Where the Standard Errors for one or more categories of an Experiment or Ranking question are either missing, or, surprisingly large, this generally means that either:
- The category is not being estimated. Note that the Standard Error in an Experiment is always NaN for the first category, as its Coefficient is zero.
- There is a numerical precision problem being caused by the independent variable having an unacceptably large or small standard deviation. The solution to this is to rescale the variable so it has a standard deviation that is, say, between 0.1 and 10 (e.g., to 1).
- Multicollinearity. In a properly conducted experiment, it should not be possible to have multicollinearity, as the standard processes used to design experiments ensure that there is no multicollinearity. Consequently, when multicollinearity appears, it is a consequence of an error in the experimental design or the setup of the Experiment in Q. See Multicollinearity for a discussion of how to investigate the problem.