Factors associated with prompt difficulty in automated essay scoring

This study explores factors related to the prompt difficulty in Automated Essay Scoring. The sample was composed of 6,924 students. For each student, there were 1-4 essays, across 20 different writing prompts, for a total of 20,243 essays. E-rater® v.2 essay scoring engine developed by the Educational Testing Service was used to score the essays. The scoring engine employs a statistical model that incorporates 10 predictors associated with writing characteristics of which 8 were used. The Rasch partial credit analysis was applied to the scores to determine the difficulty levels of prompts. In addition, the scores were used as outcomes in the series of hierarchical linear models (HLM) in which students and prompts constituted the cross-classification levels. This methodology was used to explore the partitioning of the essay score variance.^ The results indicated significant differences in prompt difficulty levels due to genre. Descriptive prompts, as a group, were found to be more difficult than the persuasive prompts. In addition, the essay score variance was partitioned between students and prompts. The amount of the essay score variance that lies between prompts was found to be relatively small (4 to 7 percent). When the essay-level, student-level-and prompt-level predictors were included in the model, it was able to explain almost all variance that lies between prompts. Since in most high-stakes writing assessments only 1-2 prompts per students are used, the essay score variance that lies between prompts represents an undesirable or "noise" variation. Identifying factors associated with this "noise" variance may prove to be important for prompt writing and for constructing Automated Essay Scoring mechanisms for weighting prompt difficulty when assigning essay score.^

The thermodynamics and statistical mechanics of protein folding

The physics of self-organization and complexity is manifested on a variety of biological scales, from large ecosystems to the molecular level. Protein molecules exhibit characteristics of complex systems in terms of their structure, dynamics, and function. Proteins have the extraordinary ability to fold to a specific functional three-dimensional shape, starting from a random coil, in a biologically relevant time. How they accomplish this is one of the secrets of life. In this work, theoretical research into understanding this remarkable behavior is discussed. Thermodynamic and statistical mechanical tools are used in order to investigate the protein folding dynamics and stability. Theoretical analyses of the results from computer simulation of the dynamics of a four-helix bundle show that the excluded volume entropic effects are very important in protein dynamics and crucial for protein stability. The dramatic effects of changing the size of sidechains imply that a strategic placement of amino acid residues with a particular size may be an important consideration in protein engineering. Another investigation deals with modeling protein structural transitions as a phase transition. Using finite size scaling theory, the nature of unfolding transition of a four-helix bundle protein was investigated and critical exponents for the transition were calculated for various hydrophobic strengths in the core. It is found that the order of the transition changes from first to higher order as the strength of the hydrophobic interaction in the core region is significantly increased. Finally, a detailed kinetic and thermodynamic analysis was carried out in a model two-helix bundle. The connection between the structural free-energy landscape and folding kinetics was quantified. I show how simple protein engineering, by changing the hydropathy of a small number of amino acids, can enhance protein folding by significantly changing the free energy landscape so that kinetic traps are removed. The results have general applicability in protein engineering as well as understanding the underlying physical mechanisms of protein folding. ^

Reduced-parameter model of head-related transfer functions for synthesized spatial audio

Digital systems can generate left and right audio channels that create the effect of virtual sound source placement (spatialization) by processing an audio signal through pairs of Head-Related Transfer Functions (HRTFs) or, equivalently, Head-Related Impulse Responses (HRIRs). The spatialization effect is better when individually-measured HRTFs or HRIRs are used than when generic ones (e.g., from a mannequin) are used. However, the measurement process is not available to the majority of users. There is ongoing interest to find mechanisms to customize HRTFs or HRIRs to a specific user, in order to achieve an improved spatialization effect for that subject. Unfortunately, the current models used for HRTFs and HRIRs contain over a hundred parameters and none of those parameters can be easily related to the characteristics of the subject. This dissertation proposes an alternative model for the representation of HRTFs, which contains at most 30 parameters, all of which have a defined functional significance. It also presents methods to obtain the value of parameters in the model to make it approximately equivalent to an individually-measured HRTF. This conversion is achieved by the systematic deconstruction of HRIR sequences through an augmented version of the Hankel Total Least Squares (HTLS) decomposition approach. An average 95% match (fit) was observed between the original HRIRs and those re-constructed from the Damped and Delayed Sinusoids (DDSs) found by the decomposition process, for ipsilateral source locations. The dissertation also introduces and evaluates an HRIR customization procedure, based on a multilinear model implemented through a 3-mode tensor, for mapping of anatomical data from the subjects to the HRIR sequences at different sound source locations. This model uses the Higher-Order Singular Value Decomposition (HOSVD) method to represent the HRIRs and is capable of generating customized HRIRs from easily attainable anatomical measurements of a new intended user of the system. Listening tests were performed to compare the spatialization performance of customized, generic and individually-measured HRIRs when they are used for synthesized spatial audio. Statistical analysis of the results confirms that the type of HRIRs used for spatialization is a significant factor in the spatialization success, with the customized HRIRs yielding better results than generic HRIRs.

Evaluation of the internal structural validity of the Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition) anxiety disorders in children and adolescents

The purpose of the present dissertation was to evaluate the internal validity of symptoms of four common anxiety disorders included in the Diagnostic and Statistical Manual of Mental Disorders fourth edition (text revision) (DSM-IV-TR; American Psychiatric Association, 2000), namely, separation anxiety disorder (SAD), social phobia (SOP), specific phobia (SP), and generalized anxiety disorder (GAD), in a sample of 625 youth (ages 6 to 17 years) referred to an anxiety disorders clinic and 479 parents. Confirmatory factor analyses (CFAs) were conducted on the dichotomous items of the SAD, SOP, SP, and GAD sections of the youth and parent versions of the Anxiety Disorders Interview Schedule for DSM-IV (ADIS-IV: C/P; Silverman & Albano, 1996) to test and compare a number of factor models including a factor model based on the DSM. Contrary to predictions, findings from CFAs showed that a correlated model with five factors of SAD, SOP, SP, GAD worry, and GAD somatic distress, provided the best fit of the youth data as well as the parent data. Multiple group CFAs supported the metric invariance of the correlated five factor model across boys and girls. Thus, the present study’s finding supports the internal validity of DSM-IV SAD, SOP, and SP, but raises doubt regarding the internal validity of GAD.^

Statistical Models for Predicting College Success

Colleges base their admission decisions on a number of factors to determine which applicants have the potential to succeed. This study utilized data for students that graduated from Florida International University between 2006 and 2012. Two models were developed (one using SAT as the principal explanatory variable and the other using ACT as the principal explanatory variable) to predict college success, measured using the student’s college grade point average at graduation. Some of the other factors that were used to make these predictions were high school performance, socioeconomic status, major, gender, and ethnicity. The model using ACT had a higher R^2 but the model using SAT had a lower mean square error. African Americans had a significantly lower college grade point average than graduates of other ethnicities. Females had a significantly higher college grade point average than males.

Comparison of Some Improved Estimators for Linear Regression Model under Different Conditions

Multiple linear regression model plays a key role in statistical inference and it has extensive applications in business, environmental, physical and social sciences. Multicollinearity has been a considerable problem in multiple regression analysis. When the regressor variables are multicollinear, it becomes difficult to make precise statistical inferences about the regression coefficients. There are some statistical methods that can be used, which are discussed in this thesis are ridge regression, Liu, two parameter biased and LASSO estimators. Firstly, an analytical comparison on the basis of risk was made among ridge, Liu and LASSO estimators under orthonormal regression model. I found that LASSO dominates least squares, ridge and Liu estimators over a significant portion of the parameter space for large dimension. Secondly, a simulation study was conducted to compare performance of ridge, Liu and two parameter biased estimator by their mean squared error criterion. I found that two parameter biased estimator performs better than its corresponding ridge regression estimator. Overall, Liu estimator performs better than both ridge and two parameter biased estimator.