Tuberculosis remains a major health challenge in Indonesia and globally. Governments and researchers rely on statistical models to identify the key drivers behind infection rates. However, real-world health data often contain overlapping variables—such as population size, density, and infrastructure—that are closely related. This creates a statistical problem known as multicollinearity, which can distort results and reduce the reliability of predictions.
Why This Research Matters
The research team from the University of Lampung focuses on a practical issue: how to generate accurate models when variables are highly correlated. Traditional methods, such as the Inverse Gaussian Maximum Likelihood (IGML), tend to become unstable in such conditions. This instability leads to higher prediction errors, making it harder for policymakers to trust the results.
To address this, the researchers tested an alternative method, the Inverse Gaussian Hybrid Estimator (IGH), designed to stabilize calculations by reducing excessive variability in model parameters.
Simple Approach to a Complex Problem
The study analyzed secondary data from 28 districts and cities in West Java between 2022 and 2024. The dataset included:
- Number of Tuberculosis cases
- Population density
- Total population
- Land area
- Number of health facilities
- Percentage of poor population
- Population growth rate
Instead of relying solely on traditional statistical assumptions, the researchers used a model suited for uneven data distributions—common in epidemiological studies where case numbers vary widely across regions.
The analysis process included:
- Checking data distribution patterns
- Estimating model parameters using IGML
- Detecting multicollinearity using correlation and VIF (Variance Inflation Factor)
- Applying the IGH method to correct instability
- Comparing model performance using error metrics
Key Findings: IGH Outperforms Conventional Method
The results clearly show that IGH provides more accurate and stable estimates than IGML.
Main findings include:
- Severe multicollinearity was detected, with VIF values exceeding 90 for some variables
- IGH significantly reduced coefficient instability by applying a “shrinkage” technique
- The IGH model produced lower prediction errors
Performance comparison:
- IGML Mean Squared Error (MSE): 1,267,407,893
- IGH Mean Squared Error (MSE): 1,254,997,456
- IGML RMSE: 35,600.67
- IGH RMSE: 35,425.94
Lower error values indicate better model performance, confirming IGH as the superior method in this case.
The study also identified which factors significantly influence TB cases. The most impactful variables were:
- Population density
- Total population
- Number of health facilities
Other variables, such as land area, poverty rate, and population growth, did not show statistically significant effects in this model.
Real-World Impact for Public Health Policy
The implications of this research extend beyond academic statistics. By improving the accuracy of TB modeling, policymakers can make more informed decisions about where to allocate resources.
For example:
- High-density areas can be prioritized for TB prevention programs
- Investment in healthcare infrastructure can be targeted more effectively
- Data-driven strategies can replace assumptions in public health planning
Misgiyati from the University of Lampung emphasizes that the IGH method produces “more stable and accurate estimates in modeling the number of Tuberculosis cases,” especially when dealing with highly correlated variables . This stability is crucial for ensuring that policy decisions are based on reliable evidence.
Beyond healthcare, the IGH approach can also be applied in economics, education, and social sciences—any field where complex, interrelated data is common.
Author Profiles
Gracia Trifena Sintauli is a researcher specializing in applied statistics and data modeling.
Netti Herawati is an academic focusing on statistical analysis and quantitative research methods.
Misgiyati, M.Si., is a lecturer at the Faculty of Mathematics and Natural Sciences, University of Lampung, with expertise in inferential statistics and regression modeling.
Nusyirwan is a researcher in statistical modeling and data analysis.
All authors are affiliated with the University of Lampung and actively contribute to research in statistical methods for real-world applications.
Source
Sintauli, G. T., Herawati, N., Misgiyati, & Nusyirwan. (2026). Application of The Inverse Gaussian Hybrid Estimator (IGH) To Address Multicollinearity in The Number of Tuberculosis Cases. International Journal of Sustainable Applied Sciences (IJSAS), Vol. 4, No. 4, 261–270.
DOI: https://doi.org/10.59890/ijsas.v4i4.411
URL: https://dmimultitechpublisher.my.id/index.php/ijsas
0 Komentar