Actuarial applications of multivariate phase-type distributions : model calibration and credibility

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal

Model Diagnostics in the presence of measurement error

The problem of using information available from one variable X to make inferenceabout another Y is classical in many physical and social sciences. In statistics this isoften done via regression analysis where mean response is used to model the data. Onestipulates the model Y = µ(X) +ɛ. Here µ(X) is the mean response at the predictor variable value X = x, and ɛ = Y - µ(X) is the error. In classical regression analysis, both (X; Y ) are observable and one then proceeds to make inference about the mean response function µ(X). In practice there are numerous examples where X is not available, but a variable Z is observed which provides an estimate of X. As an example, consider the herbicidestudy of Rudemo, et al. [3] in which a nominal measured amount Z of herbicide was applied to a plant but the actual amount absorbed by the plant X is unobservable. As another example, from Wang [5], an epidemiologist studies the severity of a lung disease, Y , among the residents in a city in relation to the amount of certain air pollutants. The amount of the air pollutants Z can be measured at certain observation stations in the city, but the actual exposure of the residents to the pollutants, X, is unobservable and may vary randomly from the Z-values. In both cases X = Z+error: This is the so called Berkson measurement error model.In more classical measurement error model one observes an unbiased estimator W of X and stipulates the relation W = X + error: An example of this model occurs when assessing effect of nutrition X on a disease. Measuring nutrition intake precisely within 24 hours is almost impossible. There are many similar examples in agricultural or medical studies, see e.g., Carroll, Ruppert and Stefanski [1] and Fuller [2], , among others. In this talk we shall address the question of fitting a parametric model to the re-gression function µ(X) in the Berkson measurement error model: Y = µ(X) + ɛ; X = Z + η; where η and ɛ are random errors with E(ɛ) = 0, X and η are d-dimensional, and Z is the observable d-dimensional r.v.

Multilevel multitrait-multimethod model: application to the measurement of egocentered social networks

Our goal in this paper is to assess reliability and validity of egocentered network data using multilevel analysis (Muthen, 1989, Hox, 1993) under the multitrait-multimethod approach. The confirmatory factor analysis model for multitrait-multimethod data (Werts & Linn, 1970; Andrews, 1984) is used for our analyses. In this study we reanalyse a part of data of another study (Kogovšek et al., 2002) done on a representative sample of the inhabitants of Ljubljana. The traits used in our article are the name interpreters. We consider egocentered network data as hierarchical; therefore a multilevel analysis is required. We use Muthen's partial maximum likelihood approach, called pseudobalanced solution (Muthen, 1989, 1990, 1994) which produces estimations close to maximum likelihood for large ego sample sizes (Hox & Mass, 2001). Several analyses will be done in order to compare this multilevel analysis to classic methods of analysis such as the ones made in Kogovšek et al. (2002), who analysed the data only at group (ego) level considering averages of all alters within the ego. We show that some of the results obtained by classic methods are biased and that multilevel analysis provides more detailed information that much enriches the interpretation of reliability and validity of hierarchical data. Within and between-ego reliabilities and validities and other related quality measures are defined, computed and interpreted

Measurement Quality in Indicators of Compositions: a Compositional Multitrait-Multimethod Approach

Compositional data, also called multiplicative ipsative data, are common in survey research instruments in areas such as time use, budget expenditure and social networks. Compositional data are usually expressed as proportions of a total, whose sum can only be 1. Owing to their constrained nature, statistical analysis in general, and estimation of measurement quality with a confirmatory factor analysis model for multitrait-multimethod (MTMM) designs in particular are challenging tasks. Compositional data are highly non-normal, as they range within the 0-1 interval. One component can only increase if some other(s) decrease, which results in spurious negative correlations among components which cannot be accounted for by the MTMM model parameters. In this article we show how researchers can use the correlated uniqueness model for MTMM designs in order to evaluate measurement quality of compositional indicators. We suggest using the additive log ratio transformation of the data, discuss several approaches to deal with zero components and explain how the interpretation of MTMM designs di ers from the application to standard unconstrained data. We show an illustration of the method on data of social network composition expressed in percentages of partner, family, friends and other members in which we conclude that the faceto-face collection mode is generally superior to the telephone mode, although primacy e ects are higher in the face-to-face mode. Compositions of strong ties (such as partner) are measured with higher quality than those of weaker ties (such as other network members)

UV index measurement and model agreement: uncertainties and limitations

En les últimes dècades, l'increment dels nivells de radiació solar ultraviolada (UVR) que arriba a la Terra (principalment degut a la disminució d'ozó estratosfèric) juntament amb l'augment detectat en malalties relacionades amb l'exposició a la UVR, ha portat a un gran volum d'investigacions sobre la radiació solar en aquesta banda i els seus efectes en els humans. L'índex ultraviolat (UVI), que ha estat adoptat internacionalment, va ser definit amb el propòsit d'informar al públic general sobre els riscos d'exposar el cos nu a la UVR i per tal d'enviar missatges preventius. L'UVI es va definir inicialment com el valor màxim diari. No obstant, el seu ús actual s'ha ampliat i té sentit referir-se a un valor instantani o a una evolució diària del valor d'UVI mesurat, modelitzat o predit. El valor concret d'UVI està afectat per la geometria Sol-Terra, els núvols, l'ozó, els aerosols, l'altitud i l'albedo superficial. Les mesures d'UVI d'alta qualitat són essencials com a referència i per estudiar tendències a llarg termini; es necessiten també tècniques acurades de modelització per tal d'entendre els factors que afecten la UVR, per predir l'UVI i com a control de qualitat de les mesures. És d'esperar que les mesures més acurades d'UVI s'obtinguin amb espectroradiòmetres. No obstant, com que els costs d'aquests dispositius són elevats, és més habitual trobar dades d'UVI de radiòmetres eritemàtics (de fet, la majoria de les xarxes d'UVI estan equipades amb aquest tipus de sensors). Els millors resultats en modelització s'obtenen amb models de transferència radiativa de dispersió múltiple quan es coneix bé la informació d'entrada. No obstant, habitualment no es coneix informació d'entrada, com per exemple les propietats òptiques dels aerosols, la qual cosa pot portar a importants incerteses en la modelització. Sovint, s'utilitzen models més simples per aplicacions com ara la predicció d'UVI o l'elaboració de mapes d'UVI, ja que aquests són més ràpids i requereixen menys paràmetres d'entrada. Tenint en compte aquest marc de treball, l'objectiu general d'aquest estudi és analitzar l'acord al qual es pot arribar entre la mesura i la modelització d'UVI per condicions de cel sense núvols. D'aquesta manera, en aquest estudi es presenten comparacions model-mesura per diferents tècniques de modelització, diferents opcions d'entrada i per mesures d'UVI tant de radiòmetres eritemàtics com d'espectroradiòmeters. Com a conclusió general, es pot afirmar que la comparació model-mesura és molt útil per detectar limitacions i estimar incerteses tant en les modelitzacions com en les mesures. Pel que fa a la modelització, les principals limitacions que s'han trobat és la falta de coneixement de la informació d'aerosols considerada com a entrada dels models. També, s'han trobat importants diferències entre l'ozó mesurat des de satèl·lit i des de la superfície terrestre, la qual cosa pot portar a diferències importants en l'UVI modelitzat. PTUV, una nova i simple parametrització pel càlcul ràpid d'UVI per condicions de cel serens, ha estat desenvolupada en base a càlculs de transferència radiativa. La parametrització mostra una bona execució tant respecte el model base com en comparació amb diverses mesures d'UVI. PTUV ha demostrat la seva utilitat per aplicacions particulars com ara l'estudi de l'evolució anual de l'UVI per un cert lloc (Girona) i la composició de mapes d'alta resolució de valors d'UVI típics per un territori concret (Catalunya). En relació a les mesures, es constata que és molt important saber la resposta espectral dels radiòmetres eritemàtics per tal d'evitar grans incerteses a la mesura d'UVI. Aquest instruments, si estan ben caracteritzats, mostren una bona comparació amb els espectroradiòmetres d'alta qualitat en la mesura d'UVI. Les qüestions més importants respecte les mesures són la calibració i estabilitat a llarg termini. També, s'ha observat un efecte de temperatura en el PTFE, un material utilitzat en els difusors en alguns instruments, cosa que potencialment podria tenir implicacions importants en el camp experimental. Finalment, i pel que fa a les comparacions model-mesura, el millor acord s'ha trobat quan es consideren mesures d'UVI d'espectroradiòmetres d'alta qualitat i s'usen models de transferència radiativa que consideren les millors dades disponibles pel que fa als paràmetres òptics d'ozó i aerosols i els seus canvis en el temps. D'aquesta manera, l'acord pot ser tan alt dins un 0.1º% en UVI, i típicament entre menys d'un 3%. Aquest acord es veu altament deteriorat si s'ignora la informació d'aerosols i depèn de manera important del valor d'albedo de dispersió simple dels aerosols. Altres dades d'entrada del model, com ara l'albedo superficial i els perfils d'ozó i temperatura introdueixen una incertesa menor en els resultats de modelització.

The effects of environmental perturbation and measurement error on estimates of the shape parameter in the theta-logistic model of population regulation

The theta-logistic is a widely used generalisation of the logistic model of regulated biological processes which is used in particular to model population regulation. Then the parameter theta gives the shape of the relationship between per-capita population growth rate and population size. Estimation of theta from population counts is however subject to bias, particularly when there are measurement errors. Here we identify factors disposing towards accurate estimation of theta by simulation of populations regulated according to the theta-logistic model. Factors investigated were measurement error, environmental perturbation and length of time series. Large measurement errors bias estimates of theta towards zero. Where estimated theta is close to zero, the estimated annual return rate may help resolve whether this is due to bias. Environmental perturbations help yield unbiased estimates of theta. Where environmental perturbations are large, estimates of theta are likely to be reliable even when measurement errors are also large. By contrast where the environment is relatively constant, unbiased estimates of theta can only be obtained if populations are counted precisely Our results have practical conclusions for the design of long-term population surveys. Estimation of the precision of population counts would be valuable, and could be achieved in practice by repeating counts in at least some years. Increasing the length of time series beyond ten or 20 years yields only small benefits. if populations are measured with appropriate accuracy, given the level of environmental perturbation, unbiased estimates can be obtained from relatively short censuses. These conclusions are optimistic for estimation of theta. (C) 2008 Elsevier B.V All rights reserved.

Measurement of acrylamide and its precursors in potato, wheat, and rye model systems

The relationship between acrylamide and its precursors, namely, free asparagine and reducing sugars, was studied in cakes made from potato flake, wholemeal wheat, and wholemeal rye, cooked at 180 degreesC, from 5 to 60 min. Between 5 and 20 min, major losses of asparagine, water, and total reducing sugars were accompanied by large increases in acrylamide, which maximized in all three products between 25 and 30 min, followed by a slow linear reduction. Acrylamide formation did not occur to a large degree until the moisture contents of the cakes fell below 5%. Linear relationships were observed for acrylamide formation with the residual levels of asparagine and reducing sugars for all three food materials.

Model-measurement comparison of mesospheric temperature inversions, and a simple theory for their occurrence

Mesospheric temperature inversions are well established observed phenomena, yet their properties remain the subject of ongoing research. Comparisons between Rayleigh-scatter lidar temperature measurements obtained by the University of Western Ontario's Purple Crow Lidar (42.9°N, 81.4°W) and the Canadian Middle Atmosphere Model are used to quantify the statistics of inversions. In both model and measurements, inversions occur most frequently in the winter and exhibit an average amplitude of ∼10 K. The model exhibits virtually no inversions in the summer, while the measurements show a strongly reduced frequency of occurrence with an amplitude about half that in the winter. A simple theory of mesospheric inversions based on wave saturation is developed, with no adjustable parameters. It predicts that the environmental lapse rate must be less than half the adiabatic lapse rate for an inversion to form, and it predicts the ratio of the inversion amplitude and thickness as a function of environmental lapse rate. Comparison of this prediction to the actual amplitude/thickness ratio using the lidar measurements shows good agreement between theory and measurements.

Current model capabilities for simulating black carbon and sulfate concentrations in the Arctic atmosphere: a multi-model evaluation using a comprehensive measurement data set

The concentrations of sulfate, black carbon (BC) and other aerosols in the Arctic are characterized by high values in late winter and spring (so-called Arctic Haze) and low values in summer. Models have long been struggling to capture this seasonality and especially the high concentrations associated with Arctic Haze. In this study, we evaluate sulfate and BC concentrations from eleven different models driven with the same emission inventory against a comprehensive pan-Arctic measurement data set over a time period of 2 years (2008–2009). The set of models consisted of one Lagrangian particle dispersion model, four chemistry transport models (CTMs), one atmospheric chemistry-weather forecast model and five chemistry climate models (CCMs), of which two were nudged to meteorological analyses and three were running freely. The measurement data set consisted of surface measurements of equivalent BC (eBC) from five stations (Alert, Barrow, Pallas, Tiksi and Zeppelin), elemental carbon (EC) from Station Nord and Alert and aircraft measurements of refractory BC (rBC) from six different campaigns. We find that the models generally captured the measured eBC or rBC and sulfate concentrations quite well, compared to previous comparisons. However, the aerosol seasonality at the surface is still too weak in most models. Concentrations of eBC and sulfate averaged over three surface sites are underestimated in winter/spring in all but one model (model means for January–March underestimated by 59 and 37 % for BC and sulfate, respectively), whereas concentrations in summer are overestimated in the model mean (by 88 and 44 % for July–September), but with overestimates as well as underestimates present in individual models. The most pronounced eBC underestimates, not included in the above multi-site average, are found for the station Tiksi in Siberia where the measured annual mean eBC concentration is 3 times higher than the average annual mean for all other stations. This suggests an underestimate of BC sources in Russia in the emission inventory used. Based on the campaign data, biomass burning was identified as another cause of the modeling problems. For sulfate, very large differences were found in the model ensemble, with an apparent anti-correlation between modeled surface concentrations and total atmospheric columns. There is a strong correlation between observed sulfate and eBC concentrations with consistent sulfate/eBC slopes found for all Arctic stations, indicating that the sources contributing to sulfate and BC are similar throughout the Arctic and that the aerosols are internally mixed and undergo similar removal. However, only three models reproduced this finding, whereas sulfate and BC are weakly correlated in the other models. Overall, no class of models (e.g., CTMs, CCMs) performed better than the others and differences are independent of model resolution.

Flood Risk Assessment in Coastal Drainage Basins through a Multivariate Analysis within a GIS-Based Model

This paper presents a GIS-based multicriteria flood risk assessment and mapping approach applied to coastal drainage basins where hydrological data are not available. It involves risk to different types of possible processes: coastal inundation (storm surge), river, estuarine and flash flood, either at urban or natural areas, and fords. Based on the causes of these processes, several environmental indicators were taken to build-up the risk assessment. Geoindicators include geological-geomorphologic proprieties of Quaternary sedimentary units, water table, drainage basin morphometry, coastal dynamics, beach morphodynamics and microclimatic characteristics. Bioindicators involve coastal plain and low slope native vegetation categories and two alteration states. Anthropogenic indicators encompass land use categories properties such as: type, occupation density, urban structure type and occupation consolidation degree. The selected indicators were stored within an expert Geoenvironmental Information System developed for the State of Sao Paulo Coastal Zone (SIIGAL), which attributes were mathematically classified through deterministic approaches, in order to estimate natural susceptibilities (Sn), human-induced susceptibilities (Sa), return period of rain events (Ri), potential damages (Dp) and the risk classification (R), according to the equation R=(Sn.Sa.Ri).Dp. Thematic maps were automatically processed within the SIIGAL, in which automata cells (""geoenvironmental management units"") aggregating geological-geomorphologic and land use/native vegetation categories were the units of classification. The method has been applied to the Northern Littoral of the State of Sao Paulo (Brazil) in 32 small drainage basins, demonstrating to be very useful for coastal zone public politics, civil defense programs and flood management.

Robust inference in an heteroscedastic measurement error model

In this paper we deal with robust inference in heteroscedastic measurement error models Rather than the normal distribution we postulate a Student t distribution for the observed variables Maximum likelihood estimates are computed numerically Consistent estimation of the asymptotic covariance matrices of the maximum likelihood and generalized least squares estimators is also discussed Three test statistics are proposed for testing hypotheses of interest with the asymptotic chi-square distribution which guarantees correct asymptotic significance levels Results of simulations and an application to a real data set are also reported (C) 2009 The Korean Statistical Society Published by Elsevier B V All rights reserved

Improved maximum likelihood estimators in a heteroskedastic errors-in-variables model

This paper develops a bias correction scheme for a multivariate heteroskedastic errors-in-variables model. The applicability of this model is justified in areas such as astrophysics, epidemiology and analytical chemistry, where the variables are subject to measurement errors and the variances vary with the observations. We conduct Monte Carlo simulations to investigate the performance of the corrected estimators. The numerical results show that the bias correction scheme yields nearly unbiased estimates. We also give an application to a real data set.

Inference and local influence assessment in skew-normal null intercept measurement error model

In this article, we discuss inferential aspects of the measurement error regression models with null intercepts when the unknown quantity x (latent variable) follows a skew normal distribution. We examine first the maximum-likelihood approach to estimation via the EM algorithm by exploring statistical properties of the model considered. Then, the marginal likelihood, the score function and the observed information matrix of the observed quantities are presented allowing direct inference implementation. In order to discuss some diagnostics techniques in this type of models, we derive the appropriate matrices to assessing the local influence on the parameter estimates under different perturbation schemes. The results and methods developed in this paper are illustrated considering part of a real data set used by Hadgu and Koch [1999, Application of generalized estimating equations to a dental randomized clinical trial. Journal of Biopharmaceutical Statistics, 9, 161-178].

Semiparametric Bayesian measurement error modeling

This work presents a Bayesian semiparametric approach for dealing with regression models where the covariate is measured with error. Given that (1) the error normality assumption is very restrictive, and (2) assuming a specific elliptical distribution for errors (Student-t for example), may be somewhat presumptuous; there is need for more flexible methods, in terms of assuming only symmetry of errors (admitting unknown kurtosis). In this sense, the main advantage of this extended Bayesian approach is the possibility of considering generalizations of the elliptical family of models by using Dirichlet process priors in dependent and independent situations. Conditional posterior distributions are implemented, allowing the use of Markov Chain Monte Carlo (MCMC), to generate the posterior distributions. An interesting result shown is that the Dirichlet process prior is not updated in the case of the dependent elliptical model. Furthermore, an analysis of a real data set is reported to illustrate the usefulness of our approach, in dealing with outliers. Finally, semiparametric proposed models and parametric normal model are compared, graphically with the posterior distribution density of the coefficients. (C) 2009 Elsevier Inc. All rights reserved.