L'agrégation de l’anémie dans un ménage est soulignée lorsque l’anémie touche l’ensemble du ménage. La littérature scientifique fait état d’une coexistence des facteurs nutritionnels et infectieux dans l’étiologie de l’anémie. Cependant, si la cause principale de l’anémie dans une population est la carence en fer, les enfants et les femmes seront beaucoup plus affectés que les hommes. Si par contre l’anémie est liée à une cause infectieuse qui touche toute la population, l’anémie atteindra également les hommes. Ce travail a été entrepris pour vérifier l’hypothèse selon laquelle l’anémie ne serait pas spécifique à la carence en fer dans le cas où elle se concentrerait à l’ensemble du ménage. Cette étude porte sur des données d’enquête collectées au Cameroun. Nos analyses sont basées sur un sous échantillon de 2331 sujets, dont 777 femmes, 777 hommes et 777 enfants. La prévalence de l’anémie était de 53,5% chez les enfants, 39,5% chez les femmes et 18,3% chez les hommes. L’anémie était concentrée dans 34% des ménages. Le programme SPSS version 17.0 et plus particulièrement l’analyse de régression logistique a servi à tester l’impact de chaque groupe de variables (facteurs liés à l’individu, au ménage et à la communauté) sur l’agrégation de l’anémie dans les ménages. Les résultats de cette étude suggèrent que l’agrégation de l’anémie s’observerait surtout quand la santé de l’enfant est compromise. Le risque d’agrégation y est 4 fois plus élevé dans les foyers où les enfants ont un déficit pondéral et 6 fois plus élevé dans les ménages où les enfants présentent une fièvre. Le fait d’appartenir au sud forestier et à un ménage de niveau socio-économique moyen constituerait également des facteurs de risque d’agrégation. / The aggregation of anemia in a household is underlined when the anemia affects the entire household. The involvement of dietary and infectious factors in the etiology of anemia was demonstrated in the literature. Nevertheless, if the main cause of anemia in a population is iron deficiency, children and women are much more affected than men. If instead of that, anemia is associated with an infectious cause that affects the entire population, anemia will also reach the men. This work was initiated to test the hypothesis that, anemia is not specific to iron deficiency when it affects the entire household. This study examines survey data collected in Cameroon. Our analysis is based on a sub sample of 2331 subjects, including 777 women, 777 men and 777 children. The prevalence of anemia was 58% in children, 48.48% for women and 20.82% for men. Anemia was found concentrated in 34% of households. The SPSS program version 17.0 and specifically the logistic regression analysis were used to test the impact of each group of variables (factors related to the individual, health, household and community) on the aggregation of anemia in the households. The results of this study suggest that the aggregation of anemia occurs mainly when the child's health is compromised. The risk of aggregation is four times higher in homes where children are underweight and 6 times higher in households where children have a fever. Being from the region of southern's forest and belong to a household with average socioeconomic level would also be the risks factors for the aggregation.

This study formed part of a larger project that is concerned with the adaptation of a test of cognitive academic language proficiency, the Woodcock Muñoz Language Survey (WMLS). The WMLS has been adapted from English into isiXhosa and the present study is located within the broader study that is concerned with establishing overall equivalence between the two language versions of the WMLS. It was primarily concerned with the Verbal Analogies (VA) scale. Previous research on this scale has demonstrated promising results, but continues to find evidence of some inequivalence. This study aimed to cross-validate previous research on the two language versions of the WMLS and improve on methodological issues by employing matched groups. It drew upon an existing dataset from the larger research project. The study employed a monolingual matched two-group design consisting of 150 mainly English speaking and 149 mainly isiXhosa learners in grades 6 and 7. This study had two sub aims. The first was to investigate item bias by identifying DIF items in the VA scale across the isiXhosa and English by conducting a logistic regression and Mantel-Haenszel procedure. Five items were identified by both techniques as DIF. The second sub aim was to evaluate construct equivalence between the isiXhosa and English versions of the WMLS on the VA scale by conducting a factor analysis on the tests after removal of DIF items. Two factors were requested during the factor analysis. The first factor displayed significant loadings across both language versions and was identified as a stable factor. This was confirmed by the Tucker’s Phi and scatter plot. The second factor was stable for the English version but not for the isiXhosa version. The Tucker’s phi and scatter plot indicated that this factor is not structurally equivalent across the two language versions / Magister Artium (Psychology) - MA(Psych)

One of the difficult issues in detection theory is to design a robust detector that takes into account the actual distribution of the original data. The most commonly used statistical detection model for blind detection is Gaussian distribution. Specifically, linear correlation is an optimal detection method in the presence of Gaussian distributed features. This has been found to be sub-optimal detection metric when density deviates completely from Gaussian distributions. Hence, we formulate a detection algorithm that enhances detection probability by exploiting the true characterises of the original data. To understand the underlying distribution function of data, we employed the estimation techniques such as parametric model called approximated density ratio logistic regression model and semiparameric estimations. Semiparametric model has the advantages of yielding density ratios as well as individual densities. Both methods are applicable to signals such as watermark embedded in spatial domain and outperform the conventional linear correlation non-Gaussian distributed.

This thesis focuses on two statistical problems related to credit scoring. In credit scoring of individuals, two classes are distinguished, namely low and high risk individuals (the so-called "good" and "bad" risk classes). Firstly, we suggest a measure which may be used to study the nature of a classifier for distinguishing between the two risk classes. Secondly, we derive a new method DOUW (detecting outliers using weights) which may be used to fit logistic regression models robustly and for the detection of outliers. In the first problem, the focus is on a measure which may be used to study the nature of a classifier. This measure transforms a random variable so that it has the same distribution as another random variable. Assuming a linear form of this measure, three methods for estimating the parameters (slope and intercept) and for constructing confidence bands are developed and compared by means of a Monte Carlo study. The application of these estimators is illustrated on a number of datasets. We also construct statistical hypothesis to test this linearity assumption. In the second problem, the focus is on providing a robust logistic regression fit and the identification of outliers. It is well-known that maximum likelihood estimators of logistic regression parameters are adversely affected by outliers. We propose a robust approach that also serves as an outlier detection procedure and is called DOUW. The approach is based on associating high and low weights with the observations as a result of the likelihood maximization. It turns out that the outliers are those observations to which low weights are assigned. This procedure depends on two tuning constants. A simulation study is presented to show the effects of these constants on the performance of the proposed methodology. The results are presented in terms of four benchmark datasets as well as a large new dataset from the application area of retail marketing campaign analysis. In the last chapter we apply the techniques developed in this thesis on a practical credit scoring dataset. We show that the DOUW method improves the classifier performance and that the measure developed to study the nature of a classifier is useful in a credit scoring context and may be used for assessing whether the distribution of the good and the bad risk individuals is from the same translation-scale family. / Thesis (Ph.D. (Risk Analysis))--North-West University, Potchefstroom Campus, 2008.

Basal Temperature of Snow (BTS) measurements were used as the primary inputs to a high resolution (30 x 30 m grid cells) empirical-statistical regional permafrost probability model for the southern and central Yukon, and northernmost British Columbia (59° - 65°N). Data from seven individual study areas distributed across the region were combined using a blended distance decay technique, with an eighth area used for validation. The model predictions are reasonably consistent with previous permafrost maps for the area with some notable differences and a much higher level of detail. The modelling gives an overall permafrost probability of 52%. North of 62°N, permafrost becomes more extensive in the lowland areas whereas farther south permafrost is typically common only above treeline. Significant differences exist between the mountain environments of the Yukon and the Swiss Alps where the BTS method originated and as a result different modelling approaches had to be developed. This work therefore: (1) develops additional explanatory variables for permafrost probability modelling, the most notable of which is equivalent elevation, (2) confirms the use of ground truthing as a requirement for empirical-statistical modelling in the Yukon and (3) uses a combination of models for the region in order to spatially predict between study areas. The results of this thesis will be of use to linear infrastructure route-planning, geohazard assessment and climate change adaptation strategies. Future work employing the model will allow the effects of scenario-based climate warming to be examined.

This thesis focuses on two statistical problems related to credit scoring. In credit scoring of individuals, two classes are distinguished, namely low and high risk individuals (the so-called "good" and "bad" risk classes). Firstly, we suggest a measure which may be used to study the nature of a classifier for distinguishing between the two risk classes. Secondly, we derive a new method DOUW (detecting outliers using weights) which may be used to fit logistic regression models robustly and for the detection of outliers. In the first problem, the focus is on a measure which may be used to study the nature of a classifier. This measure transforms a random variable so that it has the same distribution as another random variable. Assuming a linear form of this measure, three methods for estimating the parameters (slope and intercept) and for constructing confidence bands are developed and compared by means of a Monte Carlo study. The application of these estimators is illustrated on a number of datasets. We also construct statistical hypothesis to test this linearity assumption. In the second problem, the focus is on providing a robust logistic regression fit and the identification of outliers. It is well-known that maximum likelihood estimators of logistic regression parameters are adversely affected by outliers. We propose a robust approach that also serves as an outlier detection procedure and is called DOUW. The approach is based on associating high and low weights with the observations as a result of the likelihood maximization. It turns out that the outliers are those observations to which low weights are assigned. This procedure depends on two tuning constants. A simulation study is presented to show the effects of these constants on the performance of the proposed methodology. The results are presented in terms of four benchmark datasets as well as a large new dataset from the application area of retail marketing campaign analysis. In the last chapter we apply the techniques developed in this thesis on a practical credit scoring dataset. We show that the DOUW method improves the classifier performance and that the measure developed to study the nature of a classifier is useful in a credit scoring context and may be used for assessing whether the distribution of the good and the bad risk individuals is from the same translation-scale family. / Thesis (Ph.D. (Risk Analysis))--North-West University, Potchefstroom Campus, 2008.

This thesis makes significant contributions to improving the use of Airport Safety Areas (ASAs) as aviation accident risk mitigation measures by developing improved accident frequency models and risk assessment methodologies. In recent years, the adequacy of ASAs such as the Runway End Safety Area and Runway Safety Area has come under increasing scrutiny. The current research found flaws in the existing ASA regulations and airport risk assessment techniques that lead to the provision of inconsistent safety margins at airports and runways. The research was based on a comprehensive database of ASA-related accidents, which was matched by a representative sample of normal operations data, such that the exposure to a range of operational and meteorological risk factors between accident and normal flights could be compared. On this basis, the criticality of individual risk factors was quantified and accident frequency models were developed using logistic regression. These models have considerably better predictive power compared to models used by previous airport risk assessments. An improved risk assessment technique was developed coupling the accident frequency models with accident location data, yielding distributions that describe the frequency of accidents that reach specific distances beyond the runway end or centreline given the risk exposure profile of the particular runway. The application of the proposed methodology was demonstrated in two case studies. Specific recommendations on ASA dimensions were made for achieving consistent levels of safety on each side of the runway. Advances made in this study have implications on the overall assessment and management of risks at airports.

One of the difficult issues in detection theory is to design a robust detector that takes into account the actual distribution of the original data. The most commonly used statistical detection model for blind detection is Gaussian distribution. Specifically, linear correlation is an optimal detection method in the presence of Gaussian distributed features. This has been found to be sub-optimal detection metric when density deviates completely from Gaussian distributions. Hence, we formulate a detection algorithm that enhances detection probability by exploiting the true characterises of the original data. To understand the underlying distribution function of data, we employed the estimation techniques such as parametric model called approximated density ratio logistic regression model and semiparameric estimations. Semiparametric model has the advantages of yielding density ratios as well as individual densities. Both methods are applicable to signals such as watermark embedded in spatial domain and outperform the conventional linear correlation non-Gaussian distributed.

This study investigated international medical graduates (IMGs), registered between January 1, 2007 and April 14, 2011, at the Access Centre for Internationally Educated Health Professionals in Ontario. By way of logistic regression in a cross-sectional design, it was found that permanent residents who were recent immigrants had lesser chances of being employed full-time at registration (baseline). By way of survival analysis in a cohort design, it was found that younger IMGs who have been in Canada less than 5 years and who have taken the Medical Council of Canada Evaluating Exam (MCCEE) have the greatest chances of securing residency positions in Canada or the US, whereas IMGs from Eastern Europe, South Asia and Africa have lesser chances. It was revealed that registered IMGs are a vulnerable population, and certain groups may be disadvantaged due to underlying characteristics. These groups can be targeted for specific interventions.

The negative social and economic results of traffic accidents are the most serious problems within the concept of traffic safety. Every year, unfortunately, a huge number of traffic accidents result in destructive losses. Especially, when the holiness of human life is concerned, traffic safety has an invaluable role for the traffic improvement strategies. In this manner, Turkey places one of the highest ranks regarding the growing rate and severity of traffic accidents that should be immediately taken under control. In this study, an innovative model that constructs a hybrid between the spatial and statistical analyses is developed in order to examine the importance of enhancing statistical analysis with georeferenced data and so location-based studies in traffic accident analysis. Meanwhile, the effects of road characteristic and environment are considered for exploring the integral role of roadway factor to the occurrence of accidents, and consequently for emphasizing easily applicable and controllable engineering safety measures. Because of the rare and random distribution of traffic accident data, logistic regression is used for the statistical part of the study in order to find the pairwise risk factors among the roadway and environmental parameters. After unifying these relative risk factors with the logic of Analytic Hierarchy Process, the finalized accident risk factors are attached to the digitized road characteristics map through Geographic Information Systems (GIS). The abilities of GIS in mapping, displaying and overlaying different data sets ensure to visualize high risked accident areas with their corresponding potential causal factors. The integration of statistical and spatial analyses is essential for developing appropriate and effective precautions in addition to its easily understandable, applicable and modifiable structure. Finally, the model is proven to be appropriate for both interpreting the existing traffic accident problem or potential future accidents and also developing comprehensive and reliable location-based safety studies.