Button to scroll to the top of the page.

Updates

Spring 2020 Colloquium: Graduate Portfolio in Applied Statistical Modeling

Speaker

Date

Time

Title and Zoom link

Christina Caramanis

May 11

11:30am - Noon

"Income and Stable Center-Based Child Care: Evidence from the Earned Income Tax Credit"

Zoom link: https://utexas.zoom.us/j/99436683504

Hai Hang May 11 Noon - 12:30pm

"The effectiveness of a mindfulness-based intervention for promoting positive classroom behaviors and reducing negative classroom behaviors among elementary school students: A single case experimental design study"

Zoom link: https://utexas.zoom.us/j/99436683504

Trinh Hoang May 11 1:00 - 1:30 pm

"Understanding Traffic Network Behavior During Smartphone Navigation System Failures"

Zoom Link: https://utexas.zoom.us/j/99436683504

Lingzi Zhong May 11 1:30 - 2:00 pm

 "Relational Implications of Partner Confirmation and the Moderating Effects of Attachment for Individuals with Depression in Committed Romantic Relationships"

Zoom link: https://utexas.zoom.us/j/99436683504

Athula Pudhiyidath May 12 11:30am - Noon

 "Modeling temporal bias with successor representation"

Zoom link: https://utexas.zoom.us/j/95804737444

Jodie Simkoff May 12 Noon - 12:30 pm

"Stochastic scheduling and control using data-driven nonlinear dynamic models: application to demand response operation of a chlor-alkali plant" 

Zoom link: https://utexas.zoom.us/j/95804737444

Christian Vazquez May 12 1:00-1:30pm

 "Faith Communities’ Improvements in Readiness to Engage in Addictions Recovery Programming"

Zoom link: https://utexas.zoom.us/j/95804737444

Preeti Chopra May 12 1:30 - 2:00pm

 "Determining the relationship between personality and depression using MIDUS data"

Zoom link: https://utexas.zoom.us/j/95804737444

Mayowa Oyedere May 13 11:00-11:30am

 "ROP Optimization Using Machine Learning Cassification algorithms"

Zoom link: https://utexas.zoom.us/j/99977675870

Yuan-I Chen May 13 11:30am - Noon

 "Evaluation of Fitting Algorithms for Fluorescence Lifetime via Monte Carlo Simulation"

Zoom link: https://utexas.zoom.us/j/99977675870

Morgan Kelley May 13 1:00-1:30pm

 "Demand response scheduling under uncertainty: chance-constrained framework and application to an air separation unit"

Zoom link: https://utexas.zoom.us/j/99977675870

Lisa Panisch May 13 1:30 - 2:00pm

 "An Exploratory Factor Analysis of the Parent Assessment of Protective Factors Survey Measure with a Sample of Parents at risk for Child Maltreatment in the State of Texas"

Zoom link: https://utexas.zoom.us/j/99977675870

Puneet Seth May 14 11:30am - Noon

 "Using Data Modeling and Pattern Recognition to Characterize Offset Well Pressure Response"

Zoom link: https://utexas.zoom.us/j/95436449494


Christina Caramanis

Abstract

Research clearly shows that early childhood poverty is a strong predictor of academic achievement and lasting adversity throughout life. Formal child care settings, such as center-based care, can serve to mitigate early life disadvantage, yet stable access to center-based child care is often out of reach for low-income families. A key policy question that remains unaddressed in the literature is whether modest increases in the incomes of working poor families would result in increased use of center-based child care. Further, identifying the effects of income on child care selection outcomes is inherently difficult, owing to the endogeneity of income. Applying an instrumental variables strategy to the analysis of longitudinal survey data (n = 2,548), I leverage state variation in access to the Earned Income Tax Credit (EITC) to predict the extent to which differences in family income affect stability and selection of child care arrangements. Results indicate that exogenous income increases (as a result of receiving state EITC transfers) are associated with significantly higher rates of center-based child care enrollment at age 3 and a higher likelihood of remaining in formal center-based arrangements between ages 1 and 3. Given a strong evidence base linking early child care type and stability with academic outcomes and human capital development, findings from this study suggest a policy-relevant mechanism by which increased income may decrease academic inequalities among children.

Hai Hang

Abstract

MindUP is a mindfulness-based social-emotional learning program that is designed to be implemented by teachers in school settings. Studies support MindUP’s effects on academic, social-emotional, and physical outcomes. However, most of the studies on MindUP focused on populations of students in general education classrooms, and did not distinguish between students with special behavioral challenges. As such, the effectiveness of MindUP among populations of youth with special behavioral needs has yet to be determined.The purpose of this study is to examine whether MindUP, can effectively improve the classroom conduct of students with behavioral challenges. Study hypotheses are: after receiving the intervention, students will demonstrate (a) more frequent positive classroom behaviors including both active and passive on-task behaviors, and (b) less frequent negative classroom behaviors including both disruptive off-task behaviors and non-disruptive off-task behaviors. This study used a single case experimental design (A-B design) and consisted of three phases—baseline phase, intervention phase, and follow-up phase. Primary data collection was conducted through direct behavioral observation of 11 5th-grade students in their natural classroom environment using the Direct Behavior Rating Single Item Scale (DBR-SIS). Treatment outcomes included positive classroom behaviors (active on-task behaviors and passive on-task behaviors) and negative classroom behaviors (disruptive off-task behaviors and non-disruptive off-task behaviors).Visual analysis, percentage of non-overlapping data analysis, and multilevel analysis were conducted to examine treatment effects. Analyses results consistently showed changes in passive on-task behaviors and non-disruptive off-task behaviors in the expected direction—an increase in passive on-task behaviors and a decrease in non-disruptive off-task behaviors. However, contrary to the hypothesis, active on-task behaviors decreased during the intervention and follow-up phases. No significant changes in disruptive off-task behaviors were found. MindUP showed promise in effectively increasing elementary school students’ passive on-task behaviors and decreasing non-disruptive off-task behaviors. However, it demonstrated an unexpected effect on active on-task behaviors and no effect on disruptive off-task behaviors. Implications for practice and future research are provided.

Trinh Hoang

Abstract

Smartphone navigation applications providing real-time traffic information have become increasingly popular among drivers. Studies have showed that drivers’ cognitive abilities for spatial navigation would deteriorate with frequent use of such applications, which eventually leads to drivers becoming unable to navigate without the applications’ assistance. As there exist many risks for system-wide failures of navigation applications, inadequate navigating cognition among drivers can potentially induce abrupt driver reactions, causing temporary disruptions in the urban traffic networks. However, research in transportation network resilience has yet to explore the potential impacts of such disruptions. Inthis project, a survey is developed to collect drivers’ stated preferences on multiple scenarios related to failures with navigation applications. Linear regression models are applied on the collected data to understand how different factors, such as age and gender, affect driver’s decision. Following the survey results, a simulation is developed to capture the network-level effects of such short-term disruptions using the downtown Austin traffic network. Statistical methods are then applied on the simulation output to estimate mean performance measures of the traffic network during the disruptions.

Lingzi Zhong

Abstract

Given the adverse effects of depression on individual and relational well-being, depressed individuals who are in committed romantic relationships oftentimes experience heightened relational uncertainty and tend to engage in reassurance seeking (ERS) behaviors with their partners. Drawing upon confirmation theory, the current research sought to examine the associations between perceived partner confirmation (acceptance and challenge), relational uncertainty (self, partner, and relationship uncertainty), and ERS behaviors. Additionally, depressed individuals’ attachment (avoidance and anxiety) was considered as a potential moderator that shed light on the proposed associations. A total of 222 individuals who have been diagnosed with depression and are currently in a committed relationship were recruited to participate in an online study. A series of regression analyses were conducted. Results indicated that perceived partner acceptance had a positive effect on reducing depressed individuals’ relational uncertainty and ERS behaviors, whereas perceived partner challenge was positively associated with relational uncertainty and ERS. Depressed individuals’ avoidance significantly moderated the association between depressed individuals’ perceived partner acceptance and perceived partner and relationship uncertainty. A three-way interaction was also found among avoidance, partner acceptance, and partner challenge with regard to depressed individuals’ perceived relationship uncertainty. Theoretical implications for the extension of confirmation theory to the context of mental health as well as practical recommendations for depressed individuals and their romantic partners are discussed. 

Athula Pudhiyadath

Abstract

In this experiment, participants are presented with an iterative sequence of objects on a screen whilst we scan their brain activity with functional resonance magnetic imaging (fMRI). Unbeknownst to the participants, the order in which objects appeared on the screen followed a network structure called the temporal community structure (Schapiro et al., 2013). Through this network structure, transition probabilities between objects presented on the screen is uniform, but the architecture of the network resulted in three groups or temporal communities of objects, such that objects belonging to a particular community are always mutually predictive of one another, and never of the objects belonging to another community. While previous research suggests that individuals are able to successfully track temporally relevant information and code the time and frequencies with which events co-occur, it is unknown how much these temporal statistics influence domains outside of memory, such as decision-making. With the present experiment, we wanted to measure how the initial encoding of a sequence of objects mediated by temporal community structure would affect how participants’ make inferences about relationships between the objects beyond their direct temporal statistics. One computational account for how we might encode temporal statistics is called successor representation or SR (Dayan, 1993; Gershman, 2018). The main idea of the successor representation (SR) model is that we do not encode events separately per se, but instead we encode events as a predictive representation of future events or states given the current state. The SR does not model the environment by encoding each transition probability between states, but rather as the mean discounted count of future visitations to a state (Momennejad, 2020). In the present experiment, we used SR to model how participants’ neural representations of temporal relationships (captured via fMRI signals) can predict participants’ temporal biases when making reasoning decisions in subsequent inference tasks. 

Jodie Simkoff

Abstract

Modern chemical manufacturers are increasingly interested in improving the agility of their operations. This is particularly true in regions with deregulated electricity markets, where participation in demand response programs or taking advantage of real-time prices can be profitable. Processes that consume electricity as a primary feedstock and can shift operating points quickly, such as the chlor alkali process, are a prime candidate for such participation; however, process dynamics that evolve over a longer time scale typically need to be accounted for to ensure safe operation and on-spec product. This calls for an integrated scheduling and control approach. However, the resulting optimization problems are computationally challenging due to multiple time-scales and the relatively long horizons involved (on the order of days). We present a methodology for efficient representation of nonlinear process dynamics using novel parameterizations of Hammerstein-Wiener models. We carry out an extensive study concerning real-time electricity market participation of a chlor-alkali process, focusing on the optimal allocation of bids in the day ahead and real-time markets under electricity price and product demand uncertainty.

Christian Vazquez

Abstract

Spirituality and religion are well-documented and essential components of prevention, treatment, and recovery of substance use disorders. Faith communities in particular can play an essential part in this, but there can be undesirable outcomes for those navigating substance use disorders and seeking help from the faith community when their congregation is not prepared to support an individual or group in need. Research is limited in how best to support faith communities at the organizational level in undertaking this critical work. This study highlights key predictors of faith communities’ improvements in readiness to engage in addictions recovery programming. Seven faith-based congregations engaged in leadership and congregation team ministry development programming for 24 months. Pre- and post-test data was collected from the congregants of these faith-based congregations at baseline and 24-months. The outcome of interest was the change in organizational readiness from baseline to 24-months after participation in the programming. Three variables captured at pre- and post-test were included as key independent variables. Perceived importancecaptured members’ beliefs on the importance that the faith community helps those with addictions to alcohol and/or drugs. Awareness of resourcescaptured members’ knowledge of supportive resources within the congregation or community with regard to handling addiction-related issues. Supportive communitycaptured one’s perception of the supportiveness of their congregation’s environment with regard to handling addiction-related issues. Using mixed effects linear regression,  a main effects model was tested with readiness to support recovery from substance use disorders as the dependent variable; a time variable as a covariate to differentiate between pre- and post-test, the three key independent variables, and demographic variables (faith tradition and geographic region) in the model. Three separate follow-up analyses were conducted using the same main effects model as above, with the addition of a single interaction variable consisting of one of the three key independent variables and the time. Interactions were probed by comparing marginal means. A t-test indicated there was a significant difference in the scores for congregational readiness at baseline (M=1.59, SD=1.42) and 24-months (M=2.11, SD=1.56); t(1098)=-5.83, p< .001). Main effects for the mixed effects linear regression model indicated that awareness of resources (B=0.40, 95% CI: 0.23, 0.56), supportive community (B=0.64, 95% CI: 0.58, 0.71), and perceived importance (B=0.18, 95% CI: 0.06, 0.29) were significant predictors of higher congregational readiness score. Random effects analyses resulted in no significant variability among congregations (0.004, 95% CI: <.001, 212.82). Follow-up analyses examining interactions between each of the three key independent variables and time resulted in a significant interaction between time and awareness of resources (B=0.34, 95% CI: 0.005, 0.67). On average, there was a 0.31 (95% CI: 0.13, 0.48) change in congregational readiness score for those who said yes to awareness of resources at baseline and 24-months. Based on these findings, awareness of resources within the congregation or community with regard to handling addiction-related issues is a key lever of change for congregations to focus on to begin to support recovery from substance use disorders. Promotional activities to increase awareness of existing congregational recovery support resources and activities, and testimonials from members who endorse congregational engagement in the program may be important actions for congregations to take.

Preeti Chopra

Abstract

Twenty percent of the people in the U.S. experience mental illness. Depression is one of the most dominant forms of mental illness and impacts millions of lives. While interventions have been effective in treating depression, the association between personality and Depression is not well known. Personality is generally measured in terms of openness, conscientiousness, extroversion, agreeableness, and neuroticism. Significant changes in neuroticism of individuals who experience mental illness has been observed.The personality measures such as neuroticism and conscientiousness were significantly different for retired individuals who experienced depression after 4 years, as compared to individuals who did not experience depression.However, the effect of other measures of personality on depression and other mental health issues are not clear. In addition, it has not been tested if such a relationship between mental health and personality exists in younger and non-retired population. Therefore, in this paper we study the association between mental issues and personality in a representative sample. We use survey, data of the three waves ofThe Midlife in the United States (MIDUS), namely MIDUS 1 (n = 7,108), MIDUS 2 (n=4,963) and MIDUS 3 (n = 3,294). MIDUS is a longitudinal study involving participants above the age of 20 years.Logistic regression analysis was used to identify the relationship between personality and mental health.  There is a significant relationship between personality and depression. Individuals were more likely to report depression in MIDUS 2 if they exhibited higher level of neuroticism (p<0.01) on MIDUS 1. Similarly, individuals with higher level of neuroticism on MIDUS 2 were more likely to report depression (p<0.01) in MIDUS 3. Individuals were more likely to report depressive symptoms in MIDUS 3, if they reported higher level of neuroticism (p<0.01) in MIDUS 1. Personality and depression are related.Specifically, neuroticism isstrongly related to depression. Findings of the study will allow us to identify vulnerable population and improving interventions. 

Mayowa Oyedere

Abstract

Drilling optimization has consistently generated research interests over the years because of the cost saving benefits associated with improved drilling efficiency. As a result, rate of penetration (ROP) prediction has become critical to successful drilling optimization efforts. Several physics-based and data-driven models have been developed for ROP prediction and majority of the data-driven models use regression-based approaches. This paper introduces a new approach to ROP prediction by modeling it as a classification problem consisting of two regions (low and high ROP) based on a user-defined threshold. ROP is modeled as a function of weight-on-bit (WOB), flow rate, rotary speed (RPM) and unconfined compressive strength (UCS). Five different machine learning classification algorithms - logistic regression, linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), support vector machines (SVM) and random forest were implemented in this paper to develop the classification model. Using the Area Under Curve (AUC) as the classification performance metric, results from the simulations showed that the best classifier should be chosen for each formation. A parametric study showed that the choice of threshold also impacted the performance of the classifiers. Finally, for practical application of this approach to ROP prediction, a probability gradient heat map of RPM and WOB was developed as a tool to help the driller make informed decisions on the combinations of RPM and WOB values that would yield the desired regions of ROP.

Yuan-I Chen

Abstract

Fluorescence lifetime imaging microscopy (FLIM) can provide more quantitative information about molecular effects in the cellular environment as it does not depend on the fluorophore concentration. While low-light-level FLIM minimizes phototoxicity and photobleaching in biological applications, the analytical bias derived from complex data fitting makes the precise lifetime estimation challenging. Although many analysis methods have been developed to improve the estimated results, there are no standard criteria to validate the robustness of those methods over a range of different photon counts. It is necessary to establish a standard operating procedure (SOP) for testing various models. Here I proposed a Monte Carlo (MC)-based approach to generate fluorescence lifetime decay histogram in-silico.MC simulation allows us to assign the exact number of photons in the synthetic data, providing a quantitative way to investigate the limitation of the presented fluorescence lifetime analysis methods. I proved that maximum likelihood estimator (TD_MLE) generated the most accurate lifetime estimates at low photon count, matching well with the ground truth (MSE =0.07, SSIM = 0.99; 150 photon counts) using the simulated printed letter, “UTBME,” fluorescence decay imaging. Considering the potential artifacts introduced during the data acquisition, the proposed SOP provides a simple, and precise way for algorithms comparison and even model training for state-of-the-art deep learning technique.

Morgan Kelley

Abstract

Recent increases in renewable power generation challenge the operation of the power grid: generation rates fluctuate in time and are not synchronized with power demand fluctuations. Demand response (DR)  consists of adjusting user electricity demand in order to balance available power supply. Chemical plants are  appealing candidates for DR programs; they offer large, concentrated loads that can be modulated via production scheduling. Price-based DR is a common means of engaging industrial entities; its benefits increase significantly when a longer (typically, a few days) scheduling time horizon is considered. DR production scheduling comes with its own  challenges, related to uncertainty in future (i.e., forecast) electricity prices and product demand. In this work, we provide a framework for DR production scheduling under uncertainty based on a chance-constrained formulation, that also accounts for the dynamics of the production facility.  The ideas are illustrated with an air separation unit case study.

Lisa Panisch

Abstract

A parental history of adverse childhood experiences can heighten the risk of child maltreatment and neglect, thereby perpetuating intergenerational patterns of trauma transmission. Individual and family level risks are compounded by systemic factors, such as poverty and related access barriers to necessary resources, and these systemic risk factors are further exacerbated by historical patterns of racial and ethnic discrimination and oppression in the United States. While researchers, policy stakeholders, and practitioners in the area of children and family services have traditionally emphasized an understanding risk factors for child maltreatment and neglect, there is an increasing amount of focus on identifying and strengthening protective factors that could mitigate such risk. These strengths-based approaches use instruments like the Parent Assessment of Protective Factors (PAPF), a 36-item survey tool with four subscales measuring different protective factors, namely social connections, social and emotional competence of children, parental resilience, and concrete support. Thus far, however, the PAPF has only been validated with a low-risk, sample of primarily White, English-speaking, well-educated parents. In order to ensure measurement tools like the PAPF are suitable for use with parents at-risk of child maltreatment, it is important for validation studies to be conducted with individuals that reflect the population they are intended to serve. The purpose of this study is to investigate the validity of the factor structure of the PAPF measure with a group of adversity-exposed, ethnically diverse parents via an exploratory factor analysis. Data in the current analysis was gleaned from a program evaluation of a child maltreatment intervention conducted with in 28 counties throughout the State of Texas. The PAPF was completed by a group of at-risk, adversity-exposed, ethnically diverse English and Spanish speaking parents and primary caregivers (N= 773) who participated in the intervention. First, a principal component analysis was conducted using Varimax rotation. Four main factors were identified that explained 68.86% of the cumulative variance in the data. An exploratory factor analysis was then conducted with Varimax rotation, using the four identified factors. All assumptions for conducting an exploratory factor analysis were evaluated and satisfied.The original factor labels corresponding with the four subscales of the PAPF, as outlined by Kiplinger and Harper Browne (2014), fit the extracted factors. Therefore, the factor labels and subscale names remained unchanged as a result of this analysis. All 36 items had primary factor loadings between .5 - .9, and 25 items did not have any cross-loading value above three. Of the 11 remaining items, the highest cross-loading factors were between .3 - .41. A reliability analysis using Cronbach’s alpha was conducted for each of the four factors to determine internal consistency, and excellent reliability was observed with Cronbach’s alpha values for each subscale ranging from .92-.95. These preliminary findings provide evidence in support of the PAPF as a valid and reliable instrument appropriate for use with ethnically diverse, English and Spanish speaking parents living in the State of Texas who are at risk for child maltreatment. Full validation is needed via future investigations using confirmatory factor analysis. Directions for future research include additional validation studies with groups of at-risk parents with different regional and demographic characteristics. If validity of the PAPF continues to be demonstrated, it can be used to help identify and bolster protective factors among at-risk parents, thereby reducing the risk of child maltreatment and enhancing the health and well-being of vulnerable families within society.

 

Puneet Seth

Abstract

During fracturing, pressure responses are often observed in a nearby offset monitor well as hydraulic fractures propagate from the treatment well towards the monitor well. These pressure responses can be caused by, (a) purely elastic interactions between the treatment and monitor well fractures, (b) a combination of elastic interaction and hydraulic connection between the fractures (mixed response) or (c) massive direct frac-hits from the treatment into the monitor well fractures. In this work, an automated data modeling workflow to systematically identify and interpret the different types of pressure responses is demonstrated. An automated pattern recognition workflow based on Python scripting has been developed that parses offset well pressure data during fracturing from multiple wells, stage-by-stage, in each well. The script develops overlay-plots containing treatment and monitor well pressure for each stage, which can be stored in a directory of the user’s choice (for future reference). The script then automatically determines the magnitude of pressure response as well as the type of pressure interference (“purely-elastic”, “mixed” [elastic + hydraulic] or “direct frac-hit” and the output is automatically stored stage-by-stage in a user-friendly text delimited (“.txt”, “.csv” or “.xlsx”) format while the script executes. In addition, the script can also calculate the relative distance between interacting stages. In case of a purely elastic response, pressure fall-off is observed in the monitor well as soon as the nearby treatment well is shut-in (Seth et al., 2019). This is an important distinction between purely elastic responses and other types of pressure responses where a pressure increase is observed even after the nearby treatment well is shut-in. The magnitude of pressure response also varies with the type of pressure response. Typically, purely elastic pressure responses range between 1-100 psi (sometimes higher) depending upon the distance and overlap between the interacting fractures, whereas mixed pressure responses range between 10s-100s of psi. Direct frac-hits usually cause a massive increase in the offset monitor well pressure (100s-1000s of psi) and are relatively easy to spot visually as they disrupt the pressure response trend. It is crucial to correctly identify and interpret the type of pressure interference before using offset well pressure data for further analysis (such as fracture geometry estimation). This work details the different types of pressure responses typically observed in field data and provides guidelines on identifying and characterizing these responses correctly. In addition, the automated pattern recognition workflow demonstrated in this study introduces a novel tool to systematically parse and characterize offset well pressure data efficiently.