Joint Regression and Association Models for Repeated Categorical Responses

The focus of this study is on statistical analysis of categorical responses, where the response values are dependent of each other. The most typical example of this kind of dependence is when repeated responses have been obtained from the same study unit. For example, in Paper I, the response of interest is the pneumococcal nasopharengyal carriage (yes/no) on 329 children. For each child, the carriage is measured nine times during the first 18 months of life, and thus repeated respones on each child cannot be assumed independent of each other. In the case of the above example, the interest typically lies in the carriage prevalence, and whether different risk factors affect the prevalence. Regression analysis is the established method for studying the effects of risk factors. In order to make correct inferences from the regression model, the associations between repeated responses need to be taken into account. The analysis of repeated categorical responses typically focus on regression modelling. However, further insights can also be gained by investigating the structure of the association. The central theme in this study is on the development of joint regression and association models. The analysis of repeated, or otherwise clustered, categorical responses is computationally difficult. Likelihood-based inference is often feasible only when the number of repeated responses for each study unit is small. In Paper IV, an algorithm is presented, which substantially facilitates maximum likelihood fitting, especially when the number of repeated responses increase. In addition, a notable result arising from this work is the freely available software for likelihood-based estimation of clustered categorical responses.

Novel Mass Spectrometric Analysis Methods for Anabolic Androgenic Steroids in Sports Drug Testing

The feasibility of different modern analytical techniques for the mass spectrometric detection of anabolic androgenic steroids (AAS) in human urine was examined in order to enhance the prevalent analytics and to find reasonable strategies for effective sports drug testing. A comparative study of the sensitivity and specificity between gas chromatography (GC) combined with low (LRMS) and high resolution mass spectrometry (HRMS) in screening of AAS was carried out with four metabolites of methandienone. Measurements were done in selected ion monitoring mode with HRMS using a mass resolution of 5000. With HRMS the detection limits were considerably lower than with LRMS, enabling detection of steroids at low 0.2-0.5 ng/ml levels. However, also with HRMS, the biological background hampered the detection of some steroids. The applicability of liquid-phase microextraction (LPME) was studied with metabolites of fluoxymesterone, 4-chlorodehydromethyltestosterone, stanozolol and danazol. Factors affecting the extraction process were studied and a novel LPME method with in-fiber silylation was developed and validated for GC/MS analysis of the danazol metabolite. The method allowed precise, selective and sensitive analysis of the metabolite and enabled simultaneous filtration, extraction, enrichment and derivatization of the analyte from urine without any other steps in sample preparation. Liquid chromatographic/tandem mass spectrometric (LC/MS/MS) methods utilizing electrospray ionization (ESI), atmospheric pressure chemical ionization (APCI) and atmospheric pressure photoionization (APPI) were developed and applied for detection of oxandrolone and metabolites of stanozolol and 4-chlorodehydromethyltestosterone in urine. All methods exhibited high sensitivity and specificity. ESI showed, however, the best applicability, and a LC/ESI-MS/MS method for routine screening of nine 17-alkyl-substituted AAS was thus developed enabling fast and precise measurement of all analytes with detection limits below 2 ng/ml. The potential of chemometrics to resolve complex GC/MS data was demonstrated with samples prepared for AAS screening. Acquired full scan spectral data (m/z 40-700) were processed by the OSCAR algorithm (Optimization by Stepwise Constraints of Alternating Regression). The deconvolution process was able to dig out from a GC/MS run more than the double number of components as compared with the number of visible chromatographic peaks. Severely overlapping components, as well as components hidden in the chromatographic background could be isolated successfully. All studied techniques proved to be useful analytical tools to improve detection of AAS in urine. Superiority of different procedures is, however, compound-dependent and different techniques complement each other.

Generalised regression estimation for domain class frequencies

This study examines the properties of Generalised Regression (GREG) estimators for domain class frequencies and proportions. The family of GREG estimators forms the class of design-based model-assisted estimators. All GREG estimators utilise auxiliary information via modelling. The classic GREG estimator with a linear fixed effects assisting model (GREG-lin) is one example. But when estimating class frequencies, the study variable is binary or polytomous. Therefore logistic-type assisting models (e.g. logistic or probit model) should be preferred over the linear one. However, other GREG estimators than GREG-lin are rarely used, and knowledge about their properties is limited. This study examines the properties of L-GREG estimators, which are GREG estimators with fixed-effects logistic-type models. Three research questions are addressed. First, I study whether and when L-GREG estimators are more accurate than GREG-lin. Theoretical results and Monte Carlo experiments which cover both equal and unequal probability sampling designs and a wide variety of model formulations show that in standard situations, the difference between L-GREG and GREG-lin is small. But in the case of a strong assisting model, two interesting situations arise: if the domain sample size is reasonably large, L-GREG is more accurate than GREG-lin, and if the domain sample size is very small, estimation of assisting model parameters may be inaccurate, resulting in bias for L-GREG. Second, I study variance estimation for the L-GREG estimators. The standard variance estimator (S) for all GREG estimators resembles the Sen-Yates-Grundy variance estimator, but it is a double sum of prediction errors, not of the observed values of the study variable. Monte Carlo experiments show that S underestimates the variance of L-GREG especially if the domain sample size is minor, or if the assisting model is strong. Third, since the standard variance estimator S often fails for the L-GREG estimators, I propose a new augmented variance estimator (A). The difference between S and the new estimator A is that the latter takes into account the difference between the sample fit model and the census fit model. In Monte Carlo experiments, the new estimator A outperformed the standard estimator S in terms of bias, root mean square error and coverage rate. Thus the new estimator provides a good alternative to the standard estimator.

Methods in general model localization

The aim of this study was to evaluate and test methods which could improve local estimates of a general model fitted to a large area. In the first three studies, the intention was to divide the study area into sub-areas that were as homogeneous as possible according to the residuals of the general model, and in the fourth study, the localization was based on the local neighbourhood. According to spatial autocorrelation (SA), points closer together in space are more likely to be similar than those that are farther apart. Local indicators of SA (LISAs) test the similarity of data clusters. A LISA was calculated for every observation in the dataset, and together with the spatial position and residual of the global model, the data were segmented using two different methods: classification and regression trees (CART) and the multiresolution segmentation algorithm (MS) of the eCognition software. The general model was then re-fitted (localized) to the formed sub-areas. In kriging, the SA is modelled with a variogram, and the spatial correlation is a function of the distance (and direction) between the observation and the point of calculation. A general trend is corrected with the residual information of the neighbourhood, whose size is controlled by the number of the nearest neighbours. Nearness is measured as Euclidian distance. With all methods, the root mean square errors (RMSEs) were lower, but with the methods that segmented the study area, the deviance in single localized RMSEs was wide. Therefore, an element capable of controlling the division or localization should be included in the segmentation-localization process. Kriging, on the other hand, provided stable estimates when the number of neighbours was sufficient (over 30), thus offering the best potential for further studies. Even CART could be combined with kriging or non-parametric methods, such as most similar neighbours (MSN).

Kalman Filter Algorithm for Rating and Prediction in Basketball

The Thesis presents a state-space model for a basketball league and a Kalman filter algorithm for the estimation of the state of the league. In the state-space model, each of the basketball teams is associated with a rating that represents its strength compared to the other teams. The ratings are assumed to evolve in time following a stochastic process with independent Gaussian increments. The estimation of the team ratings is based on the observed game scores that are assumed to depend linearly on the true strengths of the teams and independent Gaussian noise. The team ratings are estimated using a recursive Kalman filter algorithm that produces least squares optimal estimates for the team strengths and predictions for the scores of the future games. Additionally, if the Gaussianity assumption holds, the predictions given by the Kalman filter maximize the likelihood of the observed scores. The team ratings allow probabilistic inference about the ranking of the teams and their relative strengths as well as about the teams’ winning probabilities in future games. The predictions about the winners of the games are correct 65-70% of the time. The team ratings explain 16% of the random variation observed in the game scores. Furthermore, the winning probabilities given by the model are concurrent with the observed scores. The state-space model includes four independent parameters that involve the variances of noise terms and the home court advantage observed in the scores. The Thesis presents the estimation of these parameters using the maximum likelihood method as well as using other techniques. The Thesis also gives various example analyses related to the American professional basketball league, i.e., National Basketball Association (NBA), and regular seasons played in year 2005 through 2010. Additionally, the season 2009-2010 is discussed in full detail, including the playoffs.

Non-industrial private forest landowners’ choices of timber management strategies.

The factors affecting the non-industrial, private forest landowners' (hereafter referred to using the acronym NIPF) strategic decisions in management planning are studied. A genetic algorithm is used to induce a set of rules predicting potential cut of the landowners' choices of preferred timber management strategies. The rules are based on variables describing the characteristics of the landowners and their forest holdings. The predictive ability of a genetic algorithm is compared to linear regression analysis using identical data sets. The data are cross-validated seven times applying both genetic algorithm and regression analyses in order to examine the data-sensitivity and robustness of the generated models. The optimal rule set derived from genetic algorithm analyses included the following variables: mean initial volume, landowner's positive price expectations for the next eight years, landowner being classified as farmer, and preference for the recreational use of forest property. When tested with previously unseen test data, the optimal rule set resulted in a relative root mean square error of 0.40. In the regression analyses, the optimal regression equation consisted of the following variables: mean initial volume, proportion of forestry income, intention to cut extensively in future, and positive price expectations for the next two years. The R2 of the optimal regression equation was 0.34 and the relative root mean square error obtained from the test data was 0.38. In both models, mean initial volume and positive stumpage price expectations were entered as significant predictors of potential cut of preferred timber management strategy. When tested with the complete data set of 201 observations, both the optimal rule set and the optimal regression model achieved the same level of accuracy.

Almost Stable Matchings by Truncating the Gale–Shapley Algorithm

We show that the ratio of matched individuals to blocking pairs grows linearly with the number of propose–accept rounds executed by the Gale–Shapley algorithm for the stable marriage problem. Consequently, the participants can arrive at an almost stable matching even without full information about the problem instance; for each participant, knowing only its local neighbourhood is enough. In distributed-systems parlance, this means that if each person has only a constant number of acceptable partners, an almost stable matching emerges after a constant number of synchronous communication rounds. We apply our results to give a distributed (2 + ε)-approximation algorithm for maximum-weight matching in bicoloured graphs and a centralised randomised constant-time approximation scheme for estimating the size of a stable matching.

A local 2-approximation algorithm for the vertex cover problem

We present a distributed 2-approximation algorithm for the minimum vertex cover problem. The algorithm is deterministic, and it runs in (Δ + 1)2 synchronous communication rounds, where Δ is the maximum degree of the graph. For Δ = 3, we give a 2-approximation algorithm also for the weighted version of the problem.

A simple local 3-approximation algorithm for vertex cover

We present a local algorithm (constant-time distributed algorithm) for finding a 3-approximate vertex cover in bounded-degree graphs. The algorithm is deterministic, and no auxiliary information besides port numbering is required. (c) 2009 Elsevier B.V. All rights reserved.

A local 2-approximation algorithm for the vertex cover problem

We present a distributed 2-approximation algorithm for the minimum vertex cover problem. The algorithm is deterministic, and it runs in (Δ + 1)2 synchronous communication rounds, where Δ is the maximum degree of the graph. For Δ = 3, we give a 2-approximation algorithm also for the weighted version of the problem.

An optimal local approximation algorithm for max-min linear programs

In a max-min LP, the objective is to maximise ω subject to Ax ≤ 1, Cx ≥ ω1, and x ≥ 0 for nonnegative matrices A and C. We present a local algorithm (constant-time distributed algorithm) for approximating max-min LPs. The approximation ratio of our algorithm is the best possible for any local algorithm; there is a matching unconditional lower bound.