Perceptual vs. preferential decision making:
Same models, same process?
February 2016 Suzanna van Baardewijk, 10174443 Under the supervision of Dr. Leendert van Maanen University of Amsterdam Abstract Decision making is a large part of our lives and can tell us about the cognitive processes underlying our behaviour. Two current research directions, perceptual decision making and preferential decision making, have different theoretical and experimental backgrounds, but the same type of models can be used to explain their data. This suggests that one cognitive process is giving rise to both behaviours. To find evidence for such a common underlying decision mechanism, results from both fields are compared; looking at manipulations that have influenced the Diffusion Decision Model parameters drift rate, boundary separation, starting point and non-decision time. However, an insufficient number of preferential decision making studies that used this model have been found. Overall, more research is needed to come to a unified framework of decision making, but a specific suggestion is to be more transparent about the huge variety in terminology.1. Introduction Everyday life is full of decisions. From major life decisions like where to live, to trivial things like deciding if the traffic light is red or green, there are many different aspects to decisions and many different ideas have been proposed about how we solve them. To find some order in the chaos, different scientific domains work on forming theories about different aspects and types of decision making. Experimental decision making tasks are used to test these different theories and learn about the underlying processes of decision making. In current decision making research, two different research directions can be distinguished, perceptual decision making and preferential decision making. The first comes from a history of psychophysics research, where participants have to make decisions about stimulus properties, which can be correct or incorrect. The latter originates from experimental economics, with the participant free to decide which stimulus to choose, usually asking for a risk assessment of two gambling options or a food choice. Generally, preferential decisions are considered of a higher cognitive level than perceptual decisions. Although these two types of decision making research differ in approach and response criterion, both show probabilistic choice behaviour, which supports a conceptualization of decisions where a stochastic accumulation of evidence determines the outcome, essentially random but weighed by relative measures. In this popular cognitive theory of decision making, response alternatives are represented by abstract values that can reach a threshold value at which a response is initiated. Variations of this process are formalised in sequential sampling models, which have indeed been able to capture both types of decision making. This has prompted some to suggest that the same basic mechanism could underlie both perceptual and preferential decision making, and efforts have been made to unify the two types of decision making in experimental designs. (Dutilh & Rieskamp, 2015, Krajbich et al., 2015, Gold & Shadlen, 2007). So, can we argue that these two types of decision making tasks engage the same cognitive process? In this review, we will evaluate this by looking at results from the modelling studies that capture both processes, and see if they capture them in the same way. We have limited our search to the Diffusion Decision Model (DDM, Ratcliff, 1978, Ratcliff & McKoon, 2008), a canonical cognitive model, as we expect to find results for both perceptual and preferential decision making, and which would allow for a clear comparison, as model parameters capture different concepts in different models. The DDM has been found to explain a wide range of decision making behaviours, and has been widely applied since 1978. The model states that when performing a two-alternative decision, the change in activation of the accumulator in a trial is described as follows: 𝑑𝑥 = 𝐴𝑑𝑡 + 𝑐𝑑𝑊, 𝑥 0 = 0 where x is the activation of the accumulator, A is the average increase of the activation over time t, also known as drift rate, and cdW is Gaussian distributed noise with mean 0 and standard deviation c. The accumulator starts at the starting point x(0) = 0. It subsequently increases or decreases with evidence accumulation until it reaches the upper or lower bound. Next to the decision we identify a non-decision time Ter, where the perception of the stimulus and motor execution are processed. Additionally, a Gaussian distributed drift rate variability and a uniform starting-point variability have been added to the model. Thus, this entire process is captured in these 7 parameters: drift rate, drift rate variability, noise in the diffusion process, starting point, starting-point variability, boundary separation and non-decision time, see also Figure 1.
The DDM takes two outcome measures of experimental tasks together: response times and accuracy. As participants have to make a trade-off between performing fast and accurate while performing a task, comparison can be difficult between (groups of) participants where each scores better on another measure. By capturing both measures in drift rate, DDM makes such a comparison more clear, and has proven very accurate in predicting choice behaviour (Krajbich et al., 2015). Here, we will discuss for some important model parameters, namely drift rate, boundary separation, starting point and non-decision time, which manipulations have been shown to affect them, from both perceptual decision making research and preferential decision making research. Similarities between the manipulations would point to a similar latent process. All parameters will be discussed independently from each other, so readers interested in specific parameters only have to read the concerning paragraphs. 2. Drift rate 2.1 Perceptual decision making In perceptual decision making, drift rate is associated with difficulty. When difficulty is manipulated in a random dot motion paradigm, by decreasing coherence of the moving dots, or in a letter identification paradigm, by shortening the duration the stimulus is visible before masking, drift rate increases (Ratcliff & McKoon, 2008, Ratcliff & Rouder, 2000). Also, words with a natural higher frequency in participants’ native language engaged higher drift rates in a lexical decision task (Wagenmakers et al. 20081). Furthermore, drift rate increased with practice on a lexical decision task and decreased with alcohol intake and age (Dutilh et al., 2009, van Ravenzwaaij et al., 2012, Thapar et al., 2003). Lastly, in an in-group vs. out-group implicit association test, information was processed faster in the compatible block than in the incompatible block (van Ravenzwaaij et al., 20111). 2.2 Preferential decision making
1 In these studies, the boundaries of the DDM were based on response categories instead of the usual ‘correct’ and ‘incorrect’, because each category was correct in one condition and incorrect Figure 1. Illustration of the Diffusion Decision Model. A few possible accumulation traces accumulate over time from the starting point, z, to one of the two boundaries, 0 and a. The drift rate is calculated as an average of the accumulation trace (Ratcliff & McKoon, 2008). Apart from this process, non-decision time is added to the RT.
876 R. Ratcliff and G. McKoon
0 z a correct correct error Correct RT distribution Error RT distribution Drift Rate v response response response X X Y Z Time Time A decision boundary B decision boundary Time Total RT=u+d+w u d range=st mean=Ter w decision
encoding etc. response output Nondecision components of RT=u+w High Drift Low Drift
Figure 1: The diffusion decision model. (Top panel) Three simulated paths with drift rate v, boundary separation a, and starting point z. (Middle panel) Fast and slow processes from each of two drift rates to illustrate how an equal size slowdown in drift rate (X) produces a small shift in the leading edge of the RT distribution (Y) and a larger shift in the tail (Z). (Bottom panel) Encoding time (u), decision time (d), and response output (w) time. The nondecision component is the sum of u and w with mean = Ter and with variability represented by a uniform distribution with range st.
one of the boundaries is reached, a response is initiated. The rate of accumu-lation of information is called the drift rate (v), and it is determined by the quality of the information extracted from the stimulus. In an experiment, the value of drift rate, v, would be different for each stimulus condition that differed in difficulty. For recognition memory, for example, drift rate would represent the quality of the match between a test word and memory. A word presented for study three times would have a higher degree of match (i.e., a higher drift rate) than a word presented once. The zero point of drift rate (the drift criterion, Ratcliff, 1985, 2002; Ratcliff et al., 1999) divides drift
In preferential decision making, drift rate is also associated with difficulty. In a study where participants had to select one out of two stimuli that represented a monetary gain or loss to them, a bigger gain difference between the stimuli engaged a higher drift rate (Basten et al., 2010). However, can this study really be counted as preferential decision making? It could be argued that participants were probably not responding to their preference, but to the external criterion of the stimulus with the highest monetary gain. This leaves us unable to compare drift rate across the two research domains. 3. Boundary separation 3.1 Perceptual decision making In perceptual decision making, boundary separation is associated with response caution. It is usually manipulated by giving speed or accuracy instructions on a task. With accuracy instructions, boundary separation was larger in a random dot motion task and a lexical decision task (Ratcliff & McKoon, 2008, Wagenmakers et al. 20081). This is thought to reflect caution, as taking more time is a good way to ensure correct responses. Boundary separation is also affected by practice and age, with practice on a lexical decision task decreasing caution and age increasing caution (Dutilh et al., 2009, Thapar et al., 2003). 3.2 Preferential decision making Stressing for speed or accuracy is a manipulation also often used in preferential decision making. In one study, the boundary separation was bigger under high speed stress in simple food choices (Milosavljevic et al., 2010). This shows that the response caution explanation for boundary separation directly translates to preferential decision making. 4. Starting point 4.1 Perceptual decision making In perceptual decision making, starting point is associated with a bias for one of the response alternatives. Starting point is usually halfway between the boundaries; for example, in a lexical decision study by Wagenmakers et al., 2008, and a random-dot motion study by Ratcliff & McKoon, 2008, this was the case under both speed and accuracy instructions. However, when the proportion of words vs. non-words was manipulated in the lexical decision task, the starting point shifts to favour the category with the highest proportion (Wagenmakers et al., 20081), and unbalancing the stimulus proportion in the random dot motion task so that one direction was more likely to occur, resulted in a starting point shift, with the starting point shifting in the direction of the boundary that corresponded to the stimuli with a higher probability (Ratcliff & McKoon, 2008). In a similar study, participants were shown a cue prior to the stimulus, indicating the direction of dot movement that was most likely to occur. This also shifted the starting point towards the corresponding bound (Mulder et al., 2012). Starting point shifts thus reflect the ability of participants to respond faster to the stimulus that is more likely to be presented. They will perform worse on the alternative stimulus, but as it is less likely to occur, overall performance will improve. Additionally, in a study on the DDM account of practice, response bias seemed to shift from a slight preference for words to one for non-words over the course of an accuracy stressed task, but not over the course of a speed stressed task (Dutilh et al., 2009). 4.2 Preferential decision making In the study by Mulder et al. (2012) mentioned above, a different cue that could be presented was a payoff cue, indicating a higher reward for correct responding to a stimulus of a certain direction. This also shifted the starting point towards the corresponding bound. So with both higher probability and higher payoff, the participant is more ready to respond to the cued
stimulus. Thus, a mostly straightforward relationship exists between starting point in the two decision making domains. 5. Non-decision time 5.1 Perceptual decision making Non-decision time is thought to reflect peripheral processing; the time spent encoding the stimulus prior to the accumulation process, and executing the motor response afterwards. As specific sensory modalities (vision, audition, etc.) and motor action are their own fields of research, often with their own models, not many experiments have been conducted to manipulate non-decision time. However, in some interesting cases, non-decision time was unexpectedly found to be influenced. In a study by Dutilh et al. (2009), practice on a lexical decision task caused a strong decrease in non-decision time. Also, in an in-group vs. out-group implicit association study by van Ravenzwaaij et al. (2011), non-decision time was shorter in the compatible block. The researchers think that this may be due to task-set switching, as was proposed by Klauer et al. (2007) after finding a very similar result. According to this account, the increase in peripheral processing time reflects extra time needed to map the decision outcome onto the response keys. This is because in the compatible block, the stimuli that invoke group membership and positive evaluation ask for the same response, which allows participants to respond intuitively, but in the incompatible block, task-set switching is needed to respond correctly. Lastly, non-decision time increased with higher alcohol intake and age (van Ravenzwaaij et al., 2012, Thapar et al., 2003). 5.2 Preferential decision making No study could be found where non-decision time was manipulated in a preferential decision making task. Hence no comparison is possible. 6. Discussion Out of the four factors examined, boundary separation and starting point seem to be comparable between perceptual and preferential decision making. For drift rate and non-decision time, no adequate comparison could be made. This is understandable for non-decision time, as this factor is not often object of study in either research field, but unexpected for drift rate. The one study that we found for drift rate was, in our opinion, wrongly labelled a preferential decision making study. Overall, no decisive conclusion can be reached about the underlying process of perceptual and preferential decision making. The most evident reason for this is the lack of studies in the preferential decision making field that actually use the DDM. Despite the benefits of the DDM in taking response times into account (Krajbich et al., 2015), most studies use the utility-based models of their own field. Also, quite a few studies were found that use sequential sampling models but instead of the DDM opt for variants of the DDM, often including a form of attention as a parameter. The presented analysis could be extended with some other models closely related to the DDM to still make a comparison possible. Secondly, studies may have been overlooked because of a great variety in terminology between, but also within the different decision making fields. As highlighted in Box 1, the use of different terms for the same concepts is astonishing, but may be a product of the relative young age of this type of research. It is therefore also one of our biggest recommendations that researchers and editors acknowledge this in their writing, including a glossary and the necessary keywords, to more effectively share our findings until the field is settled on what terms to use. Yet another way to improve on the current analysis would be to
Still, theoretical concerns hinder the comparison. First and foremost, the difference in response criterion is an issue. As mentioned, in perceptual decision making an external criterion determines if the response is correct or not, when participants have to decide for example if a stimulus moves left or right. With preferential decision making this is not so clear, because the participant is free to choose an alternative based on their own preference, which can be formed in many different ways. An important issue here is that of participants’ awareness of their own preferences. With perceptual decision making, memory retrieval is considered to be a large part of making the decision. For preferential decision making, this is unclear. For example, if a participant chooses between an apple and a banana, they could compare the salience of each option in that moment, or they could reference their knowledge of their preferences, which tells them that they are allergic to one of them, or that they simply always value one over the other? And even: are factors other than salience influencing their preference, when they consider for example that they already had one of the fruits this day, but they want to uphold a varied diet, or that the apple is probably grown locally, but the banana is most likely flown in with an airplane? Considerations like these might make a person choose against their salience, based on principle. Such differences in motivation for a preference would make some choice combinations of a different quality then others. A potential solution might be to divide the decision into two phases: first, a phase where the decision problem is represented, computing all courses of action while considering internal states, like salience, and external states, like the social consequences of a course of action, and second, a phase where the decision is computed, valuating all available courses of action using the accumulation process. Non-decision time would still be a separate phase (Rangel et al., 2008, Sugrue et al., 2005). Such an approach could provide a way of differentiating between choice problems and comparing the accumulation process in a less confounded way. Second, this makes it also harder to determine how difficult a given decision is. In general, difficult decisions are those where two choice options are close together in salience, but the impact of the reward, how much the participant can gain or lose, might also matter (Oud et al., 2016). Lastly, there is the technical matter of the timespan of the decision. According to Ratcliff & Smith (2004) and Ratcliff (2006), sequential sampling models should only be applied to decision making tasks which yield mean response times that are no longer than 1000-1500 ms, because longer responses might induce multiple or repeated decision processes, according to them. While the preferential decision making studies reviewed here kept to this limit, using the DDM, their results can be much slower, lasting up to many seconds. Further research is needed to confirm the necessity of this limit. Although the theoretical differences in these research domains can be tough to overcome, a unified choice architecture for perceptual and preferential decisions would greatly further both research fields. It would also lend more weight to any theories about the underlying cognitive processes, as we strive to bring together knowledge from cognitive psychology, economics and neuroimaging and discover the fundamental mechanisms that give rise to our behaviour.
Box 1 – Terminology in decision making Terminology differences often stem from different backgrounds and theoretical opinions. However, the following terms lie so close together, that they refer to the same concept, only placing different theoretical accents. It would greatly help the field if alternative terms could be included as search keywords or in a brief glossary, as with increasing numbers of papers getting published, we should not hinder each other in finding literature. • preferential decision making/value-based decision making • non-decision time/Ter (time of encoding and response)/extra-accumulatory time • Drift Diffusion Model/Diffusion Decision Model/Wiener (diffusion) process • boundary separation/threshold • bias/start point/starting point • sequential sampling models/accumulation-to-bound models • random-dot (motion) task/motion discrimination task/random-dot kinetogram/motion direction task Literature Basten, U., Biele, G., Heekerken, H.R., & Fiebach, C.J. (2010). How the brain integrates costs and benefits during decision making. Proceedings of the National Academy of Sciences, 107, 21767– 21772.
Dutilh, G., & Rieskamp, J. Comparing perceptual and preferential decision making. Psychonomic Bulletin & Review, 1-15. Dutilh, G., Krypotos, A.–M., & Wagenmakers, E.–J. (2011). Task–related vs. stimulus–specific practice: A diffusion model account. Experimental Psychology, 58, 434–442. Dutilh, G., Vandekerckhove, J., Tuerlinckx, F., & Wagenmakers, E.-J. (2009). A diffusion model decomposition of the practice effect. Psychonomic Bulletin & Review, 16, 1026–1036. Gold, J.I., & Shadlen, M.N. (2007). The neural basis of decision making. The Annual Review of Neuroscience, 30, 535–74 Krajbich, I., Hare, T., Bartling, B., Morishima, Y., & Fehr, E. (2015). A common mechanism underlying food choice and social decisions. PLOS Comput Biol, 11, e1004371. Milosavljevic, M., Malmaud, J., Huth, A., Koch, C., & Rangel, A. (2010). The Drift Diffusion Model can account for the accuracy and reaction time of value-based choices under high and low time pressure. Judgment and Decision Making, 5, 437-449. Mulder, M.J., Wagenmakers, E.-J., Ratcliff, R., Boekel, W., & Forstmann, B.U. (2012). Bias in the Brain: A Diffusion Model Analysis of Prior Probability and Potential Payoff. The Journal of Neuroscience, 32, 2335–2343. Oud, B., Krajbich, I., Miller, K., Cheong, J. H., Botvinick, M., & Fehr, E. (2016). Irrational time allocation in decision-makingIn Proceedings. Biological sciences/The Royal Society (Vol. 283, No. 1822).
neurobiology of value-based decision making. Nature Reviews Neuroscience, 9, 545-556. Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85, 59-108. Ratcliff, R. (2006). Modeling response signal and response time data. Cognitive Psychology, 53, 195-237. Ratcliff, R., & Smith, P.L. (2004). A comparison of sequential sampling models for two-choice reaction time. Psychological Review, 111, 333–367. Ratcliff, R., & McKoon, G. (2008). The diffusion decision model: Theory and data for two-choice decision tasks. Neural Computation, 20, 873-922. Ratcliff, R., & Rouder, J.N. (2000). A Diffusion Model Account of Masking in Two-Choice Letter Identification. Journal of Experimental Psychology: Human Perception and Performance, 26, 127-140. van Ravenzwaaij, D., Dutilh, G., & Wagenmakers, E.-J. (2012). A diffusion model decomposition of the effects of alcohol on perceptual decision making. Psychopharmacology, 219, 1017–1025 van Ravenzwaaij, D., van der Maas, H.J., & Wagenmakers, E.-J. (2011). Does the name-race implicit association test measure racial prejudice? Experimental Psychology, 58, 271-277. Sugrue, L.P., Corrado, G.S., & Newsome, W.T. (2005). Choosing the greater of two goods: Neural currencies for valuation and decision making. Nature Reviews Neuroscience, 6, 363-375. Thapar, A., Ratcliff, R., & McKoon, G. (2003). A diffusion model analysis of the effects of aging on letter discrimination. Psychology and Aging, 18, 415-429. Wagenmakers, E.-J., Ratcliff, R., Gomez, P., & McKoon, G. (2008). A diffusion model account of criterion shifts in the lexical decision task. Journal of Memory and Language, 58, 140-159.
