Accurate prediction of mortality is an important component of forest growth and yield prediction systems, yet mortality remains one of the least understood components of the system. Whole-stand and individual-tree mortality models were developed for loblolly pine plantations throughout its geographic range in the United States. The model for predicting stand mortality were developed using stand characteristics and biophysical variables. The models were constructed using two modeling approaches. In the first approach, mortality functions for directly predicting tree number reduction were developed using algebraic difference equation method. In the second approach, a two-step modeling strategy was used where a model predicting the probability of tree death occurring over a period was developed in the first step and a function that estimates the reduction in tree number was developed in the second step. Individual-tree mortality models were developed using multilevel logistic regression and survival analysis techniques. Multilevel data structure inherent in permanent sample plots data i.e. measurement occasions nested within trees (e.g., repeated measurements) and trees nested within plots, is often ignored in modeling tree mortality in forestry applications. Multilevel mixed-effects logistic regression takes into account the full hierarchical structure of the data. Multilevel mixed-effects models gave better predictions than the fixed effects model; however, the model fits and predictions were further improved by taking into account the full hierarchical structure of the data. Semiparametric proportional hazards regression was also used to develop model for individual-tree mortality. Shared frailty model, mixed model extension of Cox proportional hazards model, was used to account for unobserved heterogeneity not explained by the observed covariates in the Cox model. / Ph. D.

Semiparametric modeling is a combination of the parametric and nonparametric models in which some functions follow a known form and some others follow an unknown form. In this dissertation we made contributions to semiparametric modeling for matched case-crossover data. In matched case-crossover studies, it is generally accepted that the covariates on which a case and associated controls are matched cannot exert a confounding effect on independent predictors included in the conditional logistic regression model. Any stratum effect is removed by the conditioning on the fixed number of sets of the case and controls in the stratum. However, some matching covariates such as time, and/or spatial location often play an important role as an effect modification. Failure to include them makes incorrect statistical estimation, prediction and inference. Hence in this dissertation, we propose several approaches that will allow the inclusion of time and spatial location as well as other effect modifications such as heterogeneous subpopulations among the data. To address modification due to time, three methods are developed: the first is a parametric approach, the second is a semiparametric penalized approach and the third is a semiparametric Bayesian approach. We demonstrate the advantage of the one stage semiparametric approaches using both a simulation study and an epidemiological example of a 1-4 bi-directional case-crossover study of childhood aseptic meningitis with drinking water turbidity. To address modifications due to time and spatial location, two methods are developed: the first one is a semiparametric spatial-temporal varying coefficient model for a small number of locations. The second method is a semiparametric spatial-temporal varying coefficient model, and is appropriate when the number of locations among the subjects is medium to large. We demonstrate the accuracy of these approaches by using simulation studies, and when appropriate, an epidemiological example of a 1-4 bi-directional case-crossover study. Finally, to explore further effect modifications by heterogeneous subpopulations among strata we propose a nonparametric Bayesian approach constructed with Dirichlet process priors, which clusters subpopulations and assesses heterogeneity. We demonstrate the accuracy of our approach using a simulation study, as well a an example of a 1-4 bi-directional case-crossover study. / Ph. D.

There is interest in deriving case-mix adjusted standardised mortality ratios so that comparisons between healthcare providers, such as hospitals, can be undertaken in the controversial belief that variability in standardised mortality ratios reflects quality of care. Typically standardised mortality ratios are derived using a fixed effects logistic regression model, without a hospital term in the model. This fails to account for the hierarchical structure of the data - patients nested within hospitals - and so a hierarchical logistic regression model is more appropriate. However, four methods have been advocated for deriving standardised mortality ratios from a hierarchical logistic regression model, but their agreement is not known and neither do we know which is to be preferred. We found significant differences between the four types of standardised mortality ratios because they reflect a range of underlying conceptual issues. The most subtle issue is the distinction between asking how an average patient fares in different hospitals versus how patients at a given hospital fare at an average hospital. Since the answers to these questions are not the same and since the choice between these two approaches is not obvious, the extent to which profiling hospitals on mortality can be undertaken safely and reliably, without resolving these methodological issues, remains questionable.

Effective delineation and management of stream fish populations requires a thorough knowledge of dispersal patterns, because these patterns affect a number of other demographic rates such as population growth, reproduction, survival, and gene flow. Previous studies of stream fish dispersal patterns have generally established species- and stream-specific home ranges and movement rates, but have largely failed to account for the environmental variables that may cause these parameters to vary. Many fishes occupy a variety of streams across a broad spectrum of ecological conditions, and movement rates (and thus population dynamics) may respond to these environmental gradients. Furthermore, enhanced understanding of the ecological features that induce or impede dispersal will help guide future management of stream channels for population connectivity. To determine the instream features that influence the dispersal patterns of darters, I conducted a spatially intensive mark/recapture study of three darter species in the upper Roanoke River watershed. Logistic regression was used to relate observed inter-riffle movements to gradients in riffle and corridor attributes. During the first study period, habitat area loss and habitat spacing drove dispersal patterns. However, a model developed from these data transferred poorly to the second study period, in which density dependence was a more effective predictor of dispersal. Individual size did not seem to influence the probability of emigration, but did influence the distance traveled following emigration, particularly for the two more specialist species. This finding suggests a size-based dominance hierarchy for habitat selection and occupancy in darters. Predation threat had only a minor effect on the probability of traversing inhospitable corridors, but experimentally introduced structural cover significantly elevated dispersal rates through such corridors. Taken together, results of this study indicate that a complex array of ecological features interact to produce heterogeneity in dispersal rates across the stream landscape. Knowledge of these influences can be used to manage stream channels for dispersal permeability. In addition to field studies, laboratory studies were undertaken to determine the efficacy of visible implant elastomer (VIE) and injectable photonic dye (IPD) for marking darters. No previous studies have rigorously evaluated these marks in darters, and comparisons of the two technologies in any taxa are few. Results of the laboratory study indicated that VIE is preferable to IPD for marking darters, particularly when mark longevity greater than 80 days is desired. Individuals marked with VIE exhibited higher survival and mark retention rates than did individuals marked with IPD. Additionally, VIE mark retention was more consistent across body locations. Retention of both marking technologies was biased by color. My study indicates that the results of tagging efficiency studies are not applicable across taxa, and that pilot studies are necessary prior to field use of marks in previously untested species. / Master of Science

Habitat modeling can assist in managing potentially widespread but poorly known biological resources such as the federally endangered Delmarva fox squirrel (DFS; Sciurus niger cinereus). The ability to predict or identify suitable habitat is a necessary component of this species' recovery. Habitat identification is also an important consideration when evaluating impacts of land development on this species distribution, which is limited to the Delmarva Peninsula. The goal of this study was to build a predictive model of DFS occurrence that can be used towards the effective management of this species. I developed 5 a'priori global models to predict DFS occurrence based on literature review, past models, and professional experience. I used infrared photomonitors to document habitat use of Delmarva fox squirrels at 27 of 86 sites in the southern Maryland portion of the Delmarva Peninsula. All data were collected on the U.S. Fish and Wildlife Service Chesapeake Marshlands National Wildlife Refuge in Dorchester County, Maryland. Preliminary analyses of 27 DFS present (P) and 59 DFS absent (A) sites suggested that DFS use in my study area was significantly (Wilcoxon Mann-Whitney, P < 0.10) correlated with tree stems > 50 cm dbh/ha (Pmean = 16 + 3.8, Amean = 8+ 2.2), tree stems > 40 cm dbh/ha (Pmean = 49 + 8.1, Amean = 33 + 5.5), understory height (Pmean = 11 m + 0.8, Amean = 9 m + 0.5), overstory canopy height (Pmean = 31 m + 0.6, Amean = 28 m + 0.6), percent overstory cover (Pmean = 82 + 3.9, Amean = 73 + 3.1), shrub stems/ha (Pmean = 8068 + 3218, Amean = 11,119 + 2189), and distance from agricultural fields (Pmean = 964 m + 10, Amean = 1308 m + 103). Chi-square analysis indicated a correlation with shrub evenness (observed on 7% of DFS present sites and 21% of DFS absent sites). Using logistic regression and the Information Theoretic approach, I developed 7 model sets (5 a priori and 2 post hoc) to predict the probability of Delmarva fox squirrel habitat use as a function of micro- and macro-habitat characteristics. Of over 200 total model arrays tested, the model that fit the statistical, biological, and pragmatic criteria postulated was a post hoc integrated model: DFS use = percent overstory cover + shrub evenness + overstory canopy height. This model was determined to be the best of its subset (wi = 0.54), had a high percent concordance (>75%), a significant likelihood ratio (P = 0.0015), and the lowest AICc value (98.3) observed. Employing this predictive model of Delmarva fox squirrel occurrence can benefit recovery and consultation processes by facilitating systematic rangewide survey efforts and simplifying site screenings. / Master of Science

Text in Afrikaans / Hierdie studie is onderneem met die doel om 'n model te ontwikkel waarmee die wenner in 'n konvensionele landgeveg voorspel kan word. Gegewe die omvang van die vakgebied oorlog, is die studie beperk tot die taktiese vlak en fokus op landgevegte tydens konvensionele oorlogvoering. As eerste stap in die ontwikkelingpsproses, is die faktore wat wen kan bepaal krygskundig nagevors. Die sogenaamde honderdgevegte-datastel is saamgestel uit data van 100 gevegte uit die twintigste eeu en net vroeer, met die klem op gevegte waarin Suid-Afrikaanse magte betrokke was. Verskeie statistiese tegnieke is ondersoek om 'n geskikte tegniek vir die ontleding van die data te vind. Die ondersoek het aangetoon dat logistiese regressie die beste tegniek is vir die data. 'n Ontwikkelingsproses met drie voorspellers is ook saamgestel. Verskeie modelle is ondersoek, naamlik 1 'n Voorspellingsmodel met eensydige sub-modelle sonder gevegshouding, met en sonder opponentdata. I 2 'n Voorspellingsmodel met eensydige sub-modelle met gevegshouding, met en sonder opponentdata. 3 'n Voorspellingsmodel met tweesydige sub-modelle met opponentdata.. Die ontwikkelingsproses lewer verskeie modelle wat baie goed presteer sensitiwiteit > 80%). 'n Finale keuse lewer die volgende resultaat: 1 Vir die geval waar opponentdata nie beskikbaar is nie, is 'n eensydige submode! sonder gevegshouding ontwikkel waarvan die resultaat teen 'n skeidingsgrens gemeet word om die uitslag te bepaal. Die model het 'n sensitiwiteit van 85%, maar kan net 'n wen of gelykop, of, verloor of gelykop voorspel. 2 Vir die geval waar opponentdata beskikbaar is, is 'n eensydige sub-model ivsonder gevegshouding ontwikkel wat in staat is om, deur die opponente se uitslag met mekaar te vergelyk, die wenner aan te wys. Hierdie model het 'n sensitiwiteit van 83,8% Verskeie statistiese en krygskundige gevolgtrekkings word gemaak, die belangrikste waarvan dat die gekose modelle wel daartoe in staat is om gevegsvoorspellings akkuraat te kan uitvoer. Die modelle kan ook aangewend word om gevegte te ontleed en tendense te verklaar. Krygskundig bevestig die resultaat die noodsaaklikheid van die maneuvreringsbenadering en goeie leierskap. Die resultaat van die studie het wye aanwendingspotensiaal op die gebied van die krygskunde, krygsfilosofie, krygspele en militere operasionele navorsing en laat ruimte vir interessante en noodsaaklike verdere navorsing in operasionele navorsing sowel as in die krygskunde. / The aim of this study is to develop models for the efficient prediction of the outcome of a land battle. The study is confined to conventional warfare at the tactical level. The first step was to identify the variables that may determine victory. Thirty such variables enjoying the support of various military historians and philosophers were selected. The hundred-battle data set, consisting of coded data for a hundred twentieth-century battles, was compiled. The thirty variables were encoded for each combatant. Since the outcome and most of the prediction variables are binary but a few are continuous, ordinary linear regression could not be used and several statistical and other techniques were evaluated. Logistic regression was found to be the best. A formalized development and selection process was applied to a number of broad model classes. These were 1 prediction models with one-sided sub-models without combat posture and with (without) opponent data 2 prediction models with one-sided sub-models with combat posture and with (without) opponent data 3 prediction models with two-sided sub-models without combat posture and with opponent data. The process provided several very good models and the following were selected. Without opponent data. A one-sided sub-model without combat posture, utilizing a discriminator was selected. It determines the outcome with a sensitivity of 85%. However, it only predicts victory or a draw, defeat or a draw. With opponent data. A one-sided sub-model without combat posture was selected. It predicts the outcome of battle by comparing the results of the two opponents. This model vishowed a sensitivity of 83,8%. Several statistical and military scientific conclusions followed, the most important being that the chosen models can accurately predict battle outcome or post facto determine the outcome. The models can also be used to analyze battles. In this role they confirm the importance of maneuver warfare and good leadership. The results of this study can be applied in military science, military philosophy and war gaming. The work fuses military philosophy with statistical analysis, is a first in the field and offers the possibility of breaking out of the mind-set of personal views and biases prevalent in military science. The method as such can be applied to different data bases representing war at other levels or with other technologies. / Philosophy / D.Phil. (Philosophy)

由於財務報表舞弊往往足以震撼投資大眾，造成資本市場重大損失，各國監管單位無不盡力降低此事件之爆發，以維護資本市場秩序、保障投資人，是以本研究欲瞭解影響中國大陸上市公司舞弊之因素為何，以及如何建立舞弊預測模型提供財務報表使用者作為參考之用。本文利用2007年至2014年受懲罰之上市公司為研究對象，採Logistic迴歸進行實證分析，結果顯示裁決性收入與Z"-Score對於財務報表舞弊無顯著相關，相反的獨立董事比例、是否具ST壓力、存貨週轉率、應收帳款週轉率、主營業務利潤率與財務報表舞弊具顯著關係，另外利用迴歸結果中顯著變數建立財務報表舞弊模型，發現整體正確率為53.31％。 / Due to the severe impacts caused by fraudulent financial reporting, securities regulatory commissions in most countries put much emphasis on maintaining the order of the capital markets and protecting the investors’ interests. In order to realize the factors of financial statement fraud, especially for China listed companies, and build the detecting model for the financial statements users, I select some listed companies punished by the government during the period 2007-2014 as the samples in this dissertation. Then, I use logistic regression model to test which variables are significant to fraudulent financial reporting, and the results show that the discretionary revenue and Z"-Score do not have impact on it. On the contrary, the percentage of independent directors, pressure from avoiding being “ST”, inventory turnover, accounts receivable turnover, and percentage of income from main operation are significantly relevant to fraudulent financial reporting. Moreover, when including these significant variables in the detecting model, the accuracy of the model can up to 53.31 percent.

在我們建立迴歸模型時，必須針對研究的目的去探求與相依變數有關的自變數，而且 這些自變數應能合理的解釋相依變數，然而這些自變數的組合數一定很大；所以在一 般線性迴歸分析中，最重要也是最困難的問題是如何選取模式中的自變數，棄卻不太 重要的自變數，獲得最後的模型，以符合經濟原則。 而近年來非線性迴歸模型在各種領域裡廣泛地被使用，這些線性回歸模型之自變數的 選取較線性迴歸模型之自變數的選取困難，因其必須用反覆的技術來找最大概似估計 量，然後利用此最大概似估計量來做為選取自變數的基礎所以計算的成本較高。 本文將以處理相依變數為屬質變數的羅吉斯迴歸模型（LOGISTIC REGRESSION MODEL ）為主要研究對象；首先導出此模型的CP統計量，以CP來作為選取自變數的準則；其 次介紹一種透過對數概似近似函數及一些資料的轉換，將羅吉斯迴歸模型之自變數選 擇問題變換成一般線性迴歸模型的自變數選擇問題；然後作一個模擬分析比較不經變 數變換與經變數變換的方法，所選出的自變數組合是否大致相同，若其差異不大，則 表示此種變數變換方法確時有效，往後遇到類似的非線性迴歸之自變數的選取都可轉 換成一般線性迴歸的問題來解決，可簡化許多計算過程，此亦為本文研究的目的。 本文結構：本文共分六章 第一章 緒論，說明井究動機與目的 第二章 建立羅吉斯迴歸模型（LOGISTIC REGRESSION MODEL ）及定義其殘差（ RESIDUAI） 第三章 探討非線性模型之自變數選擇方法及針對LOGISTIC REGRESSION 求其CP統計 量。 第四章 介紹一重經過變數變換的自變數選擇程序及其應用的原理。 第五章 模擬分析，比較第三章與第四章所述二種方法的差異。 第六章 結論。

過去三、四十年來世界各地發生金融危機的頻率較從前高出許多，探究原因後可以發現，與各國陸續開放金融自由化以及國際金融市場快速成長有極大的關係。除此之外，在各國中，金融危機的發生通常具備一些共同特徵，諸如危機發生時會導致資金外流、匯率大幅貶值、股市重挫、產出減少、進出口減少…等影響。因此，面對這一波次級房貸風暴，本研究也即將檢視美國在總體經濟數據上各方面的表現，希望能利用1970年以來已開發國家和開發中國家歷年來所發生的貨幣、銀行危機下所代表的各種總體經濟數據，經過轉化整理後，透過Logistic迴歸模型建立一個迴歸方程式，以了解金融危機的發生與實質匯率、進出口…等其他解釋變數之間的關聯，並利用此模型探測現階段次級房貸風暴對美國可能引發金融危機的機率值，以探討其合理性。 / We find that the frequencies of the financial crises are higher for the past forty years in the world. It is due to the financial liberalization and international financial markets which grow rapidly. Besides, financial crises usually company with some common characteristics such as capital outflow, the depreciation of the foreign exchange, the shock of the stock market, the decreasing of the production and so on. Therefore, in order to understand this financial crisis of the subprime mortgage, this thesis surveys the economic data of developed countries and developing countries from 1970s and figures out the performances of these countries under balance-of -payments crises or banking crisis. We use the logistic regression model and transform the data to construct a regression model. After understanding the relationship between the explaining variables, we use this model to predict the probability of possible financial crisis in U.S. under the subprime mortgage crisis and then discuss the rationality of those predicted values.

Pagrindinis darbo tikslas – pritaikyti matematinius ir statistinius metodus lietuvių kalbos analizėje, identifikuojant ir atsižvelgiant į lietuvių kalbos ypatumus, jos heterogeniškumą, sudėtingumą ir variabilumą. / The target of the work is to apply mathematical and statistical methods in the analysis of the Lithuanian language by identifying and taking into account peculiarities of the Lithuanian language, its heterogeneity, complexity and variability.