Modeling response times I am trying to model the time lapse between when a user sees an ad and when they call the advertiser (presuming they do).  I have two issues - part of the data seems exponential, but I am wondering if there is a similar distribution with an extra parameter because I cannot quite get it to fit.  Also, the peak does not occur at t = 0 but rather at t = 30s or so - obviously people do not call immediately, as there is a necessary physical delay between seeing the ad and picking up the phone.  But some people do call at 2 seconds, or 3, etc, and as I mentioned, the peak is at 30.  Is there a simple way to model this?
 A: The Weibull distribution [1]  is a generalization of the exponential distribution that is often used in time-to-event analysis. Also note that you can use non-parametric methods like Kaplan-Meier estimates that fit the distribution you actually have, instead of trying to fit into an arbitrary category.
[1] http://en.wikipedia.org/wiki/Weibull_distribution
A: not an expert, but maybe the ex-gaussian (gaussian plus exponential distribution)?
http://en.wikipedia.org/wiki/Gaussian_minus_exponential_distribution
http://rss.acs.unt.edu/Rdoc/library/gamlss.dist/html/exGAUS.html
http://www.tqmp.org/doc/vol4-1/p35-45_Lacouture.pdf
From the last link:

In the framework of cognitive
  processes, this convolution can be
  seen as representing the overall
  distribution of RT [Response Time] resulting from two
  additive or sequential processes. As
  proposed by Luce (1986, chap. 6), the
  exponential process can be seen as the
  decision component, i.e., the time
  required to decide which response to
  make, while the Gaussian component can
  be conceptualized as the transduction
  component, i.e., the sum of the time
  required by the sensory process and
  the time required to physically make
  the response.

