logistic.regression: Logistic and Auto-logistic regression

Description

Performs a logistic (binomial) or auto-logistic (spatially lagged binomial) regression using maximum likelihood or penalized maximum likelihood estimation.

Usage

logistic.regression(
  ldata,
  y,
  x,
  penalty = TRUE,
  autologistic = FALSE,
  coords = NULL,
  bw = NULL,
  type = "inverse",
  style = "W",
  longlat = FALSE,
  ...
)

Arguments

ldata

data.frame object containing variables

Dependent variable (y) in ldata

Independent variable(s) (x) in ldata

penalty

Apply regression penalty (TRUE/FALSE)

autologistic

Add auto-logistic term (TRUE/FALSE)

coords

Geographic coordinates for auto-logistic model matrix or sp object.

Distance bandwidth to calculate spatial lags (if empty neighbours result, need to increase bandwidth). If not provided it will be calculated automatically based on the minimum distance that includes at least one neighbor.

type

Neighbour weighting scheme (see autocov_dist)

style

Type of neighbour matrix (Wij), default is mean of neighbours

longlat

Are coordinates (coords) in geographic, lat/long (TRUE/FALSE)

...

Additional arguments passed to lrm

Value

A list class object with the following components:

model lrm model object (rms class)

bandwidth If AutoCov = TRUE returns the distance bandwidth used for the auto-covariance function

diagTable data.frame of regression diagnostics

coefTable data.frame of regression coefficients

Residuals data.frame of residuals and standardized residuals

AutoCov If an auto-logistic model, AutoCov represents lagged auto-covariance term

References

Besag, J.E., (1972) Nearest-neighbour systems and the auto-logistic model for binary data. Journal of the Royal Statistical Society, Series B Methodological 34:75-83

Dormann, C.F., (2007) Assessing the validity of autologistic regression. Ecological Modelling 207:234-242

Le Cessie, S., Van Houwelingen, J.C., (1992) Ridge estimators in logistic regression. Applied Statistics 41:191-201

Shao, J., (1993) Linear model selection by cross-validation. JASA 88:486-494

Examples

Run this code

# NOT RUN {
require(sp)
require(spdep)
require(rms)                                                                       
data(meuse)
  coordinates(meuse) <- ~x+y  
    meuse@data <- data.frame(DepVar=rbinom(dim(meuse)[1], 1, 0.5), meuse@data)

#### Logistic model
lmodel <- logistic.regression(meuse, y='DepVar', x=c('dist','cadmium','copper')) 
  lmodel$model
    lmodel$diagTable
      lmodel$coefTable

#### Logistic model with factorial variable
lmodel <- logistic.regression(meuse, y='DepVar', x=c('dist','cadmium','copper', 'soil')) 
  lmodel$model
    lmodel$diagTable
      lmodel$coefTable

### Auto-logistic model using 'autocov_dist' in 'spdep' package
lmodel <- logistic.regression(meuse, y='DepVar', x=c('dist','cadmium','copper'),
                              autologistic=TRUE, coords=coordinates(meuse), bw=5000) 
  lmodel$model
    lmodel$diagTable
      lmodel$coefTable
  est <- predict(lmodel$model, type='fitted.ind')

#### Add residuals, standardized residuals and estimated probabilities
VarNames <- rownames(lmodel$model$var)[-1]
  meuse@data$AutoCov <- lmodel$AutoCov
    meuse@data <- data.frame(meuse@data, Residual=lmodel$Residuals[,1], 
                             StdResid=lmodel$Residuals[,2], Probs=predict(lmodel$model, 
                             meuse@data[,VarNames],type='fitted') )  

#### Plot fit and probabilities
resid(lmodel$model, "partial", pl="loess") 
resid(lmodel$model, "partial", pl=TRUE)                  # plot residuals    
resid(lmodel$model, "gof")                               # global test of goodness of fit
lp1 <- resid(lmodel$model, "lp1")                        # Approx. leave-out linear predictors 
-2 * sum(meuse@data$DepVar * lp1 + log(1-plogis(lp1)))  # Approx leave-out-1 deviance
spplot(meuse, c('Probs'))                                # plot estimated probabilities at points


# }