Handling hybrid and missing data in constraint-based causal discovery to study the etiology of ADHD

Sokolova, Elena; Rhein, Daniel von; Naaijen, Jilly; Groot, P.; Claassen, Tom; Buitelaar, Jan K.; Heskes, Tom

doi:10.1007/s41060-016-0034-x

Cited by 9 publications

(6 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An alternative missing value method for causal discovery not considered in this article is inverse probability weighting 66 . Likelihood‐based approaches such as expectation maximization can be used with score‐based causal discovery, 67,68 but are not straightforward to combine with constraint‐based algorithms (see Sokolova et al 69 for a first idea assuming a so‐called joint nonparanormal distribution).…”

Section: Discussionmentioning

confidence: 99%

Multiple imputation and test‐wise deletion for causal discovery with incomplete cohort data

Witte

Foraita

Didelez

2022

Statistics in Medicine

View full text Add to dashboard Cite

Causal discovery algorithms estimate causal graphs from observational data. This can provide a valuable complement to analyses focusing on the causal relation between individual treatment‐outcome pairs. Constraint‐based causal discovery algorithms rely on conditional independence testing when building the graph. Until recently, these algorithms have been unable to handle missing values. In this article, we investigate two alternative solutions: test‐wise deletion and multiple imputation. We establish necessary and sufficient conditions for the recoverability of causal structures under test‐wise deletion, and argue that multiple imputation is more challenging in the context of causal discovery than for estimation. We conduct an extensive comparison by simulating from benchmark causal graphs: as one might expect, we find that test‐wise deletion and multiple imputation both clearly outperform list‐wise deletion and single imputation. Crucially, our results further suggest that multiple imputation is especially useful in settings with a small number of either Gaussian or discrete variables, but when the dataset contains a mix of both neither method is uniformly best. The methods we compare include random forest imputation and a hybrid procedure combining test‐wise deletion and multiple imputation. An application to data from the IDEFICS cohort study on diet‐ and lifestyle‐related diseases in European children serves as an illustrating example.

show abstract

Section: Discussionmentioning

confidence: 99%

Multiple imputation and test‐wise deletion for causal discovery with incomplete cohort data

Witte

Foraita

Didelez

2022

Statistics in Medicine

View full text Add to dashboard Cite

show abstract

“…An alternative missing value method for causal discovery not considered in this paper is in-verse probability weighting (Gain and Shpitser, 2018). Likelihood-based approaches such as Expectation Maximisation can be used with score-based causal discovery (Friedman, 1997;Scutari, 2020) but are not straightforward to combine with constraint-based algorithms (see Sokolova et al, 2017, for a first idea assuming a joint nonparanormal distribution).…”

Section: Discussionmentioning

confidence: 99%

Multiple imputation and test-wise deletion for causal discovery with incomplete cohort data

Witte¹,

Foraita²,

Didelez³

2021

Preprint

View full text Add to dashboard Cite

Causal discovery algorithms estimate causal graphs from observational data. This can provide a valuable complement to analyses focussing on the causal relation between individual treatment-outcome pairs. Constraint-based causal discovery algorithms rely on conditional independence testing when building the graph. Until recently, these algorithms have been unable to handle missing values. In this paper, we investigate two alternative solutions: Test-wise deletion and multiple imputation. We establish necessary and sufficient conditions for the recoverability of causal structures under test-wise deletion, and argue that multiple imputation is more challenging in the context of causal discovery than for estimation. We conduct an extensive comparison by simulating from benchmark causal graphs: As one might expect, we find that test-wise deletion and multiple imputation both clearly outperform list-wise deletion and single imputation. Crucially, our results further suggest that multiple imputation is especially useful in settings with a small number of either Gaussian or discrete variables, but when the dataset contains a mix of both neither method is uniformly best. The methods we compare include random forest imputation and a hybrid procedure combining test-wise deletion and multiple imputation. An application to data from the IDEFICS cohort study on diet-and lifestylerelated diseases in European children serves as an illustrating example.

show abstract

“…One possible example is the work of Sokolova et al [88] where an alternative approach is proposed to deal with mixed data (a mix of continuous and discrete data) and missing values, which are very common in medical data. This is done by transforming the continuous and discrete data into a normal distribution (many of the implementations of the algorithms presented in Sections 5.1 and 5.2 can only deal with continuous data) by using a Gaussian copula, consisting in (for each variable X i ):…”

Section: Figure 5 Example Of a Causal Decision Tree [56] And Comparimentioning

confidence: 99%

Causal discovery in machine learning: Theories and applications

Nogueira¹,

Gama²,

Ferreira³

2021

JDG

View full text Add to dashboard Cite

Determining the cause of a particular event has been a case of study for several researchers over the years. Finding out why an event happens (its cause) means that, for example, if we remove the cause from the equation, we can stop the effect from happening or if we replicate it, we can create the subsequent effect. Causality can be seen as a mean of predicting the future, based on information about past events, and with that, prevent or alter future outcomes. This temporal notion of past and future is often one of the critical points in discovering the causes of a given event. The purpose of this survey is to present a cross-sectional view of causal discovery domain, with an emphasis in the machine learning/data mining area.

show abstract

Handling hybrid and missing data in constraint-based causal discovery to study the etiology of ADHD

Cited by 9 publications

References 37 publications

Multiple imputation and test‐wise deletion for causal discovery with incomplete cohort data

Multiple imputation and test‐wise deletion for causal discovery with incomplete cohort data

Multiple imputation and test-wise deletion for causal discovery with incomplete cohort data

Causal discovery in machine learning: Theories and applications

Contact Info

Product

Resources

About