There is an obvious floor-effect in my data: most of the participans achieved the lowest possible score (which is only -.74 SD from the mean score).
Is there a statistical criterion of when to speak of a floor-effect? I would be very thankful for any source.
Is the floor a problem? Many distributions have a lower bound (e.g., Poisson, Chi-Sq), so these can be used when modeling your data. Alternatively, there is lots of work, particularly in econometrics, using truncated and censored data.
I vaguely recall that what you're referring to iscalled "zero-bounding": there's a distribution, but most are at or near the zero point.
For some types of analysis you can add 1 to every score, or add the mean to every score. Then you may get a more normal distribution. Or you could try a transformation, or standardize the scores to obtain z-scores. I think whether to mention (or correct for) the floor effect depends on the statistical test you use.
- Badges
- Science topic