MCP Penalized Regression
Minimax Concave Penalty Penalized Regression · Also known as: MCP
MCP (Minimax Concave Penalty) is a variable selection method developed by Zhang (2010) that uses a concave penalty function for automated feature selection. Like SCAD, MCP addresses bias in lasso by avoiding shrinkage of large coefficients, but uses a different penalty shape that is computationally simpler than SCAD.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Apply MCP when you want automatic variable selection from high-dimensional data without lasso's coefficient shrinkage bias, when computational efficiency is important, or when you prefer a simpler penalty than SCAD. Ideal for prediction when the underlying sparse model is unknown.
Strengths & limitations
- Oracle efficiency: achieves oracle properties like SCAD with better computational stability
- Bias reduction: avoids over-shrinking large coefficients
- Simpler penalty: easier to optimize than SCAD
- Sparse solutions: automatically identifies important variables
- Computational efficiency: coordinate descent is faster than SCAD algorithms
- Tuning parameter selection: requires cross-validation to choose penalty parameters
- Concavity assumptions: assumes concave penalty is appropriate (may not be for all problems)
- Finite-sample behavior: oracle properties are asymptotic; finite-sample performance varies
Frequently asked
When should I use MCP vs. SCAD?
Both achieve oracle properties and reduce lasso bias. MCP is computationally simpler and often faster. SCAD has stronger theoretical guarantees in some settings. Start with MCP for efficiency; switch to SCAD if results are unsatisfactory.
How does MCP choose which variables to select?
MCP shrinks small coefficients to zero and keeps large ones relatively unbiased. Variables with estimated coefficients exactly at zero are excluded; others are retained. The threshold depends on the tuning parameter.
Can I use MCP for logistic regression?
Yes. MCP applies to any GLM by penalizing the likelihood. Logistic, Poisson, Cox, and other models can use MCP penalties.
How sensitive is MCP to the tuning parameter?
Very sensitive. Cross-validation is essential to choose the optimal tuning parameter. Poor tuning can lead to over-selection or under-selection of variables.
What is the relationship between MCP and lasso?
Both are penalized regression methods for variable selection. Lasso uses L1 penalty, MCP uses concave penalty. MCP avoids lasso's bias for large coefficients while maintaining sparsity.
Sources
- Zhang, C. H. (2010). Nearly unbiased variable selection under minimax concave penalty. Annals of Statistics, 38(2), 894-942. DOI: 10.1214/09-AOS729 ↗
- Breheny, P., & Huang, J. (2011). Coordinate descent algorithms for nonconvex penalized regression. Annals of Applied Statistics, 5(1), 232-253. link ↗
- Zhang, C. H., & Zhang, T. (2012). A general theory of concave regularized M-estimators. Statistical Science, 27(4), 506-537. link ↗
How to cite this page
ScholarGate. (2026, June 3). Minimax Concave Penalty Penalized Regression. ScholarGate. https://scholargate.app/en/psychometrics/mcp-penalized-regression
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Exploratory Structural Equation ModelingPsychometrics↔ compare
- Partial Least Squares Structural Equation ModelingPsychometrics↔ compare
- Redundancy AnalysisPsychometrics↔ compare
- SCAD Penalized RegressionPsychometrics↔ compare