Figure 2From: Bias magnification in ecologic studies: a methodological investigation Nonlinearity causes error during aggregation. The ecologic data point (X, Y) will generally not fall on the risk function when the latter is non-linear as shown here. The amount of error depends on the curvature of the risk function and the exposure distribution, but is bounded above by the line connecting the risks at the minimum and maximum exposures. This error can lead to pure specification bias.Back to article page