Human performance across decision making, selective attention, and working memory tasks: Experimental data and computer simulations

This article describes the data analyzed in the paper “Individual differences in the Simon effect are underpinned by differences in the competitive dynamics in the basal ganglia: An experimental verification and a computational model” (Stocco et al., 2017) [1]. The data includes behavioral results from participants performing three cognitive tasks (Probabilistic Stimulus Selection (Frank et al., 2004) [2], Simon task (Craft and Simon, 1970) [3], and Automated Operation Span (Unsworth et al., 2005) [4]), as well as simulationed traces generated by a computational neurocognitive model that accounts for individual variations in human performance across the tasks. The experimental data encompasses individual data files (in both preprocessed and native output format) as well as group-level summary files. The simulation data includes the entire model code, the results of a full-grid search of the model's parameter space, and the code used to partition the model space and parallelize the simulations. Finally, the repository includes the R scripts used to carry out the statistical analyses reported in the original paper.


a b s t r a c t
This article describes the data analyzed in the paper "Individual differences in the Simon effect are underpinned by differences in the competitive dynamics in the basal ganglia: An experimental verification and a computational model" (Stocco et al., 2017) [1]. The data includes behavioral results from participants performing three cognitive tasks (Probabilistic Stimulus Selection (Frank et al., 2004) [2], Simon task (Craft and Simon, 1970) [3], and Automated Operation Span (Unsworth et al., 2005) [4]), as well as simulationed traces generated by a computational neurocognitive model that accounts for individual variations in human performance across the tasks. The experimental data encompasses individual data files (in both preprocessed and native output format) as well as group-level summary files. The simulation data includes the entire model code, the results of a full-grid search of the model's parameter space, and the code used to partition the model space and parallelize the simulations. Finally, the repository includes the R scripts used to carry out the statistical analyses reported in the original paper. &

Value of the data
The data includes individual human performance across three common experimental tasks that measure different cognitive abilities (decision-making, cognitive control, and working memory respectively).
The ACT-R model code is available to inspect or integrate with other ACT-R models. The model provides an explanation for the existing correlation between two of the three tasks.
Simulation data provides complete overview of model behavior across a large parameter space. R code is provided to ensure reproducibility of the experimental analysis published in the original paper [1].

Data
This paper describes the human experimental data and the computer simulations reported in the paper "Individual differences in the Simon effect are underpinned by differences in the competitive dynamics in the basal ganglia: An experimental verification and a computational model" [1]. In the case of experimental data, both group-level summary tables and individual datasets for each participant are available. Each individual dataset is available in both "raw" format (the output of the software used to present the experimental stimuli) and in "analyzed" format (the Excel worksheets used to compute individual performance measures). In addition to the experimental and simulation data, the analysis scripts used to compute the statistical results presented in the paper are also provided. The overall organization of the data follows the hierarchical structure shown in Fig. 1.
All of the data reported in [1] are organized in two folders, named "data" and "model". The "data" folder contains all of the experimental data. In three subfolders. Each of the subfolders contains the individual data obtained from each participant in the Simon [3], Probabilistic Stimulus Selection (PSS) [2], and Operation Span tasks [4], respectively. All the three subfolders contain a "Raw" directory, which holds the raw data generated by the Eprime software (Psychological Software Tools, Pittsburgh, PA). In addition, the "Simon" and "PSS" folders also have an "Analyzed" subfolder, which contains the data of individual participants in the format of a Microsoft Excel (Microsoft, Redmond, WA) workbook. Each workbook contains a sheet with the original data in table format, and one sheet with the summary results by conditions. A summary of the individual experimental results is provided in Table 1  The "model" folder contains the model code and a "simulations" subfolder. The "simulations" subfolder contains the text log files of the simulations ("simulations.txt") as well as the Python code used to generate the simulations.
Finally, the base folder also contains two scripts used to analyze the experimental and simulation results of the [1] paper. Both scripts consist of code in the R programming language. The "analysis.R" script contains the code used to generate the statistical results reported in the paper, as well as its plots and figures. The "model-flexibility-analysis.R" scripts contains additional code for the analysis of the model's performance using the methods described in [5]. The results of these analyses were eventually omitted from final publication, but can be examined by running the file's code. A copy of this repository is also available on the Cognition and Cortical Dynamics' GitHub account page, at https://github.com/UWCCDL/PSS_Simon.

Participants
Fifty-eight healthy individuals were recruited for the experiment (age ¼ 18-34 years, 44 females). Data from 8 participants (5 female) were not analyzed due to an inability to attain learning criteria   ). The thick lines represent medians; the box represents the interquartile range; data points outside the range represent outliers. Fig. 3. Tukey's boxplots of the mean response times for Congruent and Incongruent trials in the Simon task [6]. The thick lines represent medians; the box represents the interquartile range; data points outside the range represent outliers.

Experimental procedures
The PSS task [2], the Simon task [3], and one test of complex working memory span (the Operation Span task [4]) were administered to participants as part of a larger cognitive battery assessing cognitive capabilities in participants who went on to participate in a training experiment not reported herein. With the exception of the three tasks described herein, no other experimental task was selected for this specific analysis, and no dataset was discarded after having been considered for this study. All of the tasks were performed on a computer, in front of a 21" LCD screen, using a standard keyboard to respond. Stimulus presentation and response collection were controlled through the E-prime software (Psychological Software Tools, Pittsburgh, PA).

The Simon task
The "Simon" sub-folder contains the data from the Simon task [3,6], a response interference task used to collect individual measures of cognitive control. During the task, participants were presented with one of two shapes, either a black square or a black circle, on a white background. Participants were instructed to respond to one shape (e.g., squares) with their right hand, and to the other shape  (e.g., circles) with their left hand. Each trial was introduced by an 800 ms fixation, followed by a 250 ms time delay, followed by a stimulus (circle or square) that remains on the screen for either 3000 ms or until a response was recorded. Trials were either congruent (e.g., a stimulus associated with a left response and presented on the left half of the screen) or incongruent (e.g., a stimulus associated with a left response and presented on the right half of the screen), with congruent trials making up 75% of the total count. The task consisted of 64 trials divided into 4 blocks of 16 trials each.

The Probabilistic Stimulus Selection (PSS) task
The "PSS" sub-folder contains the data from Probabilistic Stimulus Selection (PSS) task, an iterative decision-making task that is used to collect measures of basal ganglia function [2]. The task consists of two consecutive phases, a training and a test phase. In both phases, participants performed multiple decision-making trials in which they were asked to choose one of two Japanese Hiragana characters, placed at the left and right side of the screen. Participants indicate their response by pressing the keys "1" (for the left character) or "0" (for the right character) on a standard keyboard. A total of six stimuli are presented, each of which is associated with a unique probability of success. During the training phase, the six stimuli are presented in three fixed pairs. To ensure that participants can discriminate the relative success probability for each stimulus, the training phase is repeated until each participant's accuracy has reached a predetermined criterion. After a maximum of six repetitions of the training phase, participants move on to the test phase. The test phase presents each of the 15 possible combinations of stimuli four times, for a total of 60 trials. The decisions made during the test phase yield two distinct measures of performance: the accuracy in choosing the most rewarding stimulus against all the others (Choose accuracy, i.e., the proportion of choices in which stimulus A is preferred over C, D, E and F) and the accuracy in avoiding the least rewarding stimulus when it is paired with all the other (Avoid accuracy, or the proportion of choices in which stimuli C, D, E, and F were preferred over B).

The Operation Span task
The "OpSpan" folder contains data from the automated version of the Operation Span task [4], a non-verbal test of complex working memory span. In the Operation Span task, participants memorize a sequence of letters (e.g., "L", "Q", and "S") that are presented in alternation to the evaluation of arithmetic expressions (e.g. "(2 × 5) − 9 ¼ 2"). While memorizing the letters, participants also have to indicate whether the expression was mathematically true or false. After the presentation of the last letter, participants indicate all the letters that had been presenting since the beginning of the trial by selecting them in the order in which they appeared from an array of 4 × 3 letters. The number of letters to be memorized varies pseudo-randomly across trials, up to a maximum number of seven [4]. Working memory capacity was measured in terms of the Span Score [4], which ranges from a minimum of 0 to a maximum of 75, the latter indicating that all of the trials were correctly remembered.

Computer simulations
The computer simulations were obtained from a computational neurocognitive model based on the ACT-R cognitive architecture [7], version 7.4, and implemented in Common Lisp. All of the model code and simulations are included in the "model" subfolder of the data repository.

Model code files
The "model" subfolder contains the entire model code, divided into three files. The "simon-model. lisp" file contains the main ACT-R model code. The "simon-device.lisp" file contains the code necessary to present the Simon task to the model and collect the model responses. Specifically, this file contains the data structures necessary to implement an ACT-R "device", that is, a Common Lisp object that can provide inputs to, and receives responses from, an ACT-R model. Finally, the "simonsimulations.lisp" file contains the code to execute multiple runs of the model under different parameters and conditions.

Model simulations
The "simulations" subfolder contains the results of a full grid-search simulation of the model's behavior across five different parameters (see [1] for details). All of the simulation results are contained in a single text file, "simulations.txt". Each row of the file represents the average performance of the model over 100 runs under a specific combinations of parameter values. The "gen-simulations. py" file contains the Python code that was used to set up the simulations. The Python code generates over 100 Lisp files, each of which is parametrized to examine a different portion of the complete parameter space. The use of multiple Lisp files permits to run the simulations in parallel on multi-core computers.