Subject: Re: GLM fit or Cubic smoothing spline for categorical boundary data?? From: Mark Huckvale <m.huckvale@xxxxxxxx> Date: Fri, 4 May 2012 16:37:04 +0100 List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>Dear Pragati A disadvantage of splines is that the estimated boundary will be strongly affected by data points close to the boundary, and independent of the values at the extremes. Ideally you'd want to use all the data points, weighted by the number of trials at each point. Chapter and verse on how to fit PFs with maximum likelihoods can be found in Treutwin & Strasburger, "Fitting the psychometric function", Perception and Psychophysics, 1999 (61(1) 87-106. Mark Huckvale On 04/05/2012 12:13, Pragati Rao wrote: > Hi all, > > I have identification responses for stimuli along F2 and VOT continuums > from a group of subjects. I tried fitting glmfit() with logistic > regression in MATLAB because each subject had to choose either 'pa' or > 'ta' for F2 and 'ta' or 'da' for VOT task. However, in many subjects I > noticed that the curve leaves out many data points and gives absurd > values for categorical boundaries. So I tried fitting cubic spline using > csaps(). As expected the curve fits very well. I have attached png files > of results obtained through both techniques. Can we use cubic smoothing > spline on such a data set? > > I do not have a strong base in statistics. Any help will be greatly > appreciated. > > Regards, > Pragati Rao > Research Officer, > All India Institute of Speech and Hearing, > Mysore, India. > -- Mark Huckvale UCL Speech, Hearing & Phonetic Sciences www.phon.ucl.ac.uk/home/mark