Graphical Representation of Missing Data Problems

Thoemmes, Felix; Mohan, Karthika

doi:10.1080/10705511.2014.937378

Cited by 57 publications

(54 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A priori all three groups differ from each other and assumptions on the missing data mechanism are needed for unbiased estimation. We apply graphical models [18][19][20] to describe the assumptions on the missing data mechanism. The validity of the assumptions is evaluated using follow-up data on mortality and morbidity.…”

Section: A C C E P T E D Accepted Manuscriptmentioning

confidence: 99%

Selection bias was reduced by recontacting nonparticipants

Karvanen

Tolonen

Härkänen

et al. 2016

Journal of Clinical Epidemiology

View full text Add to dashboard Cite

Objective: One of the main goals of health examination surveys (HES) is to provide unbiased estimates of health indicators at the population level. We demonstrate how multiple imputation methods may help to reduce the selection bias if partial data on some non-participants are collected. Study Design and Setting: In the FINRISK 2007 study, a populationbased health study conducted in Finland, a random sample of 10 000 men and women aged 25-74 were invited to participate. The study included a questionnaire data collection and a health examination. 6255 individuals participated in the study. Out of 3745 non-participants, 473 returned a simplified questionnaire after a recontact. Both the participants and the nonparticipants were followed up for death and hospitalizations. The follow-up data allowed to check the assumptions on the missing data mechanism and tailored multiple imputation methods were used to handle the missing data. Results: Non-participation is a strong predictor for mortality in the fiveyear follow-up. However, the recontact response does not predict mortality or morbidity among the non-participants when adjusted for age and sex. The M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 2 result suggests that the recontact respondents can be used as proxy for all non-participants. A comparison of raw estimates and estimates adjusted for selection bias reveals clear differences in the estimated population prevalences of smoking and heavy alcohol usage. Conclusion: All efforts to collect data on non-participants are likely to be useful even if the response rate for the recontact remains low. Statistical analysis of the recontact respondents provides an indication of the extent of the selection bias, even in studies where follow-up data are not available to check the assumptions.Keywords: survey; non-response; bias; missing data; multiple imputation What is new?:• Register data on cause specific mortality and morbidity can be used to check the assumptions on the missing data mechanism.• In the FINRISK 2007 study, we found that while the participants and the non-participants clearly differ, the non-participants with and without a recontact response have similar mortality and morbidity when adjusted for age and sex.• We propose a multiple imputation approach to handle data missing not at random. The approach is applicable when data on a non-response questionnaire are available.

show abstract

Section: A C C E P T E D Accepted Manuscriptmentioning

confidence: 99%

Selection bias was reduced by recontacting nonparticipants

Karvanen

Tolonen

Härkänen

et al. 2016

Journal of Clinical Epidemiology

View full text Add to dashboard Cite

show abstract

“…The importance of such model-based decision-making and the issue of conditioning on colliders as in the M-bias structure has recently also been recognized in the domain of missing data analysis, where conventional wisdom dictated that all available covariates should always be used in the estimation of parameters in the presence of missing data. That this is not generally correct has been shown by several authors [10][11][12][13].…”

Section: Resultsmentioning

confidence: 82%

M-bias, Butterfly Bias, and Butterfly Bias with Correlated Causes – A Comment on Ding and Miratrix (2015)

Thoemmes

2015

Journal of Causal Inference

Self Cite

View full text Add to dashboard Cite

Ding and Miratrix [1] recently concluded that adjustment on a pre-treatment covariate is almost always preferable to reduce bias. I extend the examined parameter space of the models considered by Ding and Miratrix, and consider slight extensions of their models as well. Similar to the conclusion by Pearl [7], I identify constellations in which bias due to adjustment, or failing to adjust is symmetrical, but also confirm some findings of Ding and Miratrix.Keywords: M-bias, causal inference, butterfly bias Ding and Miratrix [1], henceforth DM, recently examined bias in graphical causal models. Their main interest was to explore whether it is beneficial to adjust on a variable that may have both confounder properties (being a common cause of two variables), or collider properties (being a common effect of two variables). This is an important question, as e.g., evidenced by the debate of Rubin [2], and Pearl [3,4], and I applaud the authors to tackle it, especially using the methodology expressed in Pearl [5], and Chen and Pearl [6].The main conclusion of DM was that typically adjustment is preferable, even if a variable may have bias-inducing properties. The authors justified this conclusion based on the fact that in a majority of their conditions for which they derived asymptotic bias, adjustment yielded smaller biases. DM however did note that some of their analyses were incomplete, because certain path coefficients were always fixed to be of equal magnitude, and importantly of equal sign. They explicitly left exploration of a broader parameter space (including negative correlations) to future studies.In a comment, Pearl [7] rebutted the claim of the authors, and argued that in the case of the M-bias structure with correlated causes, bias due to adjustment should be comparable in size to bias due to failing to adjust for a confounder. In the case of so-called butterfly bias, in which both confounding and M-bias are present, Pearl [7] also conjectured that bias due to adjustment and lack of adjustment would be largely symmetrical, an argument based on the equal prevalence of positive and negative correlations of unobserved causes in the M-bias structure.This comment extends the work of DM, by examining the same models, or slight extensions thereof, but also varying involved path coefficients over a much larger, and arguably more complete, parameter space. Asymptotic biasesI used the same methodology as DM to derive the asymptotic bias in the model depicted in Figure 1.1 Note that all models of DM are included in this model as special cases. If path coefficients e and f are set to 0,

show abstract

“…Finally, we suggest that the field as a whole should try to shift its norms toward a more productive engagement with causal inference on the basis of nonexperimental data. Statistics and methods teachers could dedicate some more time to the topic-it may be time spent well, as a clearer framework for causal inference makes it easier to talk about a broad range of topics, such as missing data problems (Thoemmes & Mohan, 2015) and threats to validity, which affect most types of research (Matthay & Glymour, 2020). Editors and reviewers may also encourage a shift in thinking.…”

Section: Further Recommendationsmentioning

confidence: 99%

The Taboo Against Explicit Causal Inference in Nonexperimental Psychology

Grosz¹,

Rohrer²,

Thoemmes³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Causal inference is a central goal of research. However, most psychologists refrain from explicitly addressing causal research questions and avoid drawing causal inference on the basis of nonexperimental evidence. We argue that this taboo against causal inference in nonexperimental psychology impairs study design and data analysis, holds back cumulative research, leads to a disconnect between original findings and how they are interpreted in subsequent work, and limits the relevance of nonexperimental psychology for policy making. At the same time, the taboo does not prevent researchers from interpreting findings as causal effects—the inference is simply made implicitly, and assumptions remain unarticulated. Thus, we recommend that nonexperimental psychologists begin to talk openly about causal assumptions and causal effects. Only then can researchers take advantage of recent methodological advances in causal reasoning and analysis and develop a solid understanding of the underlying causal mechanisms that can inform future research, theory, and policy makers.

show abstract

Graphical Representation of Missing Data Problems

Cited by 57 publications

References 28 publications

Selection bias was reduced by recontacting nonparticipants

Selection bias was reduced by recontacting nonparticipants

M-bias, Butterfly Bias, and Butterfly Bias with Correlated Causes – A Comment on Ding and Miratrix (2015)

The Taboo Against Explicit Causal Inference in Nonexperimental Psychology

Contact Info

Product

Resources

About