Secure Survey Design in Organizations

In this Department of Economics Seminar, Christian Zehnder, from Lausanne, discusses his paper on Secure Survey Design in Organizations

Abstract:

We study the impact of secure survey design on information transmission in organizations. We establish two theoretical results: first, popular randomized-response designs are strategically equivalent to non-secure direct-elicitation designs; second, it is possible to obtain unbiased estimates of the impact of survey design on information transmission, provided that players are rational and do not make false accusations.

Experimental investigation shows that both predictions must be qualified in practice: randomized-response can yield significantly better results than direct-elicitation or hard-garbling designs; and false accusations lead to a small but persistent bias in treatment effect estimates. This deviation from equilibrium play can be accounted for in a Quantal Response Level-k model play.

Empirical play converges to equilibrium if players are given opportunities to learn from cross-sectional data, but does not if they learn from their own experiential data alone. A practical implication is that randomized-response designs cannot be used systematically in organizational settings. Hard-garbling designs, in contrast, can facilitate information transmission, even under long-run equilibrium conditions.