Asta-Advances in Statistical Analysis最新文献

In this work, we propose an efficient implementation of mixtures of experts distributional regression models which exploits robust estimation by using stochastic first-order optimization techniques with adaptive learning rate schedulers. We take advantage of the flexibility and scalability of neural network software and implement the proposed framework in mixdistreg, an R software package that allows for the definition of mixtures of many different families, estimation in high-dimensional and large sample size settings and robust optimization based on TensorFlow. Numerical experiments with simulated and real-world data applications show that optimization is as reliable as estimation via classical approaches in many different settings and that results may be obtained for complicated scenarios where classical approaches consistently fail.

The objective of advanced topic modeling is not only to explore latent topical structures, but also to estimate relationships between the discovered topics and theoretically relevant metadata. Methods used to estimate such relationships must take into account that the topical structure is not directly observed, but instead being estimated itself in an unsupervised fashion, usually by common topic models. A frequently used procedure to achieve this is the method of composition, a Monte Carlo sampling technique performing multiple repeated linear regressions of sampled topic proportions on metadata covariates. In this paper, we propose two modifications of this approach: First, we substantially refine the existing implementation of the method of composition from the R package stm by replacing linear regression with the more appropriate Beta regression. Second, we provide a fundamental enhancement of the entire estimation framework by substituting the current blending of frequentist and Bayesian methods with a fully Bayesian approach. This allows for a more appropriate quantification of uncertainty. We illustrate our improved methodology by investigating relationships between Twitter posts by German parliamentarians and different metadata covariates related to their electoral districts, using the structural topic model to estimate topic proportions.

This paper proposes a spatial point process model on a linear network to analyse cruise passengers’ stop activities. It identifies and models tourists’ stop intensity at the destination as a function of their main determinants. For this purpose, we consider data collected on cruise passengers through the integration of traditional questionnaire-based survey methods and GPS tracking data in two cities, namely Palermo (Italy) and Dubrovnik (Croatia). Firstly, the density-based spatial clustering of applications with noise algorithm is applied to identify stop locations from GPS tracking data. The influence of individual-related variables and itinerary-related characteristics is considered within a framework of a Gibbs point process model. The proposed model describes spatial stop intensity at the destination, accounting for the geometry of the underlying road network, individual-related variables, contextual-level information, and the spatial interaction amongst stop points. The analysis succeeds in quantifying the influence of both individual-related variables and trip-related characteristics on stop intensity. An interaction parameter allows for measuring the degree of dependence amongst cruise passengers in stop location decisions.

We analyze issues related to estimation and inference for the constrained sum of squares estimator (CSS) of the k-factor Gegenbauer autoregressive moving average (GARMA) model. We present theoretical results for the estimator and show that the parameters that determine the cycle lengths are asymptotically independent, converging at rate T, the sample size, for finite cycles. The remaining parameters lack independence and converge at the standard rate. Analogous with existing literature, some challenges exist for testing the hypothesis of non-cyclical long memory, since the associated parameter lies on the boundary of the parameter space. We present simulation results to explore small sample properties of the estimator, which support most distributional results, while also highlighting areas that merit additional exploration. We demonstrate the applicability of the theory and estimator with an application to IBM trading volume.

In this paper measures of interrater absolute agreement for quantitative measurements based on the standard deviation are proposed. Such indices allow (i) to overcome the limits affecting the intraclass correlation index; (ii) to measure the interrater agreement on single targets. Estimators of the proposed measures are introduced and their sampling properties are investigated for normal and non-normal data. Simulated data are employed to demonstrate the accuracy and practical utility of the new indices for assessing agreement. Finally, an application to assess the consistency of measurements performed by radiologists evaluating tumor size of lung cancer is presented.

Non-response is a major problem for anyone collecting and processing data. A commonly used technique to deal with missing data is imputation, where missing values are estimated and filled in into the dataset. Imputation can become challenging if the variable to be imputed has to comply with a known total. Even more challenging is the case where several variables in the same dataset need to be imputed and, in addition to known totals, logical restrictions between variables have to be satisfied. In our paper, we develop an approach for a broad class of imputation methods for multivariate categorical data such that previously published totals are preserved while logical restrictions on the data are satisfied. The developed approach can be used in combination with any imputation model that estimates imputation probabilities, i.e. the probability that imputation of a certain category for a variable in a certain unit leads to the correct value for this variable and unit.

Black box machine learning models are currently being used for high-stakes decision making in various parts of society such as healthcare and criminal justice. While tree-based ensemble methods such as random forests typically outperform deep learning models on tabular data sets, their built-in variable importance algorithms are known to be strongly biased toward high-entropy features. It was recently shown that the increasingly popular SHAP (SHapley Additive exPlanations) values suffer from a similar bias. We propose debiased or "shrunk" SHAP scores based on sample splitting which additionally enable the detection of overfitting issues at the feature level.

A family of consistent tests, derived from a characterization of the probability generating function, is proposed for assessing Poissonity against a wide class of count distributions, which includes some of the most frequently adopted alternatives to the Poisson distribution. Actually, the family of test statistics is based on the difference between the plug-in estimator of the Poisson cumulative distribution function and the empirical cumulative distribution function. The test statistics have an intuitive and simple form and are asymptotically normally distributed, allowing a straightforward implementation of the test. The finite sample properties of the test are investigated by means of an extensive simulation study. The test shows satisfactory behaviour compared to other tests with known limit distribution.

This paper uses independence-type characterizations to propose a class of test statistics which can be used for testing goodness-of-fit with several classes of null distributions. The resulting tests are consistent against fixed alternatives. Some limiting and small sample properties of the test statistics are explored. In comparison with common universal goodness-of-fit tests, the new tests exhibit better power for most of the alternatives considered, while in comparison with another characterization-based procedure, the new tests provide competitive or comparable power in various simulation settings. The handiness of the proposed tests is demonstrated through several real-data examples.

Despite the popularity of feature importance (FI) measures in interpretable machine learning, the statistical adequacy of these methods is rarely discussed. From a statistical perspective, a major distinction is between analysing a variable’s importance before and after adjusting for covariates—i.e., between marginal and conditional measures. Our work draws attention to this rarely acknowledged, yet crucial distinction and showcases its implications. We find that few methods are available for testing conditional FI and practitioners have hitherto been severely restricted in method application due to mismatched data requirements. Most real-world data exhibits complex feature dependencies and incorporates both continuous and categorical features (i.e., mixed data). Both properties are oftentimes neglected by conditional FI measures. To fill this gap, we propose to combine the conditional predictive impact (CPI) framework with sequential knockoff sampling. The CPI enables conditional FI measurement that controls for any feature dependencies by sampling valid knockoffs—hence, generating synthetic data with similar statistical properties—for the data to be analysed. Sequential knockoffs were deliberately designed to handle mixed data and thus allow us to extend the CPI approach to such datasets. We demonstrate through numerous simulations and a real-world example that our proposed workflow controls type I error, achieves high power, and is in-line with results given by other conditional FI measures, whereas marginal FI metrics can result in misleading interpretations. Our findings highlight the necessity of developing statistically adequate, specialized methods for mixed data.