Skip to contents

Fit a Calibration Curve

Source: R/calibration-fit.R
measure_calibration_fit.Rd

Fits a weighted or unweighted calibration curve for quantitation. Supports linear and quadratic models with various weighting schemes.

Usage

measure_calibration_fit(
  data,
  formula,
  model = c("linear", "quadratic"),
  weights = c("none", "1/x", "1/x2", "1/y", "1/y2"),
  origin = FALSE,
  outlier_method = c("none", "studentized", "cook"),
  outlier_threshold = NULL,
  outlier_action = c("flag", "remove"),
  sample_type_col = NULL
)

Arguments

data

A data frame containing calibration data.

formula

A formula specifying the model. The left-hand side should be the response variable, and the right-hand side should be the concentration variable (e.g., response ~ nominal_conc).

model

Model type: "linear" (default) or "quadratic".

weights

Weighting scheme:

  • "none" (default): Unweighted regression

  • "1/x": Weight by 1/concentration

  • "1/x2": Weight by 1/concentration^2

  • "1/y": Weight by 1/response

  • "1/y2": Weight by 1/response^2

  • A numeric vector of custom weights (must match data rows)

origin

Logical. If TRUE, force the curve through the origin (zero intercept). Default is FALSE.

outlier_method

Method for flagging outliers:

  • "none" (default): No outlier detection

  • "studentized": Flag points with |studentized residual| > outlier_threshold

  • "cook": Flag points with Cook's distance > outlier_threshold

outlier_threshold

Threshold for outlier detection. Default is 2.5 for studentized residuals or 1 for Cook's distance.

outlier_action

What to do with outliers:

  • "flag" (default): Flag but include in fit

  • "remove": Remove from fit (with audit trail)

sample_type_col

Optional column name for sample type. If provided, only rows with sample_type == "standard" are used for fitting.

Value

A measure_calibration object containing the fitted model, diagnostics, and metadata.

Details

Weighting

Weighting is essential when response variance changes with concentration (heteroscedasticity). Common patterns:

  • Constant CV: Use "1/x2" or "1/y2"

  • Constant absolute error: Use "none"

  • Proportional error: Use "1/x" or "1/y"

Outlier Handling

By default, outliers are flagged but NOT removed. This follows the principle of "flag, don't drop" for analytical data. If removal is enabled, the removed points are stored in the result for audit purposes.

See also

measure_calibration_predict() for prediction, autoplot.measure_calibration() for diagnostic plots, tidy.measure_calibration() for extracting coefficients.

Examples

# Simple linear calibration
data <- data.frame(
  nominal_conc = c(0, 10, 25, 50, 100, 200),
  response = c(0.5, 15.2, 35.8, 72.1, 148.3, 295.7)
)
cal <- measure_calibration_fit(data, response ~ nominal_conc)
print(cal)
#> <measure_calibration>
#>   Model: linear
#>   Weighting: none
#>   Formula: response ~ nominal_conc
#>   N points: 6
#>   R²: 0.99992

# Weighted calibration (1/x^2)
cal_weighted <- measure_calibration_fit(
  data,
  response ~ nominal_conc,
  weights = "1/x2"
)

# Quadratic model
cal_quad <- measure_calibration_fit(
  data,
  response ~ nominal_conc,
  model = "quadratic"
)