Terence O'Neill

Normal Discrimination with Unclassified Observations

Abstract Fishers linear discriminant rule may be estimated by maximum likelihood estimation using unclassified observations. It is shown that the ratio of the relevant information contained in unclassified observations to that in classified observations varies from approximately one-fifth to two-thirds for the statistically interesting range of separation of the populations. Thus, more information may be obtained from large numbers of inexpensive unclassified observations than from a small classified sample. Also, all available unclassified and classified data should be used for estimating Fishers linear discriminant rule.

The role of higher oil prices: A case of major developed countries

The primary aim of this chapter is to examine whether the recent increase in world oil prices has affected inflation expectations and stock market returns in major OECD countries. The key findings are as follows. First, we found no evidence to support the presence of a long term relationship between oil prices and inflation expectations – measured by the difference between yields of inflation indexed and non-inflation indexed government bonds – over the sample between early 2003 and late 2006. Second, higher oil prices are found to lead to expectations of higher inflation. This evidence is stronger over the period where oil prices had been higher and signs of capacity constraints in the economy were emerging. Third, the impact of higher oil prices on stock market returns differs among countries. While higher oil prices are found to adversely affect stock market returns in the United States, the United Kingdom and France, the effects are positive in Canada and Australia as these countries are significant exporters of energy resources.

The General Distribution of the Error Rate of a Classification Procedure With Application to Logistic Regression Discrimination

Abstract The large-sample distribution of the error rate of an arbitrary estimator of the optimal classification rule is given. The asymptotic distribution of the logistic regression estimator is found. These results are used to show that the efficiency of logistic regression classification in some nonnormal cases is low. This suggests that maximum likelihood discrimination should be used whenever possible.

The effectiveness of air bags.

Previous research has shown that the installation of air bags in vehicles significantly reduces crash related deaths, but these analyses have used statistical techniques which have not been capable of controlling for other major determinants of crash survival. This study analysed data from the US FARS database of fatal crashes using conditional logistic regression which is simultaneously able to estimate occupant protection effects for a range of variables. Results of the analysis provided a comparative quantification of both the effect of the air bag as well as other well known determinants of occupant crash survival (age, seat belt use, and gender). When potentially confounding variables were controlled, both the driver and passenger side air bag devices were shown to significantly reduce the probability of death in direct frontal collisions, but the effect size calculated was small compared to the effect of the seat belt. The effect size may also be very small in absolute terms depending on the severity of the crash involved. Given the limited benefit of the air bag, efforts to promote air bags seem particularly difficult to justify in countries such as the United States where the vastly superior occupant protection of the seat belt is under-utilised.

Inconsistency of the misspecified proportional Hazards model

The asymptotic limits of the estimates of the proportionality factor and the underlying distribution functions are found for a misspecified Proportional Hazards Model. A realistic example is given to show that the asymptotic biases can be quite large and of practical significance.

The health expectancies of older Australians

Health expectancies of the states ‘Disability-free’ and ‘Disabled’ are estimated for Australian females and males aged 60 and over, both by cohort from 1980 and current for survey years 1981, 1988, 1993 and 1998. Modifications of recently developed logistic regression techniques are used rather than the standard 1971 method due to Sullivan. Results from the three later surveys are broadly similar and differ in important respects from those of the 1981 survey. Based on the last three surveys our estimates support the view that, depending on age, two-thirds or more of the increase in female life expectancy over the decade 1988–1998 is spent in the Disabled state. The situation is worse for elderly men, for whom all of the increased years of expected life are estimated to be spent in the Disabled state. The findings do not support rectangularization of the survival curve or Disability-free survival curve.

H-2-Linked Control of Resistance to Ectromelia Virus Infection in B10 Congenic Mice

Several B 10 strains of mice, recombinant at theH-2 locus, have been shown to differ in their resistance to infection with ectromelia virus, a natural mouse pathogen. Of 10 strains, 1310, B 10.A(2R), B10.A(4R) and B10.D2 were the most resistant, while B10.G and B 10.A(5R) were the most susceptible. Other strains were intermediate between these extremes. Several genes conferring resistance have been mapped toDb in B10.A(2R),KkI-AkI-Bk in B10.A,I-Jb in B10.A(2R) and toDd in B 10.T(6R). In general, death among susceptible strains was not a consequence of acute liver necrosis as in other non-B10 strains, and occurred randomly from 8–14 days after infection. The exact cause of death is unknown but is characterized by persisting high titers of virus in the spleen and sometimes the liver, despite an ongoing immune response indicated by strong cytotoxic T-cell activity detectable in the spleens of all mice. The most resistant B10 and B10.A(2R) strains cleared virus from the spleen and liver by 8 days after infection. Analysis of infection in chimeric mice indicates thatH-2 genes, which determine susceptibility to virus persistence in the spleen, operate via radiosensitive cells of the lymphomyeloid system. This evidence, together with several examples ofH-2-linked differences in cytotoxic T-cell responsiveness between resistant and susceptible strains, is consistent with the hypothesis that the mechanism by whichH-2 genes control resistance to ectromelia virus in B10 strain mice is by their influence on the effectiveness of a cell-mediated immune response.

Truncated logistic regression

Truncated binary data occurs when a group of individuals, who each have a binary response, are observed only if one or more of the individuals has a positive response. In this paper the group will be taken to be a motor vehicle accident and the binary response taken to be survival or death. We compare two regression techniques that can be used for truncated binary data. The first procedure, conditional logistic regression (Breslow and Day, 1980, Statistical Methods in Cancer Research. 1: The Analysis of Case-Control Studies. No. 32. Lyon: IARC) conditions on the actual number of deaths, and has been previously used for this type of data. The second procedure, truncated logistic regression, conditions on there being at least one death. It is computationally simpler than conditional logistic for groups of size greater than two and can be considerably more efficient. A major difference between the two methods is that only truncated logistic regression requires a knowledge of group level covariates and allows estimation of group level effects.

Generalised Clopper–Pearson confidence intervals for the binomial proportion

In this paper we develop some new confidence intervals for the binomial proportion. The Clopper–Pearson interval is interpreted as an outcome of randomised confidence interval theory. The problem of randomised intervals possibly being empty is solved using a new technique involving ‘tail functions’, with the offshoot being a new class of randomised and Clopper–Pearson intervals. Some of the new intervals are investigated and shown to have attractive frequentist properties. Coverage probabilities and expected widths are compared and guidelines are established for constructing the optimal generalised Clopper–Pearson interval in any given situation.

Gene signature of stromal cells which support dendritic cell development.

Spleen stromal cells are critical determinants of dendritic cell (DC) development in spleen. The spleen stromal line, namely STX3, supports DC differentiation in vitro from overlaid bone marrow cells while the lymph node stromal line, namely 2RL22, does not. Here we have characterised the hematopoietic support capacity of each stroma, and analysed lineage origin of the stromal cell lines by gene profiling using microarrays. Stromal co-culture experiments were performed using bone marrow cells as a source of hematopoietic progenitors. A characteristic immature myeloid-like CD11c(+)CD11b(+)CD86(+)MHC-II(/lo)B220()CD8alpha() DC is produced after 14 days in STX3 cocultures, while 2RL22 cocultures produce only monocyte/macrophage-like cells. No other hematopoietic cell type is produced. The STX3 and 2RL22 stroma were compared by transcriptome analysis utilising Affymetrix Murine U74Av2 genechips to identify gene expression related to differential hematopoietic support function. Data mining was used to determine cell surface marker expression reflecting endothelial cells and fibroblasts, as well as adhesion molecules contributing to the microenvironment. STX3 shows gene expression reflective of early endothelial cells, while 2RL22 expresses markers specific to fibroblasts. The expression of genes like Flt1, CD34, Mcam, and Eng distinguishes STX3 as an early immature endothelial cell lacking markers of angioblasts or hemangioblasts like Tal1/SCL, Tie1, Tie2, Kdr or Prom1/AC133. The absence of expression of genes like Vwf and Cd31 distinguishes STX3 from fully differentiated vascular endothelial cells. In contrast, the 2RL22 lymph node stroma specifically expresses genes related to fibroblastic-like cells like osteoblasts with expression of Vdr (Vitamin D receptor), and epithelial cells with expression of Krt13 (keratins). Gene expression data identifies STX3 as splenic endothelial cells, independently able to support the outgrowth of immature, myeloid DC-like cells from progenitors present in bone marrow, while 2RL22 lymph node fibroblastic cells provide support for development of monocytes/macrophages.