Home > Publications . Search All . Browse All . Country . Browse PSC Pubs . PSC Report Series

PSC In The News

RSS Feed icon

Stern, Novak, Harlow, and colleagues say compensation due Californians forcibly sterilized under eugenics laws

Burgard and Seelye find job insecurity linked to psychological distress among workers in later years

Former PSC trainee Jay Borchert parlays past incarceration and doctoral degree into pursuing better treatment of inmates

More News

Highlights

Savolainen wins Outstanding Contribution Award for study of how employment affects recidivism among past criminal offenders

Giving Blueday at ISR focuses on investing in the next generation of social scientists

Pfeffer and Schoeni cover the economic and social dimensions of wealth inequality in this special issue

PRB Policy Communication Training Program for PhD students in demography, reproductive health, population health

More Highlights

Next Brown Bag

Mon, Jan 23, 2017 at noon:
H. Luke Shaefer

Adaptive Confidence Intervals for the Test Error in Classification

Publication Abstract

Laber, Eric B., and Susan A. Murphy. 2011. "Adaptive Confidence Intervals for the Test Error in Classification." Journal of the American Statistical Association, 106(495): 904-913.

The estimated test error of a learned classifier is the most commonly reported measure of classifier performance. However, constructing a high-quality point estimator of the test error has proved to be very difficult. Furthermore, common interval estimators (e.g., confidence intervals) are based on the point estimator of the test error and thus inherit all the difficulties associated with the point estimation problem. As a result, these confidence intervals do not reliably deliver nominal coverage. In contrast, we directly construct the confidence interval by using smooth data-dependent upper and lower bounds on the test error. We prove that, for linear classifiers, the proposed confidence interval automatically adapts to the nonsmoothness of the test error, is consistent under fixed and local alternatives, and does not require that the Bayes classifier be linear. Moreover, the method provides nominal coverage on a suite of test problems using a range of classification algorithms and sample sizes. This article has supplementary material online.

DOI:10.1198/jasa.2010.tm10053 (Full Text)

PMCID: PMC3285493. (Pub Med Central)

Browse | Search : All Pubs | Next