I am not a user of ENA (until they make a version for Mac OS X I will not be) but I will have a go at this question. My guess is that the test for digit preference in ENA is based on some sort of hypothesis test. If I were doing this I would take the least significant digit of each measurement, produce a frequency table of these digits, and then do a chi-square goodness-of-fit to a uniform distribution using a p-value threshold of p < 0.05. With this approach I would expect one in twenty (0.05 = 5%) of tests to identify digit preference even when no such preference was present. With all the surveys being done we should expect a lot of false-positive results for the digit preference test. First thing is to "eyeball" the frequency table. If it looks pretty uniform then your OK. Mostly you will ant to look for two many zeros (from NNN.0) and fives (from NNN.5). What you need to do is to test the scales with some know weights for accuracy and reliability (using water is easy as 1000 ml = 1Kg). These this over the range of the survey data. If all is well then the digit preference probably arose by chance. If you let us know the manufacturer and model number of the scales being used then we can check the manufacturers quality control data (most will be audited by their local standards laboratory). I would be very surprised if there was a problem like this with clinical scales from a reputable manufacturer.