So, in absence of an answer I made the following realisation, which may be useful to someone else. Please tell me however if this is not correct.

The maximum measurable Weber function using the 75th percentile and 50% PSE is 100% as JND = the reference stimulus.

This means that if a Weber function cannot be determined by a plot not crossing the 75% mark, then one can assume that the Weber value is at least 100%. As such any stimulus would need to increase in magnitude 100% to be noticed by the participant.

This at least allows for *some* numerical comparison with other data.

Hi

A usual good practice is to do some pilots to make sure that the stimulus range will elicit saturated responses at both ends. Anyway in your case, if this is not possible at this stage, you should be able to obtain confidence intervals of the fitted parameters. If the parameters have low variability you could safely use the theoretical value predicted by the curve at 75%. I am assuming you have a yes/no paradigm (0-100% curve). If you used 2-AFC, you might want to increase the value up to 1 SD.

Joan

You should be able to compute JND if you fit the proportion of responses over stimulus magnitude and extrapolate 75% even if the JND is outside the range. I normally use psignifit (look on bootstrap-software.org for the old version and on github for version 3) which also gives you confidence intervals estimates.

There's nothing magical about the PSE -- other criteria are possible, too. Theoretically, any two values suffice to completely specify the psychometric function, even if they are below the 75% mark.

Addendum to my last answer: One point does not suffice to specify both position and steepness of the psychometric function. But it would be possible to specify two points, say, 60 and 70 %, for all participants and then analyze their spread and position along the x-axis.