Handle Selection Bias Research Articles

Information projection approach to smoothed propensity score weighting for handling selection bias under missing at random

Information projection approach to smoothed propensity score weighting for handling selection bias under missing at random

A Potential Outcomes Approach to Selection Bias.

We propose a novel definition of selection bias in analytic epidemiology using potential outcomes. This definition captures selection bias under both the structural approach (where conditioning on selection into the study opens a noncausal path from exposure to disease in a directed acyclic graph) and the traditional definition (where a given measure of association differs between the study sample and the population eligible for inclusion). This approach is nonparametric, and selection bias under the approach can be analyzed using single-world intervention graphs both under and away from the null hypothesis. It allows the simultaneous analysis of confounding and selection bias, it explicitly links the selection of study participants to the estimation of causal effects using study data, and it can be adapted to handle selection bias in descriptive epidemiology. Through examples, we show that this approach provides a novel perspective on the variety of mechanisms that can generate selection bias and simplifies the analysis of selection bias in matched studies and case-cohort studies.

Attrition in the Gothenburg H70 birth cohort studies, an 18-year follow-up of the 1930 cohort.

Longitudinal studies are essential to understand the ageing process, and risk factors and consequences for disorders, but attrition may cause selection bias and impact generalizability. We describe the 1930 cohort of the Gothenburg H70 Birth Cohort Studies, followed from age 70 to 88, and compare baseline characteristics for those who continue participation with those who die, refuse, and drop out for any reason during follow-up. A population-based sample born 1930 was examined with comprehensive assessments at age 70 (N = 524). The sample was followed up and extended to increase sample size at age 75 (N = 767). Subsequent follow-ups were conducted at ages 79, 85, and 88. Logistic regression was used to analyze baseline characteristics in relation to participation status at follow-up. Refusal to participate in subsequent examinations was related to lower educational level, higher blood pressure, and lower scores on cognitive tests. Both attrition due to death and total attrition were associated with male sex, lower educational level, smoking, ADL dependency, several diseases, poorer lung function, slower gait speed, lower scores on cognitive tests, depressive symptoms, and a larger number of medications. Attrition due to death was also associated with not having a partner. It is important to consider different types of attrition when interpreting results from longitudinal studies, as representativeness and results may be differently affected by different types of attrition. Besides reducing barriers to participation, methods such as imputation and weighted analyses can be used to handle selection bias.

A semiparametric type II Tobit model for time-to-event data with application to a merger and acquisition study

ABSTRACTMerger and acquisition is an important corporate strategy. We collect recent merger and acquisition data for companies on the China A-share stock market to explore the relationship between corporate ownership structure and speed of merger success. When studying merger success, selection bias occurs if only completed mergers are analyzed. There is also a censoring problem when duration time is used to measure the speed. In this article, for time-to-event outcomes, we propose a semiparametric version of the type II Tobit model that can simultaneously handle selection bias and right censoring. The proposed model can also easily incorporate time-dependent covariates. A nonparametric maximum likelihood estimator is proposed. The resulting estimators are shown to be consistent, asymptotically normal, and semiparametrically efficient. Some Monte Carlo studies are carried out to assess the finite-sample performance of the proposed approach. Using the proposed model, we find that higher power balance of a company is associated with faster merger success.

Health economic evaluations based on routine data in Germany: a systematic review

BackgroundImproved data access and funding for health services research have promoted the application of routine data to measure costs and effects of interventions within the German health care system. Following the trend towards real world evidence, this review aims to evaluate the status and quality of health economic evaluations based on routine data in Germany.MethodsTo identify relevant economic evaluations, a systematic literature search in the databases PubMed and EMBASE was complemented by a manual search. The included studies had to be full economic evaluations using German routine data to measure either costs, effects, or both. Study characteristics were assessed with a structured template. Additionally, the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) were used to measure quality of reporting.ResultsIn total, 912 records were identified and 35 studies were included in the further analysis. The majority of these studies was published in the past 5 years (n = 27, 77.1%) and used insurance claims data as a source of routine data (n = 30, 85.7%). The most common method used for handling selection bias was propensity score matching. With regard to the reporting quality, 42.9% (n = 15) of the studies satisfied at least 80% of the criteria on the CHEERS checklist.ConclusionsThis review confirms that routine data has become an increasingly common data source for health economic evaluations in Germany. While most studies addressed the application of routine data, this analysis reveals deficits in considering methodological particularities and in reporting quality of economic evaluations based on routine data. Nevertheless, this review demonstrates the overall potential of routine data for economic evaluations.

Productivity effects of CAP investment support: Evidence from Sweden using matched panel data

This paper studies the effects of investment support from the common agricultural policy on labour and total factor productivity of agricultural firms in Sweden. Detailed firm-level data on 34 300 firms are used to estimate a matched panel model that relates firm productivity to a series of factors reflecting internal and external characteristics. The recently developed Coarsened Exact Matching method is used to estimate matched control groups and handle selection bias. Findings show a positive and significant treatment effect of investment support on firm productivity, but only for small firms. The analysis also reveals that an increase in the size of the support in relation to firm income has a negative and significant impact on productivity for all firms. Differentiating between various types of investment supports indicates heterogeneous treatment effects. The policy instrument can improve its efficiency if targeted to small firms and investments that have a link to public good provision.

Differences in Relationship Stability Between Individuals in First and Second Marriages

Compared to first marriages, remarriages have consistently demonstrated higher rates of divorce. Theories used to explain this phenomenon include the family process and selection perspectives. Researchers have attempted to address selection bias in this area but have not yet employed propensity score analysis to bolster causal inference. The purpose of this study is to assess the efficacy of covariate control methods for handling selection bias compared to propensity score methods and to examine whether a direct causal link between marriage order and relationship stability can be inferred after using propensity score greedy matching and propensity score weighting procedures. We analyze a sample of 1,679 first married and 410 remarried individuals drawn from the Relationship Evaluation Survey. Results provide support for the selection perspective and suggest that a direct causal link between marriage order and relationship stability should not be inferred. Implications, limitations, and future directions for research are discussed.

A comparison of missing data procedures for addressing selection bias in HIV sentinel surveillance data

BackgroundSelection bias is common in clinic-based HIV surveillance. Clinics located in HIV hotspots are often the first to be chosen and monitored, while clinics in less prevalent areas are added to the surveillance system later on. Consequently, the estimated HIV prevalence based on clinic data is substantially distorted, with markedly higher HIV prevalence in the earlier periods and trends that reveal much more dramatic declines than actually occur.MethodsUsing simulations, we compare and contrast the performance of the various approaches and models for handling selection bias in clinic-based HIV surveillance. In particular, we compare the application of complete-case analysis and multiple imputation (MI). Several models are considered for each of the approaches. We demonstrate the application of the methods through sentinel surveillance data collected between 2002 and 2008 from India.ResultsSimulations suggested that selection bias, if not handled properly, can lead to biased estimates of HIV prevalence trends and inaccurate evaluation of program impact. Complete-case analysis and MI differed considerably in their ability to handle selection bias. In scenarios where HIV prevalence remained constant over time (i.e. β = 0), the estimated derived from MI tended to be biased downward. Depending on the imputation model used, the estimated bias ranged from −1.883 to −0.048 in logit prevalence. Furthermore, as the level of selection bias intensified, the extent of bias also increased. In contrast, the estimates yielded by complete-case analysis were relatively unbiased and stable across the various scenarios. The estimated bias ranged from −0.002 to 0.002 in logit prevalence.ConclusionsGiven that selection bias is common in clinic-based HIV surveillance, when analyzing data from such sources appropriate adjustment methods need to be applied. The results in this paper suggest that indiscriminant application of imputation models can lead to biased results.

Little Evidence That Time in Child Care Causes Externalizing Problems During Early Childhood in Norway

Associations between maternal reports of hours in child care and children's externalizing problems at 18 and 36months of age were examined in a population-based Norwegian sample (n=75,271). Within a sociopolitical context of homogenously high-quality child care, there was little evidence that high quantity of care causes externalizing problems. Using conventional approaches to handling selection bias and listwise deletion for substantial attrition in this sample, more hours in care predicted higher problem levels, yet with small effect sizes. The finding, however, was not robust to using multiple imputation for missing values. Moreover, when sibling and individual fixed-effects models for handling selection bias were used, no relation between hours and problems was evident.

Utility of time-dependent inverse-probability-of-treatment weights to analyze observational cohorts in the intensive care unit

Objective When analyzing observational databases, marginal structural models (MSMs) may offer an appealing approach to estimate causal effects. We aimed at evaluating MSMs, in accounting for confounding when assessing the benefit of intensive care unit (ICU) admission and on its interaction with patient age, as compared with propensity score (PS) matching. Study Design and Setting PS and inverse-probability-of-treatment weights for MSMs were derived from an observational study designed to evaluate the benefit of ICU admission on in-hospital mortality. Only first ICU triages (time-fixed weights) or whole triage history (time-dependent weights) were considered. Weights were stabilized by either the prevalence of the actual treatment or the probability of the actual treatment given baseline covariates. Risk difference (RD) was the main outcome measure. Results MSMs with time-dependent weights offered the best reduction in the baseline imbalances as compared with PS matching. No effect of ICU admission on in-hospital mortality was found (RD = 0.010; 95% confidence interval = −0.038, 0.052) with no interaction between age and treatment. Conclusion MSMs appear interesting to handle selection bias in observational studies. When confounding evolves over time, the use of time-dependent weights should be stressed out.

Handling selection bias when choosing actions in retail credit applications

In many situations one needs to know which action one should take with a customer to yield the greatest response. Typically, estimates of the response functions of different actions will be based on the responses of customers previously assigned to each action. Often, however, the previous assignments will not have been random, so that estimates of the response functions will be biased. We examine the case of two possible actions. We look at the error arising from using the simple OLS estimate ignoring the selection bias, and also explore the possibility of using the Heckman model to allow for the sample selectivity. The performance of Heckman’s model is then compared with the simple OLS through simulation.

What works for whom? An analysis of active labour market programmes in Norway

This article examines whether some active labour market programmes (ALMP) targeted at social assistance recipients are more effective for some groups than for others in increasing self-sufficiency, i.e. gross earnings. The study focuses on the interaction between two types of programmes – employment programmes and training programmes – and the social charac-teristics of the participants. The data are derived from the pool of the entire population of social assistance recipients registered in Norway in 1995. The study adopts a quasi-experimental design. To handle selection bias, a matching procedure based on a propensity score approach is applied. Training programmes yield positive outcomes overall and for subgroups, whereas employment programmes produce no significant effects overall. However, for employment programmes there is an interaction effect between the likelihood of participation and outcome: beneficiaries with social characteristics (e.g. age, education, work experience) that are associated with a medium chance of participating experience a positive and significant increase in earnings. For those with lower and higher chances, the effect is negative. This points to the importance of conducting stratified analyses in effect evaluations. Thus, the results are also likely to be more relevant to policy makers.

Active labour market programmes in Norway: are they helpful for social assistance recipients?

This paper examines whether active labour market programmes (ALMPs) directed to the most disadvantaged in Norway are helpful in moving them from social assistance to self-sufficiency (i.e. work, earnings, and decent income) . The study focuses on programme packages that integrate several components and are especially targeted at the disadvantaged unemployed. Nine such programme packages are evaluated. Thus, nine comparison groups are formed, one for each intervention group. Both groups are derived from the pool of the entire population of social assistance recipients registered in 1995. The study adopts a quasi-experimental design. To handle selection bias, a matching procedure based on a propensity-score approach is applied. The results indicate that most of the programme packages yield a positive and, in most cases, significant effect on subsequent employment and earnings, both in the short and in the long run, that is, up to five years later.

Instrumental variables and inverse probability weighting for causal inference from longitudinal observational studies.

Inferring causal effects from longitudinal repeated measures data has high relevance to a number of areas of research, including economics, social sciences and epidemiology. In observational studies in particular, the treatment receipt mechanism is typically not under the control of the investigator; it can depend on various factors, including the outcome of interest. This results in differential selection into treatment levels, and can lead to selection bias when standard routines such as least squares regression are used to estimate causal effects. Interestingly, both the characterization of and methodology for handling selection bias can differ substantially by disciplinary tradition. In social sciences and economics, instrumental variables (IV) is the standard method for estimating linear and nonlinear models in which the error term may be correlated with an observed covariate. When such correlation is not ruled out, the covariate is called endogenous and least squares estimates of the covariate effect are typically biased. The availability of an instrumental variable can be used to reduce or eliminate the bias. In public health and clinical medicine (e.g., epidemiology and biostatistics), selection bias is typically viewed in terms of confounders, and the prevailing methods are geared toward making proper adjustments via explicit use of observed confounders (e.g., stratification, standardization). A class of methods known as inverse probability weighting (IPW) estimators, which relies on modeling selection in terms of confounders, is gaining in popularity for making such adjustments. Our objective is to review and compare IPW and IV for estimating causal treatment effects from longitudinal data, where the treatment may vary with time. We accomplish this by defining the causal estimands in terms of a linear stochastic model of potential outcomes (counterfactuals). Our comparison includes a review of terminology typically used in discussions of causal inference (e.g., confounding, endogeneity); a review of assumptions required to identify causal effects and their implications for estimation and interpretation; description of estimation via inverse weighting and instrumental variables; and a comparative analysis of data from a longitudinal cohort study of HIV-infected women. In our discussion of assumptions and estimation routines, we try to emphasize sufficient conditions needed to implement relatively standard analyses that can essentially be formulated as regression models. In that sense this review is geared toward the quantitative practitioner. The objective of the data analysis is to estimate the causal (therapeutic) effect of receiving combination antiviral therapy on longitudinal CD4 cell counts, where receipt of therapy varies with time and depends on CD4 count and other covariates. Assumptions are reviewed in context, and resulting inferences are compared. The analysis illustrates the importance of considering the existence of unmeasured confounding and of checking for 'weak instruments.' It also suggests that IV methodology may have a role in longitudinal cohort studies where potential instrumental variables are available.

Prospective funding of general hospitals in Norway--incentives for higher production?

In Norway, a new system of Activity Based Financing (ABF) for general hospitals was introduced on a comprehensive basis in July 1997. The main purpose of the reform was to increase activity so that more patients could receive treatment more quickly without reducing the quality of care. In this paper we analyse whether the reform has had any significant effect using two different performance indicators: number of patients treated and production of DRG points (Diagnosis Related Group). We divide the hospitals into two groups: hospitals owned by counties that have adopted the ABF system, and hospitals owned by counties using other funding systems. The first group then becomes the experiment group, while the second serves as a comparison group. It is argued that fixed-effect models are suitable specifications for this evaluation study, handling selection bias and the influence of unobservable explanatory variables in a consistent manner. We find that the reform has had a significant effect on the number of patients treated and DRG points produced. The results are sensitive as to how the experiment and the comparison group are determined.