Physical and digital books, media, journals, archives, and databases.
Results include
  1. A modern maximum-likelihood theory for high-dimensional logistic regression

    Sur, Pragya
    Stanford, Calif. : Department of Statistics, Stanford University, March 2018.

    Online statistics.stanford.edu

  2. The likelihood ratio test in high-dimensional logistic regression is asymptotically a rescaled chi-square

    Sur, Pragya
    Stanford, California : Department of Statistics, Stanford University, June 2017.

    Online statistics.stanford.edu

  3. A modern maximum likelihood theory for high-dimensional logistic regression

    Sur, Pragya
    [Stanford, California] : [Stanford University], 2019.

    Logistic regression is arguably the most widely used and studied non-linear model in statistics. Classical maximum-likelihood theory based statistical inference is ubiquitous in this context. This theory hinges on well-known fundamental results: (1) the maximum-likelihood-estimate (MLE) is asymptotically unbiased and normally distributed, (2) its variability can be quantified via the inverse Fisher information, and (3) the log-likelihood ratio (LLR) statistic is asymptotically a Chi-Squared. This thesis uncovers that in the common modern setting where the number of features and the sample size are both large and comparable, classical results are far from accurate. In fact, (1) the MLE is biased, (2) its variability is far greater than classical results, and (3) the LLR statistic is not distributed as a Chi-Square. Consequently, p-values obtained based on classical theory are completely invalid in such settings. This thesis provides a modern perspective on classical maximum likelihood theory in the context of logistic regression (developed jointly by the author and her collaborators). The contributions here are two-fold: first, it discovers a phase transition in the existence of the MLE and explicitly pins down the phase transition curve; second, in the regime where the MLE is finite, it characterizes the asymptotic behavior of the MLE and the LLR for a class of covariate distributions, under the aforementioned high-dimensional regime. Empirical evidence demonstrates that this asymptotic theory provides accurate inference in finite samples and is robust to certain violations of the underlying assumptions. Practical implementation of these results necessitates the estimation of a single scalar, the overall signal strength---a procedure for estimating this parameter is also discussed. This asymptotic theory can be extended to characterize distributions of penalized maximum likelihood estimators in some settings. Along the way, this thesis surveys relevant works in the field of high-dimensional inference, particularly those developing methodology for valid inference in high-dimensional regression problems.

Guides

Course- and topic-based guides to collections, tools, and services.
No guide results found... Try a different search

Library website

Library info; guides & content by subject specialists
No website results found... Try a different search

Exhibits

Digital showcases for research and teaching.
No exhibits results found... Try a different search

EarthWorks

Geospatial content, including GIS datasets, digitized maps, and census data.
No earthworks results found... Try a different search

More search tools

Tools to help you discover resources at Stanford and beyond.