Causal Forests Research Articles

Research on the identification of key factors and influence mechanism of digital transformation of manufacturing enterprises based on machine learning

Background The digital transformation of the manufacturing industry has become a key factor in enhancing enterprise competitiveness and promoting high-quality economic development amid globalization and rapid technological advances, for Chinese manufacturing enterprises, it is also a vital path to achieving high-quality development. Objective This study explores the factors influencing the digital transformation of manufacturing industries through multi-dimensional analysis with advanced machine learning techniques, assisting Chinese manufacturing enterprises in achieving high-quality transformation while considering local characteristics. Methods An interpretable model integrating XGBoost and SHAP values is proposed based on the TOE model to analyze key factors. Additionally, the causal forest approach is used to explore regional variations in these factors. Results It was found that factors such as invention patents, digital-oriented management innovation, equity incentives, and profitability significantly drive digital transformation in manufacturing firms. There are also regional differences in the importance of these factors. Conclusions The empirical evidence provides a crucial reference for enterprises and decision-makers to formulate more scientifically grounded digital transformation strategies based on regional characteristics, offering strong support for transformation according to local conditions.

Heterogeneous Effects of Decreasing the Cost-Sharing for Outpatient Care on Health Outcomes in China: A Propensity Score Matching and Causal Machine Learning Approach.

To improve accessibility and financial support for outpatient services, China introduced a scheme to decrease cost-sharing for outpatient care under the Urban Employee Basic Medical Insurance. This study evaluates the health impacts of this policy and examines its heterogeneous effects. Utilising data from the 2018 China Health and Retirement Longitudinal Study, we analysed 2896 individual-level observations across 105 prefectures. Propensity score matching and a causal forest model were applied to evaluate the effects on chronic disease status, body pain, self-rated health, and hospitalisation, while accounting for various demographic, socioeconomic, residential, health-related behaviours, and prefecture-specific factors. The reduction in cost-sharing was significantly linked to decreased probabilities of chronic disease (Average Treatment Effect (ATE)=-0.0619, p<0.01), body pain (ATE=-0.0715, p<0.05), and hospitalisation (ATE=-0.0592, p<0.001), as well as improved self-rated health (ATE=0.1557, p<0.001). These benefits may be attributed to reduced out-of-pocket payments for outpatient care (ATE=-287.6112, p<0.01) and increased outpatient visits (ATE=0.0414 visits, p<0.05). Causal forest analyses revealed that older individuals, those with higher educational attainment, higher household income, urban residents, and those engaging in healthier behaviours exhibited larger treatment effects. Decreasing outpatient cost-sharing in China has beneficial health outcomes, with variations in its impact based on socio-economic status and health behaviours. It is advisable to further increase reimbursement rates and broaden benefit packages for outpatient care, while addressing the unequal distribution of benefits.

The Role of Local Government Decarbonization Pressures in Enhancing Urban Industrial Intelligence: An Analysis of Proactive and Reactive Corporate Environmental Governance

In the context of China’s accelerated “dual transition” towards industrial intelligence and green development, this paper investigates how local government decarbonization pressures affect urban industrial intelligence in China. Using the Low-Carbon City Pilot policy as a quasi-natural experiment, a staggered difference-in-differences approach and Causal Forest model reveal the following findings: (1) Local government decarbonization pressures significantly boost urban industrial intelligence. (2) Local government decarbonization pressures foster intelligent development by encouraging the introduction of intelligent policies, which motivate enterprises to adopt proactive strategies. Meanwhile, the pressures compel enterprises to engage in source-based environmental governance, resulting in a passive intelligent response. Together, these approaches enhance urban industrial intelligence. (3) Fiscal pressure negatively moderates the relationship between local government decarbonization pressures and urban industrial intelligence. (4) There is an inverted U-shaped relationship between openness to foreign trade and the Conditional Average Treatment Effect (CATE), while CATE is higher for cities with higher urban labor costs. (5) Finally, urban industrial intelligence effectively channels local government decarbonization pressures into measurable emission reductions. These findings have significant policy relevance for building a low-carbon, intelligent society.

Is the energy quota trading policy a solution to carbon inequality in China? Evidence from double machine learning.

Is the energy quota trading policy a solution to carbon inequality in China? Evidence from double machine learning.

Associations between adolescent psychiatric disorders and adulthood payment problems: a Norwegian register study of complete birth cohorts of 1995–1997

BackgroundPsychiatric disorder diagnoses are linked to long-term socioeconomic ‘shadows’ into adulthood, but little is known about how these diagnoses are associated with adulthood payment problems in Norway and whether these...

Estimating Treatment Effect Heterogeneity in Psychiatry: A Review and Tutorial With Causal Forests.

Flexible machine learning tools are increasingly used to estimate heterogeneous treatment effects. This paper gives an accessible tutorial demonstrating the use of the causal forest algorithm, available in the R package grf. We start with a brief non-technical overview of treatment effect estimation methods, focusing on estimation in observational studies; the same techniques can also be applied in experimental studies. We then discuss the logic of estimating heterogeneous effects using the extension of the random forest algorithm implemented in grf. Finally, we illustrate causal forest by conducting a secondary analysis on the extent to which individual differences in resilience to high combat stress can be measured among US Army soldiers deploying to Afghanistan based on information about these soldiers available prior to deployment. We illustrate simple and interpretable exercises for model selection and evaluation, including targeting operator characteristics curves, Qini curves, area-under-the-curve summaries, and best linear projections. A replication script with simulated data is available at https://github.com/grf-labs/grf/tree/master/experiments/ijmpr.

Econometric advances in causal inference: The machine learning revolution

This is one of the challenges that new and fast-growing econometric literature is beginning to tackle in addressing causal inference problems with machine learning methods. Yet, empirical economics still has not really made use of the strengths of these modern approaches. Here, we revisit groundbreaking empirical work through the perspective of causal machine learning methods to connect econometric theory with applied economics. In particular, we will cover double machine learning, causal forests, and more general machine learning methodologies, both in the setting of average treatment effects and heterogeneous treatment effects. We demonstrate the application of these methods in diverse settings and discuss their significance and additional benefits relative to classical approaches that were utilized in the original studies.

Tranexamic acid for trauma: optimal timing of administration based on the CRASH-2 and CRASH-3 trials.

Tranexamic acid reduces bleeding deaths in trauma patients, but the treatment benefit depends on the time from injury. It is recommended that tranexamic acid be administered immediately and only within 3 h of injury; however, the optimal criteria have not been adequately studied. We applied machine learning-based causal forest models to investigate heterogeneity in the effects of tranexamic acid on 24-hour mortality rate conditional on covariates (for example age, sex, time from injury, systolic blood pressure, and Glasgow Coma Scale, GCS). We analysed data on 28 448 trauma patients in the CRASH-2 and CRASH-3 randomized trials. We used the policytree algorithm to determine the optimal criteria for tranexamic acid treatment. The causal forest models showed heterogeneity in the effects of tranexamic acid on 24-hour mortality rate. The relative risk reduction was greatest in patients treated within 2 h of injury but thereafter decreased rapidly. The pattern was similar regardless of age or systolic blood pressure, although with decreasing GCS, the time to treatment effects were weaker, with benefits beyond 3 h. The largest absolute risk reductions were in patients with a low blood pressure and a low GCS when treated soon after injury. The optimal criterion was statistically determined as patients within 2 h of the injury or with GCS < 9. Tranexamic acid administration was found to be beneficial when given within 2 h of injury. In patients with severe traumatic brain injury, the treatment benefits may persist beyond the 2-hour window.

The impact of maternal employment on child nutritional diversity in Bangladesh: A causal forest analysis with clustered data

Understanding the impact of women’s employment on children’s nutrition is crucial for informing effective public health policies. This study examines the relationship between mothers’ employment status and the dietary diversity of their children, aged 6 months to 5 years, in Bangladesh. The Nutritional Variety Score (NVS) is used as a measure of dietary diversity, capturing the consumption of various food groups, including eggs, meat, bread, potatoes, vegetables, fruits, fish, beans, and dairy products. To explore this relationship, advanced statistical methods were employed, including causal forest models with cluster identifiers and mixed-effects multilevel logistic regression for propensity scores. The analysis utilized data from the 2022 Bangladesh Demographic and Health Survey (BDHS), a comprehensive dataset encompassing information on women’s employment, household characteristics, and children’s dietary intake. The models controlled for several confounding variables, including the number of children, partner’s education and employment status, type of residence, wealth index, and mother’s education level. The results reveal that children of employed mothers have a higher NVS than those of non-employed mothers, with an estimated average treatment effect (ATE) of 0.532 (95% CI: 0.365-0.699). This finding suggests that working mothers may have better access to resources or opportunities to provide a more diverse diet for their children. The statistically significant ATE confirms a positive causal relationship between women’s employment and children’s nutritional variety. This study contributes to the literature by offering robust evidence on how maternal employment affects child nutrition in Bangladesh. Journal of Statistical Research 2024, Vol. 58, No. 2, pp. 317-333

Affect, Not Ideology: The Heterogeneous Effects of Partisan Cues on Policy Support

Abstract How do individuals process political information? What behavioral mechanisms drive partisan bias? In this paper, we evaluate the extent to which partisan bias is driven by affect or ideology in a three-pronged approach informed by both psychological theories and recent advances in methodology. First, we use a novel survey experiment designed to disentangle the competing mechanisms of ideology and partisan affect. Second, we leverage multidimensional scaling methods for latent variable estimation for both partisan affect and ideology. Third, we employ a principled machine learning method, causal forest, to detect and estimate heterogeneous treatment effects. Contrary to previous literature, we find that affect is the sole moderator of partisan cueing processes, and only for out-party cues. These findings not only contribute to the literature on political behavior, but underscore the importance of careful measurement and robust subgroup analysis.

Environmental Sustainability and Business Profitability: Profiling Winners and Losers With Machine Learning

ABSTRACTWe utilize a rich dataset of manufacturing firms to investigate the heterogeneous effects of ISO 14001 on the financial performance of certified firms. We employ machine learning techniques, specifically causal tree and causal forest, to uncover these effects. Our findings reveal consistently positive average effects of ISO 14001 certification on sales revenue across all categories of firms. However, when it comes to profitability, we observe significant heterogeneity in the impact of ISO 14001 certification. Specifically, ISO 14001 certification spurs profitability gains among more innovative firms with lower debt‐to‐equity ratios, among firms that are more reliant on exports, and those that operate outside the electronic component industry. Conversely, firms with a large debt‐to‐equity ratio and those that are privately held experience negative effects of ISO 14001 certification on profitability. Our study contributes to the literature examining how environmental sustainability programs, such as ISO 14001, affect firm financial performance in a heterogeneous manner. By uncovering the nuanced effects of ISO 14001 certification based on firm characteristics, our research provides valuable insights that can assist in optimizing the outcomes of similar environmental programs by tailoring strategies based on specific firm attributes.

Application of Causal Forest Model to Examine Treatment Effect Heterogeneity in Substance Use Disorder Psychosocial Treatments.

Heterogeneity of treatment effect (HTE) is a concern in substance use disorder (SUD) treatments but has not been rigorously examined. This exploratory study applied a causal forest approach to examine HTE in psychosocial SUD treatments, considering multiple covariates simultaneously. Data from 12 randomized controlled trials of nine psychosocial treatments were obtained from the National Institute on Drug Abuse Clinical Trials Network. Using causal forests, we estimated the conditional average treatment effect (CATE) on drug abstinence. To assess HTE, we compared CATE variance against total outcome variability, conducted an omnibus test, and applied the Rank-Weighted Average Treatment Effect (RATE). Across nine interventions, CATE variance was lower than total outcome variability, indicating lack of strong evidence of HTE with respect to the baseline covariates considered. The omnibus test and RATE analysis generally support this finding. However, the RATE analysis identified potential HTE in a motivational interviewing trial; this could be a false positive given the multiple analyses; replication is needed to confirm this. While causal forests show utility in exploring HTE in SUD interventions, limited baseline assessments in most trials suggest a cautious interpretation. The RATE findings for motivational interviewing highlight potential subgroup-specific treatment benefits, warranting further research.

Application of causal forests to randomised controlled trial data to identify heterogeneous treatment effects: a case study

BackgroundClassical approaches to subgroup analysis in randomised controlled trials (RCTs) to identify heterogeneous treatment effects (HTEs) involve testing the interaction between each pre-specified possible treatment effect modifier and the treatment effect. However, individual significant interactions may not always yield clinically actionable subgroups, particularly for continuous covariates. Non-parametric causal machine learning approaches are flexible alternatives for estimating HTEs across many possible treatment effect modifiers in a single analysis.MethodsWe conducted a secondary analysis of the VANISH RCT, which compared the early use of vasopressin with norepinephrine on renal failure-free survival for patients with septic shock at 28 days. We used classical (separate tests for interaction with Bonferroni correction), data-adaptive (hierarchical lasso regression), and non-parametric causal machine learning (causal forest) methods to analyse HTEs for the primary outcome of being alive at 28 days. Causal forests comprise honest causal trees, which use sample splitting to determine tree splits and estimate treatment effects separately. The modal initial (root) splits of the causal forest were extracted, and the mean value was used as a threshold to partition the population into subgroups with different treatment effects.ResultsAll three models found evidence of HTE with serum potassium levels. Univariable logistic regression OR 0.435 (95%CI [0.270, 0.683]. p = 0.0004), hierarchical lasso logistic regression standardised OR: 0.604 (95% CI 0.259, 0.701), lambda = 0.0049. Hierarchical lasso kept the interaction between the treatment and serum potassium, sodium level, minimum temperature, platelet count and presence of ischemic heart disease. The causal forest approach found some evidence of HTE (p = 0.124). When extracting root splits, the modal split was on serum potassium (mean applied threshold of 4.68 mmol/L). When dividing the patient population into subgroups based on the mean initial root threshold, risk differences in being alive at 28 days were 0.069 (95%CI [-0.032, 0.169]) and − 0.257 (95%CI [-0.368, -0.146]) with serum potassium ≤ 4.68 and > 4.68 respectively.ConclusionsThe causal forest agreed with the data-adaptive and classical method of subgroup analysis in identifying HTE by serum potassium. Whilst classical and data-adaptive methods may identify sources of HTE, they do not immediately suggest subgroup splits which are clinically actionable. The extraction of root splits in causal forests is a novel approach to obtaining data-derived subgroups, to be further investigated.

Machine-learning approaches to predict individualized treatment effect using a randomized controlled trial.

Recent advancements in machine learning (ML) for analyzing heterogeneous treatment effects (HTE) are gaining prominence within the medical and epidemiological communities, offering potential breakthroughs in the realm of precision medicine by enabling the prediction of individual responses to treatments. This paperintroducesthe methodological frameworks used to study HTEs, particularly based ona single randomized controlled trial (RCT). Wefocus on methods toestimate conditional average treatment effect (CATE)for multiple covariates, aimingto predict individualized treatment effects. We explore a range of methodologies from basic frameworks like the T-learner, S-learner, and Causal Forest, to more advanced ones such as the DR-learner and R-learner, as well as cross-validation for CATE estimation to enhance statistical efficiency by estimating CATE for all RCT participants. We also provide a practical application of these approaches using the Preventing Overweight Using Novel Dietary Strategies (POUNDS Lost) trial, which compared the effects of high versus low-fat diet interventions on 2-year weight changes. We compared different sets of covariates for CATE estimation, showing that the DR- and R-learners are useful for the estimation of CATE in high-dimensional settings. This paper aims to explain the theoretical underpinnings and methodological nuances of ML-based HTE analysis withoutrelying on technical jargon, making these concepts more accessible to the clinical and epidemiological research communities.

How Do Applied Researchers Use the Causal Forest? A Methodological Review

SummaryThis methodological review examines the use of the causal forest method by applied researchers across 133 peer‐reviewed papers. It shows that the emerging best practice relies heavily on the approach and tools created by the original authors of the causal forest such as their grf package and the approaches given by them in examples. Generally, researchers use the causal forest on a relatively low‐dimensional dataset relying on observed controls or in some cases experiments to identify effects. There are several common ways to then communicate results–by mapping out the univariate distribution of individual‐level treatment effect estimates, displaying variable importance results for the forest and graphing the distribution of treatment effects across covariates that are important either for theoretical reasons or because they have high variable importance. Some deviations from this common practice are interesting and deserve further development and use. Others are unnecessary or even harmful. The paper concludes by reflecting on the emerging best practice for causal forest use and paths for future research.

Heterogenous long-term health and social outcomes of type 1 diabetes - A full population 30-year observational cohort study.

Type 1 diabetes (T1D) is known to have adverse long-term health and social outcomes, but the modifying factors are largely unknown. We investigate to what extent T1D outcomes are modified by area-, household-, and individual-level social and economic characteristics in Finland. National registers from 1987 to 2020 were used to identify all 3,048 children with T1D diagnosed at age seven to 17 and matched controls (n=78,883). Using causal forests, we estimated the average association between T1D and adult health, social, and economic outcomes at ages 28-30, and the modifying roles of more than 30 covariates. Individuals with T1D were more likely to be deceased (2.3% vs. 0.9% in the control group), to use antidepressants (17% vs. 13%), and to be unpartnered (36% vs. 32%), and had more months of unemployment (1.18 vs. 1.02) and lower annual income (25,697 euros vs. 27,453 euros), but not significantly lower educational attainment (10.8% vs. 10.3% with only basic education). T1D had a heterogenous association with all outcomes except mortality and income, but no specific population subgroup was vulnerable across all outcomes. However, women with T1D had particularly high rates of antidepressant use, and individuals from low socioeconomic families were more likely to be unpartnered.

Application of Causal Forest Model to Examine Treatment Effect Heterogeneity in Substance Use Disorder Psychosocial Treatments

Application of Causal Forest Model to Examine Treatment Effect Heterogeneity in Substance Use Disorder Psychosocial Treatments

Heterogeneity in the effect of green financing constraints on labor investment efficiency: A causal forest approach

Heterogeneity in the effect of green financing constraints on labor investment efficiency: A causal forest approach

Informing Risk Hotspots and Critical Mitigations for Rainstorms Using Machine Learning: Evidence from 268 Chinese Cities.

Climate change is exacerbating rainstorms, increasing the risk of flooding and threatening urban sustainability, which could undermine climate action. Here, we propose a machine learning-based framework to assess heterogeneous risks and identify critical mitigation measures for rainstorms across 268 Chinese cities. Nighttime light serves as a proxy for urban functionality, and meteorological, socio-economic, and infrastructural factors are incorporated to uncover underlying impact mechanisms. The Causal Forest (CF) model identifies 150 and 250 mm monthly rainstorm totals as critical thresholds, with significant negative impacts in the risk hotspots of eastern and north-central China. Additionally, Random Forest and SHAP (RF-SHAP) analysis highlight effective mitigation strategies, including well-developed drainage and bridges, expanded road networks, and sufficient dams. The Fixed Effects (FE) model reveals that the greatest negative impacts of rainstorms occur in spring, particularly in April, followed by autumn and winter for both the 50 and 150 mm thresholds. Our results demonstrate that the three models complement and validate each other, enhancing the reliability of the estimates. This novel framework leverages machine learning model to inform evidence-based mitigation, contributing to the achievement of Sustainable Development Goals 11 and 13─building resilient cities and combating climate change.

Personalized Fluid Management in Patients with Sepsis and AKI: A Policy Tree Approach.

Intravenous fluids are mainstay of management of acute kidney injury (AKI) after sepsis but can cause fluid overload. Recent literature shows that restrictive fluid strategy may be beneficial in some patients with AKI, however, identifying these patients is challenging. We aimed to develop and validate a machine learning algorithm to identify patients who would benefit from a restrictive fluid strategy. We included patients with sepsis who developed AKI within 48 hours of ICU admission and defined restrictive fluid strategy as receiving <500mL fluids within 24 hours after AKI. Our primary outcome was early AKI reversal at 24 hours of AKI onset, and secondary outcomes included sustained AKI reversal and major adverse kidney events by 30 days (MAKE30). We used a causal forest, a machine learning algorithm to estimate individual treatment effects and policy tree algorithm to identify patients who would benefit by restrictive fluid strategy. We developed the algorithm in MIMIC-IV and validated it in SICdb databases. Among 2,044 patients in the external validation cohort, policy tree recommended restrictive fluids for 66.7%. Among these, patients who received restrictive fluids demonstrated significantly higher rate of early AKI reversal (47.1% vs 31.7%,p=0.004), sustained AKI reversal (28.7% vs 17.5%, p=0.013) and lower rates of MAKE30 (23.0% vs 37.1%, p=0.011). These results were consistent in adjusted analysis. Policy tree based on causal machine learning can identify septic patients with AKI who benefit from a restrictive fluid strategy. This approach needs to be validated in prospective trials.