Concept Architecture
Concept
Theoretically, Condition Index is a diagnostic measure used to assess the severity of multicollinearity among predictor variables in regression models. It is founded on matrix algebra and the spectral decomposition of the predictor correlation or cross-product matrix. The condition index exists to identify near-linear dependencies among explanatory variables that may inflate the variance of regression coefficient estimates, reduce estimation stability and impair statistical inference.
Mathematically, the condition index is derived from the eigenvalues of the predictor matrix. It is calculated as the square root of the ratio of the largest eigenvalue to each individual eigenvalue. Large condition indices indicate that one or more dimensions of the predictor space are nearly linearly dependent. Interpretation is strengthened by examining variance decomposition proportions, which identify the regression coefficients associated with each condition index.
In practice, condition indices are calculated after fitting a regression model as part of multicollinearity diagnostics. They are interpreted alongside variance inflation factors (VIFs), tolerance statistics and variance decomposition proportions. In health economics, condition indices are used when estimating cost functions, resource utilisation models, survival models and health outcome regressions to ensure that correlated predictors do not compromise model estimation.
Purpose
Used to diagnose multicollinearity by quantifying the degree of linear dependence among predictor variables, supporting reliable regression estimation and model interpretation.
Mathematical Formulae
Primary Formula
Condition Index for dimension i:
CI? = �(?max / ??)
where:
- ?max = largest eigenvalue of the predictor matrix
- ?? = eigenvalue for dimension i
Supporting Formulae
Condition number:
? = �(?max / ?min)
Variance decomposition proportion:
VDP?? = �??� / ?? � ??(�??� / ??)
where �?? is the eigenvector element corresponding to predictor j and dimension i.
Typical interpretation:
- CI < 10: little multicollinearity
- 10 � CI < 30: moderate multicollinearity
- CI � 30: potentially serious multicollinearity
Related Mathematical Methods
- Eigenvalue Analysis
- Spectral Decomposition
- Variance Inflation Factor
- Tolerance Statistic
- Variance Decomposition Proportions
- Multiple Linear Regression
- Principal Component Analysis
Example
A regression model predicting annual healthcare expenditure contains four predictors. Eigenvalue decomposition of the predictor correlation matrix produces:
? = (4.00, 0.80, 0.18, 0.02)
The largest eigenvalue is:
?max = 4.00
The largest condition index is:
CI = �(4.00 / 0.02)
CI = �200
CI = 14.14
A condition index of 14.14 indicates moderate multicollinearity. If several regression coefficients also exhibit large variance decomposition proportions on this dimension, further investigation of predictor redundancy is warranted.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| SQRT | =SQRT($B$2/B5) | Calculates the condition index from eigenvalues. |
| MAX | =MAX(B2:B5) | Identifies the largest eigenvalue. |
| MIN | =MIN(B2:B5) | Calculates the condition number using the smallest eigenvalue. |
| IF | =IF(C5>=30,"Serious","Acceptable") | Flags potentially problematic multicollinearity. |
VBA (Optional)
Automate eigenvalue extraction, condition index calculation and multicollinearity reporting for regression models, highlighting dimensions with high condition indices and associated variance decomposition proportions.
Sources
- Belsley DA, Kuh E, Welsch RE. Regression Diagnostics: Identifying Influential Data and Sources of Collinearity.
- Montgomery DC, Peck EA, Vining GG. Introduction to Linear Regression Analysis.
- Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied Linear Statistical Models.
- Draper NR, Smith H. Applied Regression Analysis.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
Related Concepts (5)
Library
Publications
1
Statistical Analysis of Cost-Effectiveness Data — Willan & Briggs, 1st Edition ed., 2006 (John Wiley & Sons)
A synthesis of statistical methods for analysing cost-effectiveness data, including net-benefit regression, confidence intervals for the ICER, cost-effectiveness acceptability curves, and covariate adjustment. Part of the Wiley Statistics in Practice series.
BookView source →
Frequently Asked Questions (6)
What is a condition index?
A diagnostic detecting problematic collinearity among predictor variables, calculated from the eigenvalues of the predictor matrix, larger values meaning worse collinearity.
Source: Belsley, Kuh & Welsch 1980
How does a condition index flag collinearity among predictors?
A condition index flags collinearity by drawing on the eigenvalues of the predictor matrix, with a large index signalling that the predictors are nearly redundant with one another. When variables carry overlapping information, some eigenvalues shrink toward zero, and the ratio that forms the index grows, so a high value warns that the regression struggles to disentangle their effects. Paired with variance decomposition, it can even point to which variables are entangled. Quantifying how near-collinear the predictors are is its role. Kirkwood and Sterne (2003) describe this diagnostic.
Source: Kirkwood & Sterne 2003
How is a condition index calculated?
A condition index is calculated from the eigenvalues of the scaled and, usually, centred matrix of predictors: for each eigenvalue, the condition index is the square root of the ratio of the largest eigenvalue to that eigenvalue. The largest condition index, corresponding to the smallest eigenvalue, is the condition number of the matrix. So a condition index is derived from the eigenvalues of the predictor matrix, with small eigenvalues, indicating near-dependencies, producing large condition indices, which is why examining the condition indices reveals how many and how severe the collinear relationships among the predictors are.
Source: Belsley, Kuh & Welsch 1980
How is a condition index interpreted?
A condition index is interpreted by its size, with larger values indicating stronger collinearity; rough guidelines suggest that indices around thirty or above signal moderate to strong collinearity worth investigating, though such thresholds are conventions rather than fixed rules. Large indices are examined together with variance decomposition proportions to see which predictors are involved. So a condition index is interpreted as a measure of collinearity severity, with high values flagging problematic near-dependencies among predictors, and it is used alongside the variance decomposition proportions to identify which variables contribute to each collinear relationship, giving a fuller diagnosis than the index alone.
Source: Belsley, Kuh & Welsch 1980
How does a condition index relate to collinearity?
A condition index relates to collinearity as a measure of its presence and severity: collinearity means predictors are highly correlated or nearly linearly dependent, which produces small eigenvalues in the predictor matrix and correspondingly large condition indices. So the condition index detects collinearity through the eigenvalue structure, with large indices indicating the near-dependencies that define collinearity, and it complements other diagnostics such as the variance inflation factor by identifying the number and strength of collinear relationships, which helps determine whether collinearity is severe enough to threaten the reliability of the individual coefficient estimates.
Source: Belsley, Kuh & Welsch 1980
How is a condition index used with variance decomposition?
A condition index is used together with variance decomposition proportions, which show how much of each coefficient's variance is associated with each dimension: a large condition index identifies a near-dependency, and the variance decomposition proportions reveal which predictors have a high share of their variance tied to that dimension, thus which variables are involved in the collinearity. So a condition index and the variance decomposition proportions are used jointly to diagnose collinearity, the index flagging its presence and severity and the proportions identifying the predictors responsible, which together pinpoint the collinear relationships more precisely than either does alone.
Source: Belsley, Kuh & Welsch 1980
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 12 Dec 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/condition-index
- Term code
- HE-ES-SA-032
Stable URI · Machine-readable · Resolvable · CC BY 4.0