Combining Information from Multiple Sources in Bayesian Modeling

Surveys can collect important data that inform policy decisions and drive social science research. Large government surveys collect information from the U.S. population on a wide range of topics, including demographics, education, employment, and lifestyle. Analysis of survey data presents unique challenges. In particular, one needs to account for missing data, for complex sampling designs, and for measurement error. Conceptually, a survey organization could spend lots of resources getting high-quality responses from a simple random sample, resulting in survey data that are easy to analyze. However, this scenario often is not realistic. To address these practical issues, survey organizations can leverage the information available from other sources of data. For example, in longitudinal studies that suffer from attrition, they can use the information from refreshment samples to correct for potential attrition bias. They can use information from known marginal distributions or survey design to improve inferences. They can use information from gold standard sources to correct for measurement error.

This thesis presents novel approaches to combining information from multiple sources that address the three problems described above.

The first method addresses nonignorable unit nonresponse and attrition in a panel survey with a refreshment sample. Panel surveys typically suffer from attrition, which can lead to biased inference when basing analysis only on cases that complete all waves of the panel. Unfortunately, the panel data alone cannot inform the extent of the bias due to attrition, so analysts must make strong and untestable assumptions about the missing data mechanism. Many panel studies also include refreshment samples, which are data collected from a random sample of new

individuals during some later wave of the panel. Refreshment samples offer information that can be utilized to correct for biases induced by nonignorable attrition while reducing reliance on strong assumptions about the attrition process. To date, these bias correction methods have not dealt with two key practical issues in panel studies: unit nonresponse in the initial wave of the panel and in the

refreshment sample itself. As we illustrate, nonignorable unit nonresponse

can significantly compromise the analyst's ability to use the refreshment samples for attrition bias correction. Thus, it is crucial for analysts to assess how sensitive their inferences---corrected for panel attrition---are to different assumptions about the nature of the unit nonresponse. We present an approach that facilitates such sensitivity analyses, both for suspected nonignorable unit nonresponse

in the initial wave and in the refreshment sample. We illustrate the approach using simulation studies and an analysis of data from the 2007-2008 Associated Press/Yahoo News election panel study.

The second method incorporates informative prior beliefs about

marginal probabilities into Bayesian latent class models for categorical data.

The basic idea is to append synthetic observations to the original data such that

(i) the empirical distributions of the desired margins match those of the prior beliefs, and (ii) the values of the remaining variables are left missing. The degree of prior uncertainty is controlled by the number of augmented records. Posterior inferences can be obtained via typical MCMC algorithms for latent class models, tailored to deal efficiently with the missing values in the concatenated data.

We illustrate the approach using a variety of simulations based on data from the American Community Survey, including an example of how augmented records can be used to fit latent class models to data from stratified samples.

The third method leverages the information from a gold standard survey to model reporting error. Survey data are subject to reporting error when respondents misunderstand the question or accidentally select the wrong response. Sometimes survey respondents knowingly select the wrong response, for example, by reporting a higher level of education than they actually have attained. We present an approach that allows an analyst to model reporting error by incorporating information from a gold standard survey. The analyst can specify various reporting error models and assess how sensitive their conclusions are to different assumptions about the reporting error process. We illustrate the approach using simulations based on data from the 1993 National Survey of College Graduates. We use the method to impute error-corrected educational attainments in the 2010 American Community Survey using the 2010 National Survey of College Graduates as the gold standard survey.

Navigating uncertain waters: a critical review of inferring foraging behaviour from location and dive data in pinnipeds

In the last thirty years, the emergence and progression of biologging technology has led to great advances in marine predator ecology. Large databases of location and dive observations from biologging devices have been compiled for an increasing number of diving predator species (such as pinnipeds, sea turtles, seabirds and cetaceans), enabling complex questions about animal activity budgets and habitat use to be addressed. Central to answering these questions is our ability to correctly identify and quantify the frequency of essential behaviours, such as foraging. Despite technological advances that have increased the quality and resolution of location and dive data, accurately interpreting behaviour from such data remains a challenge, and analytical methods are only beginning to unlock the full potential of existing datasets. This review evaluates both traditional and emerging methods and presents a starting platform of options for future studies of marine predator foraging ecology, particularly from location and two-dimensional (time-depth) dive data. We outline the different devices and data types available, discuss the limitations and advantages of commonly-used analytical techniques, and highlight key areas for future research. We focus our review on pinnipeds - one of the most studied taxa of marine predators - but offer insights that will be applicable to other air-breathing marine predator tracking studies. We highlight that traditionally-used methods for inferring foraging from location and dive data, such as first-passage time and dive shape analysis, have important caveats and limitations depending on the nature of the data and the research question. We suggest that more holistic statistical techniques, such as state-space models, which can synthesise multiple track, dive and environmental metrics whilst simultaneously accounting for measurement error, offer more robust alternatives. Finally, we identify a need for more research to elucidate the role of physical oceanography, device effects, study animal selection, and developmental stages in predator behaviour and data interpretation.

Mapping personal trip OD from probe data

The paper analyses the expected value of OD volumes from probe with fixed error, error that is proportional to zone size and inversely proportional to zone size. To add realism to the analysis, real trip ODs in the Tokyo Metropolitan Region are synthesised. The results show that for small zone coding with average radius of 1.1km, and fixed measurement error of 100m, an accuracy of 70% can be expected. The equivalent accuracy for medium zone coding with average radius of 5km would translate into a fixed error of approximately 300m. As expected small zone coding is more sensitive than medium zone coding as the chances of the probe error envelope falling into adjacent zones are higher. For the same error radii, error proportional to zone size would deliver higher level of accuracy. As over half (54.8%) of the trip ends start or end at zone with equivalent radius of ≤ 1.2 km and only 13% of trips ends occurred at zones with equivalent radius ≥2.5km, measurement error that is proportional to zone size such as mobile phone would deliver higher level of accuracy. The synthesis of real OD with different probe error characteristics have shown that expected value of >85% is difficult to achieve for small zone coding with average radius of 1.1km. For most transport applications, OD matrix at medium zone coding is sufficient for transport management. From this study it can be drawn that GPS with error range between 2 and 5m, and at medium zone coding (average radius of 5km) would provide OD estimates greater than 90% of the expected value. However, for a typical mobile phone operating error range at medium zone coding the expected value would be lower than 85%. This paper assumes transmission of one origin and one destination positions from the probe. However, if multiple positions within the origin and destination zones are transmitted, map matching to transport network could be performed and it would greatly improve the accuracy of the probe data.

Benefits of publicly available data

The National Morbidity, Mortality, and Air Pollution Study (NMMAPS) was designed to examine the health effects of air pollution in the United States. The primary question was whether particulate matter was responsible for the associations between air pollution and daily mortality. Secondary questions concerned measurement error in air pollution and mortality displacement.1 Since then, NMMAPS has been used to answer many important questions in environmental epidemiology...

Assessing the Impact of Sensor Accuracy and Sensor Choice on Air Traffic Conflict Risk Estimation using Risk-Ratio Concepts

One of the primary desired capabilities of any future air traffic separation management system is the ability to provide early conflict detection and resolution effectively and efficiently. In this paper, we consider the risk of conflict as a primary measurement to be used for early conflict detection. This paper focuses on developing a novel approach to assess the impact of different measurement uncertainty models on the estimated risk of conflict. The measurement uncertainty model can be used to represent different sensor accuracy and sensor choices. Our study demonstrates the value of modelling measurement uncertainty in the conflict risk estimation problem and presents techniques providing a means of assessing sensor requirements to achieve desired conflict detection performance.

Individual, environmental, and meteorological predictors of daily personal ultraviolet radiation exposure measurements in a United States cohort study

Background Individual exposure to ultraviolet radiation (UVR) is challenging to measure, particularly for diseases with substantial latency periods between first exposure and diagnosis of outcome, such as cancer. To guide the choice of surrogates for long-term UVR exposure in epidemiologic studies, we assessed how well stable sun-related individual characteristics and environmental/meteorological factors predicted daily personal UVR exposure measurements. Methods We evaluated 123 United States Radiologic Technologists subjects who wore personal UVR dosimeters for 8 hours daily for up to 7 days (N = 837 days). Potential predictors of personal UVR derived from a self-administered questionnaire, and public databases that provided daily estimates of ambient UVR and weather conditions. Factors potentially related to personal UVR exposure were tested individually and in a model including all significant variables. Results The strongest predictors of daily personal UVR exposure in the full model were ambient UVR, latitude, daily rainfall, and skin reaction to prolonged sunlight (R2 = 0.30). In a model containing only environmental and meteorological variables, ambient UVR, latitude, and daily rainfall were the strongest predictors of daily personal UVR exposure (R2 = 0.25). Conclusions In the absence of feasible measures of individual longitudinal sun exposure history, stable personal characteristics, ambient UVR, and weather parameters may help estimate long-term personal UVR exposure.

Labour, social and health outcomes of immigrants in Australia : effects of language proficiency using the IV approach

Language has been of interest to numerous economists since the late 20th century, with the majority of the studies focusing on its effects on immigrants’ labour market outcomes; earnings in particular. However, language is an endogenous variable, which along with its susceptibility to measurement error causes biases in ordinary-least-squares estimates. The instrumental variables method overcomes the shortcomings of ordinary least squares in modelling endogenous explanatory variables. In this dissertation, age at arrival combined with country of origin form an instrument creating a difference-in-difference scenario, to address the issue of endogeneity and attenuation error in language proficiency. The first half of the study aims to investigate the extent to which English speaking ability of immigrants improves their labour market outcomes and social assimilation in Australia, with the use of the 2006 Census. The findings have provided evidence that support the earlier studies. As expected, immigrants in Australia with better language proficiency are able to earn higher income, attain higher level of education, have higher probability of completing tertiary studies, and have more hours of work per week. Language proficiency also improves social integration, leading to higher probability of marriage to a native and higher probability of obtaining citizenship. The second half of the study further investigates whether language proficiency has similar effects on a migrant’s physical and mental wellbeing, health care access and lifestyle choices, with the use of three National Health Surveys. However, only limited evidence has been found with respect to the hypothesised causal relationship between language and health for Australian immigrants.

Sampling designs on stream networks using the pseudo-Bayesian approach

Monitoring stream networks through time provides important ecological information. The sampling design problem is to choose locations where measurements are taken so as to maximise information gathered about physicochemical and biological variables on the stream network. This paper uses a pseudo-Bayesian approach, averaging a utility function over a prior distribution, in finding a design which maximizes the average utility. We use models for correlations of observations on the stream network that are based on stream network distances and described by moving average error models. Utility functions used reflect the needs of the experimenter, such as prediction of location values or estimation of parameters. We propose an algorithmic approach to design with the mean utility of a design estimated using Monte Carlo techniques and an exchange algorithm to search for optimal sampling designs. In particular we focus on the problem of finding an optimal design from a set of fixed designs and finding an optimal subset of a given set of sampling locations. As there are many different variables to measure, such as chemical, physical and biological measurements at each location, designs are derived from models based on different types of response variables: continuous, counts and proportions. We apply the methodology to a synthetic example and the Lake Eacham stream network on the Atherton Tablelands in Queensland, Australia. We show that the optimal designs depend very much on the choice of utility function, varying from space filling to clustered designs and mixtures of these, but given the utility function, designs are relatively robust to the type of response variable.

Spatial and temporal variability in a residual soil profile

It is well understood that that there is variation inherent in all testing techniques, and that all soil and rock materials also contain some degree of natural variability. Less consideration is normally given to variation associated with natural material heterogeneity within a site, or the relative condition of the material at the time of testing. This paper assesses the impact of spatial and temporal variability upon repeated insitu testing of a residual soil and rock profile present within a single residential site over a full calendar year, and thus range of seasonal conditions. From this repeated testing, the magnitude of spatial and temporal variation due to seasonal conditions has demonstrated that, depending on the selected location and moisture content of the subsurface at the time of testing, up to a 35% variation within the test results can be expected. The results have also demonstrated that the completed insitu test technique has a similarly large measurement and inherent variability error and, for the investigated site, up to a 60% variation in normalised results was observed. From these results, it is recommended that the frequency and timing of insitu tests should be considered when deriving geotechnical design parameters from a limited data set.

Cross-cultural adaptation, reliability and validity of the Turkish version of the spine functional index

Background The Spine Functional Index (SFI) is a patient reported outcome measure with sound clinimetric properties and clinical viability for the determination of whole-spine impairment. To date, no validated Turkish version is available. The purpose of this study is to cross-culturally adapted the SFI for Turkish-speaking patients (SFI-Tk) and determine the psychometric properties of reliability, validity and factor structure in a Turkish population with spine musculoskeletal disorders. Methods The SFI English version was culturally adapted and translated into Turkish using a double forward and backward method according to established guidelines. Patients (n = 285, cervical = l29, lumbar = 151, cervical and lumbar region = 5, 73% female, age 45 ± 1) with spine musculoskeletal disorders completed the SFI-Tk at baseline and after a seven day period for test-retest reliability. For criterion validity the Turkish version of the Functional Rating Index (FRI) was used plus the Neck Disability Index (NDI) for cervical patients and the Oswestry Disability Index (ODI) for back patients. Additional psychometric properties were determined for internal consistency (Chronbach’s α), criterion validity and factor structure. Results There was a high degree of internal consistency (α = 0.85, item range 0.80-0.88) and test-retest reliability (r = 0.93, item range = 0.75-0.95). The factor analysis demonstrated a one-factor solution explaining 24.2% of total variance. Criterion validity with the ODI was high (r = 0.71, p < 0.001) while the FRI and NDI were fair (r = 0.52 and r = 0.58, respectively). The SFI-Tk showed no missing responses with the ‘half-mark’ option used in 11.75% of total responses by 77.9% of participants. Measurement error from SEM and MDC90 were respectively 2.96% and 7.12%. Conclusions The SFI-Tk demonstrated a one-factor solution and is a reliable and valid instrument. The SFI-Tk consists of simple and easily understood wording and may be used to assess spine region musculoskeletal disorders in Turkish speaking patients.

Heritability of working memory brain activation

Although key to understanding individual variation in task-related brain activation, the genetic contribution to these individual differences remains largely unknown. Here we report voxel-by-voxel genetic model fitting in a large sample of 319 healthy, young adult, human identical and fraternal twins (mean ± SD age, 23.6 ±1.8 years) who performed an n-back working memory task during functional magnetic resonance imaging (fMRI) at a high magnetic field (4 tesla). Patterns of task-related brain response (BOLD signal difference of 2-back minus 0-back) were significantly heritable, with the highest estimates (40 - 65%) in the inferior, middle, and superior frontal gyri, left supplementary motor area, precentral and postcentral gyri, middle cingulate cortex, superior medial gyrus, angular gyrus, superior parietal lobule, including precuneus, and superior occipital gyri. Furthermore, high test-retest reliability for a subsample of 40 twins indicates that nongenetic variance in the fMRI brain response is largely due to unique environmental influences rather than measurement error. Individual variations in activation of the working memory network are therefore significantly influenced by genetic factors. By establishing the heritability of cognitive brain function in a large sample that affords good statistical power, and using voxel-by-voxel analyses, this study provides the necessary evidence for task-related brain activation to be considered as an endophenotype for psychiatric or neurological disorders, and represents a substantial new contribution to the field of neuroimaging genetics. These genetic brain maps should facilitate discovery of gene variants influencing cognitive brain function through genome-wide association studies, potentially opening up new avenues in the treatment of brain disorders.

Using axially loaded MRI to investigate the biomechanics of adolescent idiopathic scoliosis

Introduction. Spinal flexibility measurement is an important aspect of pre-operative clinical assessment in the treatment of Adolescent Idiopathic Scoliosis (AIS). Clinically, curve flexibility is a combined measure for all vertebral levels. We propose that in vivo flexibility for individual spinal joints could provide valuable additional information in planning treatment for scoliosis. Methods. Individual spinal joint flexibility in the coronal plane was measured for a series of AIS patients using axially loaded magnetic resonance imaging. Each patient underwent magnetic resonance imaging in the supine position, with no axial load, and then following application of an axial compressive load equal to half the patient’s bodyweight. Coronal plane disc wedge angles in the unloaded and loaded configurations were measured. Joint moments exerted by the axial compressive load were used to derive estimates of individual joint compliance. Results. Fifteen AIS patients were included in the study (mean clinical Cobb angle 46 degrees, mean age 15.3 years). Mean intra-observer measurement error for endplate inclination was 1.6˚. The mean increase in measured major Cobb angle between unloaded and loaded scans was 7.6˚. For certain spinal levels (+2,+1,-2 relative to the apex) there was a statistically significant relationship between change in wedge angle under load and initial wedge angle, such that initially highly wedged discs demonstrated a smaller change in wedge angle than less wedged discs. Highly wedged discs were observed near the apex of the curve, which corresponded to lower joint compliance in the apical region. Conclusion. Approaches such as this can provide valuable biomechanical data on in vivo spinal biomechanics in AIS. Knowledge of individual joint flexibility may assist surgeons to determine which spinal procedure is most appropriate for a patient, which levels should be included in a spinal fusion and the relative mobility of individual joints in the deformed region of the spine.

Childhood sexual abuse and adult sexual dysfunction: Response to commentary by Rind and Tromovitch (2007)

Rind and Tromovitch (2007) raised four concerns relating to our article (Najman, Dunne, Purdie, Boyle, & Coxeter, 2005. Archives of Sexual Behavior, 34, 517-526.) which suggested a causal association between childhood sexual abuse (CSA) and adult sexual dysfunction. We consider each of these concerns: magnitude of effect, cause and effect, confounding, and measurement error. We suggest that, while the concerns they raise represent legitimate reservations about the validity of our findings, on balance the available evidence indicates an association between CSA and sexual dysfunction that is of "moderate" magnitude, probably causal, and unlikely to be a consequence of confounding or measurement error.