An Anthology of Missingness: Simulated Missing Data, Multivariate Tilt, and Environmental Exposures in Late Adulthood

Dworak, Elizabeth M.

doi:https://doi.org/10.21985/n2-jk80-j993

Work

An Anthology of Missingness: Simulated Missing Data, Multivariate Tilt, and Environmental Exposures in Late Adulthood

Public

Download PDF

Download All Files (.zip)

Missing data are often described as an annoyance in research and generally presented as a source of error or bias during analysis. To this end, methods that center on planned missingness are an underappreciated and powerful tool that can actually be used to improve data collection and the validity of research. Using a series of four studies, this dissertation showcases how methods that allow for missing data are effective in hopes to encourage researchers to embrace missing data designs within their own research. In particular, the studies presented in this dissertation outline the strengths of using a collection of procedures called Synthetic Aperture Personality Assessment (SAPA) techniques. These techniques are two-fold in that they refer to the method of data collection and the procedures emphasized for analysis. Foremost, SAPA methods are often used to describe the way in which both items and participants are sampled. Although the method of sampling participants is less important for these studies, SAPA is distinct from other item matrix sampling procedures in that items are selected using stratified matrix sampling with unequal probabilities from a larger item bank rather than a simple random sample with equal probabilities. Secondly, SAPA techniques don’t require imputation methods in order to analyze massively missing at random data. Rather, SAPA procedures often apply linear algebra to generate synthetic covariance matrices on pairwise completed items that can then be used to do structural analyses, reliability, and form the correlations between scales. Using these methods, Chapter 2 demonstrates how data collection using experience sampling methodology are optimized from missing data designs. Using Monte Carlo simulations and real data collected, Chapter 2 shows how planned missingness can be applied to studies and reduce burden by simulating SAPA procedures on both datasets. Ultimately, this work exhibits that synthetic covariance matrices can be used to recover within and between person parameters and factor structures using only 40% to 50% of the original data. Furthermore, this study finds how may administrations and pairwise administrations are required to recover covariance matrices and factor structures within-participant and between-participant. Together, this work indicates that SAPA techniques are a viable and effective way for making data collection with experience sampling methodology more efficient without biasing parameters. As such, this study provides evidence and suggestions to help researchers determine what proportion of missing data or sampling probabilities are tolerable for their own experience sampling methods data collection in relation to the number items and within person observations they’re administering. Building on this work, the second study (Chapter 3) shows the utility and power of SAPA procedures with data collected exclusively with a planned missingness design. Outside of cognitive reserve theory, little research has been done to examine the relationship between individual differences in cognitive ability and interests. Thus, Chapter 3 examined multivariate tilt by investigating how intraindividual patterns of the hyperplane of ability relate to behavioral preferences in recreational activities. Using a new supervised statistical learning technique called Best Items Scale that is Cross-validated, Unit-weighted, Informative and Transparent (BISCUIT), this study found that while behavioral frequency items have small reliable associations in predicting cognitive ability scores, that attempts to identify a unique set of behavioral items that were predictive of multivariate patterns in cognitive ability (tilt) were inconclusive. Chapter 4 and Chapter 5 further show the power of missing data designs by linking data from the SAPA Project to publicly available regional datasets to gain further information about participants and their local environment. Together, these studies investigated how regional differences in pollution and climate exposure were associated with geospatial variance in cognitive ability (Chapter 4) and temperament (Chapter 5). Using a sample of older adults from the United States these studies investigated if cognitive ability and temperament scores are associated with regional difference in local environmental pollution and climate exposure above and beyond individual factors and regional differences in ecological exposures. Furthermore, this study explored whether any environmental determinants accounted for greater regional differences in cognitive ability and temperament scores. Findings of these studies indicate that cognitive ability and temperament scores not only show unique geospatial distributions across the United States for older adults but have varying associations with ecological exposures after controlling for regional differences in opportunities and individual covariates. Specifically, Chapter 4 found that composite scores of extreme heat exposure and regional ozone concentration were negatively associated with overall cognitive ability scores at both the regional levels of ZIP Code Tabulation Area and county. Likewise, this predictor was negatively related to letter and number series scores at the county level. Foremost, as most geospatial research on temperament relies on examining the Big Five, Chapter 5 justifies the need for researchers to use narrower measures of temperament traits in future studies. Results of Chapter 5 found the composite score of extreme heat exposure and regional ozone concentration were positively related to conscientiousness, conservatism, authoritarianism, perfectionism, order, industry, emotional stability, and conformity at the ZIP Code Tabulation Area level and conscientiousness, conservatism, authoritarianism, perfectionism, industry, and sensation seeking at the county level. This composite score was also negatively related to agreeableness, neuroticism, openness, compassion, trust, anxiety, introspection, and art appreciation at the ZIP Code Tabulation Area level and negatively related to agreeableness, openness, compassion, trust, emotional expressiveness, humor, introspection, and art appreciation at the county level. Beyond regional measures of extreme heat and ozone concentration, a handful of temperament traits were also associated with composite scores of airborne microparticles (PM≤2.5) concentration and industrial pollutants in air, water, or soil. At the ZIP Code Tabulation Area level, this predictor was negatively related to sensation seeking and positively related to conservatism and sociability. At the county level, however, composite scores of airborne microparticles (PM≤2.5) concentration and industrial pollutants in air, water, or soil were positively related to conservatism, authoritarianism, emotional stability, sociability, and conformity, and negatively associated with openness, trust, emotional expressiveness, adaptability, sensation seeking, introspection, art appreciation, and creativity.

Creator