Abstract:
In several scientific disciplines, replicability of empirical
research results reported in papers is a critical issue as failure to reproduce
results puts the reputation of researchers and journals at stake. Using a
discrete-space version of a simple principal-agent mechanism design model put forward by R.V.Vohra,
policies for acceptance and inspection for replicability of scientific papers are investigated.
The model, which is initially posed as a non-linear and non-convex optimization
problem, is in fact equivalent to a linear integer optimization model of which
the optimal solution reveals a simple optimal acceptance/inspection policy using
two key attributes associated with each submission. In the uniform probability
case, the optimal policy is a threshold-value policy which advocates
equal treatment to all submissions with a value equal to that of the threshold
or higher.
{\bf Key words:} Replicable research, principal-agent model, incentive compatibility,
integer optimization, scientific publishing.