Combinatorial search (all possible subset) for regression analysis

Share:

Description

Implements combinatorial (exhaustive) search algorithm, aka all-possible-subsets regression. As opposed to the sequential approach (stepwise, forward addition, or backward elimination) that has a potential bias resulting from considering only one variable for selection at a time, all possible combinations of the independent variables are examined, and sets satisfying designated conditions are returned.

Usage

1
2
ma.aps.reg(dv, iv, min=1, max, mad=FALSE, aic=FALSE, bic=FALSE,
           model.sig=TRUE, coeff.sig=TRUE, coeff.vif=TRUE, coeff.cor=FALSE)

Arguments

dv

Dependent variable (r by 1)

iv

Independent variable(s) (r by c)

min

Minimum number of independent variable to explore (>=1)

max

Maximum number of independent variable to explore (<=r/10)

mad

Returns mean absolute deviation when TRUE

aic

Returns Akaike's information criterion when TRUE

bic

Returns Bayesian information criterion when TRUE

model.sig

Returns models statistically significant only when TRUE

coeff.sig

Returns models with statistically significant coefficients only when TRUE

coeff.vif

Returns models with allowable level of multicollinearity only when TRUE

coeff.cor

Returns models without suppression effects only when TRUE

Author(s)

Dong-Joon Lim, PhD

References

Hair, Joseph F., et al. Multivariate data analysis. Vol. 7. Upper Saddle River, NJ: Pearson Prentice Hall, 2006.

Examples

1
2
3
4
5
6
7
8
9
# Load airplane dataset
df <- dataset.airplane.2017

# ready
dv <- subset(df, select = 2)
iv <- subset(df, select = 3 : 7)

# go
ma.aps.reg(dv, iv, 1, 3, mad = TRUE, coeff.cor = TRUE)