RTSYS: A DOS application for the analysis of reaction time data

Abstract

RTSYS is a menu-driven DOS application for the manipulation, analysis, and graphical display of reaction time data. It can be used either in a single-task environment under DOS,with access to a set operating system commands, or as an application under Windows. All functions have contextsensitive help. RTSYS fits the ex-Gaussian distribution to reaction time data without the difficulties usually associated with numerical parameter estimation. Distribution fitting and flexible censoring and rescaling options allow RTSYS to address the problems of reaction time distribution skew and outlying responses with reasonable sample sizes. RTSYS can automatically process multiple input files from experiments with arbitrary designs and produce formatted output of statistics for further processing by graphical and inferential statistical packages. The present article reviews and explains techniques used by RTSYS and provides an overview of the operation of the program.RTSYS is a DOS application, written in Turbo Pascal 6.0, that calculates distribution statistics for reaction time (RT) data. It also provides facilities for censoring and rescaling data. Statistics calculated include the number of RTs, percent of RTs censored, the median, mean, variance, and a nonparametric measure of skew [i.e., (mean -median)/ standard deviation)]. Because RT data are often scored as correct or incorrect, where incorrect RTs are relatively infrequent, percent error and mean error RT are also calculated. RTSYS fits the ex-Gaussian distribution and reports distribution parameters u, a, and !'.Fitting the ex-Gaussian also produces chi-squared and likelihood goodness-of-fit statistics. Goodness offit can be inspected visually by plotting RT histograms and superimposed fitted ex-Gaussian distributions.RTSYS will typically be used to convert raw data files collected from experiments into files of statistical parameters that are used as input to inferential and graphical applications. One ofRTSYS's most useful features is its ability to automatically process arbitrary factorial and nonfactorial between-and within-subject designs. Design cells can be collapsed both by mixing data from different factor levels or by Vincent averaging. Vincent averaging (usually performed over a subjects factor) approximately preserves distribution shape and is useful for obtaining ex-Gaussian fits where design cells contain too fewobservations for sufficiently precise parameter estimates. VinThe development of RTSYS began during the author's doctoral research, which was supported by a Canadian Commonwealth Fellowship. Thanks to Doug Mewhort and Richard Heath for comments on the first draft of this manuscript, to Trish Van Zandt, Steve Link, and an anonymous reviewer for their comments, and to the numerous people who have tested versions of RTSYS. Thanks also to Sarah Ogilvie for proofreading the manuscript and providing a quiet place in Melbourne to finish this project. Correspondence should be addressed to A. Heath-

PASTIS: A program for curve and distribution analyses

“…Since the likelihood statistic is asymptotic, it works better with a large number of observations. When analyzed using an ex-Gaussian distribution, results are exactly the same as those reported by Heathcote (1996).…”

“…Both files had 34 data points, the second having 10% of uniformly distributed random noise added to the data. For the exGaussian distribution, we used the same data set as that reported in Heathcote (1996), which contained only 10 data points. The parameters used for data generation and the results of the analysis can be seen in Table 3.…”

“…This distribution is known to provide a good fit for RT data in different tasks. The PDF is formulated as (Dawson, 1988;Heathcote, 1996) where r stands for the average of the exponential component, ).l and (1' are the mean and the standard deviation of the normal component, and <1> expresses the standard Gaussian integral (see Kennedy & Gentle, 1980, for a numerical approximation). The mean of such a distribution is u + r, while the standard deviation is ((1'2 + r 2 )' /2 .…”

Using response time distributions to examine top-down influences on attentional capture

“…Each subject's data in each cell of the cue type × cue validity design was fit to an ex-Gaussian function using RTSYS (Heathcote, 1996), which uses continuous maximum likelihood estimation to obtain ex-Gaussian parameters. All fits converged within 500 iterations.…”

The ex-Wald distribution as a descriptive model of response times

“…From a formal and computationalpoint of view, these models are often less demanding, and their application to data sets is a standard routine (cf. Cousineau & Larochelle, 1997;Dawson, 1988;Heathcote, 1996). A prominent example of this class of descriptive RT models is the "exGaussian" model, originally described by Hohle (1965).…”