Strong minimax lower bounds for learning

Minimax lower bounds for concept learning state, for example, thatfor each sample size $n$ and learning rule $g_n$, there exists a distributionof the observation $X$ and a concept $C$ to be learnt such that the expectederror of $g_n$ is at least a constant times $V/n$, where $V$ is the VC dimensionof the concept class. However, these bounds do not tell anything about therate of decrease of the error for a {\sl fixed} distribution--concept pair.\\In this paper we investigate minimax lower bounds in such a--stronger--sense.We show that for several natural $k$--parameter concept classes, includingthe class of linear halfspaces, the class of balls, the class of polyhedrawith a certain number of faces, and a class of neural networks, for any{\sl sequence} of learning rules $\{g_n\}$, there exists a fixed distributionof $X$ and a fixed concept $C$ such that the expected error is larger thana constant times $k/n$ for {\sl infinitely many n}. We also obtain suchstrong minimax lower bounds for the tail distribution of the probabilityof error, which extend the corresponding minimax lower bounds.

On prediction of individual sequences

Sequential randomized prediction of an arbitrary binary sequence isinvestigated. No assumption is made on the mechanism of generating the bit sequence. The goal of the predictor is to minimize its relative loss, i.e., to make (almost) as few mistakes as the best ``expert'' in a fixed, possibly infinite, set of experts. We point out a surprising connection between this prediction problem and empirical process theory. First, in the special case of static (memoryless) experts, we completely characterize the minimax relative loss in terms of the maximum of an associated Rademacher process. Then we show general upper and lower bounds on the minimaxrelative loss in terms of the geometry of the class of experts. As main examples, we determine the exact order of magnitude of the minimax relative loss for the class of autoregressive linear predictors and for the class of Markov experts.

Optimised search heuristic combining valid inequalities and tabu search

This paper presents an Optimised Search Heuristic that combines a tabu search method with the verification of violated valid inequalities. The solution delivered by the tabu search is partially destroyed by a randomised greedy procedure, and then the valid inequalities are used to guide the reconstruction of a complete solution. An application of the new method to the Job-Shop Scheduling problem is presented.

Ratio maps and correspondence analysis

We compare two methods for visualising contingency tables and developa method called the ratio map which combines the good properties of both.The first is a biplot based on the logratio approach to compositional dataanalysis. This approach is founded on the principle of subcompositionalcoherence, which assures that results are invariant to considering subsetsof the composition. The second approach, correspondence analysis, isbased on the chi-square approach to contingency table analysis. Acornerstone of correspondence analysis is the principle of distributionalequivalence, which assures invariance in the results when rows or columnswith identical conditional proportions are merged. Both methods may bedescribed as singular value decompositions of appropriately transformedmatrices. Correspondence analysis includes a weighting of the rows andcolumns proportional to the margins of the table. If this idea of row andcolumn weights is introduced into the logratio biplot, we obtain a methodwhich obeys both principles of subcompositional coherence and distributionalequivalence.

Fusion of data sets in multivariate linear regression with errors-in-variables

We consider the application of normal theory methods to the estimation and testing of a general type of multivariate regressionmodels with errors--in--variables, in the case where various data setsare merged into a single analysis and the observable variables deviatepossibly from normality. The various samples to be merged can differ on the set of observable variables available. We show that there is a convenient way to parameterize the model so that, despite the possiblenon--normality of the data, normal--theory methods yield correct inferencesfor the parameters of interest and for the goodness--of--fit test. Thetheory described encompasses both the functional and structural modelcases, and can be implemented using standard software for structuralequations models, such as LISREL, EQS, LISCOMP, among others. An illustration with Monte Carlo data is presented.

On a series of Goldbach and Euler

Theorem 1 of Euler s paper of 1737 'Variae Observationes Circa Series Infinitas', states the astonishing result that the series of all unit fractions whose denominators are perfect powers of integers minus unity has sum one. Euler attributes the Theorem to Goldbach. The proof is one of those examples of misuse of divergent series to obtain correct results so frequent during the seventeenth and eighteenth centuries. We examine this proof closelyand, with the help of some insight provided by a modern (and completely dierent) proof of the Goldbach-Euler Theorem, we present a rational reconstruction in terms which could be considered rigorous by modern Weierstrassian standards. At the same time, with a few ideas borrowed from nonstandard analysis we see how the same reconstruction can be also be considered rigorous by modern Robinsonian standards. This last approach, though, is completely in tune with Goldbach and Euler s proof. We hope to convince the reader then how, a few simple ideas from nonstandard analysis, vindicate Euler's work.

Associationism and electoral participation: A multilevel study of 2000 Spanish general election

This work presents an application of the multilevel analysis techniques tothe study of the abstention in the 2000 Spanish general election. Theinterest of the study is both, substantive and methodological. From thesubstantive point of view the article intends to explain the causes ofabstention and analyze the impact of associationism on it. From themethodological point of view it is intended to analyze the interaction betweenindividual and context with a modelisation that takes into account thehierarchical structure of data. The multilevel study of this paper validatesthe one level results obtained in previous analysis of the abstention andshows that only a fraction of the differences in abstention are explained bythe individual characteristics of the electors. Another important fraction ofthese differences is due to the political and social characteristics of thecontext. Relating to associationism, the data suggest that individualparticipation in associations decrease the probability of abstention. However,better indicators are needed in order to catch more properly the effect ofassociationism in electoral behaviour.

Improved nonparametric confidence intervals in time series regressions

Condence intervals in econometric time series regressions suffer fromnotorious coverage problems. This is especially true when the dependencein the data is noticeable and sample sizes are small to moderate, as isoften the case in empirical studies. This paper suggests using thestudentized block bootstrap and discusses practical issues, such as thechoice of the block size. A particular data-dependent method is proposedto automate the method. As a side note, it is pointed out that symmetricconfidence intervals are preferred over equal-tailed ones, since theyexhibit improved coverage accuracy. The improvements in small sampleperformance are supported by a simulation study.

A generalization of histogram type estimators

We introduce simple nonparametric density estimators that generalize theclassical histogram and frequency polygon. The new estimators are expressed as linear combination of density functions that are piecewisepolynomials, where the coefficients are optimally chosen in order to minimize the integrated square error of the estimator. We establish the asymptotic behaviour of the proposed estimators, and study theirperformance in a simulation study.

Subsampling the mean of heavy-tailed dependent observations

We establish the validity of subsampling confidence intervals for themean of a dependent series with heavy-tailed marginal distributions.Using point process theory, we study both linear and nonlinear GARCH-liketime series models. We propose a data-dependent method for the optimalblock size selection and investigate its performance by means of asimulation study.

The effects of a universal child benefit

I study the impact of a universal child benefit on fertility and family well-being. I exploitthe unanticipated introduction of a new, sizeable, unconditional child benefit in Spain in2007, granted to all mothers giving birth on or after July 1, 2007. The regressiondiscontinuity-type design allows for a credible identification of the causal effects. I find thatthe benefit did lead to a significant increase in fertility, as intended, part of it coming froman immediate reduction in abortions. On the unintended side, I find that families whoreceived the benefit did not increase their overall expenditure or their consumption ofdirectly child-related goods and services. Instead, eligible mothers stayed out of the laborforce significantly longer after giving birth, which in turn led to their children spending lesstime in formal child care and more time with their mother during their first year of life. Ialso find that couples who received the benefit were less likely to break up the year afterhaving the child, although this effect was only short-term. Taken together, the resultssuggest that child benefits of this kind may successfully increase fertility, as well asaffecting family well-being through their impact on maternal time at home and familystability.

A simple permutation test for clusteredness

Hierarchical clustering is a popular method for finding structure in multivariate data,resulting in a binary tree constructed on the particular objects of the study, usually samplingunits. The user faces the decision where to cut the binary tree in order to determine the numberof clusters to interpret and there are various ad hoc rules for arriving at a decision. A simplepermutation test is presented that diagnoses whether non-random levels of clustering are presentin the set of objects and, if so, indicates the specific level at which the tree can be cut. The test isvalidated against random matrices to verify the type I error probability and a power study isperformed on data sets with known clusteredness to study the type II error.

Construyendo las bases para una comparación fiable: la Encuesta Social Europea

Surveys are a valuable instrument to find out about the social and politicalreality of our context. However, the work of researchers is often limitedby a number of handicaps that are mainly two. On one hand, the samples areusually low technical quality ones and the fieldwork is not carried out inthe finest conditions. On the other hand, many surveys are not especiallydesigned to allow their comparison, a precisely appreciated operation inpolitical research. The article presents the European Social Survey andjustifies its methodological bases. The survey, promoted by the EuropeanScience Foundation and the European Commission, is born from the collectiveeffort of the scientific community with the explicit aim to establishcertain quality standards in the sample design and in the carrying out ofthe fieldwork so as to guarantee the quality of the data and allow eachcomparison between countries.

The contributions of rare objects in correspondence analysis

Correspondence analysis, when used to visualize relationships in a table of counts(for example, abundance data in ecology), has been frequently criticized as being too sensitiveto objects (for example, species) that occur with very low frequency or in very few samples. Inthis statistical report we show that this criticism is generally unfounded. We demonstrate this inseveral data sets by calculating the actual contributions of rare objects to the results ofcorrespondence analysis and canonical correspondence analysis, both to the determination ofthe principal axes and to the chi-square distance. It is a fact that rare objects are oftenpositioned as outliers in correspondence analysis maps, which gives the impression that theyare highly influential, but their low weight offsets their distant positions and reduces their effecton the results. An alternative scaling of the correspondence analysis solution, the contributionbiplot, is proposed as a way of mapping the results in order to avoid the problem of outlying andlow contributing rare objects.

The effects of employment uncertainty and wealth shocks on the labor supply and claiming behavior of older American workers

Unemployment rates in developed countries have recently reached levels not seenin a generation, and workers of all ages are facing increasing probabilities of losingtheir jobs and considerable losses in accumulated assets. These events likely increasethe reliance that most older workers will have on public social insurance programs,exactly at a time that public finances are suffering from a large drop in contributions.Our paper explicitly accounts for employment uncertainty and unexpectedwealth shocks, something that has been relatively overlooked in the literature, butthat has grown in importance in recent years. Using administrative and householdlevel data we empirically characterize a life-cycle model of retirement and claimingdecisions in terms of the employment, wage, health, and mortality uncertainty facedby individuals. Our benchmark model explains with great accuracy the strikinglyhigh proportion of individuals who claim benefits exactly at the Early RetirementAge, while still explaining the increased claiming hazard at the Normal RetirementAge. We also discuss some policy experiments and their interplay with employmentuncertainty. Additionally, we analyze the effects of negative wealth shocks on thelabor supply and claiming decisions of older Americans. Our results can explainwhy early claiming has remained very high in the last years even as the early retirementpenalties have increased substantially compared with previous periods, andwhy labor force participation has remained quite high for older workers even in themidst of the worse employment crisis in decades.