fitprob {qpcR}R Documentation

The (chi-square) fit probability

Description

Calculates the fit probability for objects of class drc, lm, glm, nls or any other models from which residuals and coef can be extracted.

Usage

fitprob(object)

Arguments

object a fitted model.

Details

Although the residual variance is a measure for the quality of a fit, it provides no information of how good a fit performs in relation to other fits in the (hypothetical) infinite population of assays performed under the same conditions. Under the assumption that the responses are approximately normally distributed and that the regression's expectation surface is approximately linear in the neighbourhood of the best fit, it can be shown that the residual sum-of-squares (RSS) obeys a chi^2 distribution with df = number of curve points - number of parameters. The p-value chi^2(RSS, df) can be viewed as the fraction of an infinite number of assays that, under the same conditions, would be expected to have a better curve fit, i.e. a smaller RSS.

Value

The p-value.

Author(s)

Andrej-Nikolai Spiess

References

Draper NR, Smith H.
Applied Regression Analysis, 3rd Ed.
Wiley, New York, 1998.

Examples

## 'good model'
m1 <- pcrfit(reps, 1, 2, l4)
fitprob(m1)

## 'bad model'
m2 <- pcrfit(reps, 1, 2, w4)
fitprob(m2)       

[Package qpcR version 1.2-4 Index]