Differential Measurement Error Research Articles

Results on Differential and Dependent Measurement Error of the Exposure and the Outcome Using Signed Directed Acyclic Graphs

Measurement error in both the exposure and the outcome is a common problem in epidemiologic studies. Measurement errors in the exposure and the outcome are said to be independent of each other if the measured exposure and the measured outcome are statistically independent conditional on the true exposure and true outcome (and dependent otherwise). Measurement error is said to be nondifferential if measurement of the exposure does not depend on the true outcome conditional on the true exposure and vice versa; otherwise it is said to be differential. Few results on differential and dependent measurement error are available in the literature. Here the authors use formal rules governing associations on signed directed acyclic graphs (DAGs) to draw conclusions about the presence and sign of causal effects under differential and dependent measurement error. The authors apply these rules to 4 forms of measurement error: independent nondifferential, dependent nondifferential, independent differential, and dependent differential. For a binary exposure and outcome, the authors generalize Weinberg et al.'s (Am J Epidemiol. 1994;140(6):565-571) result for nondifferential measurement error on preserving the direction of a trend to settings which also allow measurement error in the outcome and to cases involving dependent and/or differential error.

Respiratory Health Effects of Airborne Particulate Matter: The Role of Particle Size, Composition, and Oxidative Potential—The RAPTES Project

Background: Specific characteristics of particulate matter (PM) responsible for associations with respiratory health observed in epidemiological studies are not well established. High correlations among, and differential measurement errors of, individual components contribute to this uncertainty.Objectives: We investigated which characteristics of PM have the most consistent associations with acute changes in respiratory function in healthy volunteers.Methods: We used a semiexperimental design to accurately assess exposure. We increased exposure contrast and reduced correlations among PM characteristics by exposing volunteers at five different locations: an underground train station, two traffic sites, a farm, and an urban background site. Each of the 31 participants was exposed for 5 hr while exercising intermittently, three to seven times at different locations during March–October 2009. We measured PM10, PM2.5, particle number concentrations (PNC), absorbance, elemental/organic carbon, trace metals, secondary inorganic components, endotoxin content, gaseous pollutants, and PM oxidative potential. Lung function [FEV1 (forced expiratory volume in 1 sec), FVC (forced vital capacity), FEF25–75 (forced expiratory flow at 25–75% of vital capacity), and PEF (peak expiratory flow)] and fractional exhaled nitric oxide (FENO) were measured before and at three time points after exposure. Data were analyzed with mixed linear regression.Results: An interquartile increase in PNC (33,000 particles/cm3) was associated with an 11% [95% confidence interval (CI): 5, 17%] and 12% (95% CI: 6, 17%) FENO increase over baseline immediately and at 2 hr postexposure, respectively. A 7% (95% CI: 0.5, 14%) increase persisted until the following morning. These associations were robust and insensitive to adjustment for other pollutants. Similarly consistent associations were seen between FVC and FEV1 with PNC, NO2 (nitrogen dioxide), and NOx (nitrogen oxides).Conclusions: Changes in PNC, NO2, and NOx were associated with evidence of acute airway inflammation (i.e., FENO) and impaired lung function. PM mass concentration and PM10 oxidative potential were not predictive of the observed acute responses.

Concordance between personality disorder assessment methods

Studies have criticized the low level of agreement between the various methods of personality disorder (PD) assessment. This is an important issue for research and clinical purposes. Seven hundred and forty-two participants in the Hopkins Epidemiology of Personality Disorders Study (HEPS) were assessed on two occasions using the Personality Disorder Schedule (PDS) and the International Personality Disorder Examination (IPDE). The concordance between the two diagnostic methods for all DSM-IV PDs was assessed using standard methods and also two item response analytic approaches designed to take account of measurement error: a latent trait-based approach and a generalized estimating equations (GEE)-based approach, with post-hoc adjustment. Raw criteria counts, using the intraclass correlation coefficient (ICC), κ and odds ratio (OR), showed poor concordance. The more refined statistical methods showed a moderate to moderately high level of concordance between the methods for most PDs studied. Overall, the PDS produced lower prevalences of traits but higher precision of measurement than the IPDE. Specific criteria within each PD showed varying endorsement thresholds and precision for ascertaining the disorder. Concordance in the raw measurement of the individual PD criteria between the two clinical methods is lacking. However, based on two statistical methods that adjust for differential endorsement thresholds and measurement error in the assessments, we deduce that the PD constructs themselves can be measured with a moderate degree of confidence regardless of the clinical approach used. This may suggest that the individual criteria for each PD are, in and of themselves, less specific for diagnosis, but as a group the criteria for each PD usefully identify specific PD constructs.

Differential Measurement Errors in Zero-Truncated Regression Models for Count Data

Measurement errors in covariates may result in biased estimates in regression analysis. Most methods to correct this bias assume nondifferential measurement errors-i.e., that measurement errors are independent of the response variable. However, in regression models for zero-truncated count data, the number of error-prone covariate measurements for a given observational unit can equal its response count, implying a situation of differential measurement errors. To address this challenge, we develop a modified conditional score approach to achieve consistent estimation. The proposed method represents a novel technique, with efficiency gains achieved by augmenting random errors, and performs well in a simulation study. The method is demonstrated in an ecology application.

The part-time pay penalty in a segmented labor market

This paper is the first to examine the implications of switching to PT work for women's subsequent earnings trajectories, distinguishing by their type of contract: permanent or fixed-term. Using a rich longitudinal Spanish data set from Social Security records of over 76,000 prime-aged women strongly attached to the Spanish labor market, we find that the PT/FT hourly wage differential is larger and more persistent among fixed-term contract workers, strengthening the existent evidence that these workers can be classified as secondary. The paper discusses problems arising in empirical estimation (including a problem not discussed in the literature up to now: the differential measurement error of the LHS variable by PT status), and how to address them. It concludes with policy implications relevant for Continental Europe and its dual structure of employment protection.

Children's Reports of Parents' Education Level: Does it Matter Whom You Ask and What You Ask About?

Education researchers who study the effect of family social background on student achievement often use students’ survey reports of parental education to investigate these effects. However, past research has demonstrated that students misreport their parents’ education levels. We expand upon this research in two ways. First we use cognitive theories about the response process to develop and test hypotheses about reporting inconsistencies across these variables. Second we evaluate the impact of student misreporting on estimates of the relationship between parental education levels and student math achievement. Using data from the German administration of PISA 2000 (OECD Programme for International Student Assessment) in which both students and parents were asked to report parental variables, we show that reporting inconsistencies are a function of student achievement: students with higher math scores tend to provide reports that are more consistent with their parents' reports. This interesting case of differential measurement error has consequences for comparisons of the effects of parental background on student achievement across different subgroups of the population and across countries (a common use of PISA data and other international studies similar to PISA).

Measurement error in the timing of events: effect on survival analyses in randomized clinical trials

Presence of measurement error in the outcome can complicate the interpretation of a randomized clinical trial. The Oncologic Drugs Advisory Committee of the US Food and Drug Administration voted against recommending approval of bevacizumab for the initial treatment of metastatic breast cancer; one of their major concerns was the presence of a large amount of nondifferential measurement error in the evaluation of progression-free survival, the primary outcome of the randomized clinical trial E2100. To investigate the effects of nondifferential measurement error in time-to-event outcomes on the conclusions of a proportional hazards analysis of a randomized clinical trial. Simulations were performed showing effects of measurement error on the estimated treatment effect (hazard ratio) in a clinical trial. In some simulations, the measurement error structure from E2100 data was approximated; for other simulations, larger or smaller measurement error was considered. The bias in estimating the hazard ratio was very small using measurement error and a hazard ratio similar to E2100. Even with a larger nondifferential measurement error, the bias remained small when the hazard ratio was in a range commonly seen in clinical trials. There was no or little effect on the variability of the estimated treatment effect. Because of censoring issues, retrospective evaluation of the measurement error structure from a completed trial is difficult. Although our simulations cover a range of plausible measurement error values, in theory, a trial could have much larger measurement error than we considered. Differential measurement error is only briefly considered. Nondifferential measurement error due to variability in estimating time-to-event outcomes will typically not be a major concern in randomized clinical trials.

Causal Inference with Differential Measurement Error: Nonparametric Identification and Sensitivity Analysis

Political scientists have long been concerned about the validity of survey measurements. Although many have studied classical measurement error in linear regression models where the error is assumed to arise completely at random, in a number of situations the error may be correlated with the outcome. We analyze the impact of differential measurement error on causal estimation. The proposed nonparametric identification analysis avoids arbitrary modeling decisions and formally characterizes the roles of different assumptions. We show the serious consequences of differential misclassification and offer a new sensitivity analysis that allows researchers to evaluate the robustness of their conclusions. Our methods are motivated by a field experiment on democratic deliberations, in which one set of estimates potentially suffers from differential misclassification. We show that an analysis ignoring differential measurement error may considerably overestimate the causal effects. This finding contrasts with the case of classical measurement error, which always yields attenuation bias.

Self-reported Confidence in Recall as a Predictor of Validity and Repeatability of Physical Activity Questionnaire Data

Self-reported confidence ratings have been used in other research disciplines as a tool to assess data quality, and may be useful in epidemiologic studies. We examined whether self-reported confidence in recall of physical activity was a predictor of the validity and retest reliability of physical activity measures from the European Prospective Investigation into Cancer and Nutrition (EPIC) past-year questionnaire and the International Physical Activity Questionnaire (IPAQ) last-7-day questionnaire.During 2005-2006 in Sydney, Australia, 97 men and 80 women completed both questionnaires at baseline and at 10 months and wore an accelerometer as an objective comparison measure for three 7-day periods during the same timeframe. Participants rated their confidence in recalling physical activity for each question using a 5-point scale and were dichotomized at the median confidence value. Participants in the high-confidence group had higher validity and repeatability coefficients than those in the low-confidence group for most comparisons. The differences were most apparent for validity of IPAQ moderate activity: Spearman correlation rho = 0.34 (95% confidence interval [CI] = 0.08 to 0.55) and 0.01 (-0.17 to 0.20) for high- and low-confidence groups, respectively; and repeatability of EPIC household activity: rho = 0.81 (0.72 to 0.87) and 0.63 (0.48 to 0.74), respectively, and IPAQ vigorous activity: rho = 0.58 (0.43 to 0.70) and 0.29 (0.07 to 0.49), respectively. Women were less likely than men to report high recall confidence of past-year activity (adjusted odds ratio = 0.38; 0.18 to 0.80). Confidence ratings could be useful as indicators of recall accuracy (ie, validity and repeatability) of physical activity measures, and possibly for detecting differential measurement error and identifying questionnaire items that require improvement.

Methodological challenges in causal research on racial and ethnic patterns of cognitive trajectories: measurement, selection, and bias.

Research focused on understanding how and why cognitive trajectories differ across racial and ethnic groups can be compromised by several possible methodological challenges. These difficulties are especially relevant in research on racial and ethnic disparities and neuropsychological outcomes because of the particular influence of selection and measurement in these contexts. In this article, we review the counterfactual framework for thinking about causal effects versus statistical associations. We emphasize that causal inferences are key to predicting the likely consequences of possible interventions, for example in clinical settings. We summarize a number of common biases that can obscure causal relationships, including confounding, measurement ceilings/floors, baseline adjustment bias, practice or retest effects, differential measurement error, conditioning on common effects in direct and indirect effects decompositions, and differential survival. For each, we describe how to recognize when such biases may be relevant and some possible analytic or design approaches to remediating these biases.

A comparison of regression calibration, moment reconstruction and imputation for adjusting for covariate measurement error in regression

Regression calibration (RC) is a popular method for estimating regression coefficients when one or more continuous explanatory variables, X, are measured with an error. In this method, the mismeasured covariate, W, is substituted by the expectation E(X|W), based on the assumption that the error in the measurement of X is non-differential. Using simulations, we compare three versions of RC with two other 'substitution' methods, moment reconstruction (MR) and imputation (IM), neither of which rely on the non-differential error assumption. We investigate studies that have an internal calibration sub-study. For RC, we consider (i) the usual version of RC, (ii) RC applied only to the 'marker' information in the calibration study, and (iii) an 'efficient' version (ERC) in which the estimators (i) and (ii) are combined. Our results show that ERC is preferable when there is non-differential measurement error. Under this condition, there are cases where ERC is less efficient than MR or IM, but they rarely occur in epidemiology. We show that the efficiency gain of usual RC and ERC over the other methods can sometimes be dramatic. The usual version of RC carries similar efficiency gains to ERC over MR and IM, but becomes unstable as measurement error becomes large, leading to bias and poor precision. When differential measurement error does pertain, then MR and IM have considerably less bias than RC, but can have much larger variance. We demonstrate our findings with an analysis of dietary fat intake and mortality in a large cohort study.

Modeling the intra-urban variability of outdoor traffic pollution in Oslo, Norway—A GA 2LEN project

Traffic is a major source of air pollutants in urban environments, and exposure to these pollutants may be associated with adverse health effects. However, inconsistencies in observational epidemiological studies may be caused by differential measurement errors in various approaches in assessing exposure. We aimed to evaluate a simple method for assessing outdoor air pollutant concentrations in Oslo, Norway, through a land-use regression method. Samples of nitrogen oxides (NO x ) were collected in two different weeks using Ogawa passive diffusion samplers simultaneously at 80 locations across Oslo. Independent variables used in subsequent regression models as predictors of the pollutants were derived using the Arc 9 geographic information system (GIS) software. Indicators of land use, traffic, population density, and physical geography were tested. The final regression model yielded an adjusted coefficient of determination ( R 2) of 0.77 for nitrogen dioxide (NO 2), 0.66 for nitric oxide (NO), and 0.73 for NO x . The results suggest that a good predictive exposure model can be derived from this approach, which can be used to estimate long-term small-area variation in concentrations for individual exposure assessment in epidemiological studies in a highly cost-effective way. These small-area variations in traffic pollution are important since they may have associations with health effects.

The Relationship Between Particulate Matter (PM) Components and Survival Among Cohort Members

SAB4-PD-08 Introduction: Several studies have reported statistically significant associations between air pollution, characterized by particulate matter (PM) and/or gaseous pollutant levels, and the survivorship among cohort members. Studies to date have not systematically investigated the associations between specific components of PM and cohort survival. This study considers the relationship between PM, components (metallic ions, organic and elemental carbon, sulfate and nitrate ions), gaseous pollutants, and other indices of environmental nuisance (eg, traffic) and cohort survivorship. Methods: This study makes use of a cohort of 70,000 men who have been followed since 1976. Proportional hazard regression techniques were used to estimate the influence of several external variables, including pollution values, on the survival times of the cohort after adjusting for demographic (age, race), behavioral (smoking history), and prior health status (body mass index, blood pressure) variables among cohort members. Several model specifications were considered, including models with multiple pollution variables. Model results were compared using measures of statistical significance, effect size, and overall model fit as measured by the Akaike Information Criterion. Results: Single pollutant models indicated statistically significant associations between cohort survivorship and elemental carbon (EC), nitrate ion, vanadium, and nickel. Peak ozone and iron were positively associated with cohort survival, but the association was not statistically significant. Two pollutants models showed EC and the nitrate ion retained statistically significant associations with survivorship when they were considered jointly with nickel ion concentrations or with peak ozone. Peak ozone, iron, vanadium, and nickel were statistically significantly associated with survivorship in at least one model with multiple pollutants. When a traffic density variable was introduced into the multiple pollutants models, it appeared to dominate the pollution variables in terms of predicting cohort survivorship and remained statistically significant in many models. Discussion and Conclusions: The results indicate that some components of pollution and PM are much more highly associated with cohort survivorship than others. An indicator of traffic density trumped these variables, however, when they were considered jointly with pollution measures. It is unclear whether this result implies that 1) there is a nonmeasured component of pollution associated with traffic that may explain the observed associations or 2) some nonpollution aspect of traffic is a better predictor of cohort survivorship. This issue is complicated by considerations of differential measurement error among the variables studied.

Comparing the Roles of Physical Activity and Fitness in Arterial Stiffness: How Important Is Exposure Measurement Error?

To the Editor: The article by Boreham et al1 in the November 2004 issue of Hypertension reports the associations between physical activity and cardiorespiratory fitness with pulse-wave velocity (PWV), a measure of arterial stiffness. One of the objectives of the study was to compare the relative relationships of physical activity and fitness with the outcome PWV. Indeed, the authors’ interpretation of their data leads them to state that “arterial stiffness related benefits of exercise are most likely to accrue if exercise prescription in young adults targets improvements in cardiorespiratory fitness”, implying that fitness is more important than physical activity in preventing stiffening of the arteries. Clearly, understanding the manner in which modifiable factors such as low physical activity and poor fitness relate with cardiovascular health is important for disease prevention. However, the conclusions and recommendations in the present study are based on data in which differential exposure measurement error is probably a pivotal factor; this possibility receives little attention in the article. Cardiorespiratory fitness, as assessed through a maximal oxygen consumption test, is not a simple construct, because fitness is determined by both modifiable factors, such as habitual physical activity (volume, intensity, and mode), smoking, and cardiovascular pathology, as well as nonmodifiable factors, such as genotype and intrauterine environment. Therefore, in considering the risk predisposed by low fitness and physical inactivity, one should preferably measure both exposures to a similar level of precision and then assess them in combination, thereby accounting for any shared variance. However, this is difficult to …

Design and interpretation of studies of differential exposure measurement error.

Differential exposure measurement error can have more adverse effects on estimates of exposure-disease associations than nondifferential measurement error, yet relatively little has been written about the design and interpretation of validity and reliability studies to assess differential measurement error. In this paper, a simple approximate equation is given for the effect of differential measurement error in a continuous exposure measure on the bias in the odds ratio. From this, it is shown that two parameters need to be estimated in validity/reliability studies in order to interpret the results in terms of the bias in the odds ratio in an epidemiologic study that will use the measure. The first is the correlation between the mismeasured and true exposure. The second is the differential bias (difference between cases and controls in the difference between mean measured and true exposure) relative to the true difference in exposure between cases and controls. It is shown that this latter parameter can be estimated in a method comparison study if one has a comparison measure that is unbiased or has nondifferential bias, so a perfect criterion measure is not needed. Researchers should consider measuring and reporting this parameter in validity/reliability studies when feasible.

Cross-Cultural Sources of Measurement Error in Substance Use Surveys

We present an overview of the cross-cultural quality of surveyreports of substance use behaviors in the United States. Empiricaldata from 36 published studies (1977–2003) are examined to evaluatethe reliability and validity of substance use reports across culturalgroups. In these studies, race/ethnicity are used as proxy indicators ofrespondent culture. In general, the available research suggests that,with a few exceptions, the quality of survey data on racial andethnic disparities in substance use is often limited by differentialmeasurement error. A conceptual paradigm is presented to considera wide range of potential causes for these differences in measurementerror that includes two dimensions: emphasis on negative vs. positivebehavior patterns, and emphasis on internal vs. external causalfactors. These two dimensions yield four potential models that areuseful in understanding variations in substance use measurementerror: the cultural deficit model, the cultural conflict model, the mainstreamconformitymodel, and the cultural distrust model. Futureresearch should focus on the ability of each of these alternativemodels to account for cultural variability in the quality of substanceuse reporting.

Intergenerational Social Mobility and Assortative Mating in Britain

This paper investigates the links between the socio-economic position of parents and the socio-economic position of their offspring and, through the marriage market, the socioeconomic position of their offspring's parents-in-law. Using the Goldthorpe-Hope score of occupational prestige as a measure of status and samples drawn from the British Household Panel Survey 1991-1999, we find that the intergenerational elasticity is around 0.2 for men and between 0.17 and 0.23 for women. On average, the intragenerational correlation is lower, and of the order of 0.15 to 0.18, suggesting that the returns to human capital, which is transmitted across generations by altruistic parents, contribute more to social status than assortative mating in the marriage market. Substantially higher estimates are reported when measurement error is accounted for. We also find strong nonlinearities, whereby both inter- and intra-generational elasticities tend to increase with parental status. We offer four possible explanations for this finding, three of which â?? one based on mean-displacement shifts in the occupational prestige distribution, another based on life-cycle effects and the third based on differential measurement errors â?? do not find strong support in our data. The fourth explanation is based on the notion of intergenerational transmission of social capital and intellectual capital. The evidence supports the idea that richer parents are likely to have a larger and more valuable stock of both social capital and intellectual capital to pass on to their children.

A stochastic differential equation model of diurnal cortisol patterns.

Circadian modulation of episodic bursts is recognized as the normal physiological pattern of diurnal variation in plasma cortisol levels. The primary physiological factors underlying these diurnal patterns are the ultradian timing of secretory events, circadian modulation of the amplitude of secretory events, infusion of the hormone from the adrenal gland into the plasma, and clearance of the hormone from the plasma by the liver. Each measured plasma cortisol level has an error arising from the cortisol immunoassay. We demonstrate that all of these three physiological principles can be succinctly summarized in a single stochastic differential equation plus measurement error model and show that physiologically consistent ranges of the model parameters can be determined from published reports. We summarize the model parameters in terms of the multivariate Gaussian probability density and establish the plausibility of the model with a series of simulation studies. Our framework makes possible a sensitivity analysis in which all model parameters are allowed to vary simultaneously. The model offers an approach for simultaneously representing cortisol's ultradian, circadian, and kinetic properties. Our modeling paradigm provides a framework for simulation studies and data analysis that should be readily adaptable to the analysis of other endocrine hormone systems.

Smoking and prostate cancer.

Considering the high prevalence of smoking and the high prostate cancer morbidity this paper examines the relation between smoking and the risks of prostate cancer. In this regard a Medline search was conducted covering the period from 1966 to March 2000. Most of the prospective cohort studies and all case-control studies that used incident cases as the outcome found no association between current smoking and prostate cancer. On the other hand contrasting results were observed among majority of prospective cohort studies that used prostate cancer mortality as outcome. Findings from these studies found a positive association between current smoking and prostate cancer. Population-based case-control studies meanwhile produced null results which were attributed to differential measurement errors in mortality cohorts. However there were methodological shortcomings noted that affect the prostate cancer-smoking estimate of effect such as the surveillance bias misclassification of smoking status and uncontrolled confounding factors. Hence prospective studies should obtain detailed information on the clinical stage and histologic grade of prostate cancer and on the subjects smoking history.

A comparison of response error by adolescents and adults: findings from a health care study.

Most assessments of response error have focused only on reports by adult respondents; we know little about the propensity to misreport or the magnitude or direction of response error for self-reports by adolescents. To the extent that adults and adolescents differ in the magnitude or direction of response error, comparisons between adults and adolescents will be confounded. This study compares the quality of self-response among adolescents and adults in their reporting of health conditions and health care utilization. Approximately 2,000 members of an HMO were interviewed, and by use of medical record information, the accuracy of their responses was assessed for several measures: number of outpatient visits, date of the most recent outpatient visit, site of care, type of professional seen during the last visit, and an indicator as to whether the respondent had asthma. The findings indicate that both the magnitude and the direction of response error vary by the age of the respondent. In multivariate models predicting erroneous response, adolescents were more likely to misreport than were adult respondents. In addition, among erroneous respondents, adolescents tended to err in the direction of overreporting, whereas adult respondents tended to err in the direction of underreporting. Although the results should be viewed cautiously because of limitations in both the sample design and the survey topic, the findings suggest that substantive comparisons of health care utilization between adolescents and adults may be confounded by differential measurement error.