An official website of the United States government
ObjectiveThe objective is to estimate the relative contributions of nonresponse, coverage, and measurement biases in survey estimates of voting. MethodsWe survey 3,000 Boston‐area households sampled from an address‐based frame matched, when possible, to telephone numbers. A two‐phase sampling design was used to follow up nonrespondents from phone interviews with personal interviews. All cases were then linked to voting records. ResultsNonresponse, coverage, and measurement‐biased survey estimates at varying stages of the study design. Coverage error linked to missing telephone numbers biased estimates that excluded nonphone households. Overall estimates including nonphone households and nonrespondent interviews include 25 percent relative bias equally attributable to measurement and nonresponse. ConclusionBias in voting measures is not limited to measurement bias. Researchers should also assess the potential for nonresponse and coverage biases.
more »
« less
Direct Loss Minimization for Sparse Gaussian Processes
The paper provides a thorough investigation of Direct Loss Minimization (DLM), which optimizes the posterior to minimize predictive loss, in sparse Gaussian processes. For the conjugate case, we consider DLM for log-loss and DLM for square loss showing a significant performance improvement in both cases. The application of DLM in non-conjugate cases is more complex because the logarithm of expectation in the log-loss DLM objective is often intractable and simple sampling leads to biased estimates of gradients. The paper makes two technical contributions to address this. First, a new method using product sampling is proposed, which gives unbiased estimates of gradients (uPS) for the objective function. Second, a theoretical analysis of biased Monte Carlo estimates (bMC) shows that stochastic gradient descent converges despite the biased gradients. Experiments demonstrate empirical success of DLM. A comparison of the sampling methods shows that, while uPS is potentially more sample-efficient, bMC provides a better tradeoff in terms of convergence time and computational efficiency.
more »
« less
- Award ID(s):
- 1906694
- PAR ID:
- 10231699
- Date Published:
- Journal Name:
- Proceedings of Machine Learning Research
- Volume:
- 130
- ISSN:
- 2640-3498
- Page Range / eLocation ID:
- 2566-2574
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Individual body size distributions (ISD) within communities are remarkably consistent across habitats and spatiotemporal scales and can be represented by size spectra, which are described by a power law. The focus of size spectra analysis is to estimate the exponent () of the power law. A common application of size spectra studies is to detect anthropogenic pressures.Many methods have been proposed for estimating most of which involve binning the data, counting the abundance within bins, and then fitting an ordinary least squares regression in log–log space. However, recent work has shown that binning procedures return biased estimates of compared to procedures that directly estimate using maximum likelihood estimation (MLE). While it is clear that MLE produces less biased estimates of site‐specificλ's, it is less clear how this bias affects the ability to test for changes inλacross space and time, a common question in the ecological literature.Here, we used simulation to compare the ability of two normalised binning methods (equal logarithmic and log2bins) and MLE to (1) recapture known values of , and (2) recapture parameters in a linear regression measuring the change in across a hypothetical environmental gradient. We also compared the methods using two previously published body size datasets across a natural temperature gradient and an anthropogenic pollution gradient.Maximum likelihood methods always performed better than common binning methods, which demonstrated consistent bias depending on the simulated values of . This bias carried over to the regressions, which were more accurate when was estimated using MLE compared to the binning procedures. Additionally, the variance in estimates using MLE methods is markedly reduced when compared to binning methods.The error induced by binning methods can be of similar magnitudes as the variation previously published in experimental and observational studies, bringing into question the effect sizes of previously published results. However, while the methods produced different regression slope estimates, they were in qualitative agreement on the sign of those slopes (i.e. all negative or all positive). Our results provide further support for the direct estimation of and its relative variation across environmental gradients using MLE over the more common methods of binning.more » « less
-
In this paper, a novel way to compute derivativebased global sensitivity measures is presented. Conjugate Unscented Transform (CUT) is used to evaluate the multidimensional definite integrals which lead to the sensitivity measures. The method is compared with Monte Carlo estimates as well as the screening method of Morris. It is shown that using CUT provides a much more accurate estimate of sensitivity measures as compared to Monte Carlo (with far lesser computational cost) as well as the Morris method (with similar computational cost). Illustrations on three test functions are presented as evidence.more » « less
-
ABSTRACT We explore the radial variation of star formation histories (SFHs) in dwarf galaxies simulated with Feedback In Realistic Environments (FIRE) physics. The sample contains 26 field dwarf galaxies with Mstar = 105–109 M⊙. We find age gradients are common in our dwarfs, with older stars dominant at large radii. The strength of the gradient correlates with overall galaxy age such that earlier star formation produces a more pronounced gradient. The relation between formation time and strength of the gradient is driven by both mergers and star formation feedback. Mergers can both steepen and flatten the age gradient depending on the timing of the merger and SFHs of the merging galaxy. In galaxies without significant mergers, feedback pushes stars to the outskirts. The strength of the age gradient is determined by the subsequent evolution of the galaxy. Galaxies with weak age gradients constantly grow to z = 0, meaning that young star formation occurs at a similar radius to which older stars are heated to. In contrast, galaxies with strong age gradients tend to maintain a constant half-mass radius over time. If real galaxies have age gradients as we predict, stellar population studies that rely on sampling a limited fraction of a galaxy can give a biased view of its global SFH. Central fields can be biased young by Gyrs while outer fields are biased old. Fields positioned near the 2D half-light radius will provide the least biased measure of a dwarf galaxy’s global SFH.more » « less
-
De Vico Fallani, Fabrizio (Ed.)The exponential family random graph modeling (ERGM) framework provides a highly flexible approach for the statistical analysis of networks (i.e., graphs). As ERGMs with dyadic dependence involve normalizing factors that are extremely costly to compute, practical strategies for ERGMs inference generally employ a variety of approximations or other workarounds. Markov Chain Monte Carlo maximum likelihood (MCMC MLE) provides a powerful tool to approximate the maximum likelihood estimator (MLE) of ERGM parameters, and is generally feasible for typical models on single networks with as many as a few thousand nodes. MCMC-based algorithms for Bayesian analysis are more expensive, and high-quality answers are challenging to obtain on large graphs. For both strategies, extension to the pooled case—in which we observe multiple networks from a common generative process—adds further computational cost, with both time and memory scaling linearly in the number of graphs. This becomes prohibitive for large networks, or cases in which large numbers of graph observations are available. Here, we exploit some basic properties of the discrete exponential families to develop an approach for ERGM inference in the pooled case that (where applicable) allows an arbitrarily large number of graph observations to be fit at no additional computational cost beyond preprocessing the data itself. Moreover, a variant of our approach can also be used to perform Bayesian inference under conjugate priors, again with no additional computational cost in the estimation phase. The latter can be employed either for single graph observations, or for observations from graph sets. As we show, the conjugate prior is easily specified, and is well-suited to applications such as regularization. Simulation studies show that the pooled method leads to estimates with good frequentist properties, and posterior estimates under the conjugate prior are well-behaved. We demonstrate the usefulness of our approach with applications to pooled analysis of brain functional connectivity networks and to replicated x-ray crystal structures of hen egg-white lysozyme.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- The DOI is not currently available.
