Floating point error is far far smaller than measurement error and missing data.