Missing data mechanisms and their implications on the analysis of categorical data

We review some issues related to the implications of different missing data mechanisms on statistical inference for contingency tables and consider simulation studies to compare the results obtained under such models to those where the units with missing data are disregarded. We confirm that although, in general, analyses under the correct missing at random and missing completely at random models are more efficient even for small sample sizes, there are exceptions where they may not improve the results obtained by ignoring the partially classified data. We show that under the missing not at random (MNAR) model, estimates on the boundary of the parameter space as well as lack of identifiability of the parameters of saturated models may be associated with undesirable asymptotic properties of maximum likelihood estimators and likelihood ratio tests; even in standard cases the bias of the estimators may be low only for very large samples. We also show that the probability of a boundary solution obtained under the correct MNAR model may be large even for large samples and that, consequently, we may not always conclude that a MNAR model is misspecified because the estimate is on the boundary of the parameter space.

Predictors of nonresponse in a questionnaire-based outcome study of vocational rehabilitation patients.

OBJECTIVE: To identify predictors of nonresponse to a self-report study of patients with orthopedic trauma hospitalized for vocational rehabilitation between November 15, 2003, and December 31, 2005. The role of biopsychosocial complexity, assessed using the INTERMED, was of particular interest. DESIGN: Cohort study. Questionnaires with quality of life, sociodemographic, and job-related questions were given to patients at hospitalization and 1 year after discharge. Sociodemographic data, biopsychosocial complexity, and presence of comorbidity were available at hospitalization (baseline) for all eligible patients. Logistic regression models were used to test a number of baseline variables as potential predictors of nonresponse to the questionnaires at each of the 2 time points. SETTING: Rehabilitation clinic. PARTICIPANTS: Patients (N=990) hospitalized for vocational rehabilitation over a period of 2 years. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURE: Nonresponse to the questionnaires was the binary dependent variable. RESULTS: Patients with high biopsychosocial complexity, foreign native language, or low educational level were less likely to respond at both time points. Younger patients were less likely to respond at 1 year. Those living in a stable partnership were less likely than singles to respond at hospitalization. Sex, psychiatric, and somatic comorbidity and alcoholism were never associated with nonresponse. CONCLUSIONS: We stress the importance of assessing biopsychosocial complexity to predict nonresponse. Furthermore, the factors we found to be predictive of nonresponse are also known to influence treatment outcome and vocational rehabilitation. Therefore, it is important to increase the response rate of the groups of concern in order to reduce selection bias in epidemiologic investigations.

Quantitative genetic modeling and inference in the presence of nonignorable missing data.

Natural selection is typically exerted at some specific life stages. If natural selection takes place before a trait can be measured, using conventional models can cause wrong inference about population parameters. When the missing data process relates to the trait of interest, a valid inference requires explicit modeling of the missing process. We propose a joint modeling approach, a shared parameter model, to account for nonrandom missing data. It consists of an animal model for the phenotypic data and a logistic model for the missing process, linked by the additive genetic effects. A Bayesian approach is taken and inference is made using integrated nested Laplace approximations. From a simulation study we find that wrongly assuming that missing data are missing at random can result in severely biased estimates of additive genetic variance. Using real data from a wild population of Swiss barn owls Tyto alba, our model indicates that the missing individuals would display large black spots; and we conclude that genes affecting this trait are already under selection before it is expressed. Our model is a tool to correctly estimate the magnitude of both natural selection and additive genetic variance.

Effects of Interviewer Experience on Components of Nonresponse in the European Social Survey

We analyze interviewer related nonresponse differences in face-to-face surveys distinguishing three types of interviewers: those who have previous experience with the same high standard cross-sectional survey ("experienced"), those who were chosen by the survey agency to complete refusal conversions ("seniors"), and usual interviewers. The nonresponse components are obtaining household contact, target person contact, and target person cooperation. In addition we examine if interviewer homogeneity with respect to these components is different across the three interviewer groups. Data come from the European Social Survey (ESS) contact forms from four countries which participated during the three rounds 2002/04/06 and used the same survey agency that in turn used to some extent the same interviewers. To analyze interviewer effects, we use discrete two-level models. We find some evidence of better performance by both senior and experienced interviewers and indications of greater homogeneity for nonresponse components, especially for those that contain room for improvement. Surprisingly, the senior interviewers do not outperform those experienced. We conclude that survey agencies should make more efforts to decrease the comparatively high interviewer turnover.

Nonresponse to clozapine and ultrarapid CYP1A2 activity: clinical data and analysis of CYP1A2 gene.

Clozapine (CLO), an atypical antipsychotic, depends mainly on cytochrome P450 1A2 (CYP1A2) for its metabolic clearance. Four patients treated with CLO, who were smokers, were nonresponders and had low plasma levels while receiving usual doses. Their plasma levels to dose ratios of CLO (median; range, 0.34; 0.22 to 0.40 ng x day/mL x mg) were significantly lower than ratios calculated from another study with 29 patients (0.75; 0.22 to 2.83 ng x day/mL x mg; P < 0.01). These patients were confirmed as being CYP1A2 ultrarapid metabolizers by the caffeine phenotyping test (median systemic caffeine plasma clearance; range, 3.85; 3.33 to 4.17 mL/min/kg) when compared with previous studies (0.3 to 3.33 mL/min/kg). The sequencing of the entire CYP1A2 gene from genomic DNA of these patients suggests that the -164C > A mutation (CYP1A2*1F) in intron 1, which confers a high inducibility of CYP1A2 in smokers, is the most likely explanation for their ultrarapid CYP1A2 activity. A marked (2 patients) or a moderate (2 patients) improvement of the clinical state of the patients occurred after the increase of CLO blood levels above the therapeutic threshold by the increase of CLO doses to very high values (ie, up to 1400 mg/d) or by the introduction of fluvoxamine, a potent CYP1A2 inhibitor, at low dosage (50 to 100 mg/d). Due to the high frequency of smokers among patients with schizophrenia and to the high frequency of the -164C > A polymorphism, CYP1A2 genotyping could have important clinical implications for the treatment of patients with CLO.

Using register data to assess the impact of response enhancement methods on the risk of nonresponse bias and survey costs

The pursuit of high response rates to minimise the threat of nonresponse bias continues to dominate decisions about resource allocation in survey research. Yet a growing body of research has begun to question this practice. In this study, we use previously unavailable data from a new sampling frame based on population registers to assess the value of different methods designed to increase response rates on the European Social Survey in Switzerland. Using sampling data provides information about both respondents and nonrespondents, making it possible to examine how changes in response rates resulting from the use of different fieldwork methods relate to changes in the composition and representativeness of the responding sample. We compute an R-indicator to assess representativity with respect to the sampling register variables, and find little improvement in the sample composition as response rates increase. We then examine the impact of response rate increases on the risk of nonresponse bias based on Maximal Absolute Bias (MAB), and coefficients of variation between subgroup response rates, alongside the associated costs of different types of fieldwork effort. The results show that increases in response rate help to reduce MAB, while only small but important improvements to sample representativity are gained by varying the type of effort. These findings lend further support to research that has called into question the value of extensive investment in procedures aimed at reaching response rate targets and the need for more tailored fieldwork strategies aimed both at reducing survey costs and minimising the risk of bias.

Adverse drug reactions following nonresponse in a depressed patient with CYP2D6 deficiency and low CYP 3A4/5 activity

A 47-year-old male taxi driver experienced multiple adverse drug reactions during therapy with clomipramine (CMI) and quetiapine for major depressive disorder, after having been unsuccessfully treated with adequate doses of mirtazapine and venlafaxine. Drug serum concentrations of CMI and quetiapine were significantly increased and pharmacogenetic testing showed a poor metabolizer status for CYP2D6, low CYP3A4/5 activity and normal CYP2C19 genotype. After reduction of the CMI dose and discontinuation of quetiapine, all ADR subsided except for the increase in liver enzymes. The latter improved but did not normalize completely, even months later, possibly due to concomitant cholelithiasis.

Undercoverage and Nonresponse in a List-sampled Telephone Election Survey

For landline telephone surveys in particular, undercoverage has been a growing problem. However, research regarding the relative contributions of socio-demographic bias and other composition effects is scarce. We propose to address this issue by analyzing an election survey which used a sample from a register-based sampling frame containing basic socio-demographic information and to which telephone numbers were subsequently matched. With respect to socio-demographic representation of the final sample, we find that difficult to match groups are also difficult to contact, while those who cooperate tend to have different characteristics. We find bias due to undercoverage to be of greater magnitude than noncontact bias, while noncooperation falls between the two. As for substantive variables, both additional efforts to match missing telephone numbers and the construction of better weights are successful in closing the gap between survey estimates of voting behavior and true values from the election results.

Identifying Pertinent Variables for Nonresponse Follow-Up Surveys : Lessons Learned from 4 Cases in Switzerland

All social surveys suffer from different types of errors, of which one of the most studied is non-response bias. Non-response bias is a systematic error that occurs because individuals differ in their accessibility and propensity to participate in a survey according to their own characteristics as well as those from the survey itself. The extent of the problem heavily depends on the correlation between response mechanisms and key survey variables. However, non-response bias is difficult to measure or to correct for due to the lack of relevant data about the whole target population or sample. In this paper, non-response follow-up surveys are considered as a possible source of information about non-respondents. Non-response follow-ups, however, suffer from two methodological issues: they themselves operate through a response mechanism that can cause potential non-response bias, and they pose a problem of comparability of measure, mostly because the survey design differs between main survey and non-response follow-up. In order to detect possible bias, the survey variables included in non-response surveys have to be related to the mechanism of participation, but not be sensitive to measurement effects due to the different designs. Based on accumulated experience of four similar non-response follow-ups, we studied the survey variables that fulfill these conditions. We differentiated socio-demographic variables that are measurement-invariant but have a lower correlation with non-response and variables that measure attitudes, such as trust, social participation, or integration in the public sphere, which are more sensitive to measurement effects but potentially more appropriate to account for the non-response mechanism. Our results show that education level, work status, and living alone, as well as political interest, satisfaction with democracy, and trust in institutions are pertinent variables to include in non-response follow-ups of general social surveys. - See more at: https://ojs.ub.uni-konstanz.de/srm/article/view/6138#sthash.u87EeaNG.dpuf

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.

Estimate of Dementia Prevalence in a Community Sample from Sao Paulo, Brazil

Aims: To estimate dementia prevalence and describe the etiology of dementia in a community sample from the city of Sao Paulo, Brazil. Methods: A sample of subjects older than 60 years was screened for dementia in the first phase. During the second phase, the diagnostic workup included a structured interview, physical and neurological examination, laboratory exams, a brain scan, and DSM-IV criteria diagnosis. Results: Mean age was 71.5 years (n = 1,563) and 58.3% had up to 4 years of schooling (68.7% female). Dementia was diagnosed in 107 subjects with an observed prevalence of 6.8%. The estimate of dementia prevalence was 12.9%, considering design effect, nonresponse during the community phase, and positive and negative predictive values. Alzheimer`s disease was the most frequent cause of dementia (59.8%), followed by vascular dementia (15.9%). Older age and illiteracy were significantly associated with dementia. Conclusions: The estimate of dementia prevalence was higher than previously reported in Brazil, with Alzheimer`s disease and vascular dementia being the most frequent causes of dementia. Dementia prevalence in Brazil and in other Latin American countries should be addressed by additional studies to confirm these higher dementia rates which might have a sizable impact on countries` health services. Copyright (C) 2008 S. Karger AG, Basel