Skip to contents

Summarize, predict, and plot diagnostic accuracy

Usage

# S3 method for class 'mars_dta'
predict(
  object,
  newdata = NULL,
  level = 0.95,
  interval = c("confidence", "prediction"),
  ...
)

# S3 method for class 'mars_dta'
coef(object, ...)

# S3 method for class 'mars_dta'
vcov(object, ...)

# S3 method for class 'mars_dta'
summary(object, newdata = NULL, level = 0.95, ...)

# S3 method for class 'mars_dta'
print(x, ...)

# S3 method for class 'summary.mars_dta'
print(x, ...)

# S3 method for class 'mars_dta'
plot(
  x,
  newdata = NULL,
  level = 0.95,
  ...,
  xlim = c(0, 1),
  ylim = c(0, 1),
  xlab = "False-positive rate",
  ylab = "Sensitivity",
  sroc = FALSE
)

Arguments

object, x

A fitted mars_dta() object or, for printing a summary, the result of summary().

newdata

Data frame with one row per moderator profile. Required for prediction and plotting when the fit has moderators. A summary without profiles reports the fitted coefficients.

level

Confidence or prediction level between zero and one.

interval

Use "confidence" for the fitted mean logits or "prediction" for the true logits in a new study.

...

For plot(), graphical arguments passed to graphics::plot(). Unused by the other methods.

xlim, ylim, xlab, ylab

Axis limits and labels for the ROC plot.

sroc

Logical. Overlay the dta_sroc() curve for the plotted moderator profile.

Value

predict() returns a data frame with two rows per profile. summary() returns study counts, coefficients, pooled accuracy when profiles are available, and heterogeneity and convergence diagnostics. plot() invisibly returns the summary point and region coordinates, plus the curve coordinates when sroc = TRUE. coef() and vcov() return the logit-scale coefficients and their estimated covariance matrix.

Details

The reported operating point is the inverse logit of the fitted mean for each endpoint. Intervals are Wald intervals computed on the logit scale and back-transformed. Prediction intervals add the estimated between-study variance and describe true study accuracy. The plot takes one moderator profile and shows raw study estimates with joint confidence and prediction regions formed on the logit scale and transformed onto ROC axes. For binomial fits, the coefficient covariance comes from the observed information of the integrated likelihood. If that information cannot be inverted, interval limits are missing and the ROC plot omits the regions.