Compute accuracy estimates and maximum likelihood estimates of precision for the nonconstant bias measurement error model using original data.

1 |

`x` |
n (no. of items) x N (no. of methods) matrix or data.frame containing the measurements. N must be >= 3 and n > N. |

`beta` |
N vector of betas, either estimated by |

`M` |
Maximum no. of iterations for convergence. |

`conf.level` |
Chosen confidence level. |

Measurement Error Model:

x[i,k] = alpha[i] + beta[i]*mu[k] + epsilon[i,k]

where x[i,k] is the measurement by the ith method for the kth item, i = 1 to N, k = 1 to n, mu[k] is the true value for the kth item, epsilon[i,k] is the Normally distributed random error with variance sigma[i] squared for the ith method and the kth item, and alpha[i] and beta[i] are the accuracy parameters for the ith method. The product of the betas is constrained to equal one (equivalently, the geometric average of the beta's equals one).

The imprecision for the ith method is sigma[i]. If all alphas are zeroes and all betas are ones, there is no bias. If all betas equal 1, then there is a constant bias. If some of the betas differ from one there is a nonconstant bias. Note that the individual betas are not unique - only ratios of the betas are unique. If you divide all the betas by beta_i, then the betas represent the scale bias of the other devices/methods relative to device/method i. Also, when the betas differ from one, the sigmas are not directly comparable because the measurement scales (size of the units) differ. To make the sigmas comparable, divide them by their corresponding beta. This result is shown as bias.adj.sigma.

By using the original data values, the betas can be estimated and also the process variance.

Technically, the alphas and betas describe the measurements in terms of the unknown true values (i.e., the unknown true values can be thought of as a latent variable). The "true values" are ALWAYS unknown (unless you have a real, highly accurate reference method/device). The real goal is to calibrate one device/method in terms of another. This is easily accomplished because each measurement is a function of the same unknow true values. By solving the measurement error model (in expectation) for mu and substituting, any two devices/methods i=1 and i=2 can be be related as:

E[x[1,k]] = alpha[1] - alpha[2]*beta[1]/beta[2] + beta[1]/beta[2]*E[x[2,k]]

or equivalently

E[x[2,k]] = alpha[2] - alpha[1]*beta[2]/beta[1] + beta[2]/beta[1]*E[x[1,k]].

Use `cplot`

to display this calibration curve and the corresponding scale-bias adjusted imprecision standard deviations.

`conf.level ` |
Confidence level used. |

`sigma.table ` |
Table of accuracy and precision estimates and confidence intervals. |

`n.items ` |
No. of items. |

`N.methods ` |
No. of methods |

`sigma2 ` |
N vector of variances that measure the method imprecision. |

`alpha.cb ` |
N vector of estimated alphas for constant bias model. |

`alpha.ncb ` |
N vector of estimated alphas for nonconstant bias model. |

`beta ` |
N vector of estimated or hypothesized betas. |

`df ` |
N vector of estimated degrees of freedom. |

`lb ` |
N vector of lower bounds for confidence intervals. |

`ub ` |
N vector of upper bounds for confidence intervals. |

`bias.adj.sigma ` |
sigma adjusted for scale bias: sigma/beta. |

`H ` |
N+1 symmetric H matrix (see p. 201, Jaech). |

`errors.nb ` |
n x N matrix of estimated measurement errors for no bias model. |

`errors.cb ` |
n x N matrix of estimated measurement errors for constant bias model. |

`errors.ncb ` |
n x N matrix of estimated measurement errors for nonconstant bias model |

Richard A. Bilonick

Jaech, J. L. (1985) *Statistical Analysis of Measurement Errors*. New York: Wiley.

`cb.pd`

,`lrt`

1 2 3 |

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.