When the response for an individual unit (well, animal etc.) is a continuously measured value, such as optical density, the data are treated as quantitative.
The relationship between the mean response and the log dose is modelled using a mathematical formula, representing either a straight line (the linear model) or a sigmoid curve (the 4 parameter logistic or 5 parameter logistic models). The parameter values are estimated using the least squares method. (Note that any log base can be used for the dose, but the same base must be used when transforming back to obtain the relative potency – see below for details. QuBAS allows a log base of 2, e or 10.)
QuBAS assumptions
QuBAS assumes the following about dose groups and replicates
Replicates are independent. If they are not, then the responses should be averaged before importing to QuBAS.
Responses follow a normal distribution within each dose group.
The variance of the distributions for the dose groups is constant.
These assumptions are required for the calculations of confidence intervals and the p values resulting from statistical tests.
The variance of the raw response sometimes increases with the level of response. This can often be resolved by transforming the response to the log scale.
An alternative method for dealing with non-constant variance is to apply varying weights to the responses in the dose groups. QuBAS provides two options for the choice of weights, if required.
The following sections describe in detail the models available in QuBAS.
