UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)
UvA-DARE (Digital Academic Repository)
The hare or the tortoise? Modeling optimal speed-accuracy tradeoff settings
van Ravenzwaaij, D.
Publication date
2012
Link to publication
Citation for published version (APA):
van Ravenzwaaij, D. (2012). The hare or the tortoise? Modeling optimal speed-accuracy
tradeoff settings.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.
Chapter
1
Introduction
In everyday life, we are constantly confronted with situations that require a quick and accurate action or decision. Examples include mundane tasks such as doing the dishes (we do not want to break china, but we also do not want to spend the next hour polishing), or vacuum cleaning (we like to get as many nooks and corners as possible, without it taking forever), but also more serious activities, such as typing a letter or performing a test. For all these actions, there exists a trade–off, such that more speed comes at the expense of more errors. This phenomenon is called the speed–accuracy trade–off (Schouten & Bekker, 1967; Wickelgren, 1977).
In situations where people are performing a test that is scored on some performance measure, the speed–accuracy trade–off may complicate interpretation. For instance, sup-pose both Jack and Jill take the same aptitude test. Jack scored 75% correct and took 45 minutes to complete the test. Jill scored 80% correct and took 60 minutes to complete the test. How can we decide which of the two performed better?
In experimental psychology it is common practice to study this speed–accuracy trade– off using relatively simple tasks. More often than not, the task requires participants to make a choice between one of two alternatives as quickly and accurately as possible. Notable examples include the lexical decision paradigm (Rubenstein, Garfield, & Millikan, 1970) in which the participant classifies letter strings as English words (e.g., GRAPE) or non–words (e.g., GERAP) and the moving dots task (e.g., Newsome, Britten, & Movshon, 1989; Gold & Shadlen, 2007) in which participants have to determine whether a cloud of partly coherently moving dots appears to move to the left or to the right. Typically, the observed variables from these and other two–alternative forced choice (2-AFC) tasks are distributions of response times (RTs) for correct and incorrect answers. Traditionally, conclusions are based on both the mean of the correct RTs and the percentage of correct responses. These measures, however, do not speak directly to underlying psychological processes, such as the rate of information processing, response caution, and time needed for stimulus encoding and non–decision processes (i.e., response execution). They also do not address the speed–accuracy trade–off.
Returning to the example of the test–performance of Jack and Jill, the fact that Jack was faster than Jill could have different reasons. Jack may process information faster (i.e., be better at completing the test items), he may be less cautious (i.e., not think too long about each item), or he may have a lower non–decision time (i.e., be faster at keying in the answers), or a combination of these factors.
1. Introduction
In order for a researcher on human cognition to draw conclusions about these unob-served processes, he or she could use a cognitive process model. An example of such a model is the drift diffusion model (DDM). Over the last three decades, the DDM has grown to be the most influential model for response times and has been very successful in explaining a range of psychological phenomena. Among others, the DDM has been suc-cessfully applied to experiments on perceptual discrimination, letter identification, lexical decision–making, recognition memory, implicit association forming, and signal detection (e.g., Klauer, Voss, Schmitz, & Teige-Mocigemba, 2007; Ratcliff, 1978; Ratcliff, Gomez, & McKoon, 2004; Ratcliff, Thapar, Gomez, & McKoon, 2004; van Ravenzwaaij, van der Maas, & Wagenmakers, 2011; Wagenmakers, Ratcliff, Gomez, & McKoon, 2008).
After applying the DDM to the test performance data of Jack and Jill, it may become clear that while Jack is faster than Jill, his advantage in speed is caused by lower response caution, whereas Jill is faster at processing information. Thus, application of the DDM allows us to clear the fog of the speed–accuracy trade–off.
The main claim that I try to drive home in this dissertation is that cognitive process models, such as the DDM, are a necessity in experimental psychology. This conviction is motivated by the fact that cognitive process models yield more informative conclusions, because they use the complete range of the behavioral data (instead of just the mean RT) and because they transform relatively uninformative raw data into meaningful psy-chological processes. This mapping of model parameters onto psypsy-chological processes has been validated countless times, a point to which I will return in the conclusion.
1.1
Chapter Outline
This dissertation is subdivided in two parts. Part I comprises chapters 2, 3, 4, and 5 and deals with theoretical issues concerning cognitive process models for decision making. Part II consists of chapters 6, 7, and 8, in which various applications of cognitive process models for decision making are presented. The outline of the chapters is as follows:
In chapter 2 I begin by introducing the DDM. The chapter gives an overview of a number of psychological phenomena in the research on intelligence. These phenomena include right–skewed RT distributions, the worst performance rule, more pronounced cor-relations between intelligence and RT standard deviations than between intelligence and RT means, an almost perfect linear relation between individual differences in RT stan-dard deviations and RT means, linear Brinley plots, and stronger correlations between intelligence and inspection time than between intelligence and RT mean. It is common in the field of research on IQ to treat all of these as separately occurring, IQ–specific phenomena. This chapter shows that all of these phenomena are in fact general mani-festations of a single DDM parameter: drift rate, or the speed of information processing parameter.
Chapter 3 builds a bridge between the DDM — a decision making model that produces optimal performance — and neural inhibition models — biologically plausible models that are based on the neurophysiological architecture of the brain. The chapter challenges the claim by Bogacz, Brown, Moehlis, Holmes, and Cohen (2006) that there is a simple one–on–one correspondence between the DDM and neural inhibition models.
In chapter 4 I examine the effect of response alternatives with unequal prior prob-abilities on perceptual decision making using the DDM. In terms of DDM parameters, such a biased decision can manifest itself as a shift in starting point and as a shift in drift rate criterion. Previous research asserts that bias should manifest itself as a shift in
1.1. Chapter Outline
starting point when the difficulty of each decision is fixed over trials (e.g., W. Edwards, 1965; Bogacz et al., 2006; Hanks, Mazurek, Kiani, Hopp, & Shadlen, 2011), whereas bias should additionally manifest itself as a shift in drift rate criterion when decision difficulty varies over trials (Yang et al., 2005; Hanks et al., 2011). In this chapter, we show that just a shift in starting point always leads to optimal performance, both for fixed and variable across–trial difficulty.
In chapter 5 three different implementations for estimating parameters of the DDM are introduced. These are the EZ model (Wagenmakers, van der Maas, & Grasman, 2007), fast–dm (Voss & Voss, 2007), and DMAT (Vandekerckhove & Tuerlinckx, 2007). The three implementations are compared in their ability to capture mean differences in experimental conditions and in their ability to capture individual differences in parameter estimates. The chapter also investigates how well each of the implementations is capable of recovering DDM parameters from datasets that were generated with two different decision models: the leaky competing accumulator model (Usher & McClelland, 2001) and the linear ballistic accumulator model (Brown & Heathcote, 2008).
In chapter 6, the effects of alcohol on perceptual decision making are investigated by means of the DDM. Specifically, the effects of both a moderate dose (blood alcohol content: 0.5g/l) and a high dose (blood alcohol content: 1g/l) on information processing, response caution, and non–decision time are examined. The chapter concludes that both information processing and non–decision time are negatively affected by alcohol. The detrimental effects on both of these components get progressively larger for higher alcohol doses.
Chapter 7 challenges the claim that the Implicit Association Test (or IAT; Greenwald, McGhee, & Schwartz, 1998) is a tool that assesses latent racial prejudice. Instead, I suggest an alternative explanation: the IAT effect is due to differences in in–group/out– group status. Using three different IATs that contrasted Dutch names, Moroccan names and Finnish names in all three combinations, it was concluded using a DDM analysis that in–group/out–group status is a better explanation of the IAT–effect than latent racial prejudice.
The 8th and final chapter examines a number of decision models that aim to quantify the psychological processes underlying performance in the Balloon Analogue Risk Task (or BART; Lejuez et al., 2002). Using a parameter recovery study, the chapter shows that a two–parameter version of the model originally proposed by Wallsten, Pleskac, and Lejuez (2005) outperforms two three–parameter versions and a four–parameter version. In the second part of the chapter, the model is applied to an empirical dataset in which participants performed the BART following various amounts of alcohol consumption.
