Till Örebro universitet

I denna studie har undersökningarna Kortperiodisk Sysselsättningsstatistiks (KS) och Konjunkturstatistik över Vakansers (KV) befintliga granskningssystem utvärderats med avseende på hur effektivt det är. Processdata har framställts och analyserats. Resultaten tyder på att många av de inkomna blanketterna med misstänkt felaktiga uppgifter inte rättas upp, utan tvingas igenom trots att granskningssystemet ej accepterade uppgifterna. Det befintliga granskningssystemet har en högre träffsäkerhet avseende KS-undersökningen, men både KS och KV skulle kunnas granskas mer effektivt.

För att utvärdera det befintliga granskningssystemet ytterligare användes en poängfunktion. Till studien fanns tillgång till både helt ogranskat material och helt granskat material och dessa material användes i poängfunktionen. Det uppräknade ogranskade värdet för varje objekt jämfördes med det uppräknade granskade värdet och ställdes i relation till respektive skattade branschtotal. De poängsatta blanketterna rangordnades sedan. Därefter analyserades materialet för att försöka finna var det skulle vara lämpligt att sätta det tröskelvärde som skulle skilja det material som ”egentligen” skulle ha behövts granskas från det som kunde ha lämnats orört. Att sätta tröskelvärdet är svårt. Här gjordes det godtyckligt utifrån kriterierna att det fel som införs i skattningarna för att allt material inte granskas skulle hållas så lågt som möjligt samt att antalet blanketter som skulle behöva granskas manuellt av produktionsgruppen också skulle hållas så lågt som möjligt. Även här visade det sig att det befintliga granskningssystemet inte är så effektivt som önskas. När resultaten från denna del av utvärderingen analyserades upptäcktes problem som beror på blankettutformningen. Skulle blanketterna ses över och åtgärdas skulle det fel som införs för att allt material inte granskas kunna minskas avsevärt. Genom att minska det införda felet kan tröskelvärdet förmodligen sättas på en ny nivå vilket medför att omfattningen av granskningen skulle minska ytterligare.

Hur skulle då ett mer effektivt granskningssystem kunna se ut? I den här studien har valet fallit på att testa ”significance editing” på KS-undersökningen, det som på svenska kallas för effektgranskning. En poängfunktion användes även här, denna tilldelar de inkomna blanketterna varsin poäng och dessa poäng rangordnas därefter. Efter att poängen rangordnats bestäms en gräns, ett tröskelvärde, och de blanketter med en poäng som överstiger tröskelvärdet granskas och rättas upp av produktionsgruppen. De blanketter med en poäng som understiger det satta tröskelvärdet rättas inte upp, utan behåller sina originalvärden. Poängfunktionen jämför det inkomna ogranskade, uppräknade, värdet med ett uppräknat ”förväntat” värde och ställer denna differens i relation till den skattade branschtotalen. Svårigheten ligger ofta i att hitta ett bra förväntat värde och detta problem uppstår ideligen i urvalsundersökningar. Tanken med effektgranskning är att omfattningen av granskningen ska minska och den granskning som utförs ska ha effekt på slutresultatet.

Det var inte lätt att hitta ett bra förväntat värde på den tid som stod till förfogande. Två problem som snabbt upptäcktes var dels att i KS-undersökningen finns inte uträknade säsongs- eller trendfaktorer per variabel. Dessutom byttes en mycket stor del av urvalet ut till kvartal 2 (som denna studie har avgränsats till att behandla). Detta har fått till följd att cirka hälften av objekten i urvalet inte går att följa bakåt i tiden eftersom de inte ingått i urvalet tidigare. I studien har respektive stratums medelvärde använts som förväntat värde. Resultaten visar att det valda förväntade värdet inte skulle ha använts i praktiken, men det fungerar bra i syfte att illustrera hur det i praktiken skulle kunna gå till att införa en mer effektiv granskning.

This paper presents the computation of likelihood prediction with the generalized linear and mixed models. The method of likelihood prediction is briefy discussed and approximate formulae are provided to make easy computation of the likelihoodprediction with generalized linear models. For complicated prediction problems, simulation methods are suggested. An R add-in package is accompanied to carryout the computation of the predictive inference with the generalized linear and mixed models. The likelihood prediction is applied to the prediction of the credit defaults using a real data set. Results show that the predictive likelihood can be a useful tool to predict portfolio credit risk.

This thesis deals with developing and testing feasible computational procedures to facilitate the estimation of and carry out the prediction with the generalized linear mixed model (GLMM) with a scope of applying them to large data sets. The work of this thesis is motivated from an issue arising incredit risk modelling. We have access to a huge data set, consisting of about one million observations, on credit history obtained from two major Swedish banks. The principal research interest involved with the data analysis is to model the probability of credit defaults by incorporating the systematic dependencies among the default events. In order to model the dependent credit defaults we adopt the framework of GLMM which is apopular approach to model correlated binary data. However, existing computational procedures for GLMM did not offer us the flexibility to incorporate the desired correlation structure of defaults events.For the feasible estimation of the GLMM we propose two estimation techniques being the fixed effects (FE) approach and the two-step pseudolikelihood approach (2PL). The preciseness of the estimation techniques and their computational advantages are studied by Monte-Carlo simulations and by applying them to the credit risk modelling. Regarding the prediction issue, we show how to apply the likelihood principle to carryout prediction with GLMM. We also provide an R add-in package to facilitate the predictive inference for GLMM.

This paper presents a two-step pseudo likelihood estimation for generalized linear mixed models with the random effects being correlated between groups. The core idea is to deal with the random intractable integrals in  the likelihood function by multivariate Taylor's approximation. The accuracy of the estimation technique is assessed in a Monte-Carlo study: An application of it with binary response variable is presented using a real dara set on credit defaults from two Swedish banks. Thanks to   the use of two-step estimation technique, the proposed algorithm outperforms conventional likelihood algoritms in terms of computational time.

It is commonly agreed that the credit defaults are correlated. However, the structure and magnitude of such dependence is not yet fully understood. This paper contributes to the current understanding about the defaults comovement in the following way. Assuming that the industries provides the basis of defaults comovement it provides empirical evidence as to how such comovements can be modeled using correlated industry shocks. Generalized linear mixed model (GLMM) with correlated random effects is used to model the defaults comovement. It is also demonstrated as to how a GLMM with complex correlation structure can be estimated through a very simple way. Empirical evidences are drawn through analyzing quarterly individual borrower level credit history data obtained from two major Swedish banks between the period 1994 and 2000. The results show that, conditional on the borrower level accounting data and macro business cycle variables, the defaults are correlated both within and between industries but not over time (quarters). A discussion has also been presented as to how a GLMM for defaults correlation can be explained.

This paper presents the techniques of likelihood prediction for the generalized linear mixed models. Methods of likelihood prediction is explained through a series of examples; from a classical one to more complicated ones. The examples show, in simple cases, that the likelihood prediction (LP) coincides with already known best frequentist practice such as the best linear unbiased predictor. The paper outlines a way to deal with the covariate uncertainty while producing predictive inference. Using a Poisson error-in-variable general-ized linear model, it has been shown that in complicated cases LP produces better results than already know methods.

In this paper, we discuss how a regression model, with a non-continuous response variable, which allows for dependency between observations, should be estimated when observations are clustered and measurements on the subjects are repeated. The cluster sizes are assumed to be large.We find that the conventional estimation technique suggested by the literature on generalized linear mixed models(GLMM) is slow and sometimes fails due to non-convergence and lack of memory on standard PCs.We suggest to estimate the random effects as fixed effects by generalized linear model and to derive the covariance matrix from these estimates.A simulation study shows that our proposal is feasible in terms of mean-square error and computation time.We recommend that our proposal be implemented in the software of GLMM techniques so that the estimation procedure can switch between the conventional technique and our proposal, depending on the size of the clusters.

Realized covariance matrices are often constructed under the assumption that richness of intra-day return data is greater than the portfolio size, resulting in nonsingular matrix measures. However, when for example the portfolio size is large, assets suffer from illiquidity issues, or market microstructure noise deters sampling on very high frequencies, this relation is not guaranteed. Under these common conditions, realized covariance matrices may obtain as singular by construction. Motivated by this situation, we introduce the Singular Conditional Autoregressive Wishart (SCAW) model to capture the temporal dynamics of time series of singular realized covariance matrices, extending the rich literature on econometric Wishart time series models to the singular case. This model is furthermore developed by covariance targeting adapted to matrices and a sector wise BEKK-specification, allowing excellent scalability to large and extremely large portfolio sizes. Finally, the model is estimated to a 20-year long time series containing 50 stocks and to a 10-year long time series containing 300 stocks, and evaluated using out-of-sample forecast accuracy. It outperforms the benchmark models with high statistical significance and the parsimonious specifications perform better than the baseline SCAW model, while using considerably less parameters.

This paper investigates the problem of calibrating sensors affected by (i) heteroskedastic measurement noise and (ii) a polynomial bias, describing a systematic distortion of the measured quantity. First, a set of increasingly complex statistical models for the measurement process was proposed. Then, for each model the authors design a Bayesian parameters estimation method handling heteroskedasticity and capable to exploit prior information about the model parameters. The Bayesian problem is solved using MCMC methods and reconstructing the unknown parameters posterior in sampled form. The authors then test the proposed techniques on a practically relevant case study, the calibration of Light Detection and Ranging (Lidar) sensor, and evaluate the different proposed procedures using both artificial and field data.

One of the challenges with predictive modeling is how to quantify the reliability of the models' predictions on new objects. In this work we give an introduction to conformal prediction, a framework that sits on top of traditional machine learning algorithms and which outputs valid confidence estimates to predictions from QSAR models in the form of prediction intervals that are specific to each predicted object. For regression, a prediction interval consists of an upper and a lower bound. For classification, a prediction interval is a set that contains none, one, or many of the potential classes. The size of the prediction interval is affected by a user-specified confidence/significance level, and by the nonconformity of the predicted object; i.e., the strangeness as defined by a nonconformity function. Conformal prediction provides a rigorous and mathematically proven framework for in silico modeling with guarantees on error rates as well as a consistent handling of the models' applicability domain intrinsically linked to the underlying machine learning model. Apart from introducing the concepts and types of conformal prediction, we also provide an example application for modeling ABC transporters using conformal prediction, as well as a discussion on general implications for drug discovery.

Minskade anslag gör att den årliga undersökningen om våra levnadsförhållanden hotas att halveras. Det kan drabba redan svaga grupper som äldre, invandrare och ensamstående föräldrar.

This paper proposes several resampling algorithms suitable for error component models and evaluates them in the context of bootstrap testing. In short, all the algorithms work well and lead to tests with correct or close to correct size. There is thus little or no reason not to use the bootstrap with error component models.

The assumption of approximate normality of a pivotal used for constructing confidence intervals or tests often is violated by a substantial correlation between the point estimator at hand and the estimator of its variance. This can be caused by underlying skewness of data, generating both bias and skewness of the pivotal and leading to, for example, confidence intervals with poorer coverage properties, especially for one-sided intervals. We present in a general setting a simple, yet effective procedure that takes into account this correlation and leads to alternative adjusted intervals with more appealing coverage properties. We provide examples, including a simulation study.

We use a rich longitudinal data set following a cohort of Swedish women from age 10 to 49 to analyse the effects of birth and early-life conditions on adulthood outcomes. These latter include both well-being and the stress hormone cortisol. Employment and marital status are important adult determinants of well-being. Log family income and absence from school also predict adult well-being, although their importance falls when controlling for adult and birth characteristics. Among the birth characteristics, we find that high birth weight (>4.3kg) affects adult well-being. We predict the level of adult cortisol only poorly, and suggest that the relationship between life satisfaction and cortisol is non-monotonic: both high and low cortisol are negatively correlated with life satisfaction. The results from an OLS life satisfaction regression and a multinomial logit of high or low cortisol (as compared to medium) are more similar to each other.

Our study is, to our knowledge, the first joint analysis of subjective and objective measures of well-being. Using a rich longitudinal data from the mothers pregnancy until adulthood for a birth cohort of children who attended school in Örebro during the 1960s, we analyse in a first step how subjective (self-assessed) and objective (cortisol-based) measures of well-being are related to each other. In a second step, life-course models for these two measures are estimated and compared with each other. Despite the fact that our analysis is largely exploratory, our results suggest interesting possibilities to use objective measures to measure well-being, even though this may imply a greater degree of complexity.

Background: Current treatment of children with dental behaviour management problems (DBMP) is based on the presupposition that their difficulties are caused by dental fear, but is this always the case? Objectives: The aim of this study was to study temperamental reactivity, negative emotionality, and other personal characteristics in relation to DBMP in 8- to 12-year-old children. Methods: Forty-six children referred because of DBMP (study group) and 110 children in ordinary dental care (reference group) participated. The EASI tempramental survey assessed temperamental reactivity and negative emotionality, the Child Behaviour Questionnaire internalizing and externalizing behaviour problems, and the Children's Fear Survey Schedule general and dental fears. Cluster analyses and tree-based modelling were used for data analysis. Results: Among the five clusters identified, one could be characterized as 'balanced temperament'. Thirty-five per cent of the reference group compared to only 7% of the study group belonged to this cluster. Negative emotionality was the most important sorting variable. Conclusions: Children referred because of DBMP differed from children in ordinary dental care, not only in dental fear level, but also in personal characteristics. Few of the referred children were characterized by a balanced temperament profile. It is important to consider the dual impact of emotion dysregulation and emotional reactivity in the development of DBMP.

This paper applies the GARCH-MIDAS (mixed data sampling) model to examine whether information contained in macroeconomic variables can help to predict short-term and long-term components of the return variance. A principal component analysis is used to incorporate the information contained in different variables. Our results show that including low-frequency macroeconomic information in the GARCH-MIDAS model improves the prediction ability of the model, particularly for the long-term variance component. Moreover, the GARCH-MIDAS model augmented with the first principal component outperforms all other specifications, indicating that the constructed principal component can be considered as a good proxy of the business cycle.

Several studies have found segmentation within Real Estate markets. Real Estate markets are often modelled by hedonic models and these often rely on proxy variables to represent a number of hedonic characteristics. The rateable value of a property is designed to be a measure of its market value and is often used as a proxy for difficult to observe or unobservable characteristics in such studies. Such studies often assume that the rateable value is simply proportional to the market value of a property. The presence of segmentation may invalidate this assumption. A sample of data from the Real Property Register maintained by Lantmäteriet, the Swedish land registry, were investigated to explore the relationship between land values and rateable values. The sample studied contains the selling price of properties sold across a range of years, the year of sale and a limited number of other characteristics. However, the rateable values for these properties are only available for a given year, the calendar year 2009. Accounting for inflationary effects is therefore necessary and was attempted in the model with indicator variables and then a quadratic expression. An inconsistent relationship between the selling price and the rateable value was found when a simple non-segmented model was used. With the absence of other likely explanatory variables in the data clustering techniques were considered to anatomise the data. Finite Mixture Modelling is the most appropriate method of classifying this data into clusters which are modelled as a finite mixture of regression models. The investigation found the presence of 3 distinct clusters. None of the variables in the data set appear to cause cluster affiliation.

In this paper we consider the estimation of the weights of tangent portfolios from the Bayesian point of view assuming normal conditional distributions of the logarithmic returns. For diffuse and conjugate priors for the mean vector and the covariance matrix, we derive stochastic representations for the posterior distributions of the weights of tangent portfolio and their linear combinations. Separately we provide the mean and variance of the posterior distributions, which are of key importance for portfolio selection. The analytic results are evaluated within a simulation study, where the precision of coverage intervals is assessed. 

SCB gör en vakansundersökning varje månad, granskningssystemet för undersökningen ska eventuellt ändras till en metod som kallas för significance editing. Med den granskningsmetoden behövs ett jämförelsevärde för att kunna avgöra om enkäten är korrekt besvarad eller om ett värde är misstänkt och behöver granskas ytterligare. Uppsatsens syfte är att genom poissonregression ta fram en bra modell som kan generera detta jämförelsevärde. Ett antal hjälpvariabler togs fram och testades för att se om de passade i poissonregressionen och om de kunde förklara antalet vakanser. De hjälpvariabler som har använts är antal vakanser föregående månad, antal anställda på arbetsstället, dummyvariabel för Sveriges län och månaderna. Metoden testades på två olika branscher, pappersmassaindustrin och metallindustrin. I de resultat som togs fram kan man se att variablerna antal anställda på arbetsstället och antalet vakanser föregående månad alltid blir signifikanta och tillför till att skatta antalet vakanser. Därför kan dessa användas för att skatta ett jämförelsevärde. Län och månader behövs i modellen men det är olika län och månader som blir signifikanta för de två olika branscherna. Generellt kan man dra slutsatsen att antalet vakanser ökar på våren och sommaren. Huruvida metoden går att tillämpa på det ogranskade datamaterialet får vidare undersökningar visa.

The present commentary gives a brief overview of the person-oriented and variable-oriented approaches, how they are commonly used in longitudinal research, and what one should take into consideration before using either approach. In addition to presenting an empirical example on girls’ adjustment problems using both approaches, this commentary uses the contributions in the present issue of Merrill Palmer Quarterly to illustrate some of the main issues surrounding these two perspectives. Special attention is also given to the contrast between the person-oriented and variable-oriented approaches in terms of aggregation and disaggregation, model appropriateness and usefulness, and prediction as a goal. Future directions with regard to implementing a personoriented approach are discussed, including the importance of conceptual clarity, practical and theoretical training, and method development.

Kasta gris är ett spel där spelarna tävlar om att komma först till 100 poäng. Två grisformade tärningar kastas och beroende på hur de landar ger de olika poäng, alternativt förlust av poäng. För en spelare som har samlade poäng i en kastomgång innebär ytterligare ett kast en chans att erhålla en högre poängsumma men också en risk att förlora den redan samlade. I denna uppsats vill vi ta reda på vid vilken högsta poängsumma i en kastomgång som spelaren bör välja att fortsätta kasta. Eftersom tärningarna är grisformade och alltså inte symmetriska är sannolikheterna olika för de möjliga utfallen. Att sannolikheterna därtill är okända omöjliggör att beräkna den sökta poängsumman exakt. Vi har genomfört ett eget försök med 10 517 kast uppdelade på tre gristärningspar. Med hjälp av insamlad data och metoder inom sannolikhetslära har vi kunnat skatta de okända sannolikheterna och därmed den sökta poängsumman. För att få ett mått på osäkerheten i vår skattning av den senare har vi använt två metoder inom inferensteorin, Deltametoden och bootstrap. I vårt resultat fann vi att 21, med åtminstone 75 procents säkerhet, är den högsta poängsumma för vilken en spelare bör fortsätta sin kastomgång. Resultatet ger en spelare möjlighet att maximera sin förväntade poäng i en kastomgång men att använda detta som en spelstrategi genom hela spelet är dock ingen garanti för vinst. 

This article investigates the hypothesis that office type has an influence on workers' health status and job satisfaction and 469 employees in seven different types, defined by their unique setup of architectural and functional features, have rated their health status and job satisfaction. Multivariate regression models were used for analysis of these outcomes, with adjustment for age, gender, job rank, and line of business. Both health status and job satisfaction differed between the seven office types. Lowest health status was found in medium-sized and small open plan offices. Best health was among employees in cell offices and flex offices. Workers in these types of offices and in shared room offices also rated the highest job satisfaction. Lowest job satisfaction was in combi offices, followed by medium-sized open plan offices. The differences between employees could possibly be ascribed to variations in architectural and functional features of the office types.

Short-term exposure to dust and dust added with beta-(1,3)-d-glucan or aldehydes may cause sensory reactions. In random order, we exposed 36 volunteers in a climate chamber to clean air, office dust, dust with glucan, and dust with aldehydes. Three groups of subjects were exposed, eleven were non-atopic with nasal histamine hyperreactivity, 13 were non-atopic, and 12 were atopic. Subjective ratings of symptoms and general health were registered four times during four 6-h exposure sessions. Six symptom intensity indices were constructed. The nasal hyperreactive group had a high and time-dependent increase of mucous membrane irritations, whereas the atopic group had a low and stable rate of irritations with exposure time, close to the reference group (P = 0.02 for differences between the groups with respect to time under exposure for Weak Inflammatory Responses and P = 0.05 for Irritative Body Perception, significance mainly because of the nasal hyperreactive group). Exposure to dust, with or without glucan or aldehydes, showed increased discomfort measured by the index for Constant Indoor Climate, and dust with glucan had a similar effect for the index for Lower Respiratory Effects. For Psychological and Neurological Effects these were dependent on group affiliation, thus preventing a uniform statement of exposure effects for all three investigated groups.Opportunities for identifying persons with high or low sensitivity to low-level exposures are important in preventive medicine and will reduce intra-group variability and thus increase the power of experimental and epidemiological studies searching for correlations between exposures and health effects. The contrast between nasal hyperreactive on one side and atopic and reference subjects on the other side is particularly important. The atopic group indicated a non-homogenous reaction depending on their hyperreactive status, a finding that could be important but needs further confirmation.