Are the standard deviation estimates from quest meant to be interpreted alone (e.g. exp(QuestSD(q))]? Or are they only meant to be interpreted when added to the estimated threshold and then converted to non-logarithmic values[e.g. exp(QuestMean(q)+QuestSD(q))]. In a recent beta test of a study the standard deviation values interpreted alone are nearly 30x the estimate and outside the range of possible stimulus values, leading me to believe that they are only to be interpreted when added to the log values and only then can be converted using the exponential function. Am I correct in my understanding or are the standard deviations for this recent beta test really indicating that the standard deviation is 30x the estimate?
Your interpretation is correct, but it can be said more simply.
Quest provides a mean estimate of the threshold value of the log intensity. QuestSD is the estimated standard deviation of that estimate.
You are welcome to convert things from log to linear, but it would be a mistake to think that exp of a standard deviation of a log quantity equals the standard deviation of the plain quantity. The appropriate conversion, as you guessed, is to compute the endpoints of a confidence interval for the log quantity, and then convert each point to a plain quantity.
Hope that helps.
Denis
Denis Pelli
Professor of Psychology and Neural Science at New York University
On Tue, Oct 20, 2015 at 2:19 AM, acidnynex@... [PSYCHTOOLBOX] <PSYCHTOOLBOX@yahoogroups.com> wrote:Are the standard deviation estimates from quest meant to be interpreted alone (e.g. exp(QuestSD(q))]? Or are they only meant to be interpreted when added to the estimated threshold and then converted to non-logarithmic values[e.g. exp(QuestMean(q)+QuestSD(q))]. In a recent beta test of a study the standard deviation values interpreted alone are nearly 30x the estimate and outside the range of possible stimulus values, leading me to believe that they are only to be interpreted when added to the log values and only then can be converted using the exponential function. Am I correct in my understanding or are the standard deviations for this recent beta test really indicating that the standard deviation is 30x the estimate?