Research map / Probabilistic and scientific modeling / Causal discovery
Causal Explanation of Quality of Parent-Child Interactions with Multimodal Behavioral Features (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Combines sparse multiple canonical correlation analysis with causal discovery (GFCI in TETRAD) to uncover explainable causal links between nine categories of multimodal behavioral features and expert ratings of parent-child interaction quality.
Problem
Machine learning models predicting expert ratings of parenting behavior from video recordings are black-box, using many low-level multimodal features that offer limited explainability for domain experts.
Approach
Reduce 157 low-level multimodal features in 9 behavioral groups to sparse composite variables with sparse mCCA, then run causal discovery (GFCI, yielding an equivalence class graph) over those composites and the three rating outcomes to identify explainable causal relations.
Key results
- Random Forest models reduced RMSE vs baseline mean predictor by 6.59% (sensitivity), 0% (intrusiveness), and 21.11% (positive regard).
- Causal graph shows LIWC features causally linked to intrusiveness and positive regard, and turn frequency linked to intrusiveness.
- Audio, paralinguistic, and LIWC emotion features form one causally linked cluster; video and remaining language features form another.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Random Forest instead of a baseline mean predictor | Parenting behavior rating prediction · Sensitivity | RMSE: 6.59% | Helped |
| Use Random Forest instead of a baseline mean predictor | Parenting behavior rating prediction · Intrusiveness | RMSE: 0% | No clear effect |
| Use Random Forest instead of a baseline mean predictor | Parenting behavior rating prediction · Positive regard | RMSE: 21.11% | Helped |
Surprising findings
- Random Forest models showed no RMSE reduction over the mean predictor for intrusiveness (0%).
Limitations
- Causal directionality not established; proposed follow-up with linear Gaussian SEMs and goodness-of-fit.
- Predictive models only moderately accurate; baseline improvements small or zero for some outcomes.
- Preliminary results from a single dataset of 220 dyads.
Experimental setup
Datasets: pre-intervention at-home recordings of semi-structured mother-child play from a randomized controlled study (220 pairs)
Metrics: root mean squared error reduction vs mean predictor
Baselines: Random Forest mean predictor baseline
Builds on: sparse multiple canonical correlation analysis GFCI
Closest papers by meaning
- Local Causal Structure Learning in the Presence of Latent Variables (ICML 2024 · similarity 0.85)
- Revealing Multimodal Causality with Large Language Models (NeurIPS 2025 · similarity 0.84)
- Causal Representation Learning from Multimodal Biomedical Observations (ICLR 2025 · similarity 0.83)
- Differentiable Causal Discovery for Latent Hierarchical Causal Models (ICLR 2025 · similarity 0.83)
- Learning Mixtures of Unknown Causal Interventions (NeurIPS 2024 · similarity 0.82)
- Local Identifying Causal Relations in the Presence of Latent Variables (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.