No description available.

No description available.

No description available.

No description available.

No description available.

The shrinkage estimation has proven to be very useful when dealing with a large number of mean parameters. In this dissertation, we consider the problem of simultaneous estimation of multiple variances and construct a shrinkage type, non-parametric estimator. We take the non-parametric empirical Bayes approach by starting with an arbitrary prior on the variances. Under an invariant loss function, the resultant Bayes estimator relies on the marginal cumulative distribution function of the sample variances. Replacing the marginal cdf by the empirical distribution function, we obtain a Non-parametric Empirical Bayes estimator for multiple Variances (NEBV). The proposed estimator converges to the corresponding Bayes version uniformly over a large set. Consequently, the NEBV works well in a post-selection setting. We then apply the NEBV to construct condence intervals for mean parameters in a post-selection setting. It is shown that the intervals based on the NEBV are shortest among all the intervals which guarantee a desired coverage probability. Through real data analysis, we have further shown that the NEBV based intervals lead to the smallest number of discordances, a desirable property when we are faced with the current "replication crisis". / Statistics

Innovation is the central value of economic behavior. In this dissertation research, I attempt to explore the social and cognitive origins of firm innovation through three interrelated studies, by merging several research streams — managerial cognitions, social networks, and innovation, and collecting data through multiple methods — archives and surveys. First, I proposed a social-cognitive view to study the sources of firm innovation. In the context of firm innovation, top management teams' cognitions or an entrepreneur's cognitions shape the way they use the social structure available to them, while the social structures influence the embedded actors' cognitions and ultimately strategic actions. Managers and entrepreneurs form collaborative partnerships aimed at innovation and competitiveness. During this dynamic social learning process, cognitive differences influence the formation of social capital and its realized benefits. The impact of social capital on innovation cannot be evaluated without understanding the individual cognitive characteristic first. Next, I tested this theoretical model in two contexts. In the first empirical study, I derived firm-level hypotheses that link the top management team's cognitions, the firm's social capital, and the technological innovations. These hypotheses are tested on a sample of U.S. semiconductor firms in the years 1991-1998. In the second empirical study, I derived similar hypotheses that link entrepreneur's cognitions, social capital and startup's technological innovations. A survey was conducted in both Pennsylvania and Virginia, targeting the entrepreneurial firms in technology industries. The hypotheses were empirically tested on a final sample of 70 U.S. small and medium-sized manufacturers. Two empirical studies supported some of the derived hypotheses and the findings have significant theoretical, empirical, and practical implications. In a diverse social network, actors' knowledge structure tends to be more complex, and more centralized. In addition, these studies indicate that both social capital and cognitive structure play important roles in technological innovation. By distinguishing between cognitive structures, as well as social capital characteristics, and by investigating their effects on firm innovations, this dissertation extends the literature on organization theory, innovation research, entrepreneurship, and research methodologies. This dissertation research deepens our understanding of firm innovation, and opens a whole line of further research. / Ph. D.

Pathway diagrams, similar to the graph diagrams using a node-link representation, are used by biologists to represent complex interactions at the molecular level in living cells. The recent shift towards data-intensive bioinformatics and systems-level science has created a strong need for advanced pathway visualization tools that support exploratory data analysis. User studies suggest that an important requirement for biologists is the need to associate microarray data to pathway diagrams. A design space for visualization tools that allow analysis of microarray data in pathway context was identified for a systematic evaluation of the visualization alternatives. The design space is divided into two dimensions. Dimension 1 is based on the method used to overlay data attributes onto pathway nodes. The three possible approaches are: overlay of data on pathway nodes one data attribute at a time by manipulating a visual property (e.g. color) of the node, along with sliders or some such mechanism to animate the pathway for other timepoints. In another approach data from all the attributes in data can be overlaid simultaneously by embedding small charts (e.g., line charts or heatmap) into pathway nodes. The third approach uses miniature version of the pathways-as-glyph view for each attribute in the data. Dimension 2 decides if additional view besides pathway diagrams were used. These pathway visualizations are often linked to other type of visualization methods (e.g., parallel co-ordinates) using the concept of brushing and linking. The visualization alternatives from pathway + microarray data design space were evaluated by conducting two independent user studies. Both the studies used timeseries datasets. The first study used visualization alternatives from both dimension 1 and dimension 2. The results suggest that the method to overlay multidimensional data on pathway nodes has a non trivial influence on accuracy of participants' responses, whereas the number of visualizations affect participants' performance time for pre-selected tasks. The second study used visualization alternatives from dimension 1 that focuses on method used to overlay data attributes on pathway nodes. The study suggests that participants using pathway visualization that display data one attribute at a time on nodes have more controlled performance for all type of tasks as compared to the participants using other alternatives. Participants using pathway visualization that display data in node-as-glyphs view have better performance for tasks that require analysis for a single node, and identifying outlier nodes. Whereas, pathway visualizations with pathways-as-glyph view provide better performance on tasks that require analysis of overall changes in the pathway, and identifying interesting timepoints in the data. An insight-based method was designed to evaluate visualization tools for real world biologists' data analysis scenarios. The insight-based method uses different quantifiable characteristics of an "insight" that can be measured uniformly across participants. These characteristics were identified based on observations of the participants analyzing microarray data in a pilot study. The insight-based method provides an alternative to traditional task-based methods. This is especially helpful for evaluating visualization tools on large and complicated datasets where designing tasks can be difficult. Though, the insight-based method was developed to empirically evaluate visualization tools for short term studies, the method can also be used in real world longitudinal studies that analyzes the usage of visualization tools by the intended end-users. / Ph. D.

Since the 1960s, the Phillips curve has survived various significant changes (Kuhnian paradigm shifts) in macroeconomic theory and generated endless controversies. This dissertation revisits several important, representative papers throughout the curve's four historical, formative periods: Phillips' foundational paper in 1958, the wage determination literature in the 1960s, the expectations-augmented Phillips curve in the 1970s, and the latest New Keynesian iteration. The purpose is to provide a retrospective evaluation of the curve's empirical evidence. In each period, the preeminent role of the theoretical considerations over statistical learning from the data is first explored. To further appraise the trustworthiness of empirical evidence, a few key empirical models are then selected and evaluated for their statistical adequacy, which refers to the validity of the probabilistic assumptions comprising the statistical models. The evaluation results, using the historical (vintage) data in the first three periods and the modern data in the final one, show that nearly all of the models in the appraisal are misspecified - at least one probabilistic assumption is not valid. The statistically adequate models produced from the respecification with the same data suggest new understandings of the main variables' behaviors. The dissertations' findings from the representative papers cast doubt on the traditional narrative of the Phillips curve, which the representative papers play a crucial role in establishing. / Doctor of Philosophy / The empirical regularity of the Phillips curve, which captures the inverse relationship between the inflation and unemployment rates, has been widely debated in academic economic research and between policymakers in the last 60 years. To shed light on the debate, this dissertation examines a selected list of influential, representative studies from the Phillips curves' empirical history through its four formative periods. The examinations of these papers are conducted as a blend between a discussion on the methodology of econometrics (the primary quantitative method in economics), the role of theory vs. statistical learning from the observed data, and evaluations of the validity of the probabilistic assumptions assumed behind the empirical models. The main contention is that any departure of probabilistic assumptions produces unreliable statistical inference, rendering the empirical analysis untrustworthy. The evaluation results show that nearly all of the models in the appraisal are untrustworthy - at least one assumption is not valid. Then, an attempt to produce improved empirical models is made to produce new understandings. Overall, the dissertation's findings cast doubt on the traditional narrative of the Phillips curve, which the representative papers play a crucial role in establishing.

In this thesis, we measure the impact of reference semantics on programming and reasoning. We designed a survey to compare how well programmers perform under three different programming paradigms. Two of the paradigms, object-copying and swapping use value semantics, while the third, reference-copying, uses reference semantics. We gave the survey to over 25 people. We recorded number of questions answered correctly in each paradigm and the amount of time it took to answer each question. We looked at the overall results as well as the results within various levels of Java experience. Based on anecdotal evidence from the literature, we expected questions based on value semantics to be easier than questions based on reference semantics. The results did not yield differences that were statistically significant, but they did conform to our general expectations. While further investigation is clearly needed, we believe that this work represents an important first step in the empirical analysis of a topic that has previously only been discussed informally. / Master of Science