# Re: GLM fit or Cubic smoothing spline for categorical boundary data?? (Stuart Rosen )

```Subject: Re: GLM fit or Cubic smoothing spline for categorical boundary data??
From:    Stuart Rosen  <s.rosen@xxxxxxxx>
Date:    Mon, 7 May 2012 11:31:38 +0100
List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>

<html>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
<body bgcolor="#FFFFFF" text="#000000">
You can't make sensible statistical inferences from a least-squares
fit. A proper statistical approach (i.e., using maximum likelihood
as in logistic regression) would enable you to to answer questions
like how many parameters are necessary for an adequate description
of the data. <br>
<br>
Yours - Stuart<br>
<br>
On 07/05/2012 11:23, Pragati Rao wrote:
<blockquote
cite="mid:10401_1336386215_4FA7A2A7_10401_136_1_CAKL8Na=Oi1XKhvr4-dfBjJi7fU-TGE4pYTnjpEJddNcxhSH08w@xxxxxxxx"
type="cite">
<meta http-equiv="Content-Type" content="text/html;
charset=ISO-8859-1">
Hi everyone,<br>
<br>
Thank you for the helpful replies. Based on some of the
suggestions I tried a two parameter logistic curve fit using
lsqcurvefit(). The equation used was y(t)=1/(1+exp(-r(t-t0))). The
results obtained for the same data is attached. I have a few more
questions:<br>
<br>
1. Will the 4 parameter fit be better? And should I use&nbsp; y(t)=
k1/(1+exp(-r(t-t0)))+k2 ? <br>
<br>
2. Trueutwein and Strasberger (1999) suggest that maximum
likelihood is better for fitting psychometric function data. Has
anyone found results from a maximum likelihood fit better than a
least squares fit? <br>
<br>
Any opinions on this?<br>
<br>
Regards,<br>
Pragati<br>
</blockquote>
</body>
</html>
```

This message came from the mail archive
/var/www/postings/2012/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University