification. Be honest about the limitations while making the strongest case possible for your findings.

The mark of rigorous empirical work is not that every robustness check confirms your main result, it’s that you’ve thoughtfully considered the most important threats to validity and provided evidence about whether those threats are likely to overturn your conclusions.